Non-linear Modeling and Analysis of Solids and Structures

This page intentionally left blank

Non-linear Modeling and Analysis of Solids and Structures Steen Krenk

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Dubai, Tokyo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521830546 © Cambridge University Press 2009 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2009

ISBN-13

978-0-511-60413-3

eBook (EBL)

ISBN-13

978-0-521-83054-6

Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

To Jette

Contents

Preface 1 1.1

page ix

1.3 1.4

Introduction A simple non-linear problem 1.1.1 Equilibrium 1.1.2 Virtual work and potential energy Simple non-linear solution methods 1.2.1 Explicit incremental method 1.2.2 Newton–Raphson method 1.2.3 Modiﬁed Newton–Raphson method Summary and outlook Exercises

1 2 3 6 7 8 9 13 14 15

2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

Non-linear bar elements Deformation and strain Equilibrium and virtual work Tangent stiﬀness matrix Use of shape functions Assembly of global stiﬀness and forces Total or updated Lagrangian formulation Summing up the principles Exercises

17 18 20 24 26 31 36 39 43

3 3.1 3.2 3.3 3.4 3.5

Finite rotations The rotation tensor Rotation of a vector into a speciﬁed direction The increment of the rotation variation Parameter representation of an incremental rotation Quaternion parameter representation 3.5.1 Representation of the rotation tensor

47 49 53 55 60 63 64

1.2

3.6 3.7 3.8 4 4.1 4.2 4.3

4.4

4.5 4.6 5 5.1

5.2

5.3 5.4 6 6.1

6.2

Contents

3.5.2 Addition of two rotations 3.5.3 Incremental rotation from quaternion parameters 3.5.4 Mean and diﬀerence of two rotations Alternative representation of the rotation tensor Summary of rotations and their virtual work Exercises

65 67 68 69 72 73

Finite rotation beam theory Equilibrium equations Virtual work, strain and curvature Increment of the virtual work equation 4.3.1 Constitutive stiﬀness 4.3.2 Geometric stiﬀness 4.3.3 The load increments Finite element implementation 4.4.1 Element stiﬀness matrix 4.4.2 Loads and internal forces 4.4.3 Shear locking Summary of ‘elastica’ beam theory Exercises

76 77 78 81 82 83 85 86 87 89 91 98 99

Co-rotating beam elements Co-rotating beams in two dimensions 5.1.1 Co-rotation form of the tangent stiﬀness 5.1.2 Element deformation stiﬀness 5.1.3 Total tangent stiﬀness 5.1.4 Finite element implementation Co-rotating beams in three dimensions 5.2.1 Co-rotation form of the tangent stiﬀness 5.2.2 Element deformation stiﬀness 5.2.3 Total tangent stiﬀness 5.2.4 Finite element implementation Summary and extensions Exercises

100 101 104 107 110 112 117 120 127 130 133 139 141

Deformation and equilibrium of solids Deformation and strain 6.1.1 Non-linear strain 6.1.2 Decomposition into deformation and rigid body motion Virtual work and stresses 6.2.1 Piola–Kirchhoﬀ stress 6.2.2 Cauchy and Kirchhoﬀ stresses

145 146 148 151 154 155 158

Contents

6.3

6.4

6.5 6.6 7 7.1

7.2

7.3

7.4

7.5

7.6 7.7 7.8 8 8.1

vii

6.2.3 Stress rates Total Lagrangian formulation 6.3.1 Equilibrium and residual forces 6.3.2 Tangent stiﬀness 6.3.3 Finite element implementation Updated Lagrangian formulation 6.4.1 Transformation from total to updated format 6.4.2 Virtual work in the current conﬁguration 6.4.3 Finite element implementation Summary of non-linear motion of solids Exercises

160 165 166 167 170 174 174 176 180 185 186

Elasto-plastic solids Elastic solids 7.1.1 Stress invariants 7.1.2 Strain invariants and small strain elasticity 7.1.3 Isotropic elasticity at ﬁnite strain General plasticity theory 7.2.1 Reversible deformation 7.2.2 Maximum plastic dissipation rate 7.2.3 Evolution equations 7.2.4 Isotropic and kinematic hardening Von Mises plasticity models 7.3.1 Yield surface and ﬂow potential 7.3.2 Explicit integration 7.3.3 Radial return algorithm General aspects of plasticity models 7.4.1 Combined isotropic and kinematic hardening 7.4.2 Internal variables and non-associated ﬂow 7.4.3 General computational procedure Models for granular materials 7.5.1 Flow potential and yield surface 7.5.2 Elasticity and hardening Finite strain plasticity Summary Exercises

189 190 192 198 200 203 204 207 212 216 218 219 222 225 229 230 234 237 241 242 247 249 252 253

Numerical solution techniques Iterative solution of equilibrium equations 8.1.1 Non-linear iteration strategies

256 257 259

viii

8.2 8.3

8.4 8.5 8.6 9 9.1

9.2 9.3

9.4

9.5 9.6

Contents

8.1.2 Direction and step-size control Orthogonal residual method Arc-length methods 8.3.1 General constraint formulation 8.3.2 Hyperplane constraints 8.3.3 Hypersphere constraint Quasi-Newton methods Summary Exercises

260 263 270 272 274 278 283 287 288

Dynamic eﬀects and time integration Newmark algorithm for linear systems 9.1.1 Energy balance and stability 9.1.2 Numerical accuracy and damping Non-linear Newmark algorithm Energy-conserving integration 9.3.1 State-space formulation 9.3.2 Non-linear kinematics for Green strain 9.3.3 Energy-conserving algorithm Algorithmic energy dissipation 9.4.1 Spectral analysis of linear systems 9.4.2 Linear algorithm with energy dissipation 9.4.3 Non-linear algorithm with energy dissipation Summary and outlook Exercises

290 292 295 300 304 309 310 311 315 323 323 325 327 331 333

References Index

336 345

Preface

The aim of this book is to take the reader on a concentrated tour of some of the central issues of non-linear modeling and analysis of structures and solids. Traditionally, the non-linear theories of solids have been treated in books on continuum mechanics, while the questions of analysis have formed the focus of books on ﬁnite element techniques. The idea of the present book is to place the emphasis on modeling with a view to its numerical implementation right from the outset. Two guiding principles have determined the main style of the book: the story should be told in the form of concentrated chapters, each giving the central ideas of a speciﬁc aspect such as ‘ﬁnite rotations’ or ‘elasto-plastic solids’, and the reader should have the possibility of getting a feel for the numerical implementation by access and use of simple high-level implementations of the basic algorithms. A text based on these principles cannot provide exhaustive coverage, but aims at giving an interesting introduction to the basic ideas, which can then be studied elsewhere in greater detail as needed. It is hoped that the combination of a concise theoretical presentation in plain language supported by speciﬁc algorithms will make the text of interest to graduate students as well as professionals. The book contains nine chapters: a brief introductory chapter setting the scene by use of elementary arguments, four chapters on structures, two chapters on non-linear deformation and material behavior of solids, and ﬁnally two chapters on numerical techniques for non-linear quasi-static and dynamic analysis. The theory is combined with demonstrations and exercises using a small Matlab toolbox FemFiles providing routines for creation and assembly of element matrices and permitting the solution of non-linear ﬁnite element problems in a fairly simple script ﬁle format. The toolbox FemFiles is available from the author via the internet. Exercises that require the use of a high-level program like FemFiles are marked ∗. ix

Preface

The text started as a draft manuscript prepared for a short introductory course on non-linear aspects of the ﬁnite element method at Aalborg University in the fall of 1992. A visit to Lund Institute of Technology sponsored by NorFA provided an opportunity to include additional material on the numerical aspects. The text was later extended with material on ﬁnite rotations, co-rotating formulation of elements, potential theory of plasticity theory and plasticity models for geotechnical materials, and conservation algorithms for numerical integration of dynamic problems. Several parts of this work have been sponsored by the Danish Technical Research Council. The work on bringing it all together was initiated during a visiting appointment as Melchor Professor at the University of Notre Dame, Indiana, in the fall of 2001. The ﬁnal stage has been combined with courses at Helsinki University of Technology 2004, and at Aalborg University and Lund Institute of Technology 2005.

1 Introduction

Many problems of practical interest involve non-linear behavior of solids and structures. In the present context a solid means a body with a ﬁrm shape, as opposed to a ﬂuid, while a structure refers to a solid composed of slender elements such as beams, plates and shells. Typical problems are the motion of robots, collapse scenarios of structures, metal forming processes in industrial production, and material deformation and failure in geotechnical engineering. These problems typically involve a considerable change of shape, often accompanied by non-linear material behavior. The ﬁnite element method is an important tool for the analysis of nonlinear problems, such as geometrical and material non-linear behavior of solids and structures. The solution of non-linear problems by the ﬁnite element method involves modeling, leading to the formulation of an appropriate set of non-linear equations describing the problem, followed by an appropriate strategy for the numerical solution of these equations. In contrast to linear problems, where the solution strategy reduces to the solution of a system of linear equations, the solution phase in a non-linear problem typically involves an iterative procedure. Non-linear modeling and analysis is a very active research area with many engineering applications. The many diﬀerent aspects involved are not covered in any single text. However, some central references to general texts should be given here. A brief introduction to some of the basic problems of non-linear ﬁnite element analysis of solids and structures is included in the book by Cook et al. (1989). A general state-of-the-art presentation of the ﬁnite element method, including the non-linear aspects of solids, structures and ﬂuids, has been given in Zienkiewicz and Taylor (2000). A presentation with main emphasis on incremental formulation of geometrically non-linear problems, including details of implementation, has been given by Bathe (1996). The books by Crisﬁeld (1991, 1997) and Belytschko et al. 1

Introduction

(2000) are entirely devoted to non-linear analysis of solids and structures, combining illustrative examples with speciﬁc ﬁnite element procedures. The present text is an introduction to some of the central ideas of nonlinear modeling and ﬁnite element analysis. It covers theoretical aspects of geometric and material non-linearity and associated numerical techniques. The text proceeds from the elementary level to a fairly rigorous presentation of ideas used in current research. Only the main ideas can be covered, and the references should be consulted according to need. This ﬁrst chapter gives an illustration of geometric non-linear behavior with reference to a simple two-element truss model. The example serves as a vehicle for an informal introduction to a non-linear load–displacement relation, the tangent stiﬀness, and the relation between the equilibrium and the virtual work approach to the problem. The example also provides a simple realistic nonlinear equation on which to try diﬀerent variants of the Newton–Raphson solution technique.

1.1 A simple non-linear problem The simple two-element truss model shown in Fig. 1.1 has often been used to illustrate some basic features of geometric non-linear behavior, see e.g. Bathe (1996, p. 494) and Crisﬁeld (1991, pp. 2–13). The structure consists of two identical truss elements, loaded with a vertical force f at the center and simply supported at the other ends. The vertical displacement at the center is called u. In the initial conﬁguration the length of the bars is l0 .

Fig. 1.1. Two-element truss model.

Application of the load leads to a deformed state with vertical displacement u of the central node, Fig. 1.2. The structure is assumed to be shallow, i.e. a b. This permits series expansion of the square roots deﬁning the original bar length l0 and the bar length l corresponding to the current

1.1 A simple non-linear problem

deformed state:

1 a2 , b2 + a2 b 1 + 2 b2 1 a + u 2 b2 + (a + u)2 b 1 + . l = 2 b

l0 =

(1.1) (1.2)

Fig. 1.2. Initial length l0 and current length l.

The deformation of the bars is described by their elongation. A nondimensional measure of deformation is the engineering strain, deﬁned as the elongation relative to the original length, l − l0 au 1 u 2 ε = + . (1.3) l0 l0 l0 2 l0 The ﬁrst term is the linear part of the strain, while the second term is nonlinear. A true measure of deformation must not be inﬂuenced by any rigid body motion of the bar, and thus a true deformation measure must be a nonlinear function of the displacement component(s). If the displacement u is small relative to all characteristic lengths of the geometry – l0 and a – the linear term will constitute a fair approximation, but if this approximation is used, some of the characteristic non-linear features of the problem are lost. 1.1.1 Equilibrium The two bars are assumed to be linear elastic with axial stiﬀness EA, where E is the elastic modulus and A is the cross-section area. Thus, the axial force in each bar is expressed in terms of the strain as au 1 u 2 N = EA ε EA . (1.4) + l0 l0 2 l0 Equilibrium of the central node in the deformed state requires that the external force f is equal to the internal force g(u) generated by deformation of the structure. Projection of the normal force gives 2EA a+u au + 12 u2 (a + u). (1.5) g(u) = 2 N 3 l l0

Introduction

In non-dimensional form this is a 3 u 3 u 2 1 u 3 + , + g(u) = 2 EA l0 a 2 a 2 a

(1.6)

where the normalized displacement is u/a. The load–displacement relation (1.6) is shown in Fig. 1.3 corresponding to a downward load.

Fig. 1.3. Load–displacement curve for two-element truss.

From the unloaded state A an increasing downward load leads to a local maximum B. In this state the structure cannot support a further increase of the load. Thus, further increase of the load from B would lead to snapthrough to F . The snap-through is an unstable dynamic process, and thus the load–displacement curve in Fig. 1.3 is not fully representative. Alternatively the structure may be loaded in displacement control, in which the central node is given a controlled downward displacement −u. This would require an increasing load from A to B, and then a decreasing load from B to C, where u = −a and the two bars form a straight line. An upward force is now required to proceed to D and E, where the structure is stress-free, forming an angle symmetric to the original conﬁguration with respect to the base line. Further downward load leads through F with increasing stiﬀness of the structure. For a structure with one degree of freedom, the stiﬀness is a measure of the change in force for a given change in displacement. Thus, the tangent stiﬀness K is deﬁned as the stiﬀness corresponding to inﬁnitesimal changes in u and g: K =

dg . du

(1.7)

In the present case the tangent stiﬀness K follows from straightforward

1.1 A simple non-linear problem

diﬀerentiation of (1.6): K =

u 3 u 2 2EA a 2 + 1+3 . l0 l0 a 2 a

(1.8)

Although this expression deﬁnes the tangent stiﬀness K, it does not convey the physics of the problem very clearly. This is better accomplished by diﬀerentiation of the equilibrium equation (1.5): a + u EA a + u 2 N d 2N = 2 +2 . (1.9) K = du l0 l0 l0 l0 Here a + u is the height of the structure in the current state, while N is the current value of the axial force. The ﬁrst term is due to changes in the normal force N , while the second term is due to changes in the geometric conﬁguration with constant normal force N . Sometimes the ﬁrst term is separated into a constant corresponding to u = 0 and the rest, whereby (1.9) takes the form EA a 2 EA 2au + u2 N K = 2 +2 +2 2 l0 l0 l0 l0 l0 =

+ Kσ ,

(1.10)

where K0 is the linear stiﬀness, Ku is the initial displacement stiﬀness, and Kσ is the initial stress stiﬀness. In an incremental procedure, where the geometry is updated, the current value of u is absorbed in the updated value of a, and in that case the initial displacement stiﬀness Ku vanishes.

Fig. 1.4. Load–displacement curve for two-element truss with spring.

A family of load–displacement curves with diﬀerent degrees of non-linearity can be obtained by introducing a vertical linear elastic spring with stiﬀness

Introduction

k at the central node of the structure. The load–displacement relation (1.6) is changed to a 3 u 3 u 2 1 u 3 g(u) = 2EA + + ku (1.11) + a 2 a 2 a l0 and the tangent stiﬀness (1.8) to u 3 u 2 2EA a 2 + + k. 1+3 K = a 2 a l0 l0

(1.12)

Figure 1.4 shows the load–displacement curve for diﬀerent values of the spring stiﬀness k. For k ≥ EAa2 /l03 the variation of load with displacement is monotonic, corresponding to K ≥ 0. 1.1.2 Virtual work and potential energy The load–displacement relations (1.6) and (1.11) were obtained from equilibrium of the center node. For structures with more degrees of freedom or more complicated elements it is often convenient to make use of the principle of virtual work. Essentially, the principle of virtual work is a restatement of a set of equilibrium equations, where each equation is multiplied by a corresponding inﬁnitesimal virtual displacement component. With an appropriate deﬁnition of the force and displacement components summation of their products forms a scalar invariant, known as the virtual work. In the particular example of the two-element truss with an elastic spring the equilibrium equation can be written as a+u 2N + ku − f = 0. (1.13) l Multiplication by a virtual displacement δu gives the virtual work equation a+u δu + (ku)δu − f δu = 0. (1.14) δV = 2 N l The displacement factor in the ﬁrst term is similar to the ﬁrst variation of the strain (1.3): ∂ a u 1 u 2 a + u δu δε = + . (1.15) δu = ∂u l0 l0 2 l0 l0 l0 If, for the time being, the diﬀerence between l0 and l is neglected, the virtual work equation (1.14) can now be written as

l0 δV 2 N δε ds + (ku) δu − f δu = 0. (1.16) 0

The integral is the internal virtual work of the bar elements, the second term

1.2 Simple non-linear solution methods

is the virtual work of the elastic spring, while the last term is the external virtual work. Apart from the factor l0 /l that is somehow missing, the use of virtual work in the present case where δε is constant within the elements is almost trivial. However, for more general problems with more degrees of freedom and nontrivial displacement ﬁelds within the elements, the principle of virtual work is an important tool for establishing the balance equations of the discretized model. The question of the factor l0 /l is discussed in Chapter 2, where the theory of non-linear bar elements is discussed more rigorously. Here, the relation between virtual work and potential energy is discussed brieﬂy before turning to elementary numerical solution methods for non-linear equilibrium equations. When the internal forces such as the axial force N and the spring force ku are functions of the state of displacement given by u, and the external load is also a function of u, the virtual work δV can be considered as the diﬀerential of an energy function Φ(u) – the potential energy. In the present case (1.16) is written as

l0 δΦ(u) = 2 EA εδε ds + ku δu − f δu. (1.17) 0

This relation can be integrated with respect to the displacement u, giving the following expression for the potential energy:

l0 2 2 1 1 Φ(u) = 2 (1.18) 2 EAε ds + 2 k u − f u. 0

The potential energy is the internal strain energy of the structure, including the spring, minus the external work represented by f u. For linear elastic structures it may be simpler to derive the equilibrium equations from the potential energy by considering an incremental change δu of the displacements. However, the principle of virtual work is valid irrespective of the speciﬁc material behavior, and thus the principle of virtual work has become the method of choice for setting up equilibrium equations.

1.2 Simple non-linear solution methods For a system with only one degree of freedom non-linear behavior can often be described explicitly as a function of the displacement u, and the problem may then be considered as one of displacement control. However, in the case of several degrees of freedom the use of displacement control is non-trivial, and most problems are formulated in terms of a load history, for which

Introduction

the corresponding displacement history is to be calculated. This requires the solution of a system of non-linear equations. Here some of the simpler methods for solving non-linear equations are brieﬂy introduced, leaving more specialized techniques to Chapter 8. The methods are illustrated for a single degree of freedom and then generalized to matrix form. 1.2.1 Explicit incremental method An explicit incremental method, often called the Euler explicit method, is obtained by replacing the diﬀerentials in the deﬁnition (1.7) of the tangent stiﬀness with ﬁnite increments ∆f and ∆u: ∆u = K −1 ∆f.

(1.19)

The load–displacement history is described by a number of increments ∆fn , ∆un , n = 1, 2, . . . deﬁning the states fn = fn−1 + ∆fn ,

un = un−1 + ∆un ,

n = 1, 2, . . .

(1.20)

In the explicit incremental method the tangent stiﬀness K corresponds to the state at the beginning of the increment. Thus, the precise form of (1.19) is ∆un = K −1 (un−1 ) ∆fn ,

n = 1, 2, . . .

(1.21)

This procedure is illustrated in Fig. 1.5.

Fig. 1.5. Explicit incremental method.

It is seen that the computed states deviate more and more from the exact load–displacement curve. There are two reasons for this: the tangent stiﬀness of each increment is taken at the left end-point and in this particular case overestimates the stiﬀness, and deviations from the exact curve are

1.2 Simple non-linear solution methods

added to a cumulative error. While it is diﬃcult to use an exact representation for the stiﬀness corresponding to the full increment, the problem of increasing deviations can be countered by introducing equilibrium iterations as discussed in the following. The explicit incremental method is easily generalized to multi-degree of freedom systems. Let the displacement vector be u and the corresponding load vector f . The tangent stiﬀness matrix K is then deﬁned by df = K(u) du.

(1.22)

The corresponding explicit incremental method is ∆un = K−1 (un−1 ) ∆fn ,

n = 1, 2, . . .

(1.23)

The use of the inverse matrix K−1 in (1.23) should not be taken literally. In practice the matrix K is factored and the product K−1 ∆f found by back substitution. 1.2.2 Newton–Raphson method In order to avoid accumulating errors in each additional load step, equilibrium iterations may be used to establish equilibrium to a desired degree of accuracy at each load step. This procedure is a special instance of the Newton–Raphson method, well known from numerical analysis. In principle, the method works by applying two steps intermittently: (i) check if equilibrium is satisﬁed to within the desired accuracy; (ii) if not, make a suitable adjustment of the state of deformation. The ﬁrst step consists in checking the equilibrium equation. This is done by forming the diﬀerence between the external load f and internal force g(u), r(u, f ) = f − g(u) = 0,

(1.24)

where r(u, f ) is called the residual force. In a state of equilibrium the internal force g(u) is equal to the external load f , and thus the residual vanishes. In practice, lack of equilibrium will be produced at the beginning of each load increment, where the load f is increased, while no new displacement estimate u is yet available. Thus, the need arises for obtaining an improved estimate of the state of displacement u. In the absence of equilibrium an improved estimate of the displacement u is obtained from a linearized form of the residual r(u + δu, f ) around the known residual r(u, f ), r(u + δu, f ) = r(u, f ) + δr(u, f ) + · · · = 0.

(1.25)

Introduction

The dots indicate higher-order terms, because δr is only a linearized form of the increment of the residual. In the classic form of equilibrium iterations the load f is assumed ﬁxed within the given load step, and thus the increment of the residual only depends on the internal force g(u). The linearized increment is then given by the ﬁrst derivative of the internal force as dg(u) δu = −K(u) δu. (1.26) du Here the tangent stiﬀness K, introduced in (1.7), has been introduced. The displacement increment is now obtained from the linearized form of (1.25) by substitution of the tangent stiﬀness relation (1.26). When rearranging the terms in (1.25), the linearized equation becomes δr = −

K(u) δu = r.

(1.27)

In this equation the residual r(u, f ) is known, as it relates to the current state of load f and displacement u. The tangent stiﬀness K(u) at the current displacement state u can also be calculated. Thus, this equation permits determination of the displacement increment δu, δu = K −1 (u) r.

(1.28)

Once the displacement increment δu is determined, the current displacement state is updated to ui = ui−1 + δui .

(1.29)

In this equation the superscript is used to indicate that the iteration i changes the estimated displacement from ui−1 to ui . In a computer program the iteration superscript i is not needed, as the register containing ui−1 is simply overwritten by the new value ui according to the assignment statement u : = u + δu.

(1.30)

Here, : = is the assignment operator, implying that the variable u is assigned a new value. In this book many of the algorithms are presented in the form of pseudocode – i.e. a code format that appears like high-level programs such as Matlab. In the pseudocode presented here assignments are indicated by the normal equality sign, as all equalities are assignment statements. The Newton–Raphson equilibrium iteration procedure is illustrated in Fig. 1.6. The ﬁgure shows load step n. This load step starts from a state of equilibrium already established at the previous load fn−1 with displacement un−1 . The load step is initiated by increasing the load by ∆fn to fn . This generates the ﬁrst residual rn1 = ∆fn . This residual and the tangent stiﬀness

1.2 Simple non-linear solution methods

Fig. 1.6. Newton–Raphson equilibrium iterations.

K(un−1 ) lead to the displacement increment δu1n , shown in the ﬁgure. At the new displacement un−1 +δu1n , the internal force g – represented by the curve – is still smaller than the imposed load. The diﬀerence forms the residual rn2 , and the procedure is continued. It should be noted that the use of suband superscripts to indicate load step and iteration number, respectively, is merely for illustration in relation to the ﬁgure. These indices are not needed when programming the algorithm. The iteration process needs a termination criterion. This may be taken as the requirement that the current residual force rn should be less than a prescribed fraction of the load increment ∆f of the present load step, |r| < |∆f |.

(1.31)

The value of could be on the order of say 10−4 –10−6 . For structures developing very small stiﬀness, the criterion (1.31) may be supplemented by the displacement criterion |δu| < |∆u|,

(1.32)

where ∆u is the total displacement increment accumulated in the present load step. In the corresponding multi-component problem with displacement vector u and load vector f , the residual force vector is r(u, f ) = f − g(u).

(1.33)

The tangent stiﬀness matrix is deﬁned by the incremental change of the

Introduction Algorithm 1.1. Newton–Raphson method. Load steps

n = 1, 2, . . . , nmax

fn = fn−1 + ∆fn un = un−1 Iterations i = 1, 2, . . . , imax dg(un ) du = fn − g(un )

Kn = rn

δun = K−1 n rn un = un + δun Stop iteration when rn < ∆fn End of load step

internal forces, K(u) =

dg(u) . du

(1.34)

The tangent stiﬀness is now introduced into a linearized form of the equilibrium condition, whereby the following vector equation is obtained for the displacement sub-increment δu: K(u) δu = r.

(1.35)

In contrast to the one-dimensional case, the solution of these equations may be a non-trivial part of the procedure. The termination criteria will typically make use of the ‘length’ of the corresponding vectors, whereby (rTn rn )1/2 < (∆f T ∆f )1/2 ,

(1.36)

(δuT δu)1/2 < (∆uT ∆u)1/2 .

(1.37)

The Newton–Raphson procedure is summarized as Algorithm 1.1. Note that in the iteration loop the computer overwrites quantities like rn by their new value in the same register. Therefore, the superscript i does not appear explicitly in the algorithm. Similarly, the load step subscript n is only used to avoid the explicit indication of storing the result of each completed load step. The actual algorithm is conveniently programmed without the use of indexed variables in the iteration loops.

1.2 Simple non-linear solution methods

1.2.3 Modiﬁed Newton–Raphson method In the original Newton–Raphson method the current tangent stiﬀness matrix K(u) is computed and factored in each iteration. For non-linear problems with a single or a few degrees of freedom this is usually not a problem, but for problems with many degrees of freedom the computational cost involved in forming the stiﬀness matrix K(u) and solving the corresponding equations for δu in each iteration may be considerable. It is seen from Fig. 1.6 and Algorithm 1.1 that Kn appears within the inner loop. A simple modiﬁcation of the Newton–Raphson method consists in moving the stiﬀness matrix K outside the iteration loop. Then K = Kn−1 is only computed and factored once for each load step on the basis of the previous state of displacement un−1 , dg(un−1 ) . (1.38) du This simpliﬁes the iteration loop as shown in Fig. 1.7 and Algorithm 1.2. Kn−1 =

Fig. 1.7. The modiﬁed Newton–Raphson method.

The asymptotic convergence of the modiﬁed Newton–Raphson method is slower than that of the Newton–Raphson method, and this may oﬀset some of its computational eﬃciency. A diﬀerent, more reﬁned type of modiﬁcation makes use of a secant approximation of K. This requires a non-trivial generalization of the secant concept to multi-degree of freedom systems. The corresponding methods, called quasi-Newton methods, are described in Chapter 8. The classic Newton methods encounter problems at a load maximum. Several methods have been developed to deal with this problem. A common feature of these methods is that the load increment is also subject to changes during the iterations, e.g. by linking load and displacement increments. This

Introduction Algorithm 1.2. Modiﬁed Newton–Raphson method. Load steps

n = 1, 2, . . . , nmax

fn = fn−1 + ∆fn un = un−1 dg(un−1 ) du Iterations i = 1, 2, . . . , imax Kn−1 =

rn = fn − g(un ) δun = K−1 n−1 rn un = un + δun Stop iteration when rn < ∆fn End of load step

introduces a kind of displacement control near the maximum, as discussed in Chapter 8.

1.3 Summary and outlook Non-linear problems of structures and solids involve processes in which neighboring states are connected by non-linear relations. This is illustrated by the simple example of a two-bar truss loaded by a quasi-static force. A model of the problem requires representation of the material behavior and the formulation of suitable equilibrium equations. In this introductory chapter these issues were dealt with in an ad hoc fashion, also introducing approximations to simplify the presentation. In the following chapters these issues are dealt with in a rigorous way within the framework of ﬁnite element analysis. While the equilibrium equations of simple systems often can be formulated directly, it is generally advantageous to use the principle of virtual work, which retains its simplicity for larger and more complicated systems. Several aspects of this will appear in later chapters. The idea of the principle of virtual work is to consider the work done by the actual forces through an imagined – or virtual – displacement ﬁeld. This changes a multi-component problem into a similar number of scalar problems. A necessary requirement is that the product of the internal forces and the virtual measures of deformation constitute work. This property of the virtual work serves to identify suitable internal force and stress measures, when a displacement and strain

1.4 Exercises

representation has been selected. This is illustrated in the next chapter in connection with a general discussion of bar elements, and is used to deﬁne several stress measures for solids in Chapter 6. The virtual work also plays a key role in deﬁning the properties associated with rotations and moments as discussed in Chapter 4. In order to obtain the solution to speciﬁc problems, the material behavior must be represented in the form of a relation between the internal forces and the corresponding measures of deformation. In the problem of the two-bar truss the bars were assumed to be linear elastic, thus simplifying the presentation. In many situations non-linear material behavior is an important part of the problem. The basic form of the equations of elastoplastic material behavior is described in Chapter 7. Even the simple two-bar truss exhibits a non-monotonic relation between load and displacement. The simple Newton-type solution methods brieﬂy outlined in this chapter need to be modiﬁed in order to enable computation of non-monotonic force–displacement relations. This problem is dealt with in Chapter 8 on numerical methods. The chapter on numerical methods can be read without reference to the chapters before, and indeed reading this chapter ﬁrst will enable the reader to supplement the theory of the intermediate chapters with non-trivial numerical examples. In the ﬁnal chapter the numerical methods are extended to dynamic problems with inertial eﬀects.

1.4 Exercises Exercise 1.1 Consider the load–displacement curve of the two-element truss shown in Fig. 1.3. (a) Determine the non-dimensional coordinates to the points B and D. (b) Select a suitable load step magnitude and sketch the states produced by the explicit incremental method, the Newton–Raphson method, and the modiﬁed Newton–Raphson method. Note in particular the passage of the maximum at B. Exercise 1.2 Consider the two-element truss shown in Fig. 1.1 and add a vertical spring at the central node with spring stiﬀness k = 1.2EAa2 /l03 . (a) Introduce the non-dimensional displacement v = −u/a and the nondimensional load p = −f l03 /(EAa3 ). Give the relation between p and v and the corresponding tangent stiﬀness. (b) Organize the explicit incremental method, the Newton–Raphson method, and the modiﬁed Newton–Raphson method in tabular form and com-

Introduction

pute the p–v relation with a suitable load step, e.g. 0.2–0.4. Sketch the result for each of the three methods. Exercise 1.3* Implement the Newton–Raphson algorithm shown as Algorithm 1.1 in Matlab for analysis of the two-bar truss treated in Section 1.1. Organize the implementation in four m-ﬁles: (a) data.m: script ﬁle containing the model parameters and the load increment, e.g. b = 1.0, a = 0.1, EA = 1.0, k = 0, and ∆f = −0.0001. (b) g bar.m: the internal force in (1.24), obtained as a function of the displacement u from the expression (1.11). (c) kt bar.m: the function K(u) in (1.12), giving the tangent stiﬀness as a function of the displacement u. (d) nr bar: script ﬁle containing the Newton–Raphson algorithm from Algorithm 1.1. (Use limited loops, e.g. n = 1 : nmax and i = 1 : imax , with nmax = 20 and imax = 8.) Use the program to study the behavior of the Newton–Raphson algorithm via plots of f as a function of displacement u, particularly near the turning point of the load–displacement curve. Use e.g. ∆f = −1.0 × 10−4 and ∆f = −0.5 × 10−4 . Exercise 1.4* Make a modiﬁed version mnr bar.m of the driver routine nrbar.m from Exercise 1.3 using the modiﬁed Newton–Raphson algorithm. Use the two routines to study the behavior of the algorithms for diﬀerent values of the spring constant k. Note in particular that the modiﬁed Newton– Raphson algorithm has convergence problems for increasing stiﬀness. Sketch this problem. Exercise 1.5* Both the full and the modiﬁed form of the Newton–Raphson algorithm are based on prescribed load increments, and therefore require monotonically changing load. In the two-bar truss problem the load will increase monotonically if a vertical spring with stiﬀness k > (EA/l0 )(a/l0 )2 is introduced. The actual force transferred to the truss is feﬀ = f − ku. Introduce a spring of suﬃcient stiﬀness and solve the problem with the program developed in Exercise 1.3. Plot the eﬀective load feﬀ against the displacement u to demonstrate recovery of the full curve from Fig. 1.3. For a very stiﬀ spring equal load increments correspond to equal displacements, and the method is equivalent to the use of displacement control. This method is only directly applicable in the case of a single load component. General solution methods are discussed in Chapter 8.

2 Non-linear bar elements

The ﬁnite element method has been the method of choice for modeling and analysis of structures and solids for several decades. The basic idea is that the structure (or solid) is considered as an assembly of elements, and that each element is modeled in a standard format that permits repetitive use of the individual element formats. Bar elements only contain a single internal degree of freedom – the elongation – and they are therefore a convenient means for introducing and illustrating the basic features of geometrically non-linear ﬁnite element analysis. In a geometrically non-linear problem the ﬁrst question to arise is the definition of a non-linear measure of deformation, the strain. This is addressed in Section 2.1. When a structure is assembled from the individual elements, use is generally made of the principle of virtual work. The principle of virtual work is a restatement of the equilibrium equations in scalar form. It turns out that once a non-linear strain deﬁnition has been adopted, the corresponding deﬁnition of stress follows from the formulation of the principle of virtual work. This is the subject of Section 2.2. The tangent stiﬀness matrix of a geometrically non-linear bar element is derived in Section 2.3 in global coordinates. The derivation of the equilibrium and stiﬀness relations of the bar element is quite simple because the strain is constant within the element. In order to illustrate the relation to more complex problems involving other types of elements, the tangent stiﬀness matrix is re-derived by use of shape functions in global coordinates. This indicates the procedure followed in isoparametric solid elements. Another alternative makes use of a local coordinate system rotating together with the element during displacement. This so-called corotational approach is typically used for generalized continuum elements such as beams and shells, where the displacement representation depends on the local orientation of the beam axis, middle plane, etc. Co-rotating 17

Non-linear bar elements

beam elements are treated in Chapter 5. Bar elements have been given an extensive treatment by Crisﬁeld (1991), and Mattiasson (1983) has described the co-rotational formulation of bar elements in detail. The implementation of the ﬁnite element method involves the assembly of elements into the global structure. A brief sketch of the assembly procedure is given in Section 2.5. The actual solution of the ﬁnite element relations can follow one of two formulations. In the ﬁrst the initial conﬁguration is used as reference through the full analysis, and total displacements from this conﬁguration are used. This is the total Lagrangian formulation. In the other formulation the reference state is updated each time an equilibrium state has been established. Thus, the displacements in this formulation refer to the last equilibrium state. This is the updated Lagrangian formulation. The simple bar element is used to illustrate these two formulations in Section 2.6. In spite of its simplicity the bar element contains the main features of geometrically non-linear structural and solid elements, and Section 2.7 sums up these main features in a general form. These general features are encountered repeatedly in various settings in the following chapters.

2.1 Deformation and strain In most ﬁnite element formulations for structures and solids the displacements are the primary variables of the problem. The displacements may lead to deformation of the elements, and this in turn to internal forces. It is important to use deformation measures that vanish identically for rigid body displacements. In a theory with ﬁnite displacements this requirement can be satisﬁed by several diﬀerent strain deﬁnitions. The most common of these are brieﬂy discussed here for a simple bar element to indicate their use and limitations. In a bar element the deformation is characterized by the elongation. Figure 2.1 shows a bar element with initial length l0 . Deformation introduces the elongation u, whereby the new length is l = l0 + u.

(2.1)

A suitable measure of strain must give the relative deformation. The simplest deﬁnition is the engineering strain, εE =

u l − l0 = . l0 l0

(2.2)

This is the traditional deﬁnition used in the theory of inﬁnitesimal strain. For ﬁnite deformation it is somewhat arbitrary to refer to the initial length

2.1 Deformation and strain

Fig. 2.1. Bar element.

l0 . Alternatively, the strain increment δε can be deﬁned with reference to the current length l by the relation δl . l Integration of this relation gives the logarithmic strain

l l . δε = ln εL = l0 l0 δε =

(2.3)

(2.4)

This strain makes no distinction between initial and ﬁnal length, and interchange of l0 and l merely changes the sign of εL . For small strains εE and εL are nearly equal. Many problems involve large displacements but only small to moderate strains. In those problems the important point is to use a strain deﬁnition without ‘self straining’, i.e. a strain deﬁnition that does not produce straining for arbitrary rigid body motion. It is then convenient to use l2 instead of l as a basis for the strain deﬁnition. One reason for the use of l2 is that the square of the length a of a vector a = [a1 , a2 , a3 ]T is the sum of the squared coordinates, a2 = a21 + a22 + a23 . Another is that a deﬁnition based on l2 is simpler to generalize to two- and three-dimensional continua. A strain deﬁnition based on l2 can be obtained by rewriting (2.2) in the form εE =

l2 − l02 (l − l0 )(l + l0 ) = 2 . l0 (l + l0 ) l0 (2 + εE )

(2.5)

If εE is omitted from the denominator, the deﬁnition of the Green strain εG is obtained, l2 − l02 . (2.6) εG = 2 l02 This strain deﬁnition, and its generalization to two and three dimensions, are often used in solid mechanics. The diﬀerent strains introduced here are related by εL = ln(1 + εE ) =

1 2

ln(1 + 2εG ).

(2.7)

Non-linear bar elements

Fig. 2.2. The strain measures εE , εG and εL .

The strains εE , εL and εG are shown in Fig. 2.2 as a function of u/l0 . It is seen that εG > εE > εL . For |u/l0 | < 0.05 the deviation from εE is of the order 2–3%. For larger strains it may be necessary to distinguish between the strain to be used or account for the diﬀerence through the stress–strain relation.

2.2 Equilibrium and virtual work Figure 2.3 shows a bar element with end-points A and B. The coordinates of A and B are referred to a global Cartesian coordinate system. In the initial conﬁguration the coordinates are xA0 and xB0 , respectively. Boldface letters are used to denote vectors and matrices. The dimension corresponds to the dimension of the space, i.e. 2 or 3. It turns out that this dimension is not important in the formulation of the equilibrium and stiﬀness relations except at the very end, when individual components are written down.

Fig. 2.3. Bar element AB in initial and displaced conﬁgurations.

The points A0 and B0 are now given the displacements uA and uB , re−−→ spectively. Only the direction and length of the vector AB inﬂuence the

2.2 Equilibrium and virtual work

equilibrium and stiﬀness. The initial position is x0 = xB0 − xA0 .

(2.8)

The diﬀerence between the displacement at A and B is denoted u = uB − uA ,

(2.9)

−−→ and thus AB after displacement is x = x0 + u.

(2.10)

It is now a simple matter to calculate the initial length l0 and the length l after deformation: l02 = xT0 x0 ,

(2.11)

l2 = xT x = (x0 + u)T (x0 + u).

(2.12)

The strain of the bar can now be expressed in terms of l0 and l by any of the strain measures deﬁned in Section 2.1. The fact that the square of the length is given in simple form suggests the use of the Green strain εG . From the deﬁnition (2.6) it follows that εG =

l2 − l02 1 = 2 (xT0 u + 12 uT u). 2 2 l0 l0

(2.13)

This formula can be interpreted as a projection of the displacement u on the mean vector x1/2 = 12 (x0 + x), scaled by l02 : εG =

1 1 1 (x0 + 12 u)T u = 2 21 (x0 + x)T u = 2 xT1/2 u. 2 l0 l0 l0

(2.14)

This interpretation is illustrated in Fig. 2.4. In the general formulation of the Green strain the interpretation in terms of a mean position vector may be helpful in the evaluation of εG . A slightly more general formulation is discussed in Exercise 2.1.

Fig. 2.4. Projection of displacement u on mean vector x1/2 = 12 (x0 + x).

Non-linear bar elements

The variation of the strain (2.13) is needed for the formulation of the principle of virtual work: δεG =

1 1 (x0 + u)T δu = 2 xT δu. 2 l0 l0

(2.15)

Note that the variation of the strain consists of the projection of the displacement variation δu on the current vector x, scaled by l02 . The universal method of establishing ﬁnite element relations consists in the use of the principle of virtual work. In the case of bar elements the state of deformation within the element is hom*ogeneous, and the relations could therefore in principle be obtained directly, but in order to illustrate the general procedure and to identify precisely the internal force that is conjugate to the Green strain, the principle of virtual work is also used here. According to the principle of virtual work an arbitrary displacement variation must lead to identical internal and external work. The internal virtual work is expressed in terms of the internal force(s), while the external virtual work is expressed in terms of the external loads. Thus, the principle of virtual work establishes a relation between the internal and the external forces. For a bar element with the external forces fA and fB acting at A and B, respectively, the principle of virtual work is

δV = N δε ds − fAT δuA − fBT δuB = 0. (2.16) In order to be speciﬁc, a strain measure must be selected and an appropriate length measure must be used in the integral. Here the Green strain is used together with the initial length l0 of the element. With this choice and δεG from (2.15) the principle of virtual work takes the form

l0 1 T T T (2.17) δV = 2 (N x )(δuB − δuA ) ds − fA δuA − fB δuB = 0. l 0 0 Taking the transpose of (2.17) leads to changing the order of the factors, whereby l0 1 (N x) ds − f δV = δuTA − A 2 0 l0 l0 1 T + δuB = 0. (2.18) (N x) ds − f B 2 0 l0 This equation must be valid for any choice of the virtual displacements

2.2 Equilibrium and virtual work

δuA and δuB , and therefore the following two equations are obtained after evaluating the integrals: 1 1 fA = − N x, fB = N x. (2.19) l0 l0 These equations give a precise deﬁnition of the normal force N appearing in the principle of virtual work as conjugate to the Green strain. The vector x/l0 is in the direction of the bar after displacement with length l/l0 . Thus N l/l0 is the actual force in the bar element. For moderate strain l/l0 1, and N gives the actual force directly. For large strains the diﬀerence between N and the actual force can be absorbed in the constitutive equation for N . The main point is that with this deﬁnition of N the use of the Green strain and initial length in the principle of virtual work (2.16) is an exact representation of the equilibrium equations. The use of engineering strain is discussed in Exercise 2.2. For a linear elastic bar, in which N is proportional to εG , the constitutive relation is N = EA εG ,

(2.20)

where E is the modulus of elasticity and A is the cross-section area. The internal forces qA , qB generated by the deformation of the element can then be expressed in terms of the displacements uA and uB by substitution of (2.20) into (2.19): qA = − EA εG

1 x, l0

qB = EA εG

1 x. l0

(2.21)

With the displacements uA and uB available the displacement vector u is evaluated from (2.9), the current vector x from (2.10), and εG from (2.14). This enables the evaluation of the internal forces qA , qB in an iterative solution procedure. In order to ﬁnd the corresponding displacement increment the current stiﬀness is needed. The formulae for strain and equilibrium derived so far have been expressed in a compact form making use of the diﬀerences in position and displacement between the two ends of the bar. This form is convenient for derivation and discussion of the physical meaning of the formulae. However, in a ﬁnite element formulation it is important to have a convenient matrix format, in which all coordinates, displacements and forces appear in vector format. A tilde is introduced to denote an extended vector, containing the vectors from all element nodes. In this notation the extended position, displacement and internal force vectors of the bar element are x ˜T = [ xTA , xTB ],

u ˜ T = [ uTA , uTB ],

q ˜T = [ qTA , qTB ]

(2.22)

Non-linear bar elements

for coordinates, displacements and forces, respectively. In terms of this notation the formula (2.11) for the initial length of the element is I −I T 2 T x ˜0 , l0 = x0 x0 = x (2.23) ˜0 −I I where I is the unit matrix corresponding to the dimension of the space, typically 2 or 3. By use of (2.8) and (2.9) the formula (2.14) for the Green strain is then rewritten as 1 T 1 1 I −I I −I T u ˜ = 2x u ˜. (2.24) εG = 2 2 (˜ x0 + x ˜) ˜ −I I I l0 l0 1/2 −I The strain increment (2.15) takes the form 1 T I −I δ˜ u. δεG = 2 x ˜ −I I l0

(2.25)

The diﬀerence between the mean position x ˜1/2 in εG and the current position x ˜ in δεG is clear. The internal element forces (2.21) have the following simple matrix form: EAεG N I −I I −I x ˜ = x ˜, (2.26) q ˜ = I −I I l0 −I l0 where εG is evaluated from (2.24). Note that the strain and the internal forces q ˜ are non-linear functions of the displacements u ˜.

2.3 Tangent stiﬀness matrix The tangent stiﬀness matrix gives the changes in the internal forces qA and qB corresponding to inﬁnitesimal changes in the displacements uA and uB . It is conveniently found from diﬀerentiation of (2.21a): dN N − dx l0 l0 x dN N = − + I d(uB − uA ), l0 du l0

dqA = − x

(2.27)

where I is the unit matrix. It follows from (2.21) that dqB = − dqA .

(2.28)

2.3 Tangent stiﬀness matrix

For a linear elastic bar with the constitutive relation (2.20), diﬀerentiation gives dεG EA EA dN = EA = 2 (xT0 + uT ) = 2 xT . (2.29) du du l0 l0 When this expression is inserted into (2.27) and (2.28), the result can be written in block matrix format as dqA x xT −x xT I −I duA EA N = + . (2.30) l0 −I dqB −x xT x xT duB I l03 Note that the format of the equations (2.27) implies the identity dqB = −dqA and dqB = dqA = 0 for a rigid body translation duB = duA . The ﬁrst matrix in (2.30) represents the constitutive stiﬀness due to material deformation, while the second matrix is the initial stress stiﬀness, often called geometric stiﬀness. The separation into these two parts can be illustrated by considering the representation of a vector N by its length N and a unit vector e = N/N giving the direction, i.e. N = N e.

(2.31)

The corresponding incremental form is dN = dN e + N de.

(2.32)

The increments are illustrated in Fig. 2.5 for the case where the length of the vector e remains constant, i.e. de corresponds to a rotation of e. The rotation term N de corresponds to the initial stress term, while the change of length dN e corresponds to the constitutive stiﬀness term.

Fig. 2.5. Decomposition of increment into a rotation and a change of length.

In terms of the vector notation (2.22), the tangent stiﬀness relation (2.30) takes the form ∂˜ q d˜ q = d˜ u = K d˜ u. (2.33) ∂˜ u The tangent stiﬀness matrix K consists of the following three contributions: K = K 0 + Ku + Kσ ,

(2.34)

Non-linear bar elements

with the linear stiﬀness matrix EA K0 = 3 l0

x0 xT0 −( · · · )

−( · · · ) , +( · · · )

the initial displacement stiﬀness matrix EA x0 uT + uxT0 + uuT Ku = 3 −( · · · ) l0

−( · · · ) , +( · · · )

and the initial stress stiﬀness matrix N I −I . Kσ = I l0 −I

(2.35)

(2.36)

(2.37)

The compact products in terms of x and u can be computed by formulae similar to (2.23), but direct computation will be more eﬃcient. In the reference state the initial displacement is zero, and the initial displacement stiﬀness matrix vanishes. 2.4 Use of shape functions The bar element is so simple that the full description could be made entirely by reference to the vector x connecting the end-points A and B. For most other elements the notion of shape functions is needed. Shape functions describe the displacement ﬁeld in the element, usually in terms of a set of local, or basic, coordinates. In order to illustrate the use of shape functions in a simple case, the equilibrium equations already treated in Section 2.2 are re-derived. The position vector r(ξ) of an arbitrary point of the bar AB is deﬁned by interpolation between the end-points xA and xB . In the initial conﬁguration r0 (ξ) = hA (ξ)xA0 + hB (ξ)xB0 .

(2.38)

The shape functions are deﬁned by hA (ξ) = hB (ξ) =

1 2 (1 1 2 (1

− ξ),

−1 < ξ < 1,

+ ξ),

−1 < ξ < 1.

(2.39)

The basic coordinate ξ is traditionally normalized to the interval (−1, 1). The shape function representation (2.38) can also be written in matrix format: r0 (ξ) = [ hA (ξ)I , hB (ξ)I ] x ˜0 .

(2.40)

In the displaced conﬁguration the same relation holds, but with x ˜0 replaced by x ˜=x ˜0 + u ˜.

2.4 Use of shape functions

The Green strain at a point deﬁned by ξ is determined by using the deﬁnition (2.6) on an inﬁnitesimal vector with initial value dr0 (ξ) and ﬁnal value dr(ξ): εG (ξ) =

drT dr − drT0 dr0 . 2 drT0 dr0

(2.41)

The inﬁnitesimal vector dr0 is found by diﬀerentiation of (2.40) as dh

dhB dr0 A I, I x ˜0 = − 12 I , 12 I x = ˜0 . dξ dξ dξ

(2.42)

The inﬁnitesimal vector dr in the displaced conﬁguration follows from a similar formula with x ˜0 replaced by x ˜ = x ˜0 + u ˜ . From this the scalar products in the strain deﬁnition (2.41) are found as I −I T 1 T x ˜0 dξ 2 = 14 l02 dξ 2 dr0 dr0 = 4 x ˜0 (2.43) −I I and

dr dr =

1 x0 4 (˜

+u ˜)

I −I −I I

(˜ x0 + u ˜ ) dξ 2 =

2 1 2 4 l dξ .

(2.44)

Substitution of these expressions into the strain deﬁnition (2.41) gives 1 T 1 I −I I −I T 1 u ˜ = 2x u ˜. (2.45) x0 + 2 u ˜) ˜ εG = 2 (˜ −I I I l0 l0 1/2 −I This is the formula previously established as (2.24). The strain increment follows by diﬀerentiation as in (2.25): 1 T I −I δ˜ u. (2.46) ˜ δεG = 2 x −I I l0 Thus, the strain and strain increment have been established on a pointwise basis. The equilibrium equations can now be established directly from the principle of virtual work. In the present bar element the only external loads act at the end-points, and therefore the principle of virtual work takes the form

l0 δεG N ds − δ˜ uT ˜ f = 0. (2.47) δV = 0

This identity is expressed in matrix form by substitution of the virtual strain from (2.46). In the present case the integration is trivial. For more complicated elements the volume integral must be evaluated by approximate

Non-linear bar elements

numerical integration in terms of the basic variables (ξ, . . .). δ˜ u

N l0

I −I ˜ x ˜ − f = 0. −I I

(2.48)

The variation δ˜ u of the displacement vector is arbitrary, and thus the scalar identity (2.48) gives the vector equilibrium equation N l0

I −I x ˜ =˜ f. −I I

(2.49)

The tangent stiﬀness relation follows by considering increments in the equilibrium equation (2.48). The procedure and results are the same as in Section 2.3. Example 2.1. Equilibrium and stiffness of two-element truss. The equilibrium equations and the tangent stiﬀness of the two-element truss shown in Fig. 2.6 are derived and used to describe a lateral bifurcation instability and the associated displacement pattern (Pecknold et al., 1985). The equations are used to illustrate special numerical solution techniques in Chapter 8.

Fig. 2.6. Two-element truss supported by a lateral spring.

Figure 2.6 shows a truss consisting of two identical bar elements with stiﬀness AE. The initial geometry is speciﬁed by the lengths a, b and c, where c indicates the deviation from vertical, e.g. due to a geometric imperfection. The initial length of the elements is denoted l0 . Lateral support is provided by a linear spring with stiﬀness k. This spring remains horizontal during deformation. The load consists of a vertical force f1 and a horizontal force f2 acting at the top of the truss. In the present example the general equilibrium equations and tangent stiﬀness are derived. The speciﬁc results are here limited to the ‘perfect’ structure, c = 0 and f2 = 0.

2.4 Use of shape functions

In the present example the simplest way to obtain the equilibrium equations is by use of the potential energy Φ = 2 12 l0 EAε2G +

2 1 2 ku2

− fα uα ,

where a repeated Greek subscript implies summation over the terms, i.e. fα uα = α fα uα . By use of the stiﬀness AE and the initial length l0 = a2 + b2 + c2 , the following non-dimensional parameters are introduced: α = a/l0 ,

β = b/l0 ,

γ = c/l0 ,

vα = uα /l0 ,

κ = kl0 /EA,

pα = fα /EA.

In terms of these variables the strain εG is εG = −v1 (α − 12 v1 ) + v2 (γ + 12 v2 ) and the potential energy takes the form Φ = l0 EA ε2G + 12 κv22 − pα vα . The equilibrium equations are now established in non-dimensional form from the stationarity condition δΦ = 0 using the strain variation −(α − v1 ) δεG = −δv1 (α − v1 ) + δv2 (γ + v2 ) = (δv1 , δv2 ) . (γ + v2 ) The resulting equilibrium equations are −2εG (α − v1 )

= p1 ,

2εG (γ + v2 ) + κv2 = p2 . The tangent stiﬀness relation follows from the equilibrium equations by differentiation as

2(α − v1 )2 + 2εG −2(α − v1 )(γ + v2 ) −2(γ + v2 )(α − v1 ) 2(γ + v2 )2 + 2εG + κ

dv1 dv2

dp1 dp2

The terms 2εG correspond to the initial stress stiﬀness. These are the two sets of equations used in a non-linear analysis of the two-element truss. A brief analysis of the ideal, symmetric system with γ = 0 and p2 = 0 is now given. This analysis establishes the principal behavior of the system

Non-linear bar elements

and typical load levels to be used in Chapter 8. For the ideal, symmetric truss the transverse equilibrium equation is (2εG + κ) v2 = 0. This equation can be satisﬁed in two ways: v2 = 0 or 2εG + κ = 0. The ﬁrst solution is symmetric, while the second is non-symmetric. The symmetric branch, v2 = 0, corresponds to the introductory example presented in Section 1.1. In the present notation the load is

p1 = α2 − (α − v1 )2 (α − v1 ) = 2α2 v1 − 3αv12 + v13 , corresponding to (1.6). The maximum load on this branch is attained when dp1 /dv1 = 0, i.e. at 1 v1 = α 1 − √ , 3

p1 =

2α3 √ . 3 3

The symmetric branch is shown as ABCD in Fig. 2.7. The point C of maximum load is called a limit point.

Fig. 2.7. Symmetric and non-symmetric branches for two-element truss.

The non-symmetric solution is only possible when the strain has reached the value εG = − 21 κ. Initially this corresponds to v2 = 0 and 2εG + κ = 0. These combined conditions determine the bifurcation point B shown in Fig. 2.7. At the bifurcation point v1 = α − α2 − κ, p1 = κ α2 − κ. A comparison between the location of the bifurcation point and the limit point leads to the following conclusions regarding the inﬂuence of the spring

2.5 Assembly of global stiﬀness and forces

stiﬀness κ: 0 < κ < 32 α2

bifurcation point before limit point;

2 2 3α

bifurcation point after limit point;

< κ < α2

α2 < κ

no bifurcation point.

For κ < α2 the non-symmetric branch satisﬁes the condition 2εG + κ = 0. Clearly the strain εG remains constant, while the displacement components describe a circle with the equation (α − v1 )2 + v22 = α2 − κ. On this branch the load is determined by p1 = −2εG (α − v1 ) = κ(α − v1 ) and thus the non-symmetric branches CD and CD form parts of a plane ellipse, while D B D forms a semi-circle. For a non-symmetric truss the bifurcation and limit loads of the equivalent symmetric truss are never reached. However, for a nearly symmetric truss the load–deformation curves are qualitatively similar to those of Fig. 2.7, and thus the solution for the symmetric truss provides useful information when investigating the non-symmetric situation.

2.5 Assembly of global stiﬀness and forces So far the current internal force and the tangent stiﬀness have been determined for an individual bar element. A structure consists of an assembly of elements, joined at nodes. In the case of bar elements the loads and supports are applied at the nodes. The displacement of the complete structure is described completely in terms of the displacement vector un at each node, and similarly the resulting force at each node can be found by summation of the contributions from all elements attached to that node. Thus, the displacement at a node is shared among elements, while the resulting internal force is accumulated by summation over the elements. In the case of truss structures formed from bar elements this argument is simple and suﬃcient, but for more general elements, e.g. for shells and solids, a more general approach is needed, as the elements are joined along curves and surfaces and not just at the nodes. The current case of truss structures and bar elements is therefore used to illustrate the general procedure using the principle of virtual work. The use of the principle of virtual work also turns out to play

Non-linear bar elements

a key role in the formulation of tangent stiﬀness relations for elements with displacement ﬁelds described in terms of translations and rotations, such as beams and shells. Consider a truss structure formed by joining a number of bar elements at a set of nodes. The principle of virtual work states that equilibrium corresponds to vanishing diﬀerence between the internal and the external virtual work. The internal virtual work is the sum of the contributions from each element, and for truss structures, where the loads and supports are applied at the nodes, the external virtual work is the work of the forces at the nodes. Thus, the virtual work of the structure takes the form

δV = δε N ds − δuTn fn = 0, (2.50) elements

nodes

where the virtual strain δε in the elements is deﬁned by (2.15) in terms of the virtual displacements δu at the element nodes. The incremental virtual strains δε can be expressed by a derivative of the total strain as δε =

∂ε ∂ε δun = δuTn T , ∂un ∂un

nodes

(2.51)

nodes

where the last form brings the displacement factor δuTn to the front by transposing and interchanging the order of the two factors in the scalar product. The strain in each element only depends on the displacement vector at the nodes of the element, and thus only these nodes participate in the summation. The virtual work equation (2.50) can now be written as ∂ε T δV = (2.52) δun N ds − fn = 0. ∂uTn nodes

elements

The integrals over the elements are identiﬁed as the internal element forces qn at the particular node. In fact, it follows from diﬀerentiation of the strain deﬁnition (2.13) that

1 1 ∂ε ∂ε qA = N ds = − N x, q = N ds = N x (2.53) B T T l0 l0 ∂uA ∂uB when the element vector x is oriented from A to B. One equilibrium vector equation is recovered for each node from (2.52) by considering one non-vanishing virtual displacement component at a time. When it is realized that the integrals deﬁne the internal forces from the individual elements at the appropriate node n, these equilibrium equations

2.5 Assembly of global stiﬀness and forces

take the form

qn = fn ,

n = 1, . . .

(2.54)

elements

Only the elements connected to the node n participate in the sum relating to the equilibrium of that node. The equations (2.54) express equality between the internal forces, generated within the elements of the structure, and the external loads, applied at the nodes. In the solution process the relation between an inﬁnitesimal change in the loads and the corresponding change in the displacements is needed. This so-called tangent stiﬀness relation is obtained by considering the diﬀerence between two neighboring states of equilibrium. In each state of equilibrium the virtual work vanishes, and thus the same applies to the increment of the virtual work, symbolically written as d(δV ) = lim δV2 − δV1 . (2.55) The virtual work δV is linear in δu and can be considered as a ﬁrst variation of a functional V . In this terminology d(δV ) is the second variation. In the virtual work equation the displacement variation δu is subject to choice, and if the same displacement variations are selected for both of the neighboring states, the second variation (2.55) takes the form (2.56) δuTn dqn − dfn = 0. d(δV ) = nodes

elements

As before, this relation can be converted into a vector equation for each of the nodes: dqn = dfn , n = 1, . . . , (2.57) elements

this time for the increments of the internal element forces dqn and the external loads increment df . For each element the increment of the internal forces q ˜ and the displacement increments u ˜ of the element nodes are related by the element tangent stiﬀness matrix, d˜ q =

∂˜ q ˜ d˜ d˜ u = K u, ∂˜ u

(2.58)

˜ is introduced to denote the element tangent stiﬀness where the notation K matrix. When this expression is substituted into the incremental equilibrium equation (2.57), it takes the symbolic form ˜ du = df K (2.59) elements

Non-linear bar elements

in which the vectors du and df contain the components from all the nodes of the structure. The symbolic summation over the elements deﬁnes the global tangent stiﬀness matrix ˜ K = K. (2.60) elements

The assembly procedure for the global tangent stiﬀness matrix K and for the total internal forces q is illustrated in the following example. Example 2.2. Element formulation of two-element truss. In Example 2.1 the equilibrium equations and the tangent stiﬀness of the twoelement truss shown in Fig. 2.6 were derived directly from the elastic energy. This example shows how the current internal forces and the incremental equilibrium equations of the truss are obtained by assembling the element contributions. The left support is denoted A, the top node B, and the right support C. The truss then consists of the two elements AB and BC, connected at the node B. The displacement and internal element force vectors of the element AB are u ˜ T = [ uTA , uTB ],

q ˜T = [ qTA , qTB ],

while for the element BC they are u ˜ T = [ uTB , uTC ],

q ˜T = [ qTB , qTC ].

The current internal element forces are given by (2.26), N I −I x ˜, q ˜ = I l0 −I with the normal force N = EAεG , and the strain εG given by (2.24). The vector x ˜ describes the current position of the element nodes. At each node the total internal force is the sum of the contributions from all elements connected to that node. In the present example this can be expressed as       qA qA  qB  =  qB  +  qB  . qC qC BC AB The assembly of element forces consists of arranging each element force vector q ˜ in the global numbering scheme, and then adding the contributions. The element tangent stiﬀness matrix is given by (2.30) as N EA xxT −xxT I −I ˜ , Kδ = 3 + I xxT −xxT l0 −I l0

2.5 Assembly of global stiﬀness and forces

where x is the vector describing the current position of the element. The element stiﬀness matrices must be assembled such that the displacement vector at a shared node is common, while the element forces at that node are added. This is accomplished by the format          

KAB KBC

  duA     duB 

  dqA    =    dqB  

duC

dqC

  .  

The assembly of the tangent stiﬀness matrix consists of appropriate positioning of the element stiﬀness matrices within the global stiﬀness matrix format and then adding all the element contributions to form the global stiﬀness matrix. In the present example the two element stiﬀness matrices simply overlap with one quarter. In general an element AB is connected to nodes with global numbers i and j. The element stiﬀness matrix is then divided into four sub-matrices kAA kAB ˜ = K . kBA kBB The contribution from this element to the global tangent stiﬀness matrix can then be written as   .. .. . .    i  ··· k · · · k · · · AA AB     . .   . . K =  . .  .    · · · kBA · · · kBB · · ·  j   .. .

.. .

It is seen that the key to the assembly of the stiﬀness matrix is the identiﬁcation of the element node numbers within the global stiﬀness matrix. In the ﬁnite element method the element properties are evaluated and assembled for all nodes, and the boundary conditions are introduced subsequently in the global stiﬀness matrix and load vector. In the two-bar truss the nodes A and C are fully restrained, i.e. uA = 0 and uC = 0. This implies that the corresponding internal forces qA and qC are unknown and have the character of reactions. The fact that uA = 0 eliminates all contributions from the ﬁrst block column of the matrix equation. Similarly,

Non-linear bar elements

uC = 0 eliminates all contributions from the last block column. The ﬁrst and last block rows are not proper equations, as qA and qC are unknown. This leaves only the central block of the equation system. In the sub-matrix notation the ﬁnal equations are (AB) (BC) kBB + kAA duB = dqB . It may be shown that these equations are identical to those of Example 2.1, see Exercise 2.4.

2.6 Total or updated Lagrangian formulation Finite element formulations that make use of representations of the motion of the material points are called Lagrangian or material. This type of formulation is often used for structures and solids, and is indeed the one used here for truss structures. The alternative formulation, using a representation of the motion of material through a ﬁxed part of space, is called Eulerian or spatial. The Eulerian approach is typically used in ﬂuid mechanics. In recent years mixed Lagrangian–Eulerian methods have been developed, e.g. for modeling very large deformations of solids (Belytschko et al., 2000). The present formulation makes use of an initial conﬁguration x0 and displacements u from this conﬁguration. In practice, the solution often involves a number of load steps f1 , . . . , fn , . . . and the load step from fn−1 to fn may then be formulated in two ways. In the total Lagrangian formulation the initial conﬁguration x0 is retained as reference during the whole load history, and the total displacements un are used as illustrated in Fig. 2.8. This calls for use of the full tangent stiﬀness matrix (2.34), including the initial displacement stiﬀness matrix (2.36), during all stages of the computation.

Fig. 2.8. Total Lagrangian formulation.

2.6 Total or updated Lagrangian formulation

Algorithm 2.1. Total Lagrangian formulation. Load steps n = 1, 2, . . . , nmax fn = fn−1 + ∆fn un = un−1 Kn−1 = K0 (x0 ) + Ku (x0 , un−1 ) + Kσ (εn−1 ) Iterations εn

i = 1, 2, . . . , imax

1 = 2 l0

1 x0 2 (˜

rn = fn −

+x ˜n )T

I −I

EA εn l0

elements

−I I I −I

u ˜n −I I

x ˜n

δun = (Kn−1 )−1 rn un = un + δun xn = x0 + un Stop iteration when rn < fn End of load step

The main points of the total Lagrangian formulation are indicated in Algorithm 2.1 using a modiﬁed Newton–Raphson method. The formulae in the table are only indications of the main ideas, and no distinction is made between global and element parameters. Each load step starts with incrementing the load, initializing the displacement u0n from the last equilibrium state, and evaluating and factoring the tangent stiﬀness matrix Kn−1 at state n − 1. This tangent stiﬀness matrix is retained during the iterations according to the modiﬁed Newton–Raphson method. In each iteration the residual force rin is evaluated as the diﬀerence between the external and the internal forces, and the displacement iteration increment δuin is computed by use of the factored tangent stiﬀness matrix. Then the total displacement uin is updated. The ﬁnal evaluation of xin is just a computational convenience and does not change the original reference state x0 . As previously noted the iteration index i does not appear explicitly in the algorithm, as new values overwrite the previous value. In the updated Lagrangian formulation each load step ends with the deﬁnition of an updated conﬁguration xn = xn−1 +∆un , as illustrated in Fig. 2.9 and Algorithm 2.2. Thus, the ﬁrst evaluation of the tangent stiﬀness matrix in each load step does not contain any initial displacement contribution. In

Non-linear bar elements Algorithm 2.2. Updated Lagrangian formulation. Load steps n = 1, 2, . . . , nmax fn = fn−1 + ∆fn ∆un = 0 Kn−1 = K0 (xn−1 ) + Kσ (εn−1 ) Iterations ∆εn

i = 1, 2, . . . , imax 1 = 2 l0

1 xn−1 2 (˜

+x ˜n )T

I −I

εn = εn−1 + ∆εn EA I rn = fn − εn l0 −I elements

−I I −I I

∆˜ un

x ˜n

δun = (Kn−1 )−1 rn ∆un = ∆un + δun xn = xn−1 + ∆un Stop iteration when rn < fn End of load step

each iteration the residual force rin is evaluated and the displacement iteration increment δuin is computed by use of the factored tangent stiﬀness matrix as in the total Lagrangian formulation. The diﬀerence appears in the updates. In the updated Lagrangian formulation δuin is used to update the displacement ∆uin from the last equilibrium state, and the strain increment ∆εin also is relative to state n − 1. The change of reference conﬁguration in a Lagrangian formulation requires some care, because the strain deﬁnition (2.6) and the internal force deﬁnition

Fig. 2.9. Updated Lagrangian formulation.

2.7 Summing up the principles

(2.19) make use of a reference length from this state. If the current length ln−2 was used as reference length in load step n − 1, the actual normal force in the bar at the end of load step n − 1 is Nn−1 ln−1 /ln−2 . In the following load step the new reference length would be ln−1 and N should therefore be multiplied by the factor ln−1 /ln−2 when initiating load step n. For small strains this correction factor is without importance, and it is usually omitted. In Algorithm 2.2 the reference length l0 has been used throughout, whereby the updated method is just a reformulation of the total method. Diﬀerences arising in the total and updated Lagrangian formulations due to the use of diﬀerent reference lengths are discussed by Yang and Leu (1991).

2.7 Summing up the principles The results obtained in this chapter for bar elements are particularly simple because the deformation is one-dimensional and hom*ogeneously distributed within each bar element. However, the results are of general nature, and it is therefore worthwhile to restate the principal results in general terms, applicable also to other geometrically non-linear elements. The discussion makes use of the current coordinate vector x, the displacement vector u and the load vector f , and no explicit distinction is made between the element and the global level. The basic relation (2.50) is a statement of equilibrium formulated in terms of virtual work in the form

δV = δε N ds − δuT f = 0, (2.61) where the integral covers all elements, and the vectors u and f include all nodal displacements and the corresponding loads. The state of the structure is expressed in terms of the current geometry described by x and u, the internal state of stress described by the normal force N in each element, and the external load f . δε is the virtual strain corresponding to the virtual nodal displacements δu, i.e. δε is the strain increment corresponding to a virtual change of the nodal positions from x to x + δu. The increments δε and δu are considered as virtual – i.e. corresponding to a hypothetical change of displacement of the structure. Thus, they do not change the actual state of the structure, and the internal stresses and external loads remain unchanged. In the statement (2.61) of the virtual work principle the strain ε was deliberately stated without reference to the particular choice of strain measure. The important point is that the strain measure must be independent of any

Non-linear bar elements

rigid body displacement. For the bar this will be the case for any strain measures based only on the length of the bar. In Section 2.1 three diﬀerent strain measures were introduced, all meeting this requirement: the engineering strain εE , the Green strain εG , and the logarithmic strain εL . When formulated in two- or three-dimensional space, all of these strain measures are non-linear functions of the displacements. Indeed, any strain measure that is invariant to (large) rotations must be a non-linear function of the displacements. Also, the curve length parameter s was left unspeciﬁed in (2.61). The choice of curve parameter depends on the choice of strain. For engineering and Green strain the initial curve length s0 is used, while logarithmic strain corresponds to the use of current curve length s. Similar considerations hold for the internal force N , the interpretation of which also depends on the chosen strain measure. The strain increment is found by diﬀerentiation of the total strain ε, δε =

∂ε ∂ε δu = δuT T . ∂u ∂u

(2.62)

For bars with Green strain this relation corresponds to (2.46). For a nonlinear strain deﬁnition the derivative ∂ε/∂x depends on the current geometry x. In the virtual work equation the internal work is the product of the stresses and strains, and thus the stress measure depends on the selected strain measure, or conversely. This relation is explored further in relation to rotations in Chapters 3 and 4 and in relation to solids in Chapter 6. When the virtual strain δε in the principle of virtual work is expressed in terms of the virtual nodal displacements δu, the statement of equilibrium takes the form ∂ε δuT N ds − f = 0. (2.63) ∂uT internal force

The components of the virtual displacement vector δu can be associated with arbitrary values, and therefore the internal and external force vectors inside the parentheses must be equal. Thus, the scalar equation (2.63) is equivalent to the system of equilibrium equations

∂ε N ds = f (2.64) ∂uT corresponding to (2.54). These equilibrium equations only involve the current state of the structure, with ∂ε/∂uT appearing as a weight function on the internal state of stress N . In non-linear analysis the equilibrium equations derived from virtual work

2.7 Summing up the principles

are used to check equilibrium and to evaluate the corresponding residual, which is an expression for the unbalanced loads. When equilibrium has not been attained, it is necessary to consider a change of an equilibrium conﬁguration x, N with load f to a neighboring equilibrium conﬁguration x+ du, N +dN with load f +df . In contrast to the virtual increments considered above, the increments du, dN and df correspond to actual changes, and thus the corresponding changes in the state of deformation dε lead to changes dN in the internal state of stress. Equilibrium in the new state is expressed in terms of the principle of virtual work. The change in virtual work from the original equilibrium state to the neighboring state is found by diﬀerentiation of δV , and when the virtual displacements δu are kept constant, the resulting second variation is found to be

d(δV ) = δε dN + d(δε) N ds − δuT df = 0. (2.65) constitutive change

geometry change

Note that the load term has been obtained from d(δuT f ) = δuT df , because there is no change of the virtual nodal displacement, i.e. d(δu) = 0. It is intuitively clear that the same virtual translation δu can be used for both neighboring conﬁgurations. However, when the nodal displacements also contain ﬁnite rotations ϕ, as in the case of beams and shells, independence can no longer be assumed, and d(δϕ) = 0. This problem and its solution are considered in detail in Chapter 3 dealing with ﬁnite rotations. The ﬁrst part of the integrand in (2.65) contains the eﬀect of the change of the internal state of stress dN . The change of the state of stress is due to the change in the state of strain dε, and the constitutive relation will usually provide the incremental stiﬀness factor in the relation dN =

dN dε. dε

(2.66)

In the case of linear elasticity the stiﬀness of the bar is dN/dε = AE, but non-linear elasticity or plasticity could also provide the incremental stiﬀness needed in this format. The second part of the integrand represents the eﬀect of the change of geometry on the equilibrium of the internal state of stress. The normal force in the bar changes direction due to the incremental displacement du. Note that this eﬀect is not related to deformation as such, but is a consequence of the rotation of the internal stress state. The virtual strain increment δε is given by (2.15). In a geometrically non-linear theory the virtual strain

Non-linear bar elements

increment δε depends on the current geometry, and thus the factor ∂ε/∂u will generally be a function of the current node positions x. The incremental change is found by diﬀerentiation, d(δε) =

∂ ∂ε ∂2ε du. δu du = δuT T ∂u ∂u ∂u ∂u

(2.67)

Note that also here it is used that the virtual displacement δu remains unaﬀected by the change of state d( ). When (2.66) and (2.67) are substituted into the increment of the virtual work as given by (2.65), the following form is obtained:

∂2ε dN T d(δV ) = δε dε + δu N du ds − δuT df = 0. (2.68) dε ∂uT ∂u The ﬁrst term can be expressed in terms of the nodal displacement increments by use of (2.62) for δε and the similar formula for dε, whereby ∂2ε ∂ε dN ∂ε T +N ds du − df = 0. (2.69) δu ∂uT dε ∂u ∂uT ∂u The variation δu is arbitrary, and thus the scalar equation (2.69) is equivalent to the incremental stiﬀness relation ∂2ε ∂ε dN ∂ε +N ds du = df . (2.70) ∂uT dε ∂u ∂uT ∂u The square brackets contain the tangent stiﬀness K. It is seen that the constitutive contribution to the tangent stiﬀness depends on the ﬁrst derivatives of the strain with respect to the coordinates, while the initial stress contribution depends on the second derivative. Thus, the geometric eﬀect of rotating the internal stress state can only be represented by use of a non-linear strain measure, and this non-linearity leads to state-dependent factors in the constitutive term. Note that the form of the incremental stiﬀness in (2.70) leads to a symmetric matrix – also for non-linear material behavior. In the general continuum theory with more than one strain component at a point, symmetry depends on the symmetry properties of the constitutive matrix. This is dealt with in Chapter 7 on plasticity. For the elastic bar element the tangent stiﬀness relation was given in (2.30). The ﬁrst and second derivatives of the axial Green strain follow from the incremental strain relation (2.25), and it is easily veriﬁed that (2.30) follows as a special case of (2.70). In the special case of an elastic material, there exists a strain energy

2.8 Exercises

density function, which is an expression for the accumulated internal work,

ε ϕ(ε) = N dε. (2.71) 0

It follows from this deﬁnition that the internal stress N is the derivative of the strain energy density, dϕ(ε) N = . (2.72) dε For elastic bodies the virtual work can be considered as the ﬁrst variation of a potential energy function. Thus, the potential energy corresponding to (2.61) is

Φ(u) = ϕ(ε) ds − uT f = 0. (2.73) The ﬁrst variation of the potential energy gives the equilibrium equations in the form

dϕ ds − δuT f = 0. (2.74) δΦ(u) = δε dε The incremental stiﬀness relation follows by diﬀerentiation as above, leading to the equations dϕ ∂ 2 ε ∂ε d2 ϕ ∂ε ds du = df . (2.75) + ∂uT dεdε ∂u dε ∂uT ∂u This form of the incremental stiﬀness equations is symmetric, even for the general case of multi-dimensional stresses due to the existence of an energy potential ϕ(ε). The results of this chapter can be generalized in two directions: by introducing rotations at the nodes in addition to the translation displacements, leading to beams and shells, and by extending the notion of strain to two or three dimensions, leading to continuum models for solids. The ﬁrst route is followed in Chapters 3–5 dealing with ﬁnite rotations and beam elements. The second is taken up in Chapter 6 on the non-linear relations of solid mechanics and the following chapter on plasticity models for solids.

2.8 Exercises Exercise 2.1 Consider the bar element AB in Fig. 2.9. In the initial state B the end-points have the coordinates xA 0 and x0 . In the state n − 1 the A B coordinates are xn−1 and xn−1 . In the following state n the coordinates are A A B B B xA n = xn−1 + ∆un and xn = xn−1 + ∆un .

Non-linear bar elements

(a) Derive the mean position formula for the increment in Green strain used in the updated Lagrangian formulation in Algorithm 2.2: ∆εn = εn − εn−1 =

1 l02

1 2 (xn−1

+ xn )T ∆un .

Note that the common reference length is l0 . Exercise 2.2 The virtual work equation (2.16) was used to ﬁnd the precise deﬁnition of the internal force N corresponding to a given strain measure. (a) Find δεE corresponding to the engineering strain εE = (l − l0 )/l0 , i.e. the formula similar to (2.15). (b) Use the variation δεE and the initial length l0 in the integral of the virtual work equation (2.16) to ﬁnd fA and fB . (c) What is the physical meaning of the internal force N , when used together with engineering strain. Exercise 2.3 The general framework of geometrically non-linear bar elements outlined in Section 2.7 does not depend on the particular strain measure used in the theory. (a) Use the results of Section 2.6 to establish the equilibrium equations of the current state (2.64) and the incremental stiﬀness relation (2.70) in explicit matrix form, when formulated in terms of engineering strain εE . (b) Obtain the explicit matrix expressions when the theory is formulated in logarithmic strain εL (using current length measures ds and l). Exercise 2.4 Example 2.2 describes the assembly of the total internal forces and the global tangent stiﬀness matrix with speciﬁc reference to the two-bar truss treated in Example 2.1. In the notation introduced in Example 2.1 the current coordinates of the three nodes of the truss are A = (0, 0, −b),

B = (−c + u1 , u2 , u3 ),

C = (0, 0, b).

Consider the symmetric case, in which u3 = 0. The following computations can then be based on the element AB alone. (a) Find the total strain εG and the strain increment δεG in the element AB. (b) Form the current position vector x for the element AB, and ﬁnd the internal element forces qA and qB in terms of εG . (c) Determine the tangent stiﬀness matrix of the element AB, and ﬁnd the part of the global tangent stiﬀness matrix that enters into the equilibrium equations after elimination of the support conditions. (d) Explain how the spring shown in Fig. 2.6 enters into the stiﬀness matrix.

2.8 Exercises

Exercise 2.5* Set up a ﬁnite element model of the two-bar truss shown in Fig. 2.6. Subroutines from FemFiles can be used. The model has 2 bar elements, connecting 3 nodes, of which 2 are fully constrained. The horizontal linear spring is represented by a direct entry of the spring stiﬀness in the corresponding diagonal term. (a) Write a data script ﬁle truss01 for a non-linear ﬁnite element analysis of the truss structure. The ﬁle must deﬁne the following data arrays: - X = [x,y,z;. . .], containing node coordinates; - T = [n1,n2,matl;. . .], for topology and material identiﬁcation; - D = [A,E;. . . ], deﬁning material property table; - F = [n1,Fx,Fy,Fz;. . .], deﬁning the load increment vector; - C = [n1,dof,u0;. . .], deﬁning the boundary constraints. In addition, the ﬁle must provide some algorithmic parameters: number of load steps n max, max number of iterations i max, and tolerance of force residual unbalance TOL. (b) Write subroutines kbar(K,T,X,D,u) and gbar(g,T,X,D,u) for the global tangent stiﬀness matrix and global internal force vector, respectively. (c) Write a Newton–Raphson driver routine nrtruss that calls the data script ﬁle truss01 and makes use of the functions kbar(K,T,X,D,u) and gbar(g,T,X,D,u). (d) Use truss01 and nrtruss to study the equilibrium path of the truss from Fig. 2.6 for ideal symmetry and for a small inclination of the truss with vertical. Illustrate the results with plots of the displacement components u1 , u2 and the load f1 . Exercise 2.6* The shallow dome shown in Fig. 8.10 is made of elastic bars. The dimensions of the dome are speciﬁed by the reference length b1 = 1.0 and b2 = 2.0b1 , h1 = 0.25b1 and h2 = 1.3h1 . The dome is loaded by a concentrated downward force −f at nodes 3 and 4 and an upward force f at nodes 6 and 7. (a) Write a data script ﬁle dome01 for non-linear ﬁnite element analysis of the dome. The maximum load may be estimated around f /EA 10−3 , corresponding to downward displacement of node 3 around w3 /b1 0.1. The ﬁle must deﬁne the structure by the arrays listed in question (a) of Exercise 2.5. In addition, it should give data for a f –w plot. (b) Write a Newton–Raphson driver routine nrtruss that calls the data script ﬁle dome01 and makes use of the functions kbar(K,T,X,D,u) and gbar(g,T,X,D,u) from Exercise 2.5 for the global tangent stiﬀness matrix and global internal force vector.

Non-linear bar elements

(c) Use dome01 and nrtruss to calculate the deformation of the dome and make load–displacement plots for nodes 3 and 6. (d) Is the displacement ﬁeld anti-symmetric? Can the full load–displacement path be traced using the Newton–Raphson method with a set of vertical springs?

3 Finite rotations

Beams and shells can be considered as three-dimensional solids, in which two or one of the dimensions are small compared to the remaining dimensions. This has given rise to beam and shell models in which the representation of the transverse displacement components is simpliﬁed. These models are formulated in terms of the translation and rotation of a reference curve or surface in the case of beams and shells, respectively. As it will appear, there are fundamental diﬀerences between the representation of translations and rotations. It is important to account for the special characteristics of rotations in the formulation and analysis of kinematically non-linear theories for beams and shells, and this chapter provides a concise presentation of the special properties of rotations needed for the development of general nonlinear beam and shell theories. A detailed discussion of rotations and their various representations has been given by Argyris (1982) and by G´eradin and Cardona (2001). The translation of a point is described by a vector u, and the result of a sequence of translations ∆u1 , ∆u2 , . . . is simply the sum of the individual translation vectors. The result is independent of the order of the individual terms. Finite rotations cannot be described in this simple manner, as may be seen by considering two rotations of 90◦ about orthogonal axes, and comparing the result with that of the same rotations taken in the opposite order (Goldstein, 1980). The top sequence of Fig. 3.1 shows a box which is ﬁrst rotated 90◦ around the x1 -axis and then −90◦ around the x2 -axis. This leaves the box standing on its end. The bottom sequence shows the box when rotated in the opposite order, ﬁrst −90◦ around the x2 -axis and then 90◦ around the x1 -axis. This leaves the box lying on its side. The two results are distinctly diﬀerent, and there is a clear need for a representation of rotations that reﬂects their non-commuting character. The basic representation of ﬁnite rotations is considered in Sections 3.1 47

Finite rotations

Fig. 3.1. Rotation of box around x1 - and x2 -axes.

and 3.2. It is demonstrated that ﬁnite rotations can be expressed in the form of a rotation tensor, the components of which form a 3 × 3 matrix. The components of the rotation tensor can be expressed in terms of a unit vector, deﬁning the direction of the rotation axis, and the angle of rotation. This indicates that a rotation can be described in a vector-like format ϕ in which the direction of the vector deﬁnes the rotation axis, while the length of the vector deﬁnes the magnitude of the rotation. In structural models it is convenient to represent the rotations in this three-component pseudo-vector format. In structural theories involving rotations, small rotation increments occur in two contexts: as a virtual rotation δϕ used in the principle of virtual work δV , and as a change of rotation between neighboring points dϕ/ds, expressing curvature or deformation of the structure. The role of rotation increments in the principle of virtual work is dealt with in Section 3.3. In particular, the comparison of virtual work in neighboring states of equilibrium, required to establish the tangent stiﬀness, leads to the important result that the virtual rotation δϕ cannot be considered identical in the two states, and therefore an extra term appears in the tangent stiﬀness relation. The second role, in which the derivatives dϕ/ds are needed, is discussed in Section 3.4. In kinematically non-linear structural models rotations need repeatedly to be updated from computed rotation increments. It is therefore of interest to ﬁnd the most eﬃcient way to combine two rotations. This is accomplished by an extended pseudo-vector representation of the rotations, called quaternion parameters. The quaternion parameter representation of rotations and the associated product rule that adds the corresponding rotations are derived

3.1 The rotation tensor

in Section 3.5 from concepts of vector analysis. Finally, Section 3.6 gives a simple geometric derivation of a slightly diﬀerent pseudo-vector representation of rotations, and a direct transformation of this vector to quaternion parameters.

3.1 The rotation tensor The basic properties of ﬁnite rotations are important for the development of kinematically non-linear theories of beams and shells, where the rotation is used as an independent representation of displacement. Let a0 be a vector that is rotated into the position a as shown in Fig. 3.2(a). This can be described by a relation of the form a = R a0 ,

(3.1)

where R is the rotation tensor. In this chapter it is convenient to use the notation of matrix analysis. A vector is represented by its Cartesian components in column format. This implies that the scalar product of two vectors is written as aT b and the exterior product as abT .

(a)

(b)

Fig. 3.2. Rotation of vector a0 by the angle ϕ around the direction n.

The components of the rotation tensor constitute a 3×3 matrix. However, it is seen from the ﬁgure that a rotation can be speciﬁed by a direction n and an angle of rotation ϕ. It is therefore clear that not all the components of the rotation tensor R are independent. The relations between the components are found by using the fact that the length of any vector as well as the mutual orientation of any two vectors must be conserved by the rotation. Let a0 and b0 be two vectors. They are now rotated by the common rotation tensor R into the new vectors a and b. The scalar product of the vectors is

Finite rotations

left unchanged by the rotation, and therefore aT b = aT0 RT Rb0 = aT0 I b0 = aT0 b0 ,

(3.2)

where I is the unit tensor. Thus R must satisfy the relations RT R = R RT = I,

det(R) = +1.

(3.3)

The condition of positive determinant ensures that a right-hand vector triple retains its right-hand orientation after the rotation. These relations impose six constraints on the components of the tensor R, thus leaving three independent components to specify the magnitude and direction of the rotation. The relations (3.3) deﬁne the rotation tensor R as a proper orthogonal tensor. Figure 3.2(a) shows the rotation of a vector a0 by the angle ϕ around the direction n into the new position a. The rotation parameters may be combined into the pseudo-vector ϕ = ϕ n,

(3.4)

often called the rotation vector. The corresponding rotation tensor can be obtained directly in terms of n and ϕ by the following geometric argument, illustrated in Fig. 3.2. First the original vector is decomposed −−→ into its projection on the rotation axis OC = n (nT a0 ) and the remainder −→ −→ CA0 = [a0 − n(nT a0 )]. The vector CA0 is rotated by the angle ϕ in the −→ plane orthogonal to n. Hereby the component CA0 in the initial direction −−→ −→ −−→ is reduced to CB = cos ϕ CA0 , while a new component BA = sin ϕ (n × a0 ) is generated, using the vector product denoted by ×. This gives the rotated vector in the form a = n (nT a0 ) + cos ϕ [ a0 − n (nT a0 ) ] + sin ϕ (n × a0 ),

(3.5)

i.e. in terms of the vectors a0 and n, and trigonometric functions of the angle ϕ. This relation is often termed Rodrigues’ formula, although it was in fact derived earlier by Euler, see e.g. Cheng and Gupta (1989). The vector a0 can be extracted as a factor by introducing the notation ˆ a0 , n × a0 = n

(3.6)

ˆ is a skew-symmetric two-dimensional tensor with the component where n matrix

(ˆ n)ij = −εijk nk =

0 n3 −n2

−n3 0 n1

n2 −n1 0

(3.7)

The skew-symmetric component matrix is also indicated in compact form

3.1 The rotation tensor

by use of the summation convention and the permutation symbol εijk . The summation convention implies that a subscript that appears with the same symbol twice corresponds to a summation, and thus (3.7) is a sum of three terms corresponding to k = 1, 2, 3, respectively. The permutation symbol εijk equals +1 or −1 when ijk is an even or an odd permutation of the numbers 1, 2, 3, and equals zero if any index value occurs more than once. With the vector product notation, the rotated vector can be expressed as ˆ + (1 − cos ϕ) n nT a0 . a = cos ϕ I + sin ϕ n (3.8) In this expression the rotation tensor corresponds to the terms in the square brackets, and thus ˆ + (1 − cos ϕ) n nT . R = cos ϕ I + sin ϕ n

(3.9)

The component form can be written using Kronecker’s delta δij for the unit matrix and the permutation symbol εijk , Rij = cos ϕ δij − sin ϕ εijk nk + (1 − cos ϕ) ni nj .

(3.10)

It is noted that the ﬁrst and last terms are symmetric, while the middle term is skew-symmetric. In the relations (3.8) and (3.9) the rotation tensor is expressed in terms of the direction n and the angle of rotation ϕ. For a given rotation tensor the rotation angle and the direction can be extracted in the following way. Properties in vector and tensor analysis that do not change if the coordinate system is changed are called scalars or invariants. It is known from tensor analysis that the trace of a tensor, deﬁned as the sum of the diagonal terms, is invariant, see e.g. Malvern (1969) or Holzapfel (2000). It is clear from the deﬁnition of the angle of rotation ϕ that it is independent of any choice of coordinate system, while the components [n1 , n2 , n3 ] of the direction of the rotation axis refer to the chosen coordinate system. It is therefore to be expected that the angle of rotation can be extracted from the trace of the rotation tensor. This is conﬁrmed by direct calculation from (3.10), whereby the trace of the rotation tensor is tr(R) = Rii = 1 + 2 cos ϕ.

(3.11)

Once the rotation angle has been obtained from (3.11), the direction components can be extracted from the skew-symmetric part of Rij , giving the second term in (3.10). The principles used to extract the rotation angle ϕ and the components [n1 , n2 , n3 ] of the axis of rotation constitute a simple example of the use of invariance and symmetry. In Section 3.5 similar principles are used to derive a compact addition formula for rotations.

Finite rotations

Example 3.1. Some simple rotations. For given direction n and rotation angle ϕ the components of the rotation tensor R follow from (3.10). The rotation in Fig. 3.1 around the x1 -axis is given by 1 0 0 ϕ1 = 12 π R1 = 0 0 −1 , n1 = [1, 0, 0]T 0 1 0 while the negative rotation around the x2 -axis is given by 0 0 −1 ϕ2 = − 12 π . R2 = 0 1 0 n2 = [0, 1, 0]T 1 0 0 The rotation tensor R12 corresponding to ﬁrst rotating by R1 and then rotating by R2 , as illustrated in the top sequence of Fig. 3.1, follows from the product rule (3.1) as 0 0 1

R12 = R2 R1 =

−1 0 0 −1 0 0

It is noted that the ﬁrst rotation appears to the right in the product. The angle ϕ12 of the combined rotation is found from the trace of the combined rotation tensor by (3.11), tr(R12 ) = 1 + 2 cos(ϕ12 ) = 0

⇒

ϕ12 = ± 23 π.

The diﬀerent signs correspond to opposite choices of the positive direction of the axis of rotation. Here the positive angle ϕ12 = 23 π is chosen. The direction vector n12 of the rotation axis is extracted from the skew-symmetric part of the rotation tensor R12 , given by (3.9): 0 −1 −1 1 1 1 0 −1 , ˆ 12 = R12 − RT12 = √ n 2 sin ϕ12 3 1 1 0 √

where it has been used that sin ϕ12 = 23 . This identiﬁes the direction vector of the rotation axis of the combined rotation as n12 =

√1 [1, −1, 1]T . 3

When the order of the rotations is changed, as illustrated in the bottom sequence of Fig. 3.1, the rotation tensor components are R21 = R1 R2 =

0 0 −1 0 0 1

−1 0 0

3.2 Rotation of a vector into a speciﬁed direction

The rotation angle and the direction of the rotation axis are extracted by the process described above, yielding ϕ21 =

2 3 π,

n21 =

√1 [1, −1, −1]T . 3

3.2 Rotation of a vector into a speciﬁed direction The rotation tensor R has been expressed above in terms of the direction vector n of the rotation axis and the magnitude ϕ of the rotation. In applications the rotation may be speciﬁed as a rotation of a unit vector e0 into a unit vector e1 with rotation axis normal to the plane containing the two vectors, as illustrated in Fig. 3.3. This is the smallest rotation that accomplishes the mapping of e0 on e1 . In this case the rotation tensor can be expressed directly in terms of the two unit vectors e0 and e1 .

Fig. 3.3. Rotation of unit vector e0 into speciﬁed direction e1 .

First, the rotation tensor (3.9) is written in the form ˆ − (1 − cos ϕ)( I − n nT ) R = I + sin ϕ n

(3.12)

and the two last terms are then expressed in terms of the vectors e0 and e1 . The second term is rewritten by introducing the rotation axis via the vector product, n = e0 × e1 / sin ϕ. This gives, by the well-known double vector product formula, ˆ a = sin ϕ (n × a) = (e0 × e1 ) × a sin ϕ n = e1 (eT0 a) − e0 (eT1 a) = e1 eT0 − e0 eT1 a.

(3.13)

In the last term of (3.12) the tensor (I − n nT ) represents a projection on a plane normal to the direction n of the rotation axis. This plane contains the two vectors e0 and e1 . They can be combined into the two orthogonal vectors e0 +e1 and e0 −e1 . When these vectors are normalized the projection

Finite rotations

operator can be expressed as the sum of the projection on each of these vectors,

I − n nT =

(e0 − e1 )(e0 − e1 )T (e0 + e1 )(e0 + e1 )T . + (e0 + e1 )T (e0 + e1 ) (e0 − e1 )T (e0 − e1 )

(3.14)

Note the exterior vector products as numerators, and the scalar products as denominators. It is now used that e0 and e1 are unit vectors, and that cos ϕ = eT0 e1 , whereby |e0 ± e1 |2 = 2(1 ± cos ϕ). The rotation tensor (3.12) can then be expressed in the form R = I + 2 e1 eT0 − 2

(e0 + e1 )(e0 + e1 )T . |e0 + e1 |2

(3.15)

This constitutes the expression for the rotation tensor directly in terms of the two unit vectors e0 and e1 . It is easily veriﬁed that e1 = Re0 , as required.

(a)

(b)

Fig. 3.4. Unit vectors enj at node and eej in element. (a) Beam, (b) shell.

In the formulation of geometrically non-linear theories for beams, plates and shells, the nodes may be associated with a unit vector triple enj = [en1 , en2 , en3 ] as shown in Fig. 3.4. In the co-rotational formulation, discussed in Chapter 5, it is of interest to know the rotation of the vector triple enj relative to a coordinate system eej moving with the element. This involves rotating the triple enj such that one of the vectors, e.g. en1 , is brought into a direction ee1 , deﬁned by the element-based coordinate system. The rotation is deﬁned as a rotation of the vector en1 into the new vector ee1 about an axis orthogonal to the plane of these two vectors, i.e. by the smallest rotation of the vector en1 into ee1 . Thus, the vector triple is rotated into the new unit vector triple eej = [ee1 , ee2 , ee3 ]. The two rotated vectors ee2 and ee3 follow directly from the rotation formula (3.15). The two initial vectors en2 and en3 are orthogonal to the vector en1 , and therefore the second term in (3.15)

3.3 The increment of the rotation variation

does not contribute. Thus, the transformation formula for the original base vectors enj can be written in the simple symmetric form ¯e ¯T ) enj , eej = ( I − 2 e

j = 2, 3

with

¯ = e

en1 + ee1 . | en1 + ee1 |

(3.16)

It is seen that the transformation of the two vectors enj is described com¯. This transformation corresponds to a reﬂection pletely by the unit vector e ¯, see Exercise 3.2. This type of transin a plane normal to the unit vector e formation is used in numerical analysis under the name of a Householder transformation (Press et al., 1986). In the case of a co-rotating beam element, treated in Chapter 5, the unit vectors enα deﬁne the beam cross-section or shell surface and the rotation deﬁnes the bending of the element.

3.3 The increment of the rotation variation In the formulation of theories of structures with explicit representation of rotations, a need arises for comparing a state of rotation described by the rotation pseudo-vector ϕ = (ϕ1 , ϕ2 , ϕ3 ) with a neighboring state with rotation pseudo-vector ϕ + δϕ = (ϕ1 + δϕ1 , ϕ2 + δϕ2 , ϕ3 + δϕ3 ). According to Fig. 3.1 and the ﬁndings in Example 3.1, the components of the total rotation pseudo-vector will in general not be the algebraic sum of the components of the constituent rotations. Thus, the components (δϕ1 , δϕ2 , δϕ3 ) will in general not describe an incremental rotation. Here the notation ¯ = (δ ϕ¯1 , δ ϕ¯2 , δ ϕ¯3 ) will be used for the incremental rotation leading from δϕ the state ϕ to the state ϕ + δϕ. The relation between the three pseudo¯ is illustrated in Fig. 3.5. vectors ϕ, δϕ and δ ϕ The principle of virtual work is concerned with the change from the cur-

¯ Fig. 3.5. Total rotation ϕ + δϕ as rotation ϕ followed by incremental rotation δ ϕ.

Finite rotations

rent state to a virtual neighboring state. Thus, the principle of virtual work ¯ including rotations is formulated in terms of the incremental rotation δ ϕ. This has a profound inﬂuence on the tangent stiﬀness relations, where the ¯ has to be considered. The questions relating to the use of the change d(δ ϕ) ¯ in the principle of virtual work and its higher-order incremental rotation δ ϕ ¯ in stiﬀness relations are treated in this section, while the increment d(δ ϕ) ¯ and the corresponding general relation between an incremental rotation dϕ parameter increments dϕ, needed e.g. for describing curvature, is derived in the following section. Let a0 be a vector that is rotated into the new position a as shown in Fig. 3.2. The rotation is described by the rotation tensor R(ϕ), deﬁned as a function of the rotation pseudo-vector components ϕ. Now consider a small variation δa generated by an inﬁnitesimal rotation δR(ϕ) from the rotated position a. It follows from (3.1) and (3.2) that δa = δR a0 = δR RT a.

(3.17)

It follows from diﬀerentiation of (3.3a) that the tensor δRRT is skewsymmetric, δ(R RT ) = δR RT + R δRT = δR RT + (δR RT )T = 0.

(3.18)

¯ is An incremental rotation of the vector a by the inﬁnitesimal vector δ ϕ given by the vector product ¯ × a = δ ¯ a, δa = δ ϕ ϕ

(3.19)

where the symbol was introduced in (3.7) for the skew-symmetric matrix associated with a three-dimensional vector and representing a vector product. ¯ corresponding From this it follows that the skew-symmetric matrix δ ϕ ¯ is expressed in terms of the rotation to the incremental rotation vector δ ϕ tensor as ¯ = δ ¯ = δR RT . δ ϕ× ϕ (3.20) ¯ The explicit non-linear relation between the incremental rotation vector δ ϕ and the increment of the pseudo-vector components δϕ for an arbitrary value of the rotation parameter vector ϕ is derived in Section 3.4. However, many of the physical implications of the non-commuting property of rotations can be identiﬁed by considering the simpler case of small rotations around the state ϕ 0, and therefore this special case is considered independently ﬁrst. First, small rotations relative to a conﬁguration in which ϕ = 0 are con¯ for an sidered with the objective of studying the changes in the variation δ ϕ

3.3 The increment of the rotation variation

inﬁnitesimal change dϕ = [dϕ1 , dϕ2 , dϕ3 ]T in the pseudo-vector components. For this purpose it is suﬃcient to consider expansions of δR and RT including only constant and linear terms in the rotation pseudo-vector ϕ. While the general relation (3.9) between the rotation tensor R and the rotation pseudo-vector ϕ is rather indirect, because it relies entirely upon a split into the direction n and the magnitude of the rotation ϕ, the second-order expansion of R can be expressed directly in terms of the pseudo-vector ϕ as R = I + (ϕ×) − 12 (ϕT ϕ) I +

1 2

ϕ ϕT + O(ϕ3 ).

(3.21)

From this the following expansions are obtained: δR = (δϕ×) − (δϕT ϕ) I + 12 (δϕ ϕT + ϕ δϕT ) + O(ϕ2 )

(3.22)

and RT = I − (ϕ×) + O(ϕ2 ).

(3.23)

In these expansions only linear terms in ϕ have been retained. The expansion of the product δR RT , including linear terms in ϕ, is then δR RT

= (δϕ×) − δϕ × (ϕ×) − (δϕT ϕ) I + 12 (δϕ ϕT + ϕ δϕT ) + O(ϕ2 ).

(3.24)

Finally, this expansion is reformulated by rewriting the double vector product in terms of scalar products using the well-known vector relation

a × (b × c) = (aT c) b − (aT b) c = b cT − c bT a (3.25) and then reassembling the resulting scalar products into vector product form again. The result of this rearrangement of the terms is

(3.26) δR RT = δϕ − 12 δϕ × ϕ + O(ϕ2 ) × ¯ is related to the It then follows from (3.20) that the incremental rotation δ ϕ pseudo-vector ϕ and its increment δϕ by ¯ = δϕ − δϕ

1 2

δϕ × ϕ + O(ϕ2 ).

(3.27)

It follows from this result that in a conﬁguration with ϕ = 0 the two ﬁrstorder increments are equal, i.e. ¯ = δϕ δϕ

for ϕ = 0.

(3.28)

Thus, the inﬁnitesimal parameter increments δϕ = [δϕ1 , δϕ2 , δϕ3 ]T do rep¯ However, this resent the components of an inﬁnitesimal rotation vector δ ϕ. result is limited to the linear part of the generally non-linear relation between the rotation parameters and the incremental rotation.

Finite rotations

¯ describes the change of the incremental rotaThe second variation d(δ ϕ) ¯ when the rotation components are changed by dϕ and the variation tion δ ϕ, δϕ = [δϕ1 , δϕ2 , δϕ3 ]T of the components remains constant. It follows from diﬀerentiation of (3.27), where the linear term in ϕ gives rise to the second variation ¯ = − d(δ ϕ)

1 2

δϕ × dϕ for ϕ = 0.

(3.29)

In the second variation the increments may be replaced by their correspond¯ and dϕ ¯ ing ‘physical’ representations in terms of incremental rotations δ ϕ on account of the ‘tangency condition’ (3.28). Hereby any reference to the parameter representation in terms of ϕ disappears, and the second variation takes the form ¯ = − d(δ ϕ)

1 2

¯ × dϕ. ¯ δϕ

(3.30)

Note that the sign of the second variation is determined by the order of the ¯ is introduced, and then a change of operations: ﬁrst a rotation variation δ ϕ ¯ is considered. This eﬀect is purely three-dimensional; in conﬁguration dϕ plane problems the two rotation increments are co-linear, and their vector product vanishes. The implications of this new term are illustrated in the following two examples. Example 3.2. Changes in external moments. In ﬁnite element analysis the eﬀect of an external moment M is introduced via the contribution to external virtual work, ¯ T M. δVM = δ ϕ The contribution to the tangent stiﬀness relation is evaluated from the second variation, ¯ T M) = δ ϕ ¯ T dM + d(δ ϕ) ¯ T M. d(δVM ) = d(δ ϕ The ﬁrst term is the well-known contribution from the change of the moment, while the second term is the change in the virtual rotation because of the non-linearity of the rotation parametrization. When the second varia¯ is substituted from (3.30), and the factors in the resulting triple tion d(δ ϕ) product are rearranged, ¯ T dM − 12 (δ ϕ ¯ × dϕ) ¯ T M = δϕ ¯ T dM − 12 dϕ ¯ ×M . d(δVM ) = δ ϕ This relation can be written in the form ¯ T d∗ M, d(δVM ) = δ ϕ

3.3 The increment of the rotation variation

where dM∗ is the eﬀective change of the external moment, deﬁned as d∗ M = dM −

1 2

¯ × M. dϕ

Thus, an external moment for which d∗ M = 0 will appear as a constant moment in tangent stiﬀness relations. A similar conclusion was obtained by Christoﬀersen (1989) using the reciprocity relation for conservative systems. A case of particular interest is the motion of a beam element with an initial distribution of forces and moments. In this case the force and moment at the nodes of the beam element are rotated with the element. If the incremental ¯ the corresponding moment increment is dM = rotation of the element is dϕ, ¯ × M. However, the apparent change of the moment d∗ M as seen via the dϕ change in external virtual work is d∗ M = dM −

1 2

¯ ×M = dϕ

1 2

¯ × M, dϕ

i.e. half the physical change due to the rotation of the element. This is a crucial point in the development of geometric stiﬀness of beams and shells as discussed in Chapters 4 and 5.

Example 3.3. Semi-tangential moment. Figure 3.6(a) shows the end of a beam loaded by a moment M. The tangent of the beam is described by the unit vector n, and a rigid circular disk of radius a is mounted on the end of the beam, normal to n. The moment M = M n is generated via four long strings wound around the circular disk. Each string has the tension force P , thus producing the moment M = 4aP . When the forces are kept constant in magnitude and direction, the moment is conservative. ¯ of the end of the beam around the Now, consider a small rotation dϕ direction of one pair of forces as shown in Fig. 3.6(b). The contribution to

(a)

(b)

Fig. 3.6. Semi-tangential moment.

Finite rotations

the moment from the pair of forces parallel to the rotation vector is rotated with the disk, while the moment contribution from the other pair of forces remains unchanged. The total eﬀect is a change of the moment dM =

1 2

¯ × M. dϕ

¯ This type of Thus, the moment rotates with half the rotation increment dϕ. moment is called semi-tangential and was introduced by Ziegler (1977) in the context of follower forces and moments, and their inﬂuence on stability. Note ¯ that the expression for dM remains valid also if the rotation increment dϕ has a component along the tangent n, although the intuitive interpretation of the moment rotating through half the rotation increment becomes more indirect. For a semi-tangential moment the eﬀective moment increment, introduced in Example 3.2, is d∗ M = dM −

1 2

¯ × M = 0. dϕ

¯ of the point of application produces a change of the Thus, the rotation dϕ moment vector dM, but the non-linearity of the rotation parametrization compensates for this change in the incremental form of the principle of virtual work, making a semi-tangential moment appear as constant during the change of conﬁguration. Several authors, e.g. Argyris et al. (1979) and Yang and Kuo (1994), have made use of the concept of semi-tangential moments in the development of non-linear beam theories. However, the properties of semi-tangential moments rely on the constant direction of the forces generating the moment. In a solid body the stresses are convected with the material, and thus the essential problem of non-commutative rotations remains in the description of the stress components.

3.4 Parameter representation of an incremental rotation In Section 3.3 the implications of the non-linear parametrization of rotations were investigated for small rotations around the state ϕ = 0. In beam theory the rotations vary along the beam, and it is therefore necessary to obtain the ¯ and the analogous general relation between the rotation increment vector dϕ pseudo-vector rotation component increment dϕ. The relation plays a role in the parametrization of curved beams and shells, and has been derived by several authors, e.g. Simo (1985), Simo and Vu-Quoc (1986), Atluri and Cazzani (1995).

3.4 Parameter representation of an incremental rotation

In this section an explicit relation is obtained between the incremental ¯ and the increment of the rotation pseudo-vector components rotation dϕ dϕ from the basic relation (3.20), ¯ = d ¯ = dR RT . dϕ× ϕ

(3.31)

The rotation tensor was given in terms of the rotation pseudo-vector ϕ in (3.9) as R = cos ϕ I + sin ϕ (n×) + (1 − cos ϕ) n nT .

(3.32)

The dependence on the pseudo-vector ϕ is through its magnitude ϕ and the direction n, and it is therefore convenient ﬁrst to express the increment of R in terms of the increments dϕ and dn:

dR = − sin ϕ I + cos ϕ (n×) + sin ϕ n nT dϕ (3.33) + sin ϕ (dn×) + (1 − cos ϕ)(dn nT + n dnT ). The contribution to dRRT from the increment in magnitude dϕ is deduced using the relations ˆ = 0, nT (n×) = nT n

ˆn ˆ = ( n nT − I ). (n×)2 = n

(3.34)

This leads to the simple relation dRdϕ RT = dϕ (n×).

(3.35)

This implies that for a rotation increment dϕ proportional to the current ¯ is identical to the rotation pseudo-vector ϕ, the incremental rotation dϕ increment of the rotation parameter vector dϕ. This is the case in plane problems, where the rotation vector is always orthogonal to the plane, and thus this class of problems can be formulated without the distinction between the two types of rotation increments. In the general case also the direction changes. The contribution to dR RT from this increment can be reduced to the form

dRdn RT = sin ϕ cos ϕ (dn×) + (1 − cos ϕ) dn nT − n dnT

(3.36) + sin ϕ (1 − cos ϕ) (dn × n) nT − n (dn × n)T . In the last two terms the brackets contain exterior vector products of the form an − na with a = dn and a = dn × n, respectively. This form is equivalent to a double vector product as expressed by a reformulation of the vector product relation (3.25), (n × a)× = a nT − n aT .

(3.37)

Finite rotations

When this relation is used to reduce the last two terms, dRdn RT = sin ϕ cos ϕ (dn×) + (1 − cos ϕ) [ (n×dn)× ] + sin ϕ (1 − cos ϕ) [ (n×(dn × n))× ].

(3.38)

The vector n is a unit vector, and therefore dn is orthogonal to n. This in turn implies that n×(dn × n) = dn, whereby the last term is simpliﬁed. After combination of the ﬁrst and last term, the ﬁnal formula for the increment from the change of direction dn is dRdn RT = sin ϕ (dn×) + (1 − cos ϕ) [ (n×dn)× ].

(3.39)

Addition of the contributions from dϕ and dn then gives the total increment in the form dR RT = dϕ (n×) + sin ϕ (dn×) + (1 − cos ϕ) [ (n×dn)× ].

(3.40)

This expression consists of three terms, each corresponding to a vector prod¯ can therefore be identiﬁed directly uct. The incremental rotation vector dϕ from (3.31) as the sum of the corresponding three vectors, ¯ = n dϕ + sin ϕ dn + (1 − cos ϕ) (n×dn). dϕ

(3.41)

¯ along each of The three terms in this expression give the components of dϕ the three orthogonal directions n, dn, n×dn. The increments dϕ and dn are now expressed in terms of the parameter pseudo-vector dϕ. Diﬀerentiation of the deﬁning relations ϕ2 = ϕT ϕ and n = ϕ/ϕ gives dϕ = nT dϕ,

dn =

1 ( I − n nT ) dϕ. ϕ

(3.42)

¯ and the increThe ﬁnal relation between the incremental rotation vector dϕ ment dϕ of the rotation pseudo-vector then follows by substitution of these expressions into (3.41): sin ϕ 1 − cos ϕ ¯ = n nT + dϕ ( I − n nT ) + (n×) dϕ. (3.43) ϕ ϕ This relation is written as ¯ = T(ϕ) dϕ, dϕ

(3.44)

with the tensor T(ϕ) deﬁned by T(ϕ) = n nT +

sin ϕ 1 − cos ϕ ˆ. ( I − n nT ) + n ϕ ϕ

(3.45)

3.5 Quaternion parameter representation

Note that while the transformation tensor T(ϕ) is formed as the sum of three terms of the same structure as the rotation tensor R(ϕ), the scalar factors are diﬀerent, and T(ϕ) is not an orthogonal tensor. In the formulation of beam and shell theories, it is convenient to repre¯ directly, and to obtain the parameter sent the incremental rotation vector dϕ ¯ This requires the inverse of the transincrements dϕ subsequently from dϕ. formation tensor T(ϕ). The inverse T(ϕ)−1 is most easily obtained from an assumed vector representation in the form ˆ T(ϕ)−1 = α n nT + β ( I − n nT ) + γ n

(3.46)

similar to the form of T(ϕ), where each term represents an independent vector component in space. The parameters α, β, γ are determined from the fundamental relation T T−1 = I, giving the inverse as ˆ T(ϕ)−1 = n nT + 12 ϕ cot( 12 ϕ) ( I − n nT ) − 12 ϕ n

(3.47)

and thus conﬁrming the assumed form of the inverse. In some applications the tensor T(ϕ) or its inverse T(ϕ)−1 are needed for very small angles ϕ. In that case it may be preferable with respect to numerical accuracy to use a representation of the argument directly in terms of the rotation pseudo-vector ϕ to avoid extraction of the normalized direction n, see Exercise 3.3.

3.5 Quaternion parameter representation It is seen from the form (3.9) of the rotation tensor that it only depends on the rotation angle ϕ through trigonometric functions. This has led to several pseudo-vector formats in terms of trigonometric functions. Of these the quaternion parameter format deserves special attention, because it is non-singular for all angles, and it leads to a simple algebraic formula for the non-commutative addition of ﬁnite rotations. The concept of quaternions as four-dimensional vectors with a special product rule was originally introduced by Hamilton, and the addition formula developed by Cayley. A description of this theory has been given by e.g. Corben and Stehle (1974), and G´eradin and Cardona (2001) present a survey of diﬀerent three- and four-parameter representations of ﬁnite rotations. In this presentation the quaternion parameters will simply be used as a convenient means of representing the rotation tensor and the addition of rotations, and the necessary formulae will be developed directly in vector format, using techniques from vector analysis.

Finite rotations

3.5.1 Representation of the rotation tensor The quaternion parameter representation of a rotation ϕ about the axis n consists of a scalar r0 and a vector r, deﬁned by r0 = cos( 12 ϕ),

r = sin( 12 ϕ)n.

(3.48)

It turns out to be convenient to retain the four-parameter format, although the quaternion parameters satisfy the identity r02 + rT r = r02 + r12 + r22 + r32 = 1.

(3.49)

In terms of the four quaternion parameters r0 , r the rotation tensor (3.9) takes the hom*ogeneous quadratic form R = (r02 − rT r) I + 2 r0 ˆr + 2 r rT .

(3.50)

The corresponding component representation is Rij = (r02 − rk rk ) δij − 2εijk r0 rk + 2 ri rj or, in full matrix form, 

r02 + r12 − r22 − r32  R =  2(r2 r1 + r0 r3 ) 2(r3 r1 − r0 r2 )

2(r1 r2 − r0 r3 ) r02 − r12 + r22 − r32 2(r3 r2 + r0 r1 )

 2(r1 r3 + r0 r2 )  2(r2 r3 − r0 r1 )  . r02 − r12 − r22 + r32

(3.51)

(3.52)

The scalar quaternion parameter r02 is determined directly by the trace of the rotation matrix. From (3.52) and the normalization condition (3.49), tr(R) = Rii = 4 r02 − 1.

(3.53)

The quaternion vector components rl can be extracted from the skew-symmetric part of Rij . Multiplication of Rij in (3.51) by the permutation symbol εlij gives εlij Rij = −2εlij εijk r0 rk = −4 r0 rl ,

(3.54)

where the identity εlij εijk = 2δlk has been used. Example 3.4. Extraction of quaternion parameters. While the formulae (3.53) and (3.54) illustrate the principle behind extraction of the quaternion parameters from the rotation tensor components, they become singular at rotations of 180◦ , and lose numerical accuracy for angles around this value. A scheme that is numerically stable for all angles was proposed by Spurrier (1978) and given in simpliﬁed form by G´eradin and Cardona (2001). The current status is available in Markley (2008). In the extraction process it is necessary to divide by one of the quaternion components [r0 , r1 , r2 , r3 ],

3.5 Quaternion parameter representation

and it is numerically advantageous to select the largest. This is conveniently done by forming the quaternion component product matrix  2  r 0 r0 r1 r0 r2 r0 r3  r r   0 1 r12 r1 r2 r1 r3  S = 4   r0 r2 r2 r1 r22 r2 r3  r0 r3 r3 r1 r3 r2 r32 in terms of the rotation tensor components Rij from (3.52), 



1+R11 +R22 +R33 R32 −R23 R13 −R31 R21 −R12   R13 +R31 R32 −R32 1+R11 −R22 −R33 R12 +R21 . R=   R21 +R12 1−R11 +R22 −R33 R23 +R32 R13 −R31 1−R11 −R22 +R33 R21 −R12 R13 +R31 R23 +R32

The largest component ri is identiﬁed from the diagonal element as 4 ri2 = √ maxj Sjj , without summation of the repeated subscript. Thus, ri = 12 Sii , while the full set of components follows from the corresponding row of S as rj = 14 Sij /ri , j = 1, . . . , 4, again without summation.

3.5.2 Addition of two rotations Let two rotations be represented by the rotation tensors P and Q. If a vector a0 is ﬁrst rotated into a1 by Q and then into a2 by P, the total rotation is given by the product a2 = P a 1 = P Q a 0 .

(3.55)

Thus, the combined rotation matrix is R = PQ

(3.56)

Rij = Pik Qkj .

(3.57)

or, in component form,

It is desirable to have formulae for this matrix product directly in terms of the corresponding pseudo-vector representations, in order to compute all rotation changes directly in terms of the corresponding pseudo-vectors. The combination rule for rotations is particularly simple when expressed in terms of the quaternion parameters. Let the rotation tensors P and Q have the quaternion representations p0 , p and q0 , q, respectively. The objective then is to determine the quaternion parameters r0 , r of the combined rotation. The scalar quaternion parameter

Finite rotations

r0 is determined by the trace Rii as shown in (3.53). Substitution of the subscript form (3.51) of the quaternion representations of Pik and Qki into the trace of (3.57) gives Rii = [(2p20 − 1)δik + 2pi pk − 2εikl p0 pl ][(2q02 − 1)δki + 2qk qi − 2εkim q0 qm ]. (3.58) Products of symmetric and skew-symmetric terms cancel, and the result is therefore easily reduced to the form Rii = 4 (p0 q0 − pi qi )2 − 1

(3.59)

and thus, by (3.53), the scalar quaternion parameter of the combined rotation is r0 = p0 q0 − pi qi .

(3.60)

The vector part of the quaternion for the combined rotation follows from the skew-symmetric part of Rij by substitution of the product Pik Qkj into (3.54). After reduction of the permutation symbol products, the following result is obtained: εrij Rij = −4 (p0 q0 − pi qi )(p0 qr + q0 pr + εrij pi qj ).

(3.61)

This identiﬁes the vector part as rr = p0 qr + q0 pr + εrij pi qj .

(3.62)

In vector form the formulae for the combined rotation q followed by p are r0 = p0 q0 − pT q,

r = p0 q + q0 p + p × q.

(3.63)

The last term takes care of the eﬀect of diﬀerent directions of the two rotations p and q. If the direction is the same, (3.63) reduces to the trigonometric addition formulae for the two angles. Example 3.5. Addition of rotations by quaternions. This example demonstrates the representation and addition of the rotations shown in Fig. 3.1 by use of quaternions. The ﬁrst is a 90◦ rotation around the x1 -axis. The quaternion representation follows directly from the deﬁning √ 2 1 1 relations (3.48) with cos( 4 π) = sin( 4 π) = 2 , (1)

√

2 2 ,

r(1) = [

√

T 2 2 , 0, 0] .

The other is a 90◦ rotation around the negative x2 -axis, corresponding to the quaternion representation (2)

√

2 2 ,

r(2) = [0, −

√

T 2 2 , 0] .

3.5 Quaternion parameter representation

The quaternion of the combined rotation from ﬁrst rotating around the x1 axis and then around the x2 -axis follows from (3.63) as (12)

1 2,

r(12) = [ 12 , − 12 , 12 ]T .

The angle of rotation is then obtained from 1 2,

cos( 12 ϕ12 ) =

sin( 12 ϕ12 ) =

√

3 2

⇒

ϕ12 =

2 3π

and the direction vector follows from r(12) as n12 = √13 [1, −1, 1]T . Changing the order of the rotations just changes the sign of the vector product in (3.61), whereby (21)

1 2,

r(21) = [ 12 , − 12 , − 12 ]T .

The angle and rotation axis direction are then found to be ϕ21 = 23 π and n21 = √13 [1, −1, −1]T . As expected, the same results are obtained as in Example 3.1, but somewhat more directly.

3.5.3 Incremental rotation from quaternion parameters The problem treated in Section 3.4 of expressing an incremental rotation ¯ in terms of the increments of a suitable set of parameters describing an dϕ initial state of rotation can be solved in a compact form, when the rotation is represented in terms of quaternion parameters. The angular velocity ω ¯ normalized with the correspondis deﬁned as the incremental rotation dϕ ¯ ing time increment dt, i.e. by the relation ω = dϕ/dt, and the relation between the angular velocity and the time derivative of the quaternion parameters plays an important role in multibody dynamics, see e.g. G´eradin and Cardona (2001). Let a rotation state be described by the quaternion parameters (r0 , r). ¯ is applied, leading to a ﬁnal state From this state an inﬁnitesimal rotation dϕ described by the quaternion parameters (r0 + dr0 , r + dr). The problem is to ¯ in terms of the quaternion parameter express the incremental rotation dϕ increments (dr0 , dr). It follows from the quaternion parameter deﬁnition (3.48) that the quater¯ is (1, 12 dϕ) ¯ plus secondnion representation of an inﬁnitesimal rotation dϕ and higher-order terms. Use of the quaternion addition formulae (3.63) now gives the quaternion parameter representation of the ﬁnal rotation, ¯ T r, r0 + dr0 1 r0 − 12 dϕ

¯ + 12 dϕ×r. ¯ r + dr 1 r + r0 21 dϕ

(3.64)

Finite rotations

After cancellation of the initial rotation parameters this formula gives the quaternion parameter increments in terms of the original quaternion parameters and the incremental rotation vector, ¯ 2dr0 = −rT dϕ,

¯ − r × dϕ. ¯ 2dr = r0 dϕ

(3.65)

The inverse relation can be obtained by multiplication of the vector part by r×, yielding ¯ − r × (r × dϕ). ¯ 2r × dr = r0 r × dϕ

(3.66)

This formula is reduced by writing the double vector product in the form ¯ = −( |r|2 I − r rT ) dϕ ¯ r × (r × dϕ)

(3.67)

and substituting the product r×dϕ from (3.65b). The normalization relation (3.49) for the quaternion parameters implies that |r|2 = 1 − r02 , whereby the inverse relation is obtained from (3.66) in the form 1 ¯ 2 dϕ

= −r dr0 + r0 dr + r × dr.

(3.68)

¯ in terms This is the desired relation, giving the incremental rotation dϕ of the current quaternion parameters and their increments. It plays the same role for the quaternion representation as (3.44) plays for the rotation pseudo-vector representation. The relation also gives the angular velocity ω, when the quaternion parameter increment (dr0 , dr) is replaced by its time derivative (dr0 /dt, dr/dt). 3.5.4 Mean and diﬀerence of two rotations In the formulation of simple elements the rotation may be known at two states A and B, and the questions then arise, what is the mean rotation at some mean state C, and what is the rotation increment from A to B? In the case of displacements these questions are answered simply by the algebraic mean and diﬀerence of the corresponding displacement vector components. However, in the case of rotations that are not coaxial a slightly more elaborate formulation of the questions and answers is needed. It turns out to be convenient to pose the problem in terms of quaternion parameters. Let the rotation at A be described by the quaternion (p0A , pA ) and the rotation at B similarly by (p0B , pB ). The mean rotation, with quaternion representation (r0 , r), is now deﬁned such that the rotation from C to A is the reverse of the rotation from C to B. Let the rotation from C to B be represented by the quaternion (s0 , s). The reverse rotation from C to A is then represented by (s0 , −s), and the mean value property of C can be expressed by

3.6 Alternative representation of the rotation tensor

the quaternion addition formulae (3.63) as p0A = s0 r0 + sT r,

pA = s0 r − r0 s − s × r,

p0B = s0 r0 − sT r,

pB = s0 r + r0 s + s × r.

(3.69)

The explicit extraction of the mean rotation (r0 , r) and the increment from the mean to the original values (s0 , s) now proceeds as follows. The quaternion of the mean rotation can be expressed directly by the sum of the scalar and vector relations in the form s 0 r0 =

1 0 2 (pA

+ p0B ),

s0 r =

1 2 (pA

+ pB ),

(3.70)

where the scalar factor s0 is determined from the quaternion normalization relation (3.49), s20 =

1 0 4 (pA

+ p0B )2 + 14 |pA + pB |2 .

(3.71)

The vector part s is determined from the diﬀerence between the two sets of relations in (3.69), sT r =

1 0 2 (pA

− p0B ),

r0 s − r × s =

1 2 (pB

− pA ).

(3.72)

Pre-multiplication of the last of these equations by r× gives r0 r × s − r × (r × s) = r × 12 (pB − pA ).

(3.73)

In this equation r × s is substituted from (3.72b), the double vector product is expressed by scalar products, and r is substituted from (3.70b), r0 [ 21 (pA −pB )+r0 s ] + |r|2 s−r(rT s) =

1 1 4 (pB +pA )×(pB −pA ) s . 0

(3.74)

The ﬁnal result is obtained from this relation by introduction of |r|2 = 1−r02 , substitution of rT s from (3.72a), and multiplication by 2s0 , 2s0 s = p0A pB − p0B pA + pA × pB .

(3.75)

In the case of coaxial rotations at A and B the vector product vanishes, and the formula reduces to the trigonometric formula for the diﬀerence between the angles 12 ϕB and 12 ϕA . 3.6 Alternative representation of the rotation tensor An alternative pseudo-vector representation of the rotation tensor can be obtained by a simple geometric argument illustrated in Fig. 3.7. The ﬁgure shows the rotation of a vector a0 by the rotation pseudo-vector ϕ = ϕn into

Finite rotations

(a)

(b)

¯. Fig. 3.7. Vector increment ∆a in terms of mean vector a

its ﬁnal position a. This rotation can be characterized in simple terms by ¯ = 21 (a + a0 ). using the vector increment ∆a = a − a0 and the mean vector a It is seen from Fig. 3.7(a) that the vector increment ∆a is orthogonal to ¯ and the rotation pseudo-vector ϕ. It can therefore both the mean vector a ¯. The two be expressed as a scalar factor times the vector product n × a 1 angles at C in Fig. 3.7 are equal to 2 ϕ, and the geometry of the triangle CA0 A then implies that 1 2 (a

− a0 ) = tan( 12 ϕ) n × 12 (a + a0 ).

(3.76)

It is seen from this formulation that the rotation only appears in the form of the pseudo-vector, 1 r. (3.77) ψ = tan( 12 ϕ) n = r0 Thus, the rotation can be introduced via ψ instead of the pseudo-vector ϕ = ϕ n with length equal to the rotation angle ϕ, or the quaternion pseudovector r = sin( 12 ϕ)n with length sin( 12 ϕ). The pseudo-vector ψ leads to an algebraic representation of the rotation tensor in terms of three components, but has a singularity for ϕ = ±π, ±3π, . . . The pseudo-vector ψ, sometimes called the Rodrigues vector, has been used extensively by e.g. Argyris (1982), while here the rotation vector ϕ and the quaternion representation (r0 , r) are used as the basic variables. The relation (3.76) can be written in terms of the skew-symmetric matrix ψ = (ψ×) as ( I − ψ ) a = ( I + ψ ) a0 .

(3.78)

3.6 Alternative representation of the rotation tensor

The rotation tensor R can now be determined by solving this equation for a in terms of a0 . The solution has the form a = ( I − ψ )−1 ( I + ψ ) a0 ,

(3.79)

whereby the rotation tensor is identiﬁed as R = ( I − ψ )−1 ( I + ψ ).

(3.80)

This form, called the Cayley representation, gives the rotation tensor R as the product of the inverse of the tensor ( I − ψ ) with the tensor ( I + ψ ). This format can be used e.g. in the integration of the equations of rigid body dynamics, where the equations can be simpliﬁed by using the parameters ψ directly – see, for example, Simo and Wong (1991) and Krenk (2007b) for the basic algorithms. It has also been used with the vector ψ replaced by 12 ϕ as an approximation of the rotation tensor. The format gives an orthogonal tensor, irrespective of the speciﬁc choice of the vector denoted ψ. An explicit form of the rotation tensor can be obtained from (3.80) by introducing the identity tensor in the form ( I + ψ )−1 ( I + ψ ) between the factors. It then follows that 2

R = [(I + ψ) (I − ψ)]−1 ( I + ψ )2 = ( I − ψ )−1 ( I + ψ )2 ,

(3.81)

where the linear terms in the inverse have cancelled. It follows from the expansion of vector triple products (3.25) that ψ ψ a = ψ (ψ T a) − (ψ T ψ) a = ( ψψ T − ψ 2 I ) a.

(3.82)

When this is introduced into the ﬁrst factor of (3.81), the formula takes the form R = [ (1 + ψ 2 ) I − ψψ T ]−1 ( I + ψ )2 .

(3.83)

Now, the pseudo-vector ψ is expressed in terms of the quaternion parameters, ψ = r/r0 , and the scalar factor (1 + ψ 2 ) = cos−2 ( 12 ϕ) = r0−2 is moved to the second factor. The formula then takes the simpliﬁed form R = ( I − r rT )−1 ( r0 I + ˆr )2 .

(3.84)

The ﬁrst factor is evaluated by observing that it is the inverse of the identity tensor minus an exterior vector product. The inverse then has a similar format, and it is easily veriﬁed by direct multiplication that ( I − r rT )−1 = I +

1 r rT . r02

(3.85)

Finite rotations

Finally, substitution of this gives the rotation tensor in the explicit form R = ( r0 I + ˆr )2 + r rT = ( r02 − rT r ) I + 2r0 ˆr + 2 r rT .

(3.86)

This establishes the representation (3.50) of the rotation tensor in terms of quaternion parameters by an alternative procedure.

3.7 Summary of rotations and their virtual work Rotations occur in two diﬀerent contexts in the formulation of theories for structures and solids: as a ﬁnite rotation describing a conﬁguration at a particular instant, and as a ‘small’ change of the state of rotation typically encountered when setting up conditions of equilibrium, and in the calculation of the tangent stiﬀness relations of models described in terms of rotations. Rotations are principally diﬀerent from translations in the fact that they cannot be reduced to linear operations, and that they are non-commuting in three-dimensional space. Thus, there are two aspects of rotations in relation to mechanical theories for structures and solids. The ﬁrst is the representation of the rotation associated with a particular state, and the change of this state by a ﬁnite additional rotation. Due to the nonlinear character of motion associated with rotations, these operations are described by rotation tensors and the addition of consecutive rotations by tensor products. The basic formulae were derived in Section 3.1 in terms of the rotation pseudo-vector ϕ. An alternative representation in terms of a vector before and after rotation was presented in Section 3.2. This form is particularly relevant for determining the rotation of element nodes relative to element axes. The special character of rotations has important implications for the equilibrium equations and tangent stiﬀness of solid bodies described by use of rotations. The tangent stiﬀness is an expression for forces and moments arising from an incremental change of a state of equilibrium. When dealing with continuous bodies these relations are generally expressed via the principle of virtual work. This is very important when using rotations to describe the motion. The equilibrium equation will then contain virtual work ¯ is a virtual rotation and M is ¯ T M, where δ ϕ contributions of the form δ ϕ an internal or external moment. In the analysis of problems described in terms of translations the variation δu can be kept constant, when considering changes around the state of equilibrium. However, in the case of rotations ¯ a change of state also implies a change of the variation of the form d(δ ϕ). This ‘second variation’ was derived in Section 3.3 and the speciﬁc form given in (3.30). The implications were indicated brieﬂy in Examples 3.2 and 3.3.

3.8 Exercises

This led to the interesting conclusion that, when treated in a virtual work ¯ × M, correspondcontext, rotation of a moment will only contribute 12 dϕ ing to an apparent rotation of only half the actual incremental angle. This has important implications for theories involving rotations, and the consequences in terms of symmetry and geometric stiﬀness are demonstrated and discussed in the following chapters. Finally, a number of special issues have been discussed, notably the representation of rotations in terms of quaternion parameters presented in Section 3.5. These parameters describe a rotation in a hom*ogeneous four component format in terms of trigonometric functions of half the rotation angle. This hom*ogeneous format of order two enables a compact combination rule for rotations, expressed directly in the parameters in vector format, thereby avoiding the use of matrix multiplications.

3.8 Exercises Exercise 3.1 The ﬁnite rotation tensor formula (3.8) can also be derived from a vector diﬀerential equation. Let n be a constant unit vector, and let ϕ be the ﬁnite angle of rotation as shown in Fig. 3.2. For the ﬁxed direction n, the incremental rotation can be written as ˆa da = dϕ n × a = dϕ n or as the diﬀerential equation da ˆ a. = n dϕ The formal integral to this diﬀerential equation is a = enˆ ϕ a0 , where the exponential function of a matrix is deﬁned via its Taylor series expansion eA = I + A +

1 2 2! A

1 3 3! A

+ ···

(a) Demonstrate by substitution that the series expansion of the matrix exponential function is a solution of the diﬀerential equation. (b) Establish the vector product identity n × n × n × a = −n × a (Hint: n×a is orthogonal to n.)

ˆ3 a = − n ˆ a. n

Finite rotations

(c) Use the result from (b) to sum the even and odd power terms in the Taylor expansion of the solution to get the rotation tensor ˆ + (1 − cos ϕ) n ˆn ˆ. R = enˆ ϕ = I + sin ϕ n Exercise 3.2 Let nj be a set of orthonormal base vectors, and let nj be the orthonormal basis obtained by rotating the vector n1 into n1 . It was shown in Section 3.2 that the remaining two base vectors transform according to ¯ T ) nα , nα = (I − 2¯ nn

¯ = n

n1 + n1 . |n1 + n1 |

¯. This transformation is symmetric and deﬁned solely by the unit vector n (a) Find the inverse transformation, giving the vectors n2 and n3 in terms of the vectors n2 and n3 . (b) Show that the transformation corresponds to reﬂection of the vectors ¯ , and illustrate this by a sketch. n2 and n3 in a plane orthogonal to n (c) Show that the magnitude of the rotation angle of each of the vectors nj can be determined by sin( 12 ϕj ) =

1 2 (nj

¯ = nTj n ¯. − nj )T n

Exercise 3.3 The formula (3.9) for the rotation tensor R(ϕ), and (3.45) and (3.47) for the transformation tensor T(ϕ) and its inverse all contain potential numerical singularities for ϕ → 0. The problem is that in the limit the rotation direction vector n is not well deﬁned. Rewrite the expressions for R(ϕ), T(ϕ) and T−1 (ϕ) in terms of the rotation pseudo-vector ϕ and suitable scalar factors containing powers and trigonometric functions of ϕ or 12 ϕ with a well-deﬁned limit for ϕ → 0. Indicate a two- to three-term Taylor series expansion of the scalar factors for small ϕ. Exercise 3.4 Quaternions can be considered as four-component numbers in a similar way as complex numbers are two-component numbers. Like the complex numbers the addition rule is trivial, while the special properties are associated with the product rule. This exercise presents some of the basic properties of quaternions. (a) Let the quaternions associated with the rotation tensors P and Q be denoted (p0 , p) and (q0 , q), respectively. The product, here denoted by the asterisk ∗, is then deﬁned by the rules from (3.63), (p0 , p) ∗ (q0 , q) = ( p0 q0 − pT q , p0 q + q0 p + p × q ). (b) Ordinary vectors a and b are associated with quaternions with zero scalar part, (0, a) and (0, b). Find the corresponding quaternion product (0, a) ∗ (0, b).

3.8 Exercises

(c) The adjoint quaternion is deﬁned as (r0 , r) = (r0 , −r). Let the quaternion (r0 , r) correspond to the rotation tensor R. Show that the rotated vector a1 = R a can be evaluated by the two-sided quaternion product (0, a1 ) = (r0 , r) ∗ (0, a) ∗ (r0 , r) , where each factor is linear in the quaternion parameters r0 and r. (d) Explain how the rotation formula in (c) leads to the bi-linear product formula (3.63) for addition of rotations. Exercise 3.5* Let ϕ denote the rotation pseudo-vector ϕ = ϕn introduced in (3.4) and R the corresponding rotation tensor. (a) Use the quaternion addition formula (3.63) to write a Matlab function phi = addrot(phi1,phi2) giving the rotation pseudo-vector ϕ corresponding to the rotation ϕ1 followed by the rotation ϕ2 . (b) Write a Matlab function R = rotmat(phi) giving the rotation matrix R corresponding to the rotation ϕ. (c) Introduce the three rotations ϕ1 = [ 12 π, 0, 0]T , ϕ2 = [0, 12 π, 0]T and ϕ3 = [0, 0, 12 π]T . Obtain the rotation pseudo-vector and rotation matrix corresponding to the three cases: (i) ϕ1 followed by ϕ2 , (ii) ϕ2 followed by ϕ1 and (iii) ϕ1 followed by ϕ2 followed by ϕ3 . Exercise 3.6 Consider two coaxial rotations with pseudo-vectors ϕA = ϕA n and ϕB = ϕB n. Use the relations of Section 3.5.4 to ﬁnd the mean rotation and the rotation increment from the mean to B. Exercise 3.7 Consider the rotation tensors RA and RB associated with two points A and B, e.g. the end-points of a beam element. Let RC be the mean rotation tensor. This implies that there exists a rotation tensor S representing the rotation from A to C as well as from C to B. By the product rule of rotation tensors this implies that RC = S RA

and RB = S RC .

The rotations represented by the tensors RC and S were derived in Section 3.5.4. A more direct procedure for S is indicated here. (a) Eliminate RC to obtain an expression for S2 in terms of RA and RB . (b) Introduce the quaternion representation (s0 , s) for the rotation S, and ﬁnd the corresponding quaternion representation for S2 . (c) Use the results from (a) and (b) to establish the formula (3.75) for (s0 , s).

4 Finite rotation beam theory

Beams are slender structural elements, usually deﬁned by their axis and cross-sections. This permits two basically diﬀerent ways of modeling beams, one based on the translation and rotation of the points on the beam axis and the connected cross-sections (Fig. 4.1), the other treating the beam as a special example of a fully three-dimensional continuum. This chapter and the next develop fully non-linear beam models of the ﬁrst type. In this chapter the fully non-linear theory of a beam represented as a curve with elastic properties – a so-called elastica – is developed, and its ﬁnite element implementation discussed. This theory is in principle unique, and can therefore be called the theory of the elastica. It describes the deformation of an elastic space curve in terms of total translation and rotation, and relies heavily on the properties of ﬁnite rotations, presented in Chapter 3. An alternative formulation of non-linear beams is presented in Chapter 5. In that description the motion is decomposed into a local deformation described in a frame of reference following the beam, and a motion of the local frame of reference. The local frame of reference moves with each beam element, and this type of formulation is therefore often called co-rotating. Simple beam theories can be used within a co-rotating formulation, where ﬁnite rotation contributions are added. The application-oriented reader may want to go directly to the co-rotational formulation in Chapter 5.

Fig. 4.1. Displacement and rotation vectors in beam element.

4.1 Equilibrium equations

4.1 Equilibrium equations In this chapter a non-linear beam theory is developed by generalizing the virtual work approach of Chapter 2 to include rotations, using the results established in Chapter 3. The beam is considered as a curve segment x(s0 ), where each point is associated with a cross-section. The orientation of each cross-section is determined by two orthogonal unit vectors n2 , n3 . A third unit vector, normal to the cross-section, is deﬁned by n1 = n2 ×n3 . Thus, the triple nj = [n1 , n2 , n3 ] constitutes a local orthonormal basis, that rotates with the cross-section, see Fig. 4.2. This type of formulation, originally introduced by Dupuis (1969) and Reissner (1981), has been pursued more recently in connection with non-linear ﬁnite element analysis by Simo (1985), Simo and Vu-Quoc (1986, 1988), Cardona and G´eradin (1988), Lo (1992), Mathiesen (1993), Ibrahimbegovi´c (1995) and G´eradin and Cardona (2001).

Fig. 4.2. Beam representation as a curve with orthogonal directors.

The theory will here be developed for an elastica – i.e. a ﬂexible curve – described in terms of an initial arc-length parameter s0 . The curve is loaded by distributed forces p(s0 ) and moments m(s0 ) per unit initial length s0 . The section force and moment are N(s0 ) and M(s0 ), respectively. Consideration of a small length of the beam in its current, deformed, conﬁguration gives the force equilibrium equation dN +p = 0 ds0

(4.1)

and the moment equilibrium equation dM dx + × N + m = 0. ds0 ds0

(4.2)

Finite rotation beam theory

Note that dx/ds0 is a tangent vector, of unit length in the initial but usually not in the deformed geometry.

4.2 Virtual work, strain and curvature The equation of virtual work is now formed by introducing suitable dis¯ 0 ). Here, δ ϕ(s ¯ 0 ) is placement and rotation variation ﬁelds δu(s0 ) and δ ϕ(s a small incremental rotation as discussed in Section 3.3, and

l0 dx T dM T dN ¯ δu + p + δϕ + ×N + m ds0 = 0. (4.3) ds0 ds0 ds0 0 The equation is reformulated via integration by parts, whereby

l0 d(δu) dx T d T ¯× ¯ M ds0 − δϕ N+ (δ ϕ) ds0 ds0 ds0 0

l0 l0 T T ¯ M ¯ T m ds0 = 0. δuT p + δ ϕ − δu N + δ ϕ − 0

(4.4)

This is a statement of the principle of virtual work in the form that the internal virtual work, expressed in terms of internal forces and their conjugate deformation measures, minus the external virtual work equals zero for ¯ 0 ). arbitrary admissible variations δu(s0 ), δ ϕ(s The constitutive relations of the beam provide the components of the section force and moment in the local director basis, i.e. the components N = Nj nj ,

M = Mj nj .

The virtual work equation (4.4) then takes the form

l0 l0 ¯TM δεj Nj + δκj Mj ds0 − δuT N + δ ϕ δV = 0 0

l0 ¯ T m ds0 = 0. − δuT p + δ ϕ

(4.5)

(4.6)

In this equation δεj and δκj are the variational strain and curvature components in the local director basis, deﬁned via the virtual work equation. The use of local components in the representation of the internal work leads to a natural separation of each term into a factor representing a kinematic variation and a factor determined from the constitutive relations. In the further derivation of the incremental stiﬀness relation these factors generate the geometric and constitutive stiﬀness contributions, respectively. In the treatment of the bar element in Chapter 2, the strain deﬁnition was introduced a priori and the conjugate internal force was then determined

4.2 Virtual work, strain and curvature

from the virtual work equation. In the case of the elastica beam theory the internal forces are introduced via the equilibrium equation, and the virtual work equation then determines the appropriate virtual strain components. Thus, in general appropriate generalized stresses and their conjugate generalized virtual strains are determined by the virtual work equation, when either one of the two sets is given. The virtual strain components follow from (4.4)–(4.6) as d(δu) dx T ¯× δεj = − δϕ nj . (4.7) ds0 ds0 It is easily veriﬁed that this is the variation of the total strain components T dx εj = − n1 nj (4.8) ds0 when use is made of the fact that δ(nT1 nj ) = 0, because the directors remain orthonormal. The corresponding vector form of the total strain is dx ε = (4.9) − n1 . ds0 This deﬁnition of the total strain components leads to representation of the tangent vector as dx = n1 + εj nj = n1 + ε. (4.10) ds0 The local components [1+ε1 , ε2 , ε3 ] of the tangent vector dx/ds0 are illustrated in Fig. 4.3.

Fig. 4.3. Local components of tangent vector in terms of total strain.

The components δκj of the variation of curvature components also follow from (4.4)–(4.6) as δκj =

¯ T d(δ ϕ) nj . ds0

(4.11)

The variations δκj correspond to total curvature components κj deﬁned by

Finite rotation beam theory

the derivatives of the rotation vector along the reference curve of the beam, κj =

¯T dϕ nj . ds0

(4.12)

The fact that δκj are the variations of κj is demonstrated by diﬀerentiation of (4.12), dϕ ¯ T ¯T dϕ δκj = δ nj + δnj . (4.13) ds0 ds0 ¯ is The order of the operators δ and d in the second-order increment δ(dϕ) now interchanged by use of (3.30), and the variation of the unit vector δnj is expressed as a vector product, ¯ T ¯ T ¯T d(δ ϕ) dϕ dϕ ¯× ¯ × nj ). δκj = nj + δ ϕ nj + (δ ϕ (4.14) ds0 ds0 ds0 The factors in the triple vector products in the two last terms are identical, but occur in opposite order. These terms therefore cancel, demonstrating that δκj deﬁned by (4.11) is indeed the variation of the total curvature components κj deﬁned by (4.12). The vector form of the curvature corresponding to (4.12) is ¯ dϕ dϕ κ = = T(ϕ) , (4.15) ds0 ds0 ¯ where the tensor T(ϕ) given in (3.45) relates the incremental rotation dϕ to the increment of the rotation parameters dϕ. Remark: It should be noted that δεj and δκj are the variations of the local components εj and κj of strain and curvature. The variations do not include contributions from rotation of the directors, and thus they are not the components of the variations of the corresponding vectors ε and κ. The vectorial form of the variations can be expressed as δε = δεj nj = R (δεj RT nj ) = R δ(RT ε), δκ = δκj nj = R (δκj RT nj ) = R δ(RT κ), where the role of RT is to rotate from the local basis nj back to the ﬁxed global basis ej = RT nj . When the component variation has been taken, the result is then rotated back into the local basis nj by R. These operations are termed pull-back and push-forward by Marsden and Hughes (1983). However, the correct component form arises naturally from the principle of virtual work, and the present remark merely serves to identify the physical nature of δεj and δκj and to explain a terminology sometimes used in the computational mechanics literature.

4.3 Increment of the virtual work equation

4.3 Increment of the virtual work equation The virtual work equation (4.6) with the deﬁnitions (4.7) and (4.11) of the virtual strain and curvature components permits a check of equilibrium by evaluation of residual forces and moments. In order to prescribe a desired change of the conﬁguration of the beam, an incremental change of conﬁguration must be introduced in the virtual work equation as discussed in Section 2.7. The change of conﬁguration is described by the incremental ¯ the changes dNj and dMj in interdisplacements and rotations du and dϕ, nal forces and moments, and the changes dp, dm and dN, dM in distributed and end-point forces and moments. In the derivation of the incremental ¯ is in the relations it is important to note that the rotation increment dϕ form of a small rotation and is not generally equal to the increment dϕ of the components in the rotation pseudo-vector ϕ. The change of the internal virtual work is

l0 d(δVint ) = d δεj Nj + δκj Mj ds0 0 (4.16)

l0 δεj dNj + δκj dMj + d(δεj )Nj + d(δκj )Mj ds0 . = 0

The ﬁrst two terms of the integrand come from constitutive changes, while the two last represent the eﬀect of the change of geometry on the current stress state. For simplicity of notation it is assumed that the constitutive relations for the section forces Nj are uncoupled from the constitutive relations for the moments Mj . More general formulations may include the coupling between torsion and extension found in twisted beams, e.g. Krenk (1983a,b). By the assumed independence, the incremental constitutive relations are of the form ∂Nj ∂Mj dNj = dεk , dMj = dκk . (4.17) ∂εk ∂κk Substitution into the incremental form (4.16) of the internal virtual work then gives d(δVint ) =

l0 ∂Nj ∂Mj δεj dεk + δκj dκk + d(δεj )Nj + d(δκj )Mj ds0 . ∂εk ∂κk 0

(4.18)

For symmetric constitutive relations (4.17) the ﬁrst two terms are seen to be symmetric in the virtual and actual increments (δεj , δκj ) and (dεk , dκk ). In particular, symmetric stiﬀness terms would follow from an internal energy density function.

Finite rotation beam theory

4.3.1 Constitutive stiﬀness The incremental constitutive relations are given in terms of the local tensor components ∂Mj ∂Nj κ ε Cjk , Cjk = , (4.19) = ∂εk ∂κk ε and C κ refer to the local base vectors n followwhere the components Cjk j jk ing the beam. A typical example is the linear elastic relation expressed in principal axes, ε Cjk =

GA2 GA3

κ Cjk =

EI2

(4.20)

EI3

If the displacement and rotation ﬁelds are represented by global components, the constitutive tensors must be introduced via their global form ε n ε T Cε∗ = nj Cjk k = RC R ,

(4.21) κ n κ T Cκ∗ = nj Cjk k = RC R ,

where the rotation tensor R represents the current orientation of the axes, and the subscript ∗ indicates the global form of the constitutive parameters. The global vector form of the virtual strain increment follows from (4.7) as δε =

dx d(δu) dx d(δu) ¯× ¯ − δϕ = + × δ ϕ. ds0 ds0 ds0 ds0

(4.22)

Similarly, the virtual curvature increment vector follows from (4.11) as δκ =

¯ d(δ ϕ) . ds0

(4.23)

With these expressions and the notation ( ) = d( )/ds0 the constitutive contribution to the incremental stiﬀness can be expressed in the following block matrix format: ε dε + δκ C κ dκ = δεj Cjk j jk k k    ˆ 0 Cε∗ x du Cε∗    ¯ δϕ ¯ ] 0 ¯  , [ δu δ ϕ Cκ∗ 0   dϕ ˆ T Cε∗ x ¯ ˆ ˆ T Cε∗ 0 x dϕ x

(4.24)

ˆ = (x×) from (3.7) has been used for the skewwhere the notation x symmetric matrix equivalent of a vector. The constitutive stiﬀness contribution to a non-linear beam element follows by substitution of a suitable ¯ , δϕ ¯ and the increments du , dϕ ¯ , dϕ ¯ representation of the variations δu , δ ϕ in terms of shape functions, as discussed in Section 4.4.

4.3 Increment of the virtual work equation

4.3.2 Geometric stiﬀness The geometric stiﬀness is represented by the last two terms of the integrand in (4.18). The second variations d(δκj ) and d(δεj ) are now evaluated using the results of Section 3.3 on second-order rotation increments. It follows directly from the deﬁnition (4.11) of the curvature component variations that ¯ T ¯ T d(dδ ϕ) d(δ ϕ) ¯ × nj ). d(δκj ) = nj + (dϕ (4.25) ds0 ds0 As explained in Section 3.3, the parameter variation δϕ is kept constant when changing the conﬁguration. The non-linear form of the rotation para¯ must have the secondmetrization then implies that the virtual rotation δ ϕ order variation given by (3.30). Substitution of the second-order variation into the ﬁrst term of (4.25) gives d(δκj ) =

T ¯ T d(δ ϕ) d 1 ¯ × dϕ ¯ nj + ¯ × nj ). − 2 δϕ (dϕ ds0 ds0

(4.26)

Diﬀerentiation of the ﬁrst factor in the ﬁrst term leads to cancellation of half of the last term, leaving ¯ T ¯ T d(dϕ) d(δ ϕ) ¯ × nj ). ¯× d(δκj ) = − 12 δ ϕ nj + 12 (dϕ (4.27) ds0 ds0 Upon rearrangement of the order of the triple products this result can be written in the tensor format ¯ d(δκj ) = δ ϕ

d(dϕ) ¯ ¯ T 1 d(δ ϕ) ¯ n × − n × dϕ. j j 2 2 ds0 ds0

(4.28)

The tensor in the parentheses is skew-symmetric, and the second variation ¯ and dϕ. ¯ d(δκj ) therefore is a symmetric function of the increments δ ϕ The contribution d(δκj )Mj to the internal virtual work can be given in tensor form, by observing that the internal moment vector is M = Mj nj , and it then follows from (4.28) that ¯T d(δκj )Mj = δ ϕ

d(dϕ) ¯ ¯ T 1 d(δ ϕ) ¯ − M× dϕ. 2 ds0 ds0

2 M×

(4.29)

The invariant form of (4.29) in terms of tensors makes it valid for computations in global, local or element system components, provided the differentiation of the base vectors is included. The component format of the skew-symmetric tensor was given in (3.6). The ﬁrst step in the evaluation of the second variation of the strain components follows from (4.7) by taking diﬀerentials of all factors, and recalling

Finite rotation beam theory

that the variation of the displacement is selected such that d(δu) vanishes, d(δu) T d(du) T d(δεj ) = (dϕ ¯ × nj ) − δ ϕ ¯× nj ds0 ds0 dx T dx T ¯ × nj ). ¯× ¯ × (dϕ nj − δ ϕ − d(δ ϕ) ds0 ds0

(4.30)

The ﬁrst two terms are rearranged into a form similar to (4.28), and the ¯ is substituted from (3.30), whereby second variation d(δ ϕ) d(du) d(δu) T ¯ − ( nj ×) dϕ ds0 ds0 dx T dx T ¯ × dϕ) ¯ × ¯× ¯ × nj ). + 12 (δ ϕ nj − δ ϕ (dϕ ds0 ds0 ¯ T ( nj ×) d(δεj ) = δ ϕ

(4.31)

The last two terms are reduced by use of the vector product identities (a × b) · (c × d) = a · [b × (c × d)] = (a · c)(b · d) − (a · d)(b · c). (4.32) One of the terms with factor the fully symmetric form

1 2

cancels half of a similar term, resulting in

d(du) d(δu) T ¯ − (nj ×) dϕ ds0 ds0 dx T dx T dx ¯ − (δ ϕ ¯ T dϕ) ¯ nTj ¯ T 12 nj , + 12 nj dϕ + δϕ ds0 ds0 ds0

¯ T (nj ×) d(δεj ) = δ ϕ

(4.33)

where the terms with factor 12 are exterior products. Also in this case the introduction of the vector form of the internal force N = Nj nj reduces the result to the convenient tensor format d(du) d(δu) T ¯ − (N×) dϕ ds0 ds0 dx 1 dx T T ¯ + 2 N − (N ) I dϕ. ds0 ds0

¯ T (N×) d(δεj ) Nj = δ ϕ ¯T + δϕ

1 2N

dx T ds0

(4.34)

The format of this contribution is similar to (4.29) from the moments. The initial stress contributions (4.34) and (4.29) from section force and

4.3 Increment of the virtual work equation

moment can be combined in block matrix format, where the notation ( ) = d( )/ds0 is used, d(δεj ) Nj + d(δκj ) Mj =  0 0   0

T 0 ¯ T δ ϕ ¯T  δu δ ϕ  ˆ ˆ 1M N 2



   du   2 ¯  .   dϕ 1 T T  2 (N x + x N ) ¯ dϕ T −(N x ) I ˆT N 1 ˆ T M

(4.35)

This formula provides the initial stress, or geometric stiﬀness, contribution ¯ , δϕ ¯ and the increments to a beam element, when the variations δu , δ ϕ ¯ , dϕ ¯ are represented by suitable shape functions. In particular, if du , dϕ identical shape functions are used for each of the three vector components, the formula enables an eﬃcient global implementation of a general element for large deformation of curved beams as discussed in the following section.

Fig. 4.4. Incremental rotation of force and moment at rotating node.

There is an interesting diﬀerence between the eﬀect of rotating the internal force N and rotating the internal moment M. While rotation of the internal ¯ × N corresponding to the classic force gives a contribution of the form dϕ result for inﬁnitesimal rotation of a vector, the similar contribution from ¯ × M, i.e. reduced to precisely half. The reason for the moment is 12 dϕ the diﬀerence lies in a reduction introduced via the second variation of the rotation d(δϕ) as discussed in Examples 3.2 and 3.3. In connection with convected internal forces in structures, it appears as if the moment is only rotated by half the angle as illustrated in Fig. 4.4. Similar eﬀects may occur in relation to applied moment load, and these should be included via their contribution to the virtual work. 4.3.3 The load increments The external virtual work serves to deﬁne the external loads. In the case of the ‘elastica’ beam theory the external virtual work is the last term in (4.6):

l0 ¯ T m ds0 , δVext = δuT p + δ ϕ (4.36) 0

Finite rotation beam theory

where p(s0 ) and m(s0 ) are the distributed force and moment, respectively, per unit original length measured by s0 . A discretized representation of ¯ 0 ) will provide the the virtual translation δu(s0 ) and virtual rotation δ ϕ(s equivalent concentrated loads via the integral δVext . A change of state, e.g. by change of the loads, may also lead to change in ¯ the external virtual work. In this process δu remains unchanged, while δ ϕ changes as described in detail in Section 3.3. Thus the change in δVext is

l0 T T ¯ T ( 21 m×) dϕ ¯ ds0 . (4.37) ¯ dm ds0 + d(δVext ) = δϕ δu dp + δ ϕ 0

The ﬁrst integral represents the usual change in the external loads, while the second integral is a byproduct of the use of the incremental rotation δ ϕ¯ in the virtual work as demonstrated in Example 3.2. This second term is linear in the rotation increment dϕ¯ and therefore has the character of a skewsymmetric contribution to the stiﬀness matrix. As illustrated in Example 3.3 a conservative moment may include a dependence of the moment on dϕ, ¯ whereby this term is cancelled.

4.4 Finite element implementation The beam theory can be implemented directly in global form as proposed by Ibrahimbegovi´c (1995). The idea is to represent the beam by a space curve interpolating a given set of nodes, and then describe the displacement and rotation along the beam in terms of global vector components, referring to a ﬁxed Cartesian coordinate system, common to all the elements in the analysis. This approach represents a logical extension of the use of shape functions for bar elements illustrated in Section 2.5, but is quite diﬀerent from the traditional element formulation of beams, in which the properties of the beam element are obtained in a local coordinate system associated with the beam. The present beam theory is formulated in terms of displacements u(s0 ) and rotations ϕ(s0 ), and implementation into the ﬁnite element format implies some kind of representation in terms of given interpolation or shape functions. In the following the procedure is illustrated for the element with linear interpolation of translation and rotation increments, but the procedure is fairly easily generalized to higher-order interpolation. First the stiﬀness matrix is derived and then the evaluation of the internal forces and moments is discussed. When the translations and rotations are represented by shape functions, it is important to realize that the total strain and the strain increments are formulated in terms of a diﬀerence between the in-

4.4 Finite element implementation

cremental rotation vector and the derivative of the incremental translation vector. If translations and rotations are represented by polynomials of the same degree, this may lead to spurious strains in some deformation modes. This important problem, called locking, is discussed in Section 4.4.3.

4.4.1 Element stiﬀness matrix The fully non-linear beam theory derived in the previous sections can be implemented in terms of two-node elements with linear interpolation. The shape functions of an element with nodes A and B and initial length l0 are hA (s0 ) = 1 − s0 /l0 ,

hB (s0 ) = s0 /l0 .

(4.38)

The virtual and actual increments are represented in terms of these shape functions in the form du(s0 ) = hA (s0 ) duA + hB (s0 ) duB , ¯ 0 ) = hA (s0 ) dϕ ¯ B, ¯ A + hB (s0 ) dϕ dϕ(s

(4.39)

where subscripts A and B identify the corresponding end-point value of the translation and rotation vector increments. Similar representations are used ¯ 0 ). for the virtual increments δu(s0 ) and δ ϕ(s In order to facilitate the use of these representations it is convenient to introduce the block matrix format   duA du(s0 ) hA (s0 ) I 0 hB (s0 ) I 0 ¯   dϕ =  du A . (4.40) B ¯ 0) dϕ(s 0 hA (s0 ) I 0 hB (s0 ) I ¯B dϕ Following the notation introduced in Chapter 2, the components of all the displacements associated with nodes of the element are denoted u ˜ . For the present element these components are found on the right-hand side of (4.40), and thus ¯ TA , duTB , dϕ ¯ TB ]. d˜ uT = [ duTA , dϕ

(4.41)

In this notation the representation of the displacement increments takes the form du(s0 ) u (4.42) = H(s0 ) d˜ ¯ 0) dϕ(s with the shape functions contained in the matrix H(s0 ). It follows from the expressions (4.24) and (4.35) for the increment of the virtual work that the stiﬀness matrix requires the increments du (s0 ),

Finite rotation beam theory

¯ (s0 ) and dϕ(s ¯ 0 ). The representation of these in terms of the nodal values dϕ follows from diﬀerentiation of (4.40) as   du    A 0 hB (s0 ) I 0 du (s0 ) hA (s0 ) I ¯  d ϕ A . (4.43)  dϕ ¯ (s0 )  =  0 hA (s0 ) I 0 hB (s0 ) I  duB  ¯ 0) 0 hA (s0 ) I dϕ(s 0 hB (s0 ) I ¯ dϕ B

In the compact notation corresponding to (4.42), this relation is   du (s0 )  dϕ ¯ (s0 )  = G(s0 ) d˜ u ¯ 0) dϕ(s

(4.44)

where the matrix G(s0 ) is deﬁned by (4.43). The element tangent stiﬀness matrix K can now be expressed in a convenient matrix format. As seen from (4.18), the element stiﬀness consists of two contributions, one due to the constitutive changes of the internal forces in the beam and another due to the convection of the internal forces with the beam, the so-called geometric stiﬀness. The constitutive part of the element stiﬀness matrix is denoted Kc . It is determined by substitution of the representation (4.44) of the virtual and actual increments into the constitutive part of the virtual work increment, deﬁned by the ﬁrst two terms in (4.18),

l0 T ε κ δ˜ u Kc d˜ u = dεk + δκj Cjk dκk ds0 . (4.45) δεj Cjk 0

When using the expression (4.24) for the integrand, the constitutive part of the element stiﬀness matrix takes the form

l0 Kc = GT Dc G ds0 , (4.46) 0

where the matrix Dc follows from (4.24) as   ˆ 0 Cε∗ x Cε∗   Dc =  0 Cκ∗ 0 . T ε T ε ˆ C∗ 0 x ˆ C∗ x ˆ x

(4.47)

Similarly, the geometric part of the element stiﬀness matrix Kg is determined by the last two terms of the virtual work increment (4.18),

l0 δ˜ uT Kg d˜ u = (4.48) d(δεj ) Nj + d(δκj ) Mj ds0 . 0

4.4 Finite element implementation

This leads to the geometric element stiﬀness matrix

l0 GT Dg G ds0 , Kg =

(4.49)

where the matrix Dg follows from (4.35) as  ˆT 0 0 N  1 ˆ T 0  0 2M Dg =  1 T T  2 (N x + x N ) ˆ ˆ 1M N 2 −(NT x ) I

   . 

(4.50)

It is seen that the expressions for the constitutive and geometric stiﬀness matrices Kc and Kg are similar. This is because the incremental strain is not used explicitly in the formulation of the constitutive stiﬀness matrix, but is obtained by combination of terms within the matrix Dc . Thus, in the present formulation the element stiﬀness matrix can be written as

l0 K e = K c + Kg = GT (Dc + Dg )G ds0 . (4.51) 0

Care should be exercised to control any spurious strain contributions arising from the assumed shape functions when using this formula. This problem is discussed in Section 4.4.3.

4.4.2 Loads and internal forces The equilibrium condition, and thereby the residual, is determined from the virtual work (4.6). The contribution of distributed loads on the elements is expressed in terms of the equivalent nodal loads ˜ T = [ pTA , mTA , pTB , mTB ]. p

(4.52)

These nodal loads are determined from the virtual work equation (4.6),

l0 T ˜ p ¯ T m ds0 . ˜ = δu δuT p + δ ϕ (4.53) 0

Substitution of the representation of the virtual displacement in the form (4.42) then gives the equivalent nodal loads

l0 p ˜ = p ds0 . HT (4.54) m 0 Concentrated loads acting at the nodes can be added directly into the global formulation and will not be associated with the elements.

Finite rotation beam theory

The internal forces are arranged in the same format as the equivalent loads, ˜ T = [ −NTA , −MTA , NTB , MTB ], q

(4.55)

where the minus signs are due to the sign convention of section forces. They are determined from the internal virtual work given by the ﬁrst integral in (4.6),

l0 ˜ = ˜T q δεj Nj + δκj Mj ds0 . (4.56) δu 0

The components in this integral are local, as explained in Section 4.2. For ε and a linear elastic beam the constitutive tangent stiﬀness matrices Cjk κ Cjk introduced in (4.19) are constant. The internal virtual work (4.53) can therefore be written as

l0 ε κ T ˜ q ˜ = δu δεj Cjk (εk − ε0k ) + δκj Cjk (κk − κ0k ) ds0 , (4.57) 0

where ε0k and κ0k are the strain and curvature components in the reference state, respectively. In the present formulation the global strain and curvature components ε and κ are used together with the global stiﬀness component matrices Cε∗ and Cκ∗ introduced in (4.21). Hereby, the internal virtual work takes the form

l0 ˜T q ˜ = δu δεCε∗ (ε − ε0 ) + δκCκ∗ (κ − κ0 ) ds0 0 (4.58) ε

l0 ε − ε C 0 0 ∗ ds0 . = [ δεT , δκT ] 0 Cκ∗ κ − κ0 0 The global components of the virtual strain and curvature follow from (4.7) and (4.11) in the form ¯ δε = δu + x × δ ϕ,

¯ . δκ = δ ϕ

(4.59)

The virtual strain and curvature can be represented in block matrix format as   δu ˆ  δε I 0 x ¯  . (4.60) = δϕ 0 I 0 δκ ¯ δϕ The vector on the right-hand side has already been represented in terms of shape functions in (4.44) by the matrix G(s0 ). When this expression is used

4.4 Finite element implementation

in (4.50) the internal virtual work formula (4.58) deﬁnes the internal force ˜ as vector q

l0 ε − ε0 ˜ = q GT Dc κ − κ0 ds0 , (4.61) 0

where the zero vector 0 in the combined strain curvature block vector is introduced to eliminate the last block column of the constitutive matrix Dc , which is not used here. The global form of the total strain and curvature is given by (4.9) and (4.15) respectively, ε x − n1 = . (4.62) κ T(ϕ) ϕ In principle, these expressions are to be represented by interpolation of the current nodal values of x and ϕ. Due to the non-linear relation between ¯ and the increment of the rotation parameters the incremental rotation dϕ dϕ, linear interpolation of the rotation parameters ϕ may introduce some inconsistency. In the case of the two-node element this can be avoided by use of representative mean values of ε and κ obtained directly from the ‘mean ¯ over the element, as discussed in rotation’ and the rotation increment ∆ϕ Section 3.5.4. The idea is to consider a point of ‘mean rotation’, described e.g. by the rotation pseudo-vector ϕ. ¯ The mean is deﬁned such that further ¯ produces the rotation ϕB , while the reverse rotation − 12 ∆ϕ ¯ rotation by 12 ∆ϕ produces the rotation ϕA . The mean strain is obtained from the vector n1 as determined by the mean rotation ϕ, ¯ while the mean curvature is determined from the rotation increment over the element as ¯ ∆ϕ κ = . (4.63) l0 ¯ determined by this proceThe direction n1 and the rotation increment ∆ϕ dure are independent of the parametrization of the rotations and directly related to the element.

4.4.3 Shear locking The representation of the displacement and rotation increments by linear interpolation introduces a number of approximations, of which the most important probably is shear locking. This phenomenon, which is known for many beam and shell elements formulated in terms of translations and rotations, will here be illustrated in a simple two-dimensional setting. The following discussion relates to the linearized theory of a straight beam of

Finite rotation beam theory

length L. The base vectors (n1 , n2 , n3 ) deﬁned by the beam cross-sections can therefore be used as a ﬁxed global frame of reference. Extension of the beam is disregarded, and thus bending in the n2 –n3 plane is described by the transverse displacement u2 and the rotation ϕ = ϕ3 . The only measures of deformation are the shear strain and the curvature, ε2 = u2 − ϕ3 ,

κ3 = ϕ3 .

(4.64)

With these deformation measures the elastic energy of the beam is

1 L 2 U = EIϕ3 + GA(u2 − ϕ3 )2 ds. 2 0

(4.65)

The problem of shear locking arises when the representation of the shear strain in the last term of the energy is not able to accommodate a state of vanishing shear stress. Example 4.1. Shear locking in hom*ogeneous bending. The simple example in Fig. 4.5 illustrates the problem. A beam of length L is modeled by n identical elements of length l = L/n and loaded in hom*ogeneous bending by application of two opposite bending moments of magnitude M . Within each element the translation u(s0 ) and the rotation ϕ(s) are represented by linear interpolation between the end-point values. The curvature ϕ is represented correctly, giving ϕ =

M , EI

ϕB = −ϕA =

L M . 2 EI

The linear graph of ϕ(s) is shown in Fig. 4.5(b). The shear strain is given by u −ϕ. Due to the linear representation of the translation u, the derivative u is constant within each element. In the present case each element deforms symmetrically, and thus the graph of u intersects the graph of ϕ in the midpoint of each element. As a result of the diﬀerent degrees of representation of the two terms, the shear strain is given by the ‘sawtooth’ diﬀerence between the two graphs, while in the exact solution it is identically zero.

(a)

(b)

Fig. 4.5. (a) hom*ogeneous bending, (b) distribution of ϕ and u .

4.4 Finite element implementation

The elastic strain energy of the beam as represented by the n elements with linear representation of u and ϕ follows by integration from (4.65), Uapprox =

1 2

L EI (2ϕA /L)2 +

1 6

L GA (ϕA /n)2 .

The ﬁrst term represents the exact value, while the second term is an undesirable consequence of the inconsistent representation of the shear strain. It is convenient to write the result in the form 1 Uexact , Uapprox = 1 + 2 n ΦL where ΦL is the non-dimensional shear parameter of the beam 12 EI . L2 GA For a classical Bernoulli beam the shear stiﬀness is inﬁnite, whereby ΦL = 0. It is seen that linear interpolation of both translation and rotation does not permit representation of Bernoulli beams. The shear locking eﬀect in hom*ogeneous bending is seen to depend on the parameter n2 ΦL , which is the shear factor of the individual element, ΦL =

12 EI , l2 GA where l = L/n is the element length. In a displacement representation the factor on the energy corresponds directly to a factor on the stiﬀness. Thus, the stiﬀness in hom*ogeneous bending is increased by the linear interpolation of the displacements. The error decreases with n−2 , but it is undesirable to use many elements to obtain a satisfactory solution to simple problems of beam theory. Solutions to the shear locking problem are described in the following. Φ = n 2 ΦL =

There are three basically diﬀerent ways of countering the shear locking problem of displacement-based ﬁnite elements. For simple elements, like the present beam element, the locking problem can be resolved by use of interpolation functions that lead to consistent representation of the shear stress. For the beam element this implies that the transverse components of the rotation vector ϕ(s0 ) and the displacement derivative u (s0 ) should be interpolated to the same order, e.g. with a linear representation of ϕ(s0 ) and a quadratic representation of u(s0 ). For beam elements this can be accomplished by introducing a center-node for the translation degrees of freedom or the introduction of an additional internal mode not associated with a particular node, followed by elimination of the internal degrees of freedom

Finite rotation beam theory

at the element level, see e.g. Ibrahimbegovi´c (1995). The second method consists in evaluating the element properties by reduced integration. The term ‘reduced integration’ is typically used in the ﬁnite element literature to denote a form of numerical integration that is deliberately chosen not to include certain contributions, see e.g. Hughes (1987) and Zienkiewicz and Taylor (2000). In the beam example this amounts to evaluating the shear strain contributions by their value at the element midpoint. As seen from Fig. 4.5(b), the undesired shear strain contribution vanishes here. Many shell elements make use of reduced integration. A third method consists in the use of modiﬁed stiﬀness parameters that are selected to compensate for the errors introduced by the interpolation. In the following, modiﬁed bending and shear stiﬀness parameters EI and GA are derived for the beam elements with linear interpolation. The derivation is based on small deformation bending of a simple straight element, but the resulting modiﬁed parameters can subsequently be used in the implementation of the general theory for large displacement beam theory.

(a)

(b)

Fig. 4.6. Bending of ‘linear’ beam element. (a) Anti-symmetric, (b) symmetric.

For beams, modiﬁed bending and shear parameters EI and GA can be determined from the anti-symmetric and symmetric bending modes of an element. The anti-symmetric and symmetric bending modes of an element with linear interpolation of displacement and rotation are shown in Fig. 4.6. In the anti-symmetric mode of Fig. 4.6(a) the rotation ϕ must be equal at both ends of the element, and for linear representation the rotation must therefore be constant. In the symmetric mode of Fig. 4.6(b) the rotation ϕ must be of equal magnitude but opposite sign at the ends of the element. In both modes the displacement vanishes at both ends, and therefore identically. The modiﬁed stiﬀness parameters are now determined such that the two bending modes of the element attain the correct stiﬀness. The anti-symmetric mode is treated ﬁrst. Its elastic energy is evaluated by observing that the moment varies linearly between the values −M and M , while there is a constant shear force of magnitude Q = 2M/l. For a linear elastic beam the energy follows directly from the stress state as

Q(s)2 l 1 M2 4M 2 1 l M (s)2 Uexact = + ds = + 2 . (4.66) 2 0 EI GA 2 3 EI l GA

4.4 Finite element implementation

This expression takes the form l M2 6 EI when using the shear parameter of the element Uexact = (1 + Φ)

(4.67)

12 EI (4.68) l2 GA already introduced in Example 4.1. The energy is expressed in terms of the end-point rotation ϕ by use of the ﬂexibility relation 2ϕ = ∂U/∂M , where the factor 2 arises from the fact that both end loads contribute equally to the energy. Substitution of this expression for ϕ gives the energy as Φ =

Uexact =

6 EI 2 ϕ . 1+Φ l

(4.69)

The energy of the beam element with linear interpolation is calculated from (4.65), using the parameter GA. Due to the linear interpolation ϕ = const. and the bending contribution vanishes, leaving only

l 1 l Uapprox = GA ϕ(s)2 ds = GA ϕ2 . (4.70) 2 2 0 The parameter GA is now selected to make the two energy expressions (4.69) and (4.70) identical. This gives the equivalent shear stiﬀness GA =

Φ GA. 1+Φ

(4.71)

This modiﬁed value of GA will produce the theoretically correct stiﬀness of the linearly interpolated beam element in anti-symmetric bending. The procedure for symmetric bending, shown in Fig. 4.6(b), is similar. The symmetry of the problem and the linear interpolation imply that u ≡ 0 over the full element and ϕ = const. The exact elastic energy is evaluated from statics, observing that the moment is constant, whereby

l M2 1 l M (s)2 ds = . (4.72) Uexact = 2 0 EI 2 EI The angle at the ends of the beam is again calculated by use of the relation 2ϕ = ∂U/∂M , and substitution of the result gives the exact energy as 2 EI 2 ϕ . (4.73) l The energy of symmetric bending of the beam element with linear interpolation is calculated from (4.65), using the modiﬁed parameters EI and GA, Uexact =

Finite rotation beam theory

whereby Uapprox

1 = 2

l 0

2 EI l + GA ϕ2 . (4.74) EI ϕ (s)2 + GA ϕ(s)2 ds = l 6

The parameter EI is now selected to make the two expressions (4.73) and (4.74) identical, when using the value for GA already determined by (4.71). This gives the modiﬁed bending stiﬀness EI =

Φ EI 1+Φ

(full integration).

(4.75)

Note that by this procedure the correction factors for EI and GA are identical and less than unity. In the implementation of ﬁnite elements it is common to obtain the stiﬀness matrices by numerical integration. In the present case of the beam element with linear interpolation the anti-symmetric bending can be integrated exactly by use of one integration point located in the middle of the element. For the case of symmetric bending one-point integration will give the exact result for the constant curvature term, while the shear strain term vanishes at the midpoint, and therefore does not contribute to the result. This means that the second term, containing GA, vanishes from Uapprox in (4.74). Thus, the approximate energy will give the correct result for EI = EI

(reduced integration).

(4.76)

This particular form of modiﬁed parameters has been discussed, e.g. by Hughes (1987). Linear element stiﬀness matrix ˜ in When the node displacements and rotations are collected in the array u the form ˜ T = [ uA , ϕA , uB , ϕB ], u

(4.77)

the element stiﬀness matrix K can be found from the elastic energy expression ˜. ˜T K u U = 21 u

(4.78)

When linear interpolation is introduced, full integration of the energy expression (4.65) gives     0

EI  0 K = 0 l 0

0 1 0 −1

6 −3l 0 0 GA  −3l 2l2 0 −1  0 0  + 6 l  −6 3l 0 1 −3l l2

−6 3l 6 3l

−3l l2  3l  2l2

(full). (4.79)

4.4 Finite element implementation

The terms containing l2 arise from integration of ϕ2 . These terms are changed, if one-point integration is used, and the stiﬀness matrix obtained by this reduced integration is     4 0 0 0 1 0 −1  GA  −2l 0 0 0  + 4l  −4 0 −1 0 1 −2l 0

EI  0 K = 0 l

−2l l2 2l l2

−4 2l 4 2l

−2l l2  2l  (reduced). l2

(4.80) By substituting the equivalent shear stiﬀness GA from (4.71) and the appropriate equivalent bending stiﬀness EI from (4.75) or (4.76), both expressions for the stiﬀness matrix lead to the common form   12

−6 l

EI  −6 l (4 + Φ)l2 K =  −12 3 6l (1 + Φ)l 2 −6 l

(2 − Φ)l

−12 6l 12 6l

−6 l (2 − Φ)l2  . 6l 2 (4 + Φ)l

(4.81)

This is the well-known exact stiﬀness matrix for a beam with shear ﬂexibility, see e.g. Krenk (2001) for a simple derivation from static equilibrium states. Example 4.2. Shear locking in cantilever beam. The shear locking eﬀects discussed above can be illustrated by considering a cantilever beam of length L with a transverse end force P . The exact solution for the translation u and rotation ϕ of the loaded end is u P L2 P P L2 4 + ΦL = + = , L 3EI GA EI 12

ϕ = −

P L2 . 2EI

In this example P = EI/L2 and the shear modulus is G = 0.4E. The cross-section is rectangular with h = 2b, eﬀective shear area A = 56 bh and 1 moment of inertia I = 12 bh3 .

Fig. 4.7. Cantilever with transverse end force, n = 8, 16: – analytical, + exact element, × reduced integration, ∇ full integration with original stiﬀness parameters.

Finite rotation beam theory

Figure 4.7 shows the translation and rotation for the case L = 10h, where the shear parameter of the beam is ΦL = 0.03. This is a fairly slender beam in which the translation contribution due to shear ﬂexibility is less than 1%. The fully integrated beam element (4.79) without correction is approximately 40% too stiﬀ even with 8 elements, and an increase to 16 elements still leaves a stiﬀness error of about ( 12 )2 40% = 10% in both translation and rotation. By reduced integration the mean bending within each element is exact, and only the shear ﬂexibility eﬀect needs further correction by (4.71). In the present example there is no contribution from shear ﬂexibility to the rotation, which is modeled correctly by reduced integration. However, the translation still requires correction of the shear modulus by (4.71) to permit accurate analysis with few elements.

4.5 Summary of ‘elastica’ beam theory An elegant way of formulating a general large deformation beam theory consists in considering the beam as a space curve, on which the cross-sections are attached. The generalized strain ε and curvature κ then appear as three-component vectors, conjugate to the internal force N and the internal moment M, all functions of the length-coordinate s0 along the beam. The principle of virtual work combines the deformation measures ε, κ and the internal forces N, M and provides equilibrium equations in terms of a balance between the external load and the corresponding work of the internal forces. An important consequence is that a concise statement of the tangent stiﬀness relation can be obtained by manipulating the internal work by the rules developed in Chapter 3. By carefully following the rules for incremental rotations it is demonstrated that the tangent relation is symmetric, and explicit expressions are obtained for the constitutive stiﬀness and the geometric stiﬀness. Thus, non-symmetric terms will only occur in the tangent stiﬀness relations if there are load components with directions depending on the motion of the structure, so-called ‘follower-forces’. The theory leads to an internal shear force, deﬁned by a combination of the rotation and the lengthwise derivative of the transverse displacement. Consistent representation of the shear in the beam requires special attention to enable the beam to retain vanishing shear strain under rigid body motion and uniform bending. Three measures to ensure this are discussed: higher-degree representation of the rotation components, use of selective integration, and use of modiﬁed stiﬀness parameters.

4.6 Exercises

4.6 Exercises Exercise 4.1 As shown in Example 3.2 the increment of the virtual work of a moment is ¯ T M) = δ ϕ ¯ T ( 12 M × dϕ). ¯ ¯ T dM + δ ϕ d(δ ϕ If M is an external load and dM is given, the second term has the character of a geometric stiﬀness term. How will this term enter the geometric stiﬀness format (4.50)? Give both the matrix format and the full component format. Exercise 4.2 Example 4.2 treated a cantilever beam with transverse end load, illustrating the eﬀect of shear locking. (a) Implement the three stiﬀness matrices corresponding to full integration, reduced integration, and the modiﬁed parameter form. Make an error analysis similar to Fig. 4.7 for the case of a concentrated moment applied to the end of the beam. (b) How do these results relate to the general discussion of shear locking in Example 4.1?

5 Co-rotating beam elements

The beam theory of Chapter 4 and the corresponding ﬁnite element implementation was formulated in a ﬁxed global frame of reference using the total displacements and rotations. In many cases it may be advantageous to consider the beam element with reference to a local, element-based, coordinate system. Motion of the beam then implies motion of the local frame of reference as well as deformation of the beam element within this frame. The separation of the motion of the element into two parts – a rigid body motion associated with the element-based frame of reference and a deformation of the element within this frame of reference – is called a co-rotating formulation. The co-rotating formulation has a number of advantages, provided it can be demonstrated that the tangent stiﬀness can be decomposed into the sum of a part associated with the rotation of the element-based frame and a part associated solely with the deformation of the element within this frame of reference. The ﬁrst advantage is that displacements and rotations within the local frame of reference are small or at most moderate. Therefore, the deformation of the beam can be modeled by approximate beam theory. Secondly, the co-rotating formulation is closely associated with the idea of ‘natural modes’, advocated by Argyris et al. (1979a,b). The idea of the ‘natural modes’ is to consider any increment of the motion of an element as made up of a set of rigid body modes – typically translation and rotation – and a set of deformation modes – representing extension, bending and torsion of the beam element. In the co-rotating formulation the rigid body motion is associated with the motion of the local frame of reference, while the deformation is described within this frame. Considerable simpliﬁcations can be obtained by representing the deformation in terms of suitable ‘natural modes’. An important aspect of the co-rotating formulation is to establish that the contributions to the tangent stiﬀness involving the motion of the local frame can be expressed in a unique symmetric form, independent of 100

5.1 Co-rotating beams in two dimensions

101

the particular beam theory used to represent the deformation of the beam. For two-dimensional beam problems a simple and direct procedure can be used to obtain this part of the tangent stiﬀness. However, due to the nonadditive nature of three-dimensional rotations, this procedure does not apply to three-dimensional co-rotational formulations. Therefore, in the following the two-dimensional problem is ﬁrst presented using the classical direct method, see e.g. Crisﬁeld (1991). The general three-dimensional problem is then formulated rigorously by introducing rigid body and deformation modes in the general theory of Chapter 4. The co-rotating formulation is demonstrated using both the classic cubic bending interpolation and a nonlinear beam–column formulation originally proposed by Oran (1973a,b).

5.1 Co-rotating beams in two dimensions Figure 5.1 shows a beam element located in a plane deﬁned by a ﬁxed frame of reference {x1 , x2 }, indicated in the lower left corner. The beam itself is described with respect to a local frame of reference {x, y}, following the motion of the beam. The local frame of reference is deﬁned by the position of the end-points A and B of the beam element: the x-axis passes through the points A and B, and the origin is located with equal distance from A and B.

Fig. 5.1. Motion of beam element in local co-rotating frame of reference.

In the ﬁxed global frame of reference the motion of the beam element is described by the translation and rotation of the beam end-points A and B, A A dpTA = [ duA 1 , du2 , dφ ],

B B dpTB = [ duB 1 , du2 , dφ ].

(5.1)

Thus, the motion of the beam element is described by six components, and naturally there are also six conjugate generalized force components. The

102

Co-rotating beam elements

conjugate element forces constituting the force and moment at A and B are denoted qTA = [f1A , f2A , mA ],

qTB = [f1B , f2B , mB ].

(5.2)

It is convenient to introduce the notation dpT = [ dpTA , dpTB ],

qT = [ qTA , qTB ]

(5.3)

for the complete description of the motion and the generalized forces in the ﬁxed frame of reference. When these arrays are expressed in the local element-based frame, they are denoted dpe and qe . In this notation the virtual work of the generalized forces at the end-points of the beam element takes the form δV = δpT q.

(5.4)

This relation will be used to express equilibrium of the element, using δpe and qe in the local frame, and to obtain the element tangent stiﬀness matrix, relating the global increments dp and dq.

(a)

(b)

Fig. 5.2. Incremental rigid body modes: (a) translation, (b) rotation.

The idea of the co-rotating formulation is to separate the motion into two parts: a rigid body motion associated with the motion of the local frame of reference, and a deformation of the beam within this frame. The motion of the local frame of reference is described by the translation uT = [u1 , u2 ] of its origin and the rotation ϕ of the axes. The incremental rigid body motions corresponding to du and dϕ are illustrated in Fig. 5.2. The full description of the motion of the beam element requires six components, and when three components are used for the rigid body motion three are left for description of the deformation of the beam. These three components deﬁne three modes of deformation of the beam element. These modes can be selected in diﬀerent ways, but it turns out to be convenient to use the set of ‘natural deformation modes’ illustrated in Fig. 5.3. The extension mode shown in Fig. 5.3(a) consists of an axial translation of magnitude 12 du of the end-points A and B, increasing their distance by du. For a straight beam this corresponds to an extension of du. The symmetric bending mode shown in Fig. 5.3(b) is deﬁned by a rotation 12 dϕs of the end-points, clockwise

5.1 Co-rotating beams in two dimensions

103

at A and counterclockwise at B. Finally, the anti-symmetric bending mode shown in Fig. 5.3(c) is deﬁned by a counterclockwise rotation 21 dϕa of both end-points.

(a)

(b)

(c)

Fig. 5.3. Deformation modes: (a) extension, (b) symmetric bending, (c) antisymmetric bending.

The six generalized force components of the beam element must satisfy three equilibrium equations. This leaves three generalized force components, conveniently selected as the generalized forces conjugate to the displacement components of the natural deformation modes of Fig. 5.3. The generalized forces corresponding to the natural deformation modes are illustrated in Fig. 5.4. The extension mode corresponds to a normal force N , shown in Fig. 5.4(a). The symmetric bending mode corresponds to a moment Ms at the end-points, clockwise at A and counterclockwise at B, Fig. 5.4(b). The anti-symmetric bending mode corresponds to a counterclockwise moment Ma at both end-points. It is easily seen that the generalized forces N and Ms are equilibrium systems, while the moment Ma in the anti-symmetric bending mode must be complemented by shear forces Q = −2Ma /l as shown in Fig. 5.4(c), where l is the current distance between the end-points A and B.

(a)

(b)

(c)

Fig. 5.4. Equilibrium force systems: (a) normal force, (b) constant moment, (c) constant shear.

It is also convenient to have a matrix notation for the components of the deformation modes and the corresponding equilibrium force systems, and therefore the following notation is introduced: dvT = [ du, dϕs , dϕa ],

tT = [ N, Ms , Ma ].

(5.5)

104

Co-rotating beam elements

In this notation the external virtual work of the generalized forces at the end-points of the beam element is δV = δvT t.

(5.6)

This relation is similar to (5.4) in the ﬁxed frame of reference, but has only three components. It is seen that the factor 12 in the deﬁnition of the deformation modes shown in Fig. 5.3 is necessary to make the generalized forces t conjugate to the incremental displacements dv.

5.1.1 Co-rotation form of the tangent stiﬀness The co-rotating formulation involves two transformations: one from the reduced set of internal variables dv and t to a complete set of variables dpe and qe in a coordinate system aligned with the element, and then a transformation of these components to the ﬁxed frame of reference by a rotation. The rotation of components qe in a local frame of reference to the ﬁxed frame of reference is described by the rotation matrix

R =

− sin ϕ cos ϕ

cos ϕ sin ϕ

(5.7)

When using the combined format (5.3), the compound rotation matrix R (5.8) Re = R is introduced to give the relation q = Re qe .

(5.9)

Clearly, this relation rotates the components at A and the components at B by the rotation matrix R. The ﬁrst transformation, giving the full set of generalized forces in a local frame, is expressed as qe = S t,

(5.10)

where the 6 × 3 transformation matrix S is 

S =

S1 S2

   =   

−1 0 0 1 0 0

0 0 −1 0 0 1

0 2/l 1 0 −2/l 1

    .  

(5.11)

5.1 Co-rotating beams in two dimensions

105

It follows from equality of the virtual work δV expressed in a set of full components by (5.4) and in internal components by (5.6) that δV = δpTe qe = δpTe S t = δvT t.

(5.12)

The last equality must hold for arbitrary internal ‘stresses’ t, and the internal incremental displacements dv must therefore be related to the displacement components dpe by dv = ST dpe = ST RTe dp.

(5.13)

Thus, the transformation matrix S serves to expand the reduced set of internal forces to full format by (5.10), while the transpose ST serves to extract the modal deformation components from the full displacement representation by (5.13). As discussed in Chapters 2 and 4, the standard procedure for obtaining the tangent stiﬀness consists in considering the increment of the virtual work used to express equilibrium. In the case of co-rotational elements only the external virtual work is used. In the present case this leads to consideration of the increment of the external virtual work δV = δpT q. When calculating this increment the virtual displacement vector δp can be considered constant, and thus d(δV ) = d(δpT q) = δpT dq.

(5.14)

Thus, the virtual work increment d(δV ) is of the same form as the virtual work δV , when the generalized forces q are replaced by their increment dq. This implies that the incremental equilibrium equations for dq are of the same form as those for the total forces q. This is often taken for granted, but does not hold for general three-dimensional rotations, where the increment of the virtual rotation gives a separate contribution that must also be included in the incremental equilibrium equations as described in Section 5.2. For the two-dimensional problems considered here the tangent stiﬀness can be derived from the increment of the generalized force relation q = Re S t.

(5.15)

The incremental relation involves a change in the internal forces t, a change of S due to a change in l, and ﬁnally a change of Re due to a change in ϕ: dq = Re S dt + Re dS t + dRe S t.

(5.16)

The change of the internal force vector is related to a change in the state of deformation, and it is therefore given by a relation of the form dt = Kd dv,

(5.17)

106

Co-rotating beam elements

where Kd is the tangent stiﬀness matrix of the deformation modes of the element, expressed in the reduced 3 × 3 internal format. When this relation is substituted into the ﬁrst term of (5.16), and dv is expressed by (5.13), the increment of the force takes the form dq = Re S Kd ST dpe + dS + RTe dRe S t . (5.18) Multiplication with RTe gives the stiﬀness relation in terms of the full components in the local frame of reference, dqe = S Kd ST dpe + dS + RTe dRe S t. (5.19) The ﬁrst term represents the stiﬀness due to the local deformation, while the second term containing the internal forces t represents the combined eﬀect of the co-rotating frame of reference and the fact that the shear forces, determined from the moments by equilibrium, may change due to a change of the distance l between the end-points of the beam element. The rotation and extension increments are expressed in terms of the displacement components in the local frame of reference, A dϕ = (duB y − duy )/l,

A dl = duB x − dux .

(5.20)

The local tangent stiﬀness relation can then be rearranged into the form dqe = Ke dpe ,

(5.21)

where the element stiﬀness matrix Ke is given by Ke = S Kd ST + Kr .

(5.22)

The ﬁrst part is the stiﬀness of the deformation modes of the element, given by Kd , while the second part Kr represents the combined eﬀect of the corotating frame of reference and the change of the shear force due to changes in l. The co-rotation stiﬀness matrix is evaluated from the second term in (5.19), see Exercise 5.2. The result is conveniently given in block matrix format as Kr =

Kr11 Kr21

Kr12 Kr22

with Kr11

Kr22

−Kr12

−Kr21

1 = l

0 −Q −Q N 0 0

(5.23)

0 0 0

(5.24)

It is interesting to note that the stiﬀness matrix Kr includes the eﬀect of extension dl on the shear force in addition to the eﬀect of the rotation dϕ

5.1 Co-rotating beams in two dimensions

107

of the frame of reference. Kr would not be symmetric if the extension eﬀect were omitted. 5.1.2 Element deformation stiﬀness In the co-rotation procedure the element stiﬀness matrix Ke is composed of two parts: one from incremental deformation of the element represented in terms of the deformation modes by Kd , and the matrix Kr caused by the corotation of the frame of reference. It is important to note that this separation is diﬀerent from the previously used separation of the total stiﬀness into a constitutive and a geometric part. In the co-rotational formulation the geometric stiﬀness associated with the rotation of the local frame of reference and changes of the shear force due to extension are accounted for by Kr . Whether or not there are additional geometric stiﬀness contributions in the deformation stiﬀness matrix Kd depends on the level of approximation of the beam model. At the lowest level is a linear beam theory without initial stress terms, the next level includes geometric stiﬀness contributions via assumed shape functions, and the last level incorporates a non-linear deformation model as discussed in Section 5.3. Constitutive stiﬀness The constitutive properties of the beam element are contained in the deformation stiﬀness matrix Kd , introduced in (5.17) as the coeﬃcient matrix of the incremental relation between the static and kinematic variables of the displacement modes,      dN du  dMs  =  Kd   dϕs  . (5.25) dMa dϕa This relation involves the constitutive stiﬀness and may in addition include a geometric contribution. Here the constitutive contribution of a straight hom*ogeneous elastic beam element is considered. In this case the stiﬀnesses of the deformation modes do not couple, whereby Kd becomes a diagonal matrix and each of the three deformation modes can be considered separately. The ﬁrst deformation mode is extension. For a straight hom*ogeneous elastic beam the extension stiﬀness is EA dN = du, (5.26) l giving the ﬁrst diagonal element of Kd .

108

Co-rotating beam elements

The bending modes are illustrated in Fig. 5.5. As the beam is assumed to be linear elastic, the total moments and angles are used for simplicity of notation. The stiﬀness of the symmetric deformation mode follows from evaluation of the complementary virtual work of a constant moment distribution

l M2 M (s)2 Ms ϕs = ds = l s , (5.27) EI EI 0 corresponding to the incremental stiﬀness relation dMs =

EI dϕs . l

(5.28)

This gives the second diagonal term of Kd .

(a)

(b)

Fig. 5.5. Bending modes: (a) symmetric, (b) anti-symmetric.

The anti-symmetric bending mode is illustrated in Fig. 5.5(b). The static components are a bending moment of linear variation from −Ma to Ma and a constant shear force Q = −2Ma /l

(5.29)

determined from equilibrium. There is no transverse displacement at the ends, and complementary virtual work then gives

l 1 M2 Q2 Q2 M (s)2 a Ma ϕa = + ds = l + . (5.30) EI GA 3 EI GA 0 When introducing the shear ﬂexibility parameter ψa =

1 , 1+Φ

Φ =

12 EI , l2 GA

(5.31)

substitution of the shear force from (5.29) gives the incremental stiﬀness relation EI dMa = 3ψa dϕa . (5.32) l

5.1 Co-rotating beams in two dimensions

109

This is the ﬁnal diagonal element in Kd . Thus, the constitutive part of the stiﬀness matrix of the deformation modes is 1 = l

(5.33)

3ψa EI

This matrix enters the local element stiﬀness matrix Ke in the form SKd ST , which may be evaluated numerically or explicitly, depending on programming taste. In essence, the constitutive stiﬀness was calculated from the statics of the beam element. This procedure is easily generalized to non-hom*ogeneous and curved elastic elements, see e.g. Krenk (1994). Local geometric stiﬀness In addition to the constitutive part of the deformation stiﬀness matrix Ke evaluated above there may be local geometric stiﬀness contributions. The geometric contribution to the stiﬀness of the local deformation modes can be evaluated under the simplifying assumption of vanishing shear strain, whereby the results become very simple and usually quite representative. In the local xy-frame, bending of a beam with vanishing shear strain and with a normal force N is governed by the diﬀerential equation (EI uy ) − (N uy ) = 0.

(5.34)

The corresponding internal virtual work is found by multiplication with δuy , followed by integration by parts. The result is

l δVin = δuy EI uy + δuy N uy ds. (5.35) 0

In the absence of three-dimensional rotation eﬀects, the incremental form follows directly by d(δuy ) = 0 as

l d(δVin ) = δuy EI duy + δuy N duy ds. (5.36) 0

The local stiﬀness matrix follows from this expression by substitution of suitable shape function representations of δuy and duy . It is convenient to represent the arc-length s by the non-dimensional coordinate ξ via s =

1 2 l(1

+ ξ),

−1 ≤ ξ ≤ 1.

(5.37)

In terms of this coordinate the symmetric bending mode is duy = − 18 l(1 − ξ 2 ) dϕs ,

(5.38)

110

Co-rotating beam elements

while the anti-symmetric mode is duy = − 18 l(1 − ξ 2 )ξ dϕa .

(5.39)

The local stiﬀness matrix follows from substitution of the representations (5.38) and (5.39) into the incremental virtual work d(δV ) given by (5.36). The displacement representations do not couple as they are even and odd functions of ξ, respectively, and the stiﬀness of symmetric and anti-symmetric bending deformation modes can therefore be calculated independently. It is easily veriﬁed – see e.g. Exercise 5.3 – that the constitutive stiﬀness contribution from (5.36) gives the special case ψa = 1 of the result already obtained by a direct method including shear ﬂexibility in (5.33). The geometric stiﬀness of the symmetric bending mode follows from substitution of the derivative of uy , given by (5.38), into the second term of the integrand in (5.36), whereby

1 d 1 kss = 12 l ( 12 ξ) N ( 12 ξ) dξ = 12 l N. (5.40) −1

The geometric stiﬀness of the anti-symmetric mode follows similarly from substitution of the derivative of uy , given by (5.39), as

1 d 2 2 1 1 1 1 kaa = 2 l (5.41) 4 (1 − 3ξ ) N 4 (1 − 3ξ ) dξ = 20 l N. −1

Thus the geometric stiﬀness from the deformation bending modes with cubic interpolation is 

Kd = l 



1 12 N

1 20 N

.

(5.42)

This matrix should be added to the constitutive part (5.33) to give the full stiﬀness associated with the deformation modes.

5.1.3 Total tangent stiﬀness In the co-rotating format the element stiﬀness matrix in the local frame of reference is given by (5.22). In this format the stiﬀness of the deformation modes, with a constitutive and possibly a geometric part, is transformed into the local frame, and a contribution from the co-rotation of the local frame is added. It is simple to program this basic relation directly, using the stiﬀness matrix of the deformation modes Kd , the rotation stiﬀness matrix Kr , and the transformation matrix S relating the components of the deformation modes to a full set of variables in the local element frame. However, the

5.1 Co-rotating beams in two dimensions

111

transformation matrix is quite simple and involves many zero entries, and the full element stiﬀness matrix is therefore easily evaluated analytically by matrix multiplication. In addition to its potential use in a computer program, this form also facilitates comparison with other formulations and derivations. The local element stiﬀness matrix has the block matrix format

Ke11 Ke21

Ke =

Ke12 Ke22

(5.43)

already introduced for the rotation matrix in (5.23). When using the block matrix format (5.11) of the transformation matrix S, the co-rotation format (5.22) can be written in sub-matrix format as Keij = Si Kd STj + Krij

(5.44)

with i = 1, 2 and j = 1, 2. The sub-matrices Krij were given in (5.24), and the stiﬀness matrix of the modes Kd is the sum of the constitutive part (5.33) and the geometric part (5.42). When carrying out the matrix multiplications in (5.44), the constitutive part of the element stiﬀness matrix is found to be 

2 1 EAl Ke11 = 3  0 l 0

 EAl2 1 e  0 K22 = 3 l 0

Ke12

KeT 21

 2 1  −EAl 0 = 3 l 0

0 12ψa EI 6ψa EIl

 0 , 6ψa EIl (3ψa +1)EIl2

0 12ψa EI −6ψa EIl

 0 −6ψa EIl , (3ψa +1)EIl2  0 . 6ψa EIl (3ψa −1)EIl2

0 −12ψa EI −6ψa EIl

(5.45)

(5.46)

(5.47)

Similarly the sub-matrices of the geometric stiﬀness matrix of the element are 

0 1 Ke11 =  −Q l 0

−Q 6 5N 1 10 N l

 0 1 , 10 N l 2 2 15 N l



0 −Q 1 6 Ke22 =  −Q 5N l 1 Nl 0 − 10



1 − 10 N l , 2 2 15 N l

 0 Q 1 e eT  Q − 65 N K12 = K21 = l 1 Nl 0 − 10

(5.48)

 0 1 , 10 N l 1 2 − 30 N l

(5.49)

(5.50)

112

Co-rotating beam elements

where the shear force is determined from the antisymmetric bending moment as Q = −2Ma /l.

(5.51)

It is seen by comparison with (5.24) that the deformation modes only contribute modestly to the geometric stiﬀness matrix.

5.1.4 Finite element implementation In the co-rotation format the element properties are ﬁrst obtained in a local element-based frame of reference and are subsequently transformed into the global frame. For two-dimensional co-rotating beam element problems the formation of element forces and stiﬀness matrix can be arranged as described in the following. In order to illustrate the procedure in the simplest possible setting a total displacement description is used, and the beam element is assumed to be hom*ogeneous and linear elastic.

Fig. 5.6. Deformed beam element in two-dimensional problem.

The current state of the beam element is illustrated in Fig. 5.6, showing the current coordinate vectors xA , xB and rotation angles ϕA , ϕA at the element end-points. The element angle ϕ with the global axes is conveniently calculated from the trigonometric relations A cos ϕ = (xB 1 − x1 )/l,

A sin ϕ = (xB 2 − x2 )/l

by using the relation for tan( 12 ϕ), whereby l − ∆x 1 − cos ϕ 1 = 2 arctan . ϕ = 2 arctan sin ϕ ∆x2

(5.52)

(5.53)

This expression is singular for ∆x2 = 0, where ϕ = 0 for ∆x1 = l and ϕ = π for ∆x1 = −l. The next step is to calculate the kinematic measures of deformation: the elongation u, the angle ϕs of the symmetric deformation mode, and the angle

5.1 Co-rotating beams in two dimensions

113

Algorithm 5.1. Two-dimensional co-rotating beam elements. Element angle: ϕ = 2 arctan

l − ∆x 1

(5.53)

∆x2 Deformation parameters v: u = l − l0 ϕs = ϕ B − ϕA ϕa = ϕB + ϕA − 2(ϕ − ϕ0 ) Modulus for absolute angle: ϕa = mod(ϕa + π) − π

(5.54)

(5.55)

2π

Internal ‘stresses’: t = Kd v Element forces in global frame: q = Re S t Element tangent stiﬀness in global frame: Kij = R Keij RT

(5.33) (5.9), (5.10) (5.58)

ϕa of the anti-symmetric deformation mode. These quantities are calculated from the relations u = l − l0 ,

ϕs = ϕB − ϕA ,

ϕa = ϕB + ϕA − 2(ϕ − ϕ0 ).

(5.54)

In order to be robust for arbitrary angles, the expression for the antisymmetric rotation should be calculated using the modulus function ϕa : = mod(ϕa + π) − π.

(5.55)

2π

The modulus function places the argument in the interval [0, 2π[, and the last term re-establishes symmetry with respect to zero. Omission of this step may lead to problems, when the beam element has rotated ±π, ±2π, . . . The internal ‘stresses’ of the beam element are the normal force N , the symmetric moment Ms , and the anti-symmetric moment Ma . For the present linear theory the total relations follow from (5.25) and (5.33) as N =

EA u, l

Ms =

EI ϕs , l

Ma = 3ψa

EI ϕa . l

(5.56)

In the factors no distinction has been made between current length l and initial length l0 . With the internal ‘stresses’ t determined by (5.56), the full set of element forces qe is determined from (5.10) and the local element

114

Co-rotating beam elements

stiﬀness matrix Ke from (5.43) and (5.45)–(5.51), or the underlying product format (5.22). The ﬁnal step before assembling the element contributions into the global matrices is the transformation from the local element frame to the global frame of reference. This is accomplished by use of the compound rotation matrix Re given by (5.7)–(5.8). The element force vector is transformed by (5.9), while the element matrix is transformed by K = Re Ke RTe .

(5.57)

This transformation contains many zero entries, and the diagonal block matrix structure of the transformation matrix Re implies that the sub-matrices of the stiﬀness matrix can be transformed directly by the rotation matrix R, Kij = R Keij RT

(5.58)

with i = 1, 2 and j = 1, 2. The element sub-matrices were given in explicit form in (5.45)–(5.51). The computational procedure for obtaining the element forces and the element tangent stiﬀness matrix in the global frame is summarized in Algorithm 5.1. Example 5.1. Bending of a beam. A simple example that illustrates the beam element’s ability to handle large rotations is the roll-up of a cantilever beam by a bending moment applied to its free end. Ideally the beam will curve into a circular arc with curvature κ = 1/R = M/EI. Thus, the beam ends meet, when M = 2πEI/L. In the graphics of Fig. 5.7 the 10 elements are plotted as straight, although they are curved due to the deformation described by the shape functions. However, in the element developed here the distance between the two end-points of the element is not changed by

Fig. 5.7. Roll-up of cantilever beam.

5.1 Co-rotating beams in two dimensions

115

curvature, and thus the nodes are located at the apexes of a polygon with sides corresponding to the initial length of the beam. The rotation of the end node is proportional to the load, and thus Newton– Raphson iteration is fully adequate. The full circle is reached in 10 steps, and with a relative tolerance on the residual of 10−6 each step required six equilibrium iterations. In this example the only internal force is the moment M , and thus there is no contribution from geometric stiﬀness.

Example 5.2. Large deformation of cantilever. An example that illustrates very large deformation as well as the inﬂuence of the geometric stiﬀness matrix is a cantilever beam of length L with a transverse force P applied at the free end, shown in Fig. 5.8(a). An accurate table of the displacement and rotation of the end-point has been obtained by Mattiasson (1981) using elliptic integrals. The force is assumed to retain its direction while following the position of the end-point. The load is described by the non-dimensional parameter P L2 /EI, and Fig. 5.8(b) shows the nondimensional displacements u/l, v/L and ϕ at the end of the cantilever as functions of the load. The load is applied in 20 equal steps leading to a ﬁnal load of P L2 /EI = 10. At this load level the end-point has attained a horizontal displacement of more than half the length of the beam.

(a)

(b)

Fig. 5.8. Cantilever with conservative transverse end force.

Four models were used with 2, 4, 6, and 8 identical elements, respectively. With a relative tolerance on the residual of 10−6 the average number of iterations was around 7.5. The results at the ﬁnal load level and half of this are shown in Table 5.1 with the analytical results in the last row. It is seen that good accuracy, even in this highly deformed state, is obtained with four elements. In this problem the geometric stiﬀness matrix is important, and

116

Co-rotating beam elements

Table 5.1. Cantilever beam with conservative end force.

Nelem 2 4 6 8 –

P L2 /EI = 5 −u/L v/L 0.38941 0.38711 0.38732 0.38744 0.38763

0.73858 0.71915 0.71607 0.71506 0.71379

ϕ 1.21137 1.22314 1.21873 1.21723 1.21537

P L2 /EI = 10 −u/L v/L 0.56761 0.55572 0.55506 0.55498 0.55500

0.84771 0.81878 0.81400 0.81247 0.81061

1.47759 1.44037 1.43460 1.43268 1.43029

the convergence is lost around half the ﬁnal load if the geometric stiﬀness matrix is omitted in the iterations.

Example 5.3. Shallow angle beam. The angle beam shown in Fig. 5.9(a) includes an additional eﬀect, the shortening of the beam due to bending. This is not included in the present beam element formulation, and it is therefore necessary to use several beam elements for each of the two straight beams. In this way the ﬁnite displacement of the nodes leads to a representation of the shortening eﬀect.

(a)

(b)

Fig. 5.9. Angle beam with clamped supports: (×) 5 elements, (+) 10 elements.

An approximate analytical solution based on second-order theory has been obtained by Williams (1964). A more direct solution can be obtained by considering each of the beams as a beam-column with ﬁnite displacement connecting the support and the apex in a state of anti-symmetric bending. The corresponding beam-column solution is well known, see e.g. Timoshenko and

5.2 Co-rotating beams in three dimensions

117

Gere (1961) or Krenk (2001). The solution depends on the inclination h/b of the two beams and their slenderness, represented by the non-dimensional parameter L20 A/I. Figure 5.9(b) shows the non-dimensional vertical load P L20 /EI as a function of the vertical displacement w/h of the central node for h/b = 0.024 and L20 A/I = 3.0 × 104 . Two ﬁnite element models with straight beam elements were made, using 5 and 10 elements, respectively, for each straight beam. With equal load steps, Newton–Raphson iteration, and a relative tolerance on the residual of 10−6 , the average numbers of iterations were 5.6 and 4.8, respectively. It is seen that the error is negligible for w h, but for larger displacements even the model with 10 elements per beam exhibits some error. This indicates the usefulness of elements in which the shortening due to bending is included in the individual element, as discussed in Section 5.3. For more slender beams or larger inclination h/b a snap-through mechanism will develop, and bifurcation into a non-symmetric deformation mode may also occur. These phenomena can be computed by the more general path-tracing methods described in Chapter 8.

5.2 Co-rotating beams in three dimensions The derivation of the co-rotating beam element format for two-dimensional problems presented above is attractive because it only requires a constitutive relation for the deformation modes of the element, and a relation between these modes and the global set of element forces. Thus, ﬁnite rotations are accounted for without the need for a fully non-linear beam theory. This attractive feature is due to the fact that the tangent stiﬀness is derived from the external virtual work of the element, and not from the internal work of a fully non-linear theory as in the beam theory of Chapter 4. Furthermore, the derivation was arranged in such a way that only incremental relations between internal and global displacement components were needed. When working with ﬁnite rotations in three dimensions it is important to note ¯ the diﬀerence between a small rotation represented by the pseudo-vector δ ϕ and the corresponding set of component increments δϕ of the total rotation ϕ. This diﬀerence was discussed in detail in Section 3.3, where it was demonstrated that it is indeed possible to work directly with the small rota¯ provided that the increment of the variation d(δ ϕ) ¯ tion pseudo-vector δ ϕ, is properly accounted for by use of (3.30). When this term is included the need for an explicit parametrization of the ﬁnite rotation, used e.g. by Pacoste and Eriksson (1997) and Battini (2002), is avoided, and the procedure

118

Co-rotating beam elements

Fig. 5.10. Three-dimensional motion of beam in co-rotating frame.

of the previous section for two-dimensional beam elements can be extended directly. Figure 5.10 shows a beam element in three-dimensional space described by the ﬁxed frame of reference {x1 , x2 , x3 }. The beam is described in a local co-rotating frame of reference {x, y, z} with base vectors {nx , ny , nz }. The x-axis passes through the end-points A and B of the beam element. The y- and z-axes are deﬁned by the mean rotation of A and B as described in detail later. The incremental motion is described in the global frame of reference by the array ¯ TA , duTB , dϕ ¯ TB ] dpT = [ duTA , dϕ

(5.59)

and the corresponding conjugate element forces are qT = [ fTA , mTA , fTB , mTB ].

(5.60)

The components of dp and q refer to the ﬁxed frame of reference. In terms of these components the external virtual work of the beam element is δV = δpT q =

¯ T∗ m∗ . δuT∗ f∗ + δ ϕ

(5.61)

A,B

The increment of this relation is used to obtain the tangent stiﬀness. The components of the displacement increments and the element forces in the local co-rotating frame are denoted dpe and qe . The current orientation of the local frame of reference is deﬁned by the rotation matrix R. The relation between the local and global components is then given by the

5.2 Co-rotating beams in three dimensions

compound rotation matrix

 



Re = 

119

 

R R

(5.62)

as dp = Re dpe ,

q = Re qe .

(5.63)

While it is necessary to have an expression for the rotation R of the local frame relative to the ﬁxed frame in order to transform force components and stiﬀness matrix from the local to the global frame, the expressions for both local element forces and the local element stiﬀness matrix are independent of R.

Fig. 5.11. The six natural deformation modes of a beam element.

The six natural deformation modes of a symmetric beam element are illustrated in Fig. 5.11. The top row shows three modes with constant moment, while the bottom row shows three modes with constant internal force. The constant moment modes are: a constant torsion moment M corresponding to opposing incremental angles of twist ± 12 dϕsx , a constant moment produced by symmetric end moments Mys with incremental end-point rotations ± 12 dϕsy , and similar symmetric moments Mzs with incremental end-point rotations ± 12 dϕsz . The constant internal force modes are: a constant normal force N corresponding to the incremental extension du, opposing bending moments ±Mya corresponding to incremental end-point rotations 12 dϕay , and opposing bending moments ±Mza corresponding to incremental end-point

120

Co-rotating beam elements

rotations 12 dϕaz . This corresponds to the incremental deformation mode vector dvT = [ dϕsx , dϕsy , dϕsz , du, dϕay , dϕaz ]

(5.64)

and the conjugate internal ‘stress’ vector tT = [ M, Mys , Mzs , N, Mya , Mza ].

(5.65)

In later computations it will be convenient to use also the shear forces Qy = −2Mza /l,

Qz = 2Mya /l

(5.66)

following from the equilibrium conditions. The relation between the internal deformation variables and the full set of element variables is conveniently expressed in terms of a set of unit vectors {nx , ny , nz } constituting the base vectors of the local {x, y, z} co-ordinate system. When using these vectors it is seen directly from Fig. 5.11 that the element forces can be expressed as     Ms   fA 0 0 0 −nx −2nz /l 2ny /l  My   −nx −ny −nz   Mzs  0 n n  mA  y z  . (5.67)   f  =   0 0 0 nx 2nz /l −2ny /l   N  B  Ma  mB nx ny nz 0 ny nz y Mza S Thus the element forces q are given in terms of the internal ‘stresses’ by q = S t,

(5.68)

where the 12 × 6 transformation matrix S is deﬁned by (5.67). This relation represents the global components q if the base vectors nx , ny , nz are represented in the global frame, and the local components qe if the base vectors nx , ny , nz are represented in the local frame as nTx = [1, 0, 0], etc. This means that the rotation of the element is accounted for directly via the rotation of the unit vectors nx , ny , nz . This simpliﬁes the derivation of the tangent stiﬀness, because an incremental rotation of a vector can be represented by a vector cross product as discussed in Chapter 3.

5.2.1 Co-rotation form of the tangent stiﬀness In the co-rotation context the tangent stiﬀness is derived from the increment of the external virtual work of the element, given by (5.61). When allowance

5.2 Co-rotating beams in three dimensions

121

¯ A and δ ϕ ¯ B, is made for the increment of the variation of the rotations δ ϕ this variation is ¯ A )T mA + d(δ ϕ ¯ B )T mB . d(δV ) = δpT dq + d(δ ϕ

(5.69)

Note that in this relation the transpose is just an alternative way of writing a scalar product. The tangent stiﬀness matrix is identiﬁed by writing this in the form d(δV ) = δpT K dp.

(5.70)

For ease of notation the ﬁxed coordinate system is temporarily assumed to coincide with the element coordinate system at the start of the increment. The element stiﬀness matrix Ke will be described in the block matrix format 

Ke11  Ke21 Ke =   Ke31 Ke41

Ke12 Ke22 Ke32 Ke42

Ke13 Ke23 Ke33 Ke43

 Ke14 Ke24  , Ke34  Ke44

(5.71)

where Keij are 3 × 3 sub-matrices. A similar block matrix format is used for the rotation part Kr of the stiﬀness matrix. It follows from the formula (3.28) for the increment of the rotation variation ¯ = − 12 δ ϕ ¯ × dϕ ¯ d(δ ϕ)

(5.72)

that the two last terms of (5.69) give contributions Kr22 and Kr44 , respectively, of the form   mA 0 −mA z y 1  0 −mA Kr22 =  mA z x , 2 A A −my mx 0

  mB 0 −mB z y 1 . 0 −mB Kr44 =  mB z x 2 B B −my mx 0

(5.73)

The remaining contributions to the stiﬀness matrix can be calculated directly from dq. The increment of the element forces q in the ﬁxed frame of reference is obtained by taking the increment of (5.68), dq = S dt + (dSdl + dSdϕ¯ ) t.

(5.74)

In this expression the ﬁrst term corresponds to change in the internal ‘stresses’ due to a change in the deformation modes of the beam, while the last term ¯ accounts for a change of element length dl and element orientation dϕ. Although seemingly innocent, it is important to note that this step does not account for the eﬀect of the deformation modes for constant internal ‘stresses’. However, there is no means of representing this eﬀect without a

122

Co-rotating beam elements

ﬁnite deformation model of the beam, so it is left out for the moment to be clariﬁed at the end of the derivation of the tangent stiﬀness. The change in internal ‘stresses’ is governed by an incremental relation of the form dt = Kd dv,

(5.75)

where Kd is the tangent stiﬀness matrix of the deformation modes of the element, expressed in the reduced 6 × 6 internal format. As in the twodimensional case a virtual work argument shows that dv = ST dp,

(5.76)

where S and p are either both local or global. Substitution of this into (5.74) gives the element stiﬀness matrix in the form Ke = S Kd ST + Kr ,

(5.77)

where the ‘rotation’ matrix Kr contains the contributions (5.73) and the contributions from change of length and orientation given by the last term in (5.74). The format (5.77) is the same as that of the two-dimensional case, but here Kr contains additional contributions from the increment of the variational rotations δϕA and δϕB given by (5.73). The eﬀect of a change of length dl follows by diﬀerentiation of (5.67) as 

−nz 2 0 dSdl t = − 2  n z l 0

 ny 0  Mya dl. −ny  Mza 0

(5.78)

The shear force components Qy and Qz are now introduced from (5.66) to give 

   Qy ny + Qz nz Q   dl  0 0  = dl  dSdl t =  , l  −Qy ny − Qz nz  l −Q 0 0

(5.79)

where the shear force vector has been introduced as Q = Qy ny + Qz nz . A When the beam extension is introduced as dl = duB x − dux , the elongation is seen to contribute to the ﬁrst and seventh columns of the element stiﬀness matrix via the sub-matrix contributions Kr11

Kr33

−Kr13

−Kr31

1 = − l

0 Qy Qz

0 0 0 0 . 0 0

(5.80)

The shear force is seen to constitute the ﬁrst column of these sub-matrices. The ﬁnal contribution to the element stiﬀness matrix is from the rotation

5.2 Co-rotating beams in three dimensions

123

¯ of the local frame, denoted dSdϕ¯ t in (5.74). This implies that each of dϕ the unit vectors in the block format (5.67) of the transformation matrix S is rotated. However, rotating each of the base vectors while retaining their coeﬃcients corresponds to rotating the resulting vectors. Thus, the contribution from rotation may be expressed directly as  ¯ × fA dϕ ¯ × mA   dϕ dSdϕ¯ t =  dϕ . ¯ × fB  ¯ × mB dϕ 

(5.81)

The computation is similar for each of the four terms. It is illustrated by the force vector fB . The vector fB is expressed in an axial and a transverse component fB = N nx + Q,

(5.82)

where N is the normal force and Q is the shear force vector, orthogonal ¯ is also expressed in terms of its axial to nx . The incremental rotation dϕ and transverse components, ¯ = dϕ nx + l−1 nx ×∆u⊥ . dϕ

(5.83)

The axial component is determined by the average rotation about the x-axis, B dϕ = 12 (dϕA x + dϕx ), while the transverse rotation component is determined from the transverse component of the displacement diﬀerence between B A and A, ∆u⊥ = uB ⊥ − u⊥ . The representations (5.82) and (5.83) are now substituted into the cross product, and when using the triple product formula (3.25), the result takes the form ¯ × fB = nx×Q dϕ + l−1 N ∆u⊥ − l−1 nx (QT ∆u⊥ ). dϕ

(5.84)

When using the components of dϕ and ∆ u⊥ this expression directly gives the rotation contribution to the sub-matrices Kr31 , Kr32 , Kr33 and Kr34 . When it is observed that fA = −fB this also determines the contributions to Kr11 , Kr12 , Kr13 and Kr14 . The representation of the moment vector mB in terms of an axial and a transverse component is mB = M nx + mB ⊥.

(5.85)

The rotation cross product then follows from (5.84) as −1 −1 BT ¯ × mB = nx×mB dϕ ⊥ dϕ + l M ∆u⊥ − l nx (m⊥ ∆u⊥ ).

(5.86)

This determines the rotation contribution to the sub-matrices Kr41 , Kr42 , Kr43 and Kr44 . The similar result for mA follows by changing the superscript

124

Co-rotating beam elements

from B to A, and replacing M with −M . This determines the rotation contribution to the sub-matrices Kr21 , Kr22 , Kr23 and Kr24 . Initial non-symmetric form of Kr The matrix Kr representing the eﬀect of co-rotation of the local frame of reference can now be assembled from the three contributions: the increment of the variation of the nodal rotations (5.73), the extension of the element ¯ In this (5.80), and the eﬀect of the incremental rigid body rotation dϕ. process it is convenient to collect the sub-matrices in groups. The ﬁrst sub-matrix column Kri1 represents the eﬀect of the translation duA of node A, while the third sub-matrix column Kri3 represents the eﬀect of the translation duB of node B. It follows from invariance to a rigid body translation that Ki3 = −Ki1 . The contributions to the force increments dfA and dfB from the translations are given by the sub-matrices   0 −Qy −Qz 1 0 . Kr11 = Kr33 = −Kr13 = −Kr31 =  −Qy N l −Q 0 N z

(5.87)

This group of sub-matrices is seen to satisfy symmetry. The contributions to the moment increment dmA from translations are given by the sub-matrices Kr21

−Kr23

 0 mA y 1 = 0 M l 0 0

while the contributions to dmB are given by

 0 mB y 1 r r  K41 = −K43 = 0 −M l 0 0

 mA z 0 , M

(5.88)

 mB z 0 . −M

(5.89)

The sub-matrix columns Kri2 and Kri4 represent the eﬀect of the incremental rotation of node A and node B, respectively. The contributions to the force increments are given by the sub-matrices 

0 1 Kr12 = Kr14 = −Kr32 = −Kr34 =  Mya l Ma z

 0 0 0 0 , 0 0

(5.90)

where the anti-symmetric moments Mya and Mza have been introduced in place of the shear forces by use of (5.66). The contributions to dmA are given by  0 −mA z 1 r  K22 = 0 0 2 0 −M

 mA y M , 0

 0 1 Kr24 =  −mA z 2 mA y

 0 0 0 0 , 0 0

(5.91)

5.2 Co-rotating beams in three dimensions

while the contributions to dmB are given by  0 −mB z 1 r  K44 = 0 0 2 0 M

 mB y −M , 0

 0 1 Kr42 =  −mB z 2 mB y

125

 0 0 0 0 . 0 0

(5.92)

It is seen that the sub-matrix group given by (5.91)–(5.92) does not satisfy symmetry, and neither does the group (5.90) satisfy symmetry in combination with (5.88)–(5.89). The lack of symmetry is a consequence of the use of the virtual work principle on the rigid body motion. In rigid body motion only the total force and moment on the body contribute to equilibrium, and thus the distribution of forces and moments on the individual nodes of the element is not uniquely determined. This problem is addressed in the following section. Symmetric form of Kr The derivation of the tangent stiﬀness relation from the external virtual work may lead to a non-symmetric formulation, because the matrix Kr is constructed from extension and rigid body rotations alone. The need to use the internal virtual work of a full non-linear beam theory would to some extent defy the purpose of the co-rotating format, namely to derive a universal stiﬀness matrix Kr for a given class of elements, such as e.g. beam elements with two nodes, independent of the speciﬁc details of the local element model. It is therefore interesting that a systematic procedure can be devised to establish full symmetry by introducing terms corresponding to the missing eﬀect of the deformation modes in the derivation of Kr . The procedure relies on the fact that in a rigid body rotation the rotation ¯ A = dϕ ¯ B . Thus, for a rigid body increments at both ends are equal, dϕ motion any contribution of an element force or moment of the form ¯ A − dϕ ¯ B) Kdi (dϕ

(5.93)

will vanish. If the two rotations are not equal, their diﬀerence describes a deformation mode, and the result would represent a non-constitutive contribution from the deformation modes. Thus, non-constitutive contributions from the deformation modes can be identiﬁed by modifying the second and fourth sub-matrix columns of the non-symmetric Kr according to the format   r r d r r d K11

K12 + K1

K13

K14 − K1

 Kr Kr + Kd Kr Kr − Kd  22 2 23 24 2 Kr : =  21  Kr31 Kr32 + Kd3 Kr33 Kr34 − Kd3 Kr41

Kr42 + Kd4

Kr43

Kr44 − Kd4

  . 

(5.94)

126

Co-rotating beam elements

Here the assignment operator : = has been used to indicate the assignment of a new value in terms of the original deﬁnition. The four 3 × 3 matrices Kdi are deﬁned such that the ﬁnal matrix Kr becomes symmetric. The matrices Kd1 and Kd3 are determined directly by symmetry. The ﬁnal two matrices Kd2 and Kd4 are determined such that they make Kr24 symmetric with respect to Kr42 and make both the diagonal matrices Kr22 and Kr44 symmetric. The result is most easily obtained directly from (5.91)–(5.92) by moving the content of the matrices Kr24 and Kr42 to the matrices Kr22 and K344 on the diagonal, and then moving the anti-symmetric torsion moment terms the other way. The result is the fully symmetric form of the co-rotation matrix Kr given by the sub-matrices  0 1 Kr11 = Kr33 = −Kr13 = −Kr31 =  −Qy l −Q z

−Qy N 0

 0 1 r rT r rT  mA K12 = K21 = −K32 = −K23 = y l mA z  0 1 r rT r rT  mB K14 = K41 = −K34 = −K43 = y l mB z  0 1 Kr22 =  −mA z 2 mA y

−mA z 0 0

 mA y 0 , 0

0 M 0 0 −M 0

 0 1 Kr44 =  −mB z 2 mB y

 0 1  0 = Kr24 = KrT 42 2 0

0 0 −M

 −Qz 0 , N

 0 0 , M

(5.96)

 0 0 , −M

(5.97)

−mB z 0 0

 0 M . 0

(5.95)

 mB y 0 , 0

(5.98)

(5.99)

It is interesting to note that the symmetric form (5.95)–(5.99) is not uniquely determined. In fact, any symmetric matrix Kd0 can be added to Kr22 and Kr44 if it is also subtracted from Kr24 and Kr42 . The co-rotation procedure, in which the stiﬀness matrix is determined from the external virtual work of the element without speciﬁc account of the element deformation properties, does not provide any means of determining such a matrix Kd0 . However, a derivation of the geometric matrix from the general beam theory derived in Chapter 4 demonstrates the consistency of the present symmetric form.

5.2 Co-rotating beams in three dimensions

127

5.2.2 Element deformation stiﬀness The element deformation stiﬀness matrix Kd introduced in (5.75) contains the tangent stiﬀness of the deformation modes and their conjugate ‘stresses’. It may be represented with diﬀerent orders of accuracy, the simplest being the constitutive relation alone, the next including geometric terms. In two dimensions it was easy to obtain the linear geometric matrix from partial diﬀerentiation of the diﬀerential equation for bending. In three dimensions a complete linear geometric matrix must include coupling between bending and torsion, and therefore requires a more elaborate system of diﬀerential equations as a basis. In the following the geometric stiﬀness of the deformation modes is extracted from the general formulation of Chapter 4. Constitutive stiﬀness In the case of a straight hom*ogeneous beam the stiﬀness of the modes is uncoupled, and in terms of linear elastic parameters it may be written as 

  GJ dM  dMys     1  dMzs    dN  =     l  dM a   y a dMz



EIy EIz EA 3ψya EIy

 dϕ  dϕsy   s   dϕz   du .    dϕa  y dϕaz 3ψza EIz

(5.100)

GJ is the St. Venant torsion stiﬀness, and EA the axial stiﬀness. EIy and EIz represent the stiﬀness of symmetric bending about the y- and the zaxis, respectively. The last two coeﬃcients 3ψya EIy and 3ψza EIz are the stiﬀness of the anti-symmetric bending modes about the y- and the z-axis, respectively. The theory of shear ﬂexibility was given in Section 5.1.2. In the present notation for the three-dimensional problem the shear coeﬃcient of anti-symmetric bending about the y-axis is given by ψya =

1 , 1 + Φy

Φy =

12 EIy , l2 GAz

(5.101)

while the shear coeﬃcients for anti-symmetric bending about the z-axis are ψza =

1 , 1 + Φz

Φz =

12 EIz . l2 GAy

(5.102)

Here Ay is the eﬀective area of the shear force Qy , and Az is the eﬀective area of the shear force Qz . For curved or non-hom*ogeneous beams, coupling terms will appear in Kd . These terms are conveniently calculated by use of the complementary energy,

128

Co-rotating beam elements

using equilibrium moment distributions (Krenk, 1994). Coupling between torsion and extension for twisted beams can be accounted for by a coupling term (Krenk, 1983a,b). Local geometric stiﬀness In the computation of geometric stiﬀness it is often useful to consider the local motion as composed of rotation and strain. The rotation changes the orientation of the internal stresses in space while the strain mainly plays a role by permitting the geometric stiﬀness to be expressed in the form of internal virtual work. In the case of beams with ﬁnite rotations and small strains it is convenient to calculate the geometric stiﬀness based on the assumption of negligible shear strains. According to the general ‘elastica’ formulation in (4.7), the relation between the displacement derivative du = ¯ is given by d(du)/ds0 and the rotation of the beam cross-sections dϕ ¯ + dε. du = dϕ×x

(5.103)

When the transverse components of the incremental strain dε are neglected, the displacement derivative can be written in the form of an axial component du and a transverse component du⊥ , du = nx du + du⊥ ,

(5.104)

where the transverse displacement component du⊥ is expressed in terms of the transverse components of the rotation, ¯ du⊥ = dϕ×x .

(5.105)

The derivatives are with respect to the initial arc length s0 , and x(s0 ) describes the current conﬁguration. Thus, the unit vector nx is deﬁned by lnx = l0 x . The geometric stiﬀness follows from that part of the incremental virtual work that is proportional to the current value of the internal forces. This part of the incremental internal work was given in (4.35) for the general elastica beam theory. When the assumption of vanishing shear strains is introduced, this expression can be reduced by use of the relations for triple cross products. The following notation is introduced for a combination of internal forces, [ N ] = N I − 12 (N nTx + nx NT ),

(5.106)

where N = NT nx is the axial force of the beam. The relation (4.35) can

5.2 Co-rotating beams in three dimensions

129

then be written as d(δεj ) Nj + d(δκj ) Mj = − δu⊥T N (l0 /l) du − δu (l0 /l) NT du⊥ 1 ˆ T 0 M ¯ d ϕ 2 ¯ δϕ ¯] 1 + [ δϕ , ˆ (l/l0 )[ N ] dϕ ¯ M

(5.107)

where the terms with explicit dependence on the axial displacements have ˆ has been used for the equivalent skewbeen extracted, and the notation M symmetric matrix. In the present connection the integral over the length of the beam is needed for displacements and rotations representing the deformation modes of the beam. The internal forces N as well as the derivatives δu and du of the axial displacement are constant. Furthermore, the transverse displacements δu⊥ and du⊥ of the deformation modes of the beam vanish by deﬁnition at the beam ends, and therefore the two ﬁrst terms containing the displacements do not contribute to the integral. It is convenient to express the shape functions in terms of the non-dimensional coordinate ξ deﬁned by s0 =

1 2 l0 (1

+ ξ),

−1 ≤ ξ ≤ 1.

(5.108)

¯ 0 ) and its derivative with respect to s0 are The incremental rotation dϕ(s ¯ s and dϕ ¯ a , representing symmetric and antinow expressed in terms of dϕ symmetric deformation, respectively. There are three symmetric modes ¯ Ts = [dϕ¯sx , dϕ¯sy , dϕ¯sz ] corresponding to torsion and bending, respectively, dϕ ¯ Ta = [0, dϕ¯ay , dϕ¯az ]. while there are only two anti-symmetric bending modes, dϕ The polynomial shape function representation is ¯ (s0 ) dϕ ¯ 0) dϕ(s

I/l0 1 2ξ I

3ξ I/l0 1 2 4 (3ξ

− 1) I

¯s dϕ . ¯a dϕ

(5.109)

The bending modes can be represented as derivatives of the transverse displacement by (5.105), and it therefore follows that their integral over the element must vanish. For the symmetric mode this follows immediately from the representation as an odd function of ξ, while for the anti-symmetric mode it determines the relative magnitude of the quadratic and the constant term. The integration of the rotation terms follows after substitution of the representations (5.109). When it is observed that the internal forces N are constant within the beam element, while the moments M may contain linear

130

Co-rotating beam elements

and constant terms, the integration gives

l0 d(δεj ) Nj + d(δκj ) Mj ds0 0

¯ s δϕ ¯a ] = [ δϕ

1 12 l [ N ] 1 ˆ ¯T 4M

1 ˆ ¯ 4M

1 20 l [ N ]

¯s dϕ , ¯a dϕ

(5.110)

¯ is the mean value of the internal moment M(ξ) over the beam where M element, i.e. the value at its center. It is noted that the contributions of the internal forces N through the 1 1 and 20 already found for deformation modes appear with the coeﬃcients 12 the two-dimensional element in (5.42). In addition, there are non-trivial coupling terms between the symmetric and anti-symmetric deformation modes. These are important, e.g. for phenomena like torsional buckling.

5.2.3 Total tangent stiﬀness The element stiﬀness matrix is conveniently described in the block matrix format (5.71) in terms of sub-matrices Keij , i, j = 1, . . . , 4. When the individual four-block rows in the deﬁnition (5.67) of the transformation matrix S are denoted Si , i = 1, . . . , 4, the transformation from deformation modes to local element forces can be written in the form Keij = Si Kd STj + Krij

(5.111)

with i, j = 1, . . . , 4. The sub-matrices Krij were given in (5.95)–(5.99), and the deformation stiﬀness matrix Kd contains the constitutive terms given in (5.100) and the geometric contributions given by (5.110). The relation (5.111) is easily programmed directly, but the transformation contains a large number of zero entries and for the simple straight beam considered here the result takes a quite simple form when computed analytically. For the diagonal deformation stiﬀness matrix (5.100), the block matrices containing the constitutive stiﬀness in full element force format are   EAl2 0 0 1 , (5.112) 0 Ke11 = Ke33 = −Ke13 = −Ke31 = 3  0 12ψza EIz l a 0 0 12ψy EIy  GJ 1 Ke22 = Ke44 =  0 l 0

0 (3ψya +1)EIy 0

 0 , 0 (3ψza +1)EIz

(5.113)

5.2 Co-rotating beams in three dimensions   −GJ 0 0 1  (3ψya −1)EIy 0 Ke24 = Ke42 =  0 l 0 0 (3ψza −1)EIz

131

(5.114)

and Ke12 = Ke14 = Ke23 = Ke43 =



6 0 −Ke21 = −Ke41 = −Ke32 = −Ke34 = 2  0 l 0

0 0 −ψya EIy

 0 ψza EIz . 0

(5.115)

These relations are well known from the linear theory of beam bending. The total geometric contribution to the element stiﬀness matrix is found as the sum of the contributions from Kr given in (5.95)–(5.99) and the contribution (5.110) from the deformation modes. The result is expressed in terms of the normal force N , the shear forces [Qy , Qz ], the torsion moment A B B M , and the external bending moments at the nodes [mA y , mz ] and [my , mz ]:  0 1 Ke11 = Ke33 = −Ke13 = −Ke31 =  −Qy l −Q z

 0 1 e eT A  m = −K = −K = Ke12 = KeT y 21 32 23 l mA z 

0 1 B e eT m = −K = −K = Ke14 = KeT y 41 34 43 l mB z

Ke24

KeT 42

 0 1 lQy = 6 lQz

lQy − 15 lN −3M

−Qy 6 5N 0

 −Qz 0 , 6 5N

(5.116)

0 M 1 − 10 lN

 0 1 , 10 lN M

(5.117)

0 −M 1 − 10 lN

 0 1 , 10 lN −M

(5.118)

 lQz 3M , − 15 lN

(5.119)

 0 1 B Ke22 =  −2mA z + mz 6 B 2mA y − my

B −2mA z + mz 4 5 lN 0

B  2mA y − my , 0 4 lN 5

(5.120)

 0 1 B −2mz + mA = z 6 A − m 2mB y y

A −2mB z + mz 4 5 lN 0

A  2mB y − my . 0 4 5 lN

(5.121)

Ke44

In these expressions the shear force components Qy and Qz appear explicitly,

132

Co-rotating beam elements

although they can be expressed in terms of the external moments at the nodes as B Qy = −(mA z + mz )/l,

B Qz = (mA y + my )/l.

(5.122)

In particular, these relations have been used to obtain the present form of Kg22 and Kg44 . Example 5.4. Lateral buckling. Figure 5.12 shows a slender beam of length L with simple supports at both ends. The supports prevent translation and rotation about the axis of the beam. The load consists of identical bending moments of magnitude M , applied at the ends. For a slender beam the bending stiﬀness about a horizontal axis is large, and the deformation in bending is therefore small. However, as the bending stiﬀness about a vertical axis and the torsional stiﬀness are much smaller, the beam may fail by lateral buckling. For suﬃciently slender beams only little vertical deformation will develop, and the stability problem may be solved to a good approximation in linearized form as an eigenvalue problem. This example illustrates the ability of the cubic bending element to capture this phenomenon in a model with two elements. The shear ﬂexibility for bending in the horizontal plane is assumed negligible, i.e. ψza = 1.

Fig. 5.12. Lateral buckling of slender beam with simple supports.

Let the beam be represented by two identical elements of length l = 12 L. Due to symmetry of the problem only the element AB representing the left half of the beam need be considered. The buckling mode involves a combination of torsion with bending out of the plane of the beam. This deformation B B is described by the nodal displacement components [dϕA 3 , du2 , dϕ1 ]. These components correspond to the full element displacement components with index 6, 8 and 10, respectively. The reduced constitutive matrix follows from (5.112)–(5.115) as 

4EI3 /l Ke =  −6EI3 /l2 0

−6EI3 /l2 12EI3 /l3 0

 0 0 . GJ/l

5.2 Co-rotating beams in three dimensions

133

There is no axial or shear force in the beam, and the reduced geometric stiﬀness matrix then follows from (5.116)–(5.121) as

Kg =

0 0 0 0 0 M/l

0 M/l . 0

When combining these matrices into an eigenvalue problem it is seen that B the ﬁrst row constitutes a relation between dϕA 3 and du2 that does not involve the load M , duB 2 =

A 2 3 l dϕ3 .

This relation deﬁnes the shape of the lateral bending. When this relation is used to eliminate the component dϕA 3 , the remaining eigenvalue problem is B 3EI3 /l M/l du2 0 = . B dϕ1 M/l GJ/l 0 The eigenvalue is √ M = ±

EI3 GJ = ±

1.732 EI3 GJ. l

The analytical solution of the linearized lateral buckling problem gives the factor as π/L = 1.571/l, see e.g. Timoshenko and Gere (1961). Thus, in this case a single element gives a fair result. However, the accuracy is limited by the low-order representation of the torsion angle, and for accurate solution of torsional and lateral buckling problems several elements should be used.

5.2.4 Finite element implementation In three dimensions rotations are no longer algebraically additive. The current rotation of the nodes is calculated by tracing the equilibrium path of the structure as in the case of truss structures or two-dimensional beams. The rotation of a node A can be represented in several ways, either by the pseudo-vector ϕA , by a set of quaternion parameters (rA , rA ), or by the rotation matrix RA . The ﬁnite element formulation presented here and in ¯ A in the form of an inﬁnitesimal Chapter 4 is based on rotation increments dϕ vector. In the computation this is replaced by a small but ﬁnite increment ¯ A . The individual components of this representation are non-additive dϕ as explained in Chapter 3, and it is therefore convenient to represent the accumulated current rotation by its quaternion parameters (rA , rA ). The rotation increment is represented by quaternion parameters (drA , drA ), and

134

Co-rotating beam elements

it is consistent with the linearization implied by the tangent stiﬀness formulation to introduce the ﬁrst-order approximation 1/2 ¯ A, drA = 12 dϕ drA = 1 − drTA drA (5.123) following from the quaternion deﬁnition (3.48). The quaternion representation of the accumulated rotation is then updated by the non-linear quaternion addition formula (3.63) as rA : = drA rA − drTA rA ,

rA : = drA rA + rA drA + drA × rA . (5.124)

The quaternion representation (rA , rA ) of the node A can then be used to extract the total rotation components ϕA or to form the corresponding rotation tensor RA by use of (3.50) or (3.52). In the two-dimensional co-rotating beam problem the local element frame of reference is fully determined by the end-points of the element. In the three-dimensional problem the end-points only determine the location of the local x-axis, while the orientation of the y- and the z-axis must be determined by use of the rotation of the nodes. The orientation of the beam coordinate system in space can be formulated in two ways: by updating its position incrementally, or by placing it with reference to the total accumulated rotations of the beam element nodes. In the incremental procedure a linearized form of the rotation increments can be used. However, this in turn implies the use of steps with limited magnitude of the rotation increments. The formulation in terms of total accumulated rotation is independent of the size of the individual load steps, but must then be formulated in terms of ﬁnite rotations. For most cases of practical analysis the incremental method will be suﬃcient. Incremental formulation The co-rotating beam is described in a local frame of reference, deﬁned by the base vectors [nx , ny , nz ]. In the incremental procedure this set of base vectors is updated from its value in the last step by use of the displacement ¯ A , dϕ ¯ B of the beam increments duA , duB and the rotation increments dϕ nodes A and B. The symmetric deformation modes, shown in the top row of Fig. 5.11, are described in terms of the local components of the diﬀerence between the rotations of node B and node A. When the current element base vectors are used to deﬁne the rotation matrix R = [nx , ny , nz ], the linearized form of the symmetric rotation increment is ¯ B − dϕ ¯ A ). dϕs = RT (dϕ

(5.125)

5.2 Co-rotating beams in three dimensions

135

The pre-multiplication by RT performs projection of the global components on the local base vectors, thereby yielding the local components dϕTs = [dϕsx , dϕsy , dϕsz ]. The corresponding anti-symmetric rotation components are found from the sum of the nodal rotation increments by subtracting twice the incremental rotation of the beam axis, ¯ B + dϕ ¯ A ) − 2nx ×(duB − duA )/l]. dϕa = RT [(dϕ

(5.126)

¯ = The last term deﬁnes the incremental rotation of the beam axis, dϕ T nx ×(duB − duA )/l. Pre-multiplication by R gives the local components dϕTa = [dϕax , dϕay , dϕaz ]. The component dϕax does not have a counterpart in the two-dimensional theory, as it does not contribute to the deformation of the beam, but describes the rigid body rotation around the beam axis.

Fig. 5.13. Rotation of element base vectors [nx , ny , nz ] to element axis ∆x.

The element base vectors [nx , ny , nz ] are updated by a rotation of magnitude dϕax around the beam axis, followed by a rotation of the full basis to align the base vector nx with the updated beam axis as shown in Fig. 5.13. The rotation of a set of base vectors through the minimum angle bringing one of the vectors into a given new direction was described in Section 3.2. In the present case the vector nx is to be rotated into the direction of the beam axis, described by the unit vector ∆x/l. First mean direction is deﬁned by the unit vector n = (nx + ∆x/l)/|nx + ∆x/l|.

(5.127)

The full basis can then be updated by a reﬂection in the plane orthogonal to n, [ nx , ny , nz ] = ( I − 2 n nT )[ −nx , ny , nz ].

(5.128)

For the vector nx being realigned the transformation also involves a change of sign. When the deformation parameters vT = [ ϕsx , ϕsy , ϕsz , u, ϕay , ϕaz ] have been determined in the element frame of reference, the internal forces t follow from the constitutive equation (5.100). The nodal forces q are found from

136

Co-rotating beam elements Algorithm 5.2. Incremental update of 3D co-rotating beam elements. Element extension: u = l − l0 Symmetric and anti-symmetric rotation increments: R = [ nx , ny , nz ] ¯ B − dϕ ¯ A) dϕs = RT (dϕ

(5.125)

¯ B + dϕ ¯ A ) − 2nx ×(duB − duA )/l] dϕa = R [(dϕ

(5.126)

ϕs = ϕs + dϕs ,

ϕa = ϕa + dϕa

Rotate element basis around axis: cos(dϕax ) [ ny , nz ] = [ ny , nz ] sin(dϕax )

− sin(dϕax ) cos(dϕax )

Rotate basis to new element axis and update R: n = nx + ∆x/l,

n = n/|n|

[ nx , ny , nz ] = ( I − 2 n nT )[ −nx , ny , nz ]

(5.128)

R = [ nx , ny , nz ] Internal element forces: t = Kd v

(5.75)

Nodal element forces in global frame: q = Re S t

(5.63), (5.68)

Element tangent stiﬀness in global frame: Kij = R Keij RT

(5.129)

(5.68) and transformed to global components by use of the updated element basis R = [nx , ny , nz ]. The local components of the element stiﬀness matrix are given in terms of 3×3 block matrices by (5.112)–(5.122) and transformed to global components by use of the updated element basis R, Kij = R Keij RT ,

i, j = 1, . . . , 4.

(5.129)

This completes the incremental update of element deformations, local frame of reference, nodal element forces and element stiﬀness matrix. The procedure is summarized in pseudo-code format as Algorithm 5.2. The update is based on linearized rotation increments, and can therefore be carried out independently from the calculation of the total rotation of the nodes.

5.2 Co-rotating beams in three dimensions

137

Total formulation At the end of each load increment the new orientation in space of each beam element must be determined. In the incremental procedure described above this was done in terms of linearized rotation increments. This leads to a fairly simple procedure that is independent of any description of the accumulated nodal rotations. The price of this independence is the need for suﬃciently small steps to justify the linearization of the rotation increments. Alternatively, the orientation of the beam element in space as well as its deformation parameters can be determined from the current position and total rotation of the nodes. In such a procedure the element axes are deﬁned by coincidence of the base vector nx with the element axis, and orientation of the base vectors ny , nz via a suitable mean value of the nodal rotations. A procedure based on total displacements and rotations can be implemented in several ways. The present procedure generalizes the incremental formulation by using a quaternion representation of the rotation of the nodes. Thus, the ﬁrst step is to update the quaternions (rA , rA ) and (rB , rB ) for the rotations of the nodes by use of (5.123)–(5.124). The updated node quaternions are used to ﬁnd a mean rotation quaternion (r, r) and the quaternion (s, s) deﬁning the rotation from node A to the mean, and from the mean to B. The mean and diﬀerence quaternions were discussed in Section 3.5, and explicit formulae were given in (3.70), (3.71) and (3.75). They are included in the pseudo-code Algorithm 5.3 for the total formulation. The updated set of base vectors [nx , ny , nz ] is found from the initial element basis [n0x , n0y , n0z ] in two steps. First the initial basis is rotated by the mean rotation of the nodes [nx , ny , nz ] = R(r, r)[n0x , n0y , n0z ],

(5.130)

where the rotation tensor R(r, r) is deﬁned in terms of the mean quaternion (r, r) in (3.50). This step produces the intermediate base vectors [nx , ny , nz ] shown in Fig. 5.13. This intermediate basis is then rotated through the smallest angle that aligns the vector nx with the beam axis ∆x. This step is identical to that of the incremental formulation. First a unit vector n is deﬁned by (5.127) as the mean between the initial and ﬁnal position of nx , and the rotation of the basis is then set up as a reﬂection in the plane orthogonal to n by (5.128). This gives the updated element base vectors, and a corresponding updated rotation tensor R = [nx , ny , nz ]. The deformation of the element consists of the extension u = l − l0 , three symmetric and two anti-symmetric rotation components. The extension follows directly from the displacements of the element nodes, while the ﬁve

138

Co-rotating beam elements Algorithm 5.3. Total update of 3D co-rotating beam elements. Scalar part of diﬀerence quaternion:

1/2 s = 21 (rA + rB )2 + | rA + rB |2

(3.71)

Mean rotation quaternion: r =

1 2 (rA

+ rB )/s,

r =

1 2 (rA

+ rB )/s

(3.70)

Vector part of diﬀerence quaternion: s =

1 2 (rA rB

− rB rA + rA ×rB )/s

(3.75)

Rotate initial element basis: [ nx , ny , nz ] = R(r, r)[ n0x , n0y , n0z ]

(3.50), (5.130)

Rotate basis to element axis and redeﬁne R: n = nx + ∆x/l,

n = n/|n|

[ nx , ny , nz ] : = ( I − 2 n nT )[ −nx , ny , nz ]

(5.128)

R = [ nx , ny , nz ] Symmetric and anti-symmetric rotation components: ϕs = 4 RT s

(5.131)

ϕa = 4 R (nx × n)

(5.132)

Element extension: u = l − l0 Internal element forces: t = Kd v

(5.100)

Nodal element forces in global frame: q = Re S t

(5.63), (5.68)

Element tangent stiﬀness in global frame: Kij = R Keij RT

(5.129)

remaining components depend on the rotations. The symmetric deformation modes shown in the top row of Fig. 5.11 are described by the diﬀerence between the rotation of nodes A and B. The vector part of the diﬀerence quaternion (s, s) describes half the angle 21 ϕs in the global frame of reference. This angle is a relative angle, describing the element deformation and the sine function involved in the deﬁnition of the quaternion can therefore be linearized, whereby ϕs 4 s.

(5.131)

The local components are obtained by pre-multiplication with RT . The anti-

5.3 Summary and extensions

139

symmetric part of the rotations arises because the intermediate base vectors, deﬁned by the mean of the node rotations and shown in Fig. 5.13, are generally not aligned with the beam axis. In the algorithm it is inconvenient to store the intermediate basis, and the rotation required for alignment is therefore calculated from the mean vector n, deﬁned by (5.127). When linearizing the involved sine function, the anti-symmetric rotation components are deﬁned by the vector product ϕa = 4 nx × n.

(5.132)

The local components of the anti-symmetric rotations are found by premultiplication with RT . The vector product (5.132) produces a vector that is orthogonal to the beam axis, and thus the local component ϕax vanishes, leaving only the two components describing anti-symmetric bending. The beam deformations are included in Algorithm 5.3, together with the calculation of nodal forces and the tangent stiﬀness matrix.

5.3 Summary and extensions In the co-rotating formulation the total motion is separated into two parts: the motion of the element-based local co-rotating frame of reference, and the deformation of the element in this local frame. The advantage of the method is that the contributions from the motion of the local frame are independent of the particular element formulation, and the element formulation is limited to the deformation modes in the local frame of reference. These local deformation modes can be modeled at diﬀerent levels of sophistication, ranging from simple linear theory, over small ﬁnite deformation theories with simple geometric stiﬀness, to fairly advanced theories with non-linear column eﬀects and shortening due to bending. In Sections 5.1.2 and 5.2.2 the local deformation modes were modeled at the intermediate level corresponding to including geometric stiﬀness eﬀects based on cubic shape functions. This puts certain restrictions on the eﬀects that can be represented accurately within an individual element. Several extensions of the co-rotating beam theory have found application for oﬀshore structures where column eﬀects, initial imperfections, etc. are important (Skallerud and Amdahl, 2002). A brief indication of the basic idea of a beam-column with special eﬀects like shortening due to bending and the option of initial curvature is given here. The basic idea is to consider the axial force N , or the associated axial strain ε, as the source of geometric eﬀects (Oran, 1973a,b). This suggests an equilibrium formulation for a

140

Co-rotating beam elements

beam-column element of the form N = EA ε,

M = A(ε) ϕ,

(5.133)

where N and ε are the normal force and the corresponding axial strain, while M and ϕ are arrays containing the moments and the corresponding rotation angles describing the natural deformation modes shown in Fig. 5.11. The axial stiﬀness EA is assumed constant, while the moment stiﬀness matrix A(ε) is a function of the normal force or strain. It is an important consequence of the format (5.133) that it implies a relation between the axial strain ϕ and the elongation u of the beam element of the form u = l ε − 12 ϕT B(ε) ϕ,

(5.134)

where the last term represents the apparent shortening due to bending. The change of separation of the end-points of the beam with strain then follows as ∂u dB = l − 12 ϕT ϕ = L, (5.135) ∂ε dε where L appears as the ‘apparent length’ of the beam-column element with respect to the extension–strain relation. It is an important aspect of the present beam-column theory that the ‘shortening matrix’ B(ε) is determined explicitly by the moment stiﬀness matrix A(ε) (Krenk, 1995b). The energy of the beam element follows from ﬁrst increasing the normal force to its ﬁnal value while maintaining ϕ = 0, and then bending the beam with ε constant. This gives the internal energy Φ(ε, ϕ) =

1 2l ε N

+ 12 ϕT M − 12 ϕT B ϕ N.

(5.136)

The normal force N is conjugate to the extension given by u, and thus 1 ∂Φ ∂Φ ∂ε = . (5.137) ∂ε ∂u L ∂ε Diﬀerentiation of the expression (5.136) gives dA ∂Φ − EA B ϕ. (5.138) = L N + 12 ϕT dε ∂ε It follows from a comparison of the two last equations that consistency requires the matrix B to be deﬁned by N =

dA 1 dA = . (5.139) dN EA dε Thus, the beam-column theory is completely deﬁned by the constitutive relations (5.133) and the kinematic relation (5.134) with the matrix B deﬁned as the derivative of the moment stiﬀness matrix A(N ). B =

5.4 Exercises

141

The constitutive relations (5.133) are in total form. In the computation also the tangent stiﬀness relation is needed. It follows by diﬀerentiation of the total relations, using the strain–elongation relation (5.135), EA dN 0 0T 1 du T = [1, ϕ B] . (5.140) + dM Bϕ dϕ 0 A L It is seen that the apparent length L deﬁned by (5.135) acts as the length of the beam-column in the incremental stiﬀness relation. The matrix in the parentheses corresponds to the element deformation matrix Kd for which the linearized theory was given in Section 5.2.2. The components have been reordered to accommodate the special status of the normal force in the present beam-column theory. The details of the formulation for the classic beam-column shape functions with added shear ﬂexibility have been given by Krenk (1995b) and Krenk et al. (1999). A number of special eﬀects can be incorporated into the beam-column theory. Initial imperfections in the form of ‘out of straightness’ can be included explicitly in the formulation by adding a hypothetical step in which the beam-column element is straightened at zero normal force by application of suitable end moments. This is then accounted for by explicit terms in the bending relation (5.133b) and the strain–elongation relation (5.134). This feature enables representation of the column eﬀect for design purposes without additional degrees of freedom. Many frame structures are designed to accommodate a certain amount of yielding – typically at joints. This eﬀect can be incorporated into the individual beam-column element by including the extra local deformation at the nodes in the form of plastic hinges (Ueda and Yao, 1982; Powell and Chen, 1986). Furthermore, plastic hardening can be accounted for with the extent of the plastic yield zone estimated from the variation of the moments along the element (Fujikubo et al., 1991). A particularly simple formulation is obtained when using a ﬂexibility formulation for the local element properties as in Section 5.1.2. In that case the local plastic straining appears as a purely additive contribution to the ﬂexibility matrix (Krenk et al., 1999). By including these additional features the corotating beam formulation can be developed into a highly eﬃcient analysis tool for frame structures.

5.4 Exercises Exercise 5.1 Consider a two-dimensional beam element in the x1 x2 -plane as shown in Fig. 5.1. The bending deformation modes are described in terms of the parameters ϕs and ϕa as illustrated in Fig. 5.3. Denote the end-points

142

Co-rotating beam elements

of the beam by A and B, and let their incremental transverse displacement A B B and rotation be given by [duA y , dϕ ] and [duy , dϕ ], respectively. These four displacement components include a rigid body translation du02 and a rigid body rotation dϕ03 . Show that the incremental rigid body motion is found as 0 A duy 1 1 duy = . 0 B −1/l

dϕ3

1/l

duy

Show that the deformation modes are found by subtracting the rigid body rotation of the beam ends to give  A duy 1 1 1 A 0 − 0 dϕ dϕ s   2 2 2 = .  B 1 1 1 2 dϕa

1/l

−1/l

duy dϕB

These relations are special cases of the transformation relations (5.13a) between deformation modes and local nodal displacements. Exercise 5.2 Use the deﬁnition of the co-rotation stiﬀness matrix Kr dpe = dS + RTe dRe S t to evaluate its components in the two-dimensional case and obtain (5.23)– (5.24). It is convenient ﬁrst to express the ﬁrst term in terms of (N, Ms , Ma ) and dl and the second term in terms of (N, Ms , Ma ) and dϕ. The increments dl and dϕ are then introduced from (5.20) and the expression rearranged in matrix format. Exercise 5.3 Consider a plane straight beam element with end-points A and B. Let n1 denote a unit vector in the direction from A to B and n2 a transverse unit vector. In the local frame of reference these vectors have the components nT1 = [1, 0, 0] and nT2 = [ 0, 1, 0]. Show that the basic block matrix Kr11 of the co-rotation stiﬀness matrix Kr , given in (5.24), can be expressed in the form N Q n2 nT2 − (n1 nT2 + n2 nT1 ) l l when the unit vectors n1 and n2 are expressed in the local coordinate system. Explain why the formula remains valid and gives the co-rotation part of the stiﬀness matrix in a general coordinate system, when the unit vectors are expressed in this coordinate system. Exercise 5.4 Show that the bending part of the constitutive stiﬀness matrix for the deformation modes given by (5.33) with ψa = 1 can be derived from the expression (5.36) for the incremental virtual work with the shape Kr11 =

5.4 Exercises

143

functions for the symmetric and anti-symmetric bending modes given by (5.38) and (5.39). The advantage of the ﬂexibility method based on energy is that it allows inclusion of the shear ﬂexibility eﬀect as a simple additional term. Exercise 5.5* The geometric stiﬀness matrix is formulated for a state of equilibrium. During iterations the state of the individual elements does not correspond to equilibrium of the structure. In some cases it is found that retaining the last established equilibrium stress state in the geometric stiﬀness matrix during the equilibrium iterations reduces the number of iterations. Perform the calculations of Example 5.2 with full update of the internal stresses and with update only at equilibrium. Compare and comment on the resulting number of iterations. Exercise 5.6* Figure 5.14 shows an initially square frame with side length L and member bending stiﬀness EI. Use the two-dimensional co-rotating element formulation of Section 5.1 to analyze the tension and compression problem in the interval 0 ≤ P L2 /EI ≤ 10. Make plots of u/L, v/L and ϕ as a function of the load.

(a)

(b)

Fig. 5.14. Diamond shaped frame: (a) tension, (b) compression.

Table 5.2. Displacements of diamond-shaped frame. P L2 /EI

u/L

Tension v/L

1 2 5 10

0.13960 0.23184 0.37322 0.46601

0.11252 0.16429 0.21931 0.24380

1.05144 1.20263 1.40209 1.50351

u/L 0.17046 0.24224 0.07735 −0.12724

Compression v/L 0.24754 0.58236 1.08927 1.30578

ϕ 0.32789 −0.19539 −0.98149 −1.34277

144

Co-rotating beam elements

This problem was initially analyzed by Jenkins et al. (1966) using elliptic integrals. Accurate results, tabulated by Mattiasson (1981), are given in Table 5.2. Discuss the accuracy of the ﬁnite element solution with two and four elements per side member of the frame. Exercise 5.7* The results for the angle beam in Fig. 5.9 were obtained for 1 the parameters A = 1 and I = 12 corresponding to a square cross-section of unit side length and a frame half-width of b = 50. If the height at the center of the angle beam is increased above the value h = 1.2 used in Example 5.3 the load–displacement curve becomes non-monotonic, indicating a snap-through. Make a ﬁnite element model for the case h = 1.6, and obtain the solution by applying a vertical spring at the apex as explained in Exercise 1.5. More eﬃcient solution techniques are developed in Chapter 8. Exercise 5.8 The procedure described in Algorithm 5.2 for the calculation of the element rotation matrix R and the deformation parameters ϕ, ϕsy , ϕsz , u, ϕay , ϕaz can be formulated in several alternative ways. The important thing is that the total rotation of the element is described in terms of accurate formulae for ﬁnite rotations, while diﬀerences within the element may be handled approximately, assuming small angles. Describe and illustrate an alternative procedure, in which the rotation maA A B B B trices RA and RB deﬁne nodal base vectors [nA x , ny , nz ] and [nx , ny , nz ] B that are rotated such that nA x and nx coincide with the beam axis, represented by nx , see Fig. 5.12. In this procedure the element base vectors and the deformation parameters are then deﬁned by mean values and diﬀerences between these base vectors at the nodes A and B. Exercise 5.9 Deﬁne the nodal displacement vector du = [ duA , dϕA , duB , dϕB ] corresponding to a rigid body rotation dϕ0 and ﬁnd the corresponding nodal forces Kr du generated by the co-rotation stiﬀness matrix given by (5.95)–(5.99). Show that the eﬀect corresponds to rotating the force at each end by dϕ0 and the moment at each end by 12 dϕ0 , corresponding to the semi-tangential property of the moment. Recall the explanation given in Section 3.3 of the fact that moments appear to rotate by only half when treated within an incremental form of a virtual work equation. Exercise 5.10 In the co-rotating formulation the total geometric stiﬀness matrix is considered as consisting of two parts Kg = Kr +Kl , where Kr is the co-rotating part determined by the procedure of Section 5.2. Determine the local part Kl , and show that the contribution from the local part vanishes for a rigid body displacement.

6 Deformation and equilibrium of solids

Kinematic non-linearity is a recognized part of continuum mechanics often termed ‘large’ or ‘ﬁnite’ displacement theory. The non-linearity arises because equilibrium is considered in the current, and initially unknown, state of the body. In order to describe ﬁnite deformation of a continuous body it is necessary to have a non-linear measure of strain and a stress deﬁnition that can be used in the deformed state. It turns out that the Green strain, introduced for axial strain in Chapter 2, can be generalized to multi-component form describing the deformation of a continuous body. This is the subject of Section 6.1. For any continuum mechanics theory it is very desirable to use stresses and strains that satisfy some form of virtual work principle. It was demonstrated in Chapter 2 that the use of the Green strain, which was a convenient quadratic strain measure with exact invariance with respect to arbitrary rigid body motion, led to a slightly modiﬁed interpretation of the normal force N appearing in the principle of virtual work. In a similar way the use of the Green strain for a continuous body leads to a special stress deﬁnition, the second Piola–Kirchhoﬀ stress. For small strains this stress deﬁnition has a simple physical interpretation, precisely as N in the case of a bar element. This stress is introduced in Section 6.2, and it is demonstrated how it serves as a convenient reference for other stress measures of practical importance. In particular the change of stress, the so-called stress rate, is discussed with a view to its use in plasticity theory in Chapter 7. The change of virtual work as an instrument to study a change of state in a solid body is presented in Section 6.3. This establishes the equilibrium equations in a form suitable for ﬁnite element representation with the external forces deﬁned via the external work and the internal forces via the internal work. For elastic materials an integral to the virtual work can be found and identiﬁed as the potential energy of the body. The principle of virtual work 145

146

Deformation and equilibrium of solids

is the instrument for checking equilibrium and deﬁning the ‘out of balance’ forces if equilibrium has not been established. In order to approach equilibrium in the case of unbalance a representative stiﬀness is needed. Again the tangent stiﬀness is particularly useful because it represents the exact solution for very small changes, and because the tangent stiﬀness matrix for elastic bodies is symmetric. This is discussed in Section 6.4. Up until this stage all relations have been kept in general form with arbitrary displacement ﬁelds. In practice, the discretization of the problem in terms of ﬁnite elements is a very important step. In Section 6.4.3 the introduction of shape functions and their role in the formulation of geometrically non-linear isoparametric solid elements is discussed. The results appear as a systematic generalization of the linear isoparametric element. There are numerous books on continuum mechanics, many of which cover stresses and strains in ﬁnite deformation theory. Malvern (1969), Gurtin (1981) and Truesdell (1991) are standard references presenting a broad overview, while Holzapfel (2000) provides an introduction to more recent developments. The presentation of deformation and equilibrium of solids given here is inﬂuenced by Washizu (1974).

6.1 Deformation and strain Figure 6.1 shows a continuous body in the initial undeformed conﬁguration, serving as reference conﬁguration. In this conﬁguration each point of the body has a position vector x0 . A set of orthogonal unit base vectors i1 , i2 , i3 is introduced, forming a Cartesian coordinate system. The initial position vector can then be written in component form as x0 = x01 i1 + x02 i2 + x03 i3 .

(6.1)

In the following the components will either be given in vector notation as xT0 = [x01 , x02 , x03 ], or in index form as x0α , where the range of the subscript α is 1, 2, 3 for three-dimensional problems and 1, 2 for two-dimensional problems. Greek subscripts will be reserved for spatial components, and thus always have a range corresponding to the spatial dimension. In terms of the index notation the coordinate decomposition (6.1) is written as x0 = x0α iα .

(6.2)

The fact that the subscript α occurs twice in the same term is used to indicate a sum over the range of α. This so-called summation convention is a part of tensor analysis, the theory of vectors and their combination in space, see e.g. Malvern (1969), but its use here will be quite straightforward.

6.1 Deformation and strain

147

Fig. 6.1. Initial conﬁguration with orthogonal unit base vectors iα .

Each coordinate set x0α deﬁnes a material particle by its position in the initial conﬁguration. Now, each material particle is displaced from its initial position x0 to its current position x by the displacement vector u, see Fig. 6.2, x = x(x0 ) = x0 + u.

(6.3)

In the following each point is identiﬁed by its material coordinates x0γ . Thus, the coordinates xα of the current position are considered as functions of the material coordinates x0γ , and similarly for the displacement components. The coordinate relation corresponding to (6.3) then is xα (x0γ ) = x0α + uα (x0γ ).

(6.4)

This type of formulation in which the displacement, and thereby the current position, is expressed in terms of the coordinates of the initial position is called material or Lagrangian. It is the commonly used formulation in solid mechanics, whereas ﬂuid mechanics problems are often formulated in terms of the current position giving a spatial or Eulerian description.

Fig. 6.2. Displacement u from initial conﬁguration x0 to current conﬁguration x.

148

Deformation and equilibrium of solids

6.1.1 Non-linear strain A strain measure in a solid body must characterize the state of deformation at each point of the body. This includes the ability to describe diﬀerent elongations in diﬀerent directions and also that vanishing strain should correspond to undeformed rigid body motion. It turns out that such a strain measure can be obtained by generalizing the simple length formula (2.13) for the axial Green strain. In the case of a continuum strain is a point property, and therefore only inﬁnitesimal lengths can be used. Also the direction in which the length is measured must be incorporated. Consider a vector dx0 of inﬁnitesimal length ds0 in the initial conﬁguration, shown in Fig. 6.3. The length ds0 is determined by ds20 = dxT0 dx0 .

(6.5)

The deﬁnition of ds0 as the length of the vector dx0 implies that the normalized vector dx0 /ds0 is a unit vector, indicating the direction being considered.

Fig. 6.3. Motion of inﬁnitesimal vector dx.

The current coordinates dx of the initial vector dx0 are deﬁned by the partial derivatives of the function x(x0 ) in (6.3) as dx = F dx0 ,

(6.6)

where F is the deformation gradient tensor, with components   0 0 0 ∂x1 /∂x1

∂x1 /∂x2

∂x1 /∂x3

∂x3 /∂x01

∂x3 /∂x02

∂x3 /∂x03

F =  ∂x2 /∂x01 ∂x2 /∂x02 ∂x2 /∂x03  .

(6.7)

The partial derivatives of the displacement vector u are arranged similarly

6.1 Deformation and strain

149

in the displacement gradient tensor D with components   0 0 0 ∂u1 /∂x1

∂u1 /∂x2

∂u1 /∂x3

∂u3 /∂x01

∂u3 /∂x02

∂u3 /∂x03

D =  ∂u2 /∂x01 ∂u2 /∂x02 ∂u2 /∂x03  .

(6.8)

It follows from the linear relation (6.3) that the deformation gradient tensor can be expressed in terms of the displacement gradient tensor as F = I + D,

(6.9)

where I is the unit tensor with components given by the unit matrix. The deformation gradient components (6.7) have a very direct interpretation. Consider the unit vector i1 = dx0 /dx01 = [1, 0, 0]T along the x1 -axis. It follows from (6.6) that the current position of this vector is given by the ﬁrst column in the deformation gradient matrix (6.7). Thus, the three columns ∂x/∂x0α of the deformation gradient matrix F give the current components of the three original unit vectors i1 , i2 and i3 . This relation is illustrated in Fig. 6.4, showing displacement and deformation of the three edges of a unit cube, originally aligned with the coordinate axes. It is seen that the displacement vector describes a translation, while the nine components of the deformation gradient combine a rotation and deformation of the cube. While the deformation depends on the loads and the constitutive behavior of the material, the rotation does not change the state of the material, and it is therefore important to separate these two parts of the motion.

Fig. 6.4. Displacement vector u and deformation gradient F.

Clearly, a change of the length between two points of a solid body is independent of any rigid body motion. It therefore appears natural to investigate the axial Green strain εG using a generic line element with the initial vector representation dx0 of inﬁnitesimal length ds0 . The current vector representation is dx with length ds. The axial Green strain introduced in Chapter 2

150

Deformation and equilibrium of solids

is εG =

ds2 − ds20 dxT dx − dxT0 dx0 = . 2 ds20 2 ds20

(6.10)

Substitution of the current increment dx from (6.6) into this expression yields the axial Green strain as dxT0 ds0

εG =

1 2

dx0 T . F F−I ds0

(6.11)

The ﬁrst and last factor deﬁne the initial direction, and the factor in the middle is now deﬁned as the Green strain tensor, E =

T 1 2 (F F

− I) =

1 2 (D

+ DT ) + 21 DT D.

(6.12)

The second expression in terms of the displacement gradient follows from use of (6.9). It is observed that the Green strain tensor E can be written as the symmetric part of 12 (F + I)T D = FT1/2 D, similar to the notion of a mean state introduced in Section 2.2. Substitution of the components of the displacement gradient tensor (6.8) gives the components of the Green strain tensor, ! " ∂uβ 1 ∂uγ ∂uγ 1 ∂uα + + . (6.13) Eαβ = 0 0 2 ∂xβ ∂xα 2 ∂x0α ∂x0β The components of the Green strain tensor constitute a symmetric matrix where each term is a sum of a linear and a quadratic term. The linear term is identical to that of ‘small’ displacement theory. The components have a simple interpretation as the change of the scalar products of the base vectors before and after the motion. As illustrated in Fig. 6.4, the initial base vectors are ∂x0 /∂x0α while their current coordinates are ∂x/∂x0α . The components of the Green strain tensor express the change in the scalar products of the corresponding base vectors, 2Eαβ =

∂xT ∂x ∂xT0 ∂x0 − . ∂x0α ∂x0β ∂x0α ∂x0β

(6.14)

Hereby the diagonal elements of 2Eαβ correspond to the increase of the square of the length of the unit vectors, while the oﬀ-diagonal elements describe the cosines to the angles between the base vectors after the motion. However, this relation is somewhat indirect due to the change of the length of the base vectors. For small deformation the Green strain tensor components correspond to those of linear strain. In order to establish the principle of virtual work, the variation δE of the

6.1 Deformation and strain

151

Green strain tensor E is also needed. The variation follows immediately from (6.12), δE =

T 1 2 (F δD

+ δDT F).

(6.15)

This deﬁnes the variation of the strain components δE as the symmetric part of FT δD. The corresponding component form is ! " ∂(δuγ ) ∂xγ 1 ∂xγ ∂(δuγ ) . (6.16) + δEαβ = 2 ∂x0α ∂x0β ∂x0α ∂x0β It is seen that both the current Green strain tensor E and its variation δE are symmetric and therefore have only six independent components, while the deformation gradient and its variation are in general non-symmetric and therefore have nine independent components. The extra three components describe the rotation of the material at the point, as described in the next section.

6.1.2 Decomposition into deformation and rigid body motion The deformation gradient tensor F relates the initial and current local geometry around a point as illustrated in Fig. 6.4. It follows from the interpretation of the columns of the component matrix F as the coordinates of the base vectors after the motion that det(F) > 0, and thus the component matrix is invertible. The deformation gradient matrix has nine components, while the local deformation is described by only six components. The explanation is given by the polar decomposition theorem, which states that a positive deﬁnite three-dimensional tensor can be factored in two ways as F = R U = V R,

(6.17)

where U and V are positive deﬁnite symmetric tensors, and R is a proper orthogonal tensor, describing a rotation as discussed in Section 3.1. Each of the symmetric tensors has six independent components, while the rotation tensor is deﬁned in terms of three independent components, for example in the form of a rotation pseudo-vector. The factors in (6.17) can be interpreted by considering a line element with initial components dx0 . After the motion the current line element is dx = F dx0 = R (U dx0 ) = V (R dx0 ).

(6.18)

In the right-hand decomposition F = RU the line element ﬁrst participates in a deformation described by the symmetric matrix U, after which it is rotated by the rotation tensor R. Conversely, in the left-hand decomposition

152

Deformation and equilibrium of solids

F = VR the line element is ﬁrst rotated by the rotation tensor R, and then participates in a deformation described by the symmetric tensor V. The rotation is the same in the two cases, while the deformation tensors U and V will in general not be identical. Proof of the decompositions (6.17) and the relation between the tensors U and V are established as follows (Gurtin, 1981). The symmetric tensor U is determined from the deformation gradient tensor via the product FT F = (R U)T R U = U U = U2 .

(6.19)

This product deﬁning U2 , and thereby U, is determined by the Green strain tensor as shown by (6.12). As F is invertible, the same applies to U, and thus the rotation tensor is determined as R = F U−1 .

(6.20)

It is easily veriﬁed that RT R = I, which completes the proof of the righthand decomposition. The left-hand decomposition follows from introducing the symmetric tensor V in the form V = R U RT .

(6.21)

The product of the left-hand decomposition (6.17) then takes the form VR = (R U RT ) R = R U = F.

(6.22)

This completes the proof of the decomposition relations and provides explicit deﬁnitions for the tensors U, R and V. The decomposition of a motion into a deformation and a rotation part is illustrated by the two-dimensional examples shown in Fig. 6.5. In both cases there is no displacement in the x3 -direction, i.e. u3 = 0, and only a 2 × 2 sub-matrix of the components of F needs to be considered. Example 6.1. Biaxial extension. Figure 6.5(a) shows a state of biaxial extension, in which a unit vector along the x1 -axis is extended by ε1 , and a unit vector along the x2 -axis is extended by ε2 . There is no change of angle of the axes, and the deformation gradient therefore is 1 + ε1 0 F = . 0 1 + ε2 The right deformation matrix U2 follows from (2.19) as (1 + ε1 )2 0 2 T . U = F F = 0 (1 + ε2 )2

6.1 Deformation and strain

(a)

153

(b)

Fig. 6.5. (a) Biaxial extension, (b) shear at constant volume.

The diagonal form of U2 immediately gives 1 + ε1 0 U = . 0 1 + ε2 It is seen that the diagonal of U contains the so-called stretches of the axes. As a consequence of the diagonal form of U there is no rotation, and V = U. Example 6.2. Shear at constant volume. Figure 6.5(b) shows a state of shear at constant volume, in which the top side of a unit square is shifted γ to the right. Within linear deformation theory γ is the angular shear strain, but in a rigorous large deformation formulation matters are slightly more complicated. The deformation gradient for this case is 1 γ F = 0 1 and U2 is then deﬁned by the product 1 γ 2 T U = F F = . γ 1 + γ2 It is easy to verify – but somewhat more complicated to derive – that the matrix U then is 1 2 γ U = 2 , 4 + γ2 γ 2 + γ where it should be noted that the factor applies to all components in the matrix. The inverse of this matrix is 1 2 + γ 2 −γ −1 U = −γ 2 4 + γ2

154

Deformation and equilibrium of solids

and the rotation matrix R then follows from (6.20) as −1

R = FU

= 4 + γ2

1 γ 0 1

2 + γ 2 −γ −γ 2

= 4 + γ2

2 γ −γ 2

The rotation matrix is orthogonal with unit determinant, and can be written in terms of the rotation angle ϕ as cos ϕ − sin ϕ R = , tan ϕ = − 12 γ. sin ϕ cos ϕ For small deformation ϕ − 12 γ but for increasing values of γ the angle approaches 12 π, and the deformation loses its resemblance to shear. Finally, the left deformation tensor V follows from (6.21). Note that the determinant of U and V is equal to unity, because the volume is unchanged by the motion.

6.2 Virtual work and stresses The traditional way of introducing stresses is via an inﬁnitesimal volume element. First a rectangular element is considered to establish the diﬀerential equations of equilibrium, and then a tetrahedral element is used to establish the transformation rule needed for the surface stress vector. As already mentioned, the choice of a strain measure and the requirement that a principle of virtual work must exist leads to the stress measure that is conjugate to the chosen strain, i.e. the stress that combines with the strain variation into a virtual work equation. Here, the procedure already introduced in Chapter 2 is followed, starting out from the principle of virtual work, deriving the equilibrium equations and the formula for the surface stress vector, and then giving the physical interpretation of the stress obtained in this way. The Green strain leads to the Piola–Kirchhoﬀ stress, described in Section 6.2.1. The deﬁnition of the Piola–Kirchhoﬀ stress makes explicit reference to the initial state, and there are theoretical as well as practical reasons to investigate stresses deﬁned solely with reference to the current state. This leads to the Cauchy and the Kirchhoﬀ stress measures introduced in Section 6.2.2. These stresses refer to current directions and areas. However, the material properties – and thereby the current state of stress – are convected with the material, and it is therefore necessary to introduce a special description of the stress changes that accounts for the motion of the material. These so-called objective stress rates are discussed in Section 6.2.3.

6.2 Virtual work and stresses

155

6.2.1 Piola–Kirchhoﬀ stress The stress components associated with the Green strain components δEαβ are denoted Sαβ . The virtual work identity is formulated using the initial coordinates x0γ and the initial volume and surface elements dV0 and dS0 , respectively. When p0γ dV0 are the volume force components and t0γ dS0 the surface traction components corresponding to the Cartesian base vectors iγ , the virtual work equation is

0 δEαβ Sαβ dV0 = δuγ tγ dS0 + δuγ p0γ dV0 . (6.23) V0

The virtual strains δEαβ are symmetric. The product with the stress components will therefore only depend on the symmetric part, and the stress components appearing in the virtual work can therefore be introduced in symmetric form, i.e. Sβα = Sαβ . Due to the symmetry of the stress components the stress integral can be expressed by substituting the simple nonsymmetric form of the virtual strain expression (6.16), whereby

∂(δuγ ) ∂xγ δEαβ Sαβ dV = Sαβ dV0 . (6.24) 0 0 V0 V0 ∂xβ ∂xα By use of the divergence theorem the derivative ∂/∂x0β is moved to the product of the second and third factors,

∂(δuγ ) ∂xγ Sαβ dV0 = 0 0 V0 ∂xβ ∂xα (6.25)

∂xγ ∂xγ ∂ 0 δuγ Sαβ nβ dS0 − δuγ Sαβ dV0 . ∂x0α ∂x0β ∂x0α S0 V0 When this form is introduced into the virtual work identity (6.23), and it is observed that the displacement variation δuγ is arbitrary, it is found that the equilibrium equation ∂xγ ∂ + p0γ = 0, S γ = 1, 2, 3 (6.26) αβ ∂x0β ∂x0α must be satisﬁed in V0 , and the surface traction must be deﬁned by ∂xγ 0 tγ = Sαβ n0β , γ = 1, 2, 3 (6.27) ∂x0α on the surface S0 . The stresses Sαβ introduced in this way are the components of the second Piola–Kirchhoﬀ stress tensor. These stress components have the advantage of being symmetric. However, as seen from the equilibrium equation (6.26)

156

Deformation and equilibrium of solids

and the deﬁnition of the surface traction (6.27) they are used in the form Pγβ = (∂xγ /∂x0α )Sαβ . This combination is the ﬁrst Piola–Kirchhoﬀ stress tensor. The ﬁrst Piola–Kirchhoﬀ stress tensor Pγβ appears directly in the vector relations (6.26) and (6.27), while the second stress tensor Sαβ plays the direct role in the principle of virtual work. The symmetry and the direct relation to the principle of virtual work have led to general use of the second Piola–Kirchhoﬀ stress tensor.

Fig. 6.6. Initial unit cube and deformed shape with face tractions S1 , S2 , S3 .

The physical meaning of the two Piola–Kirchhoﬀ stress tensors can be explained with reference to Fig. 6.6. The ﬁgure shows a small unit cube with sides iγ located at the point x0γ in the initial conﬁguration. After the displacement the faces have the traction vectors S1 , S2 , S3 as shown. In the initial conﬁguration the cube had unit side length, and therefore the change of the traction vectors from one side of the volume to the other can be expressed by diﬀerentiation with respect to the original Cartesian coordinates x0α . With the body force p = p0γ iγ acting on the original unit volume the equilibrium equation is ∂S1 ∂S2 ∂S3 + + + p0 = 0, 0 0 ∂x1 ∂x2 ∂x03

(6.28)

∂Sβ + p0 = 0, ∂x0β

(6.29)

or in index notation

where the subscript β identiﬁes the face of the volume in the original conﬁguration. The similar component form was given in (6.26). When the base vectors iγ are introduced into (6.26), the following identity is obtained: ∂xγ ∂(xγ iγ ) Sβ = iγ Sαβ = Sαβ , β = 1, 2, 3. (6.30) 0 ∂xα ∂x0α These equalities show that the ﬁrst Piola–Kirchhoﬀ stress tensor corresponds

6.2 Virtual work and stresses

157

to resolving the traction vectors Sβ in components along the original unit base vectors iγ , while the second Piola–Kirchhoﬀ stress tensor corresponds to resolving the traction vectors Sβ in components along the vectors ∂x/∂x0α . These vectors form the sides of the deformed cube and thus correspond to a deformed non-orthogonal set of base vectors. The curious fact is that the use of this non-orthogonal set of base vectors leads to symmetric stress components, see e.g. Washizu (1974, p. 57). Example 6.3. Piola–Kirchhoff stress in beam theory. For small strain but arbitrarily large displacements the second Piola–Kirchhoﬀ stress components act as convected stress components, as illustrated in Fig. 6.7 for a beam.

Fig. 6.7. Stresses S11 , S21 , S31 on beam cross-section.

In the initial conﬁguration the axis of the beam is in the direction of the x1 -axis, and the cross-section of the beam therefore described in the x2 –x3 plane. The main stress components of a beam theory are those acting on the cross-section. In the present case the initial position of the cross-section is identiﬁed by the normal i1 and thus the relevant stress components correspond to the traction vector S1 . The stress components Sα1 correspond to a resolution in the deformed basis. In the present case there are two vectors ﬁxed in the cross-section giving the shear stress components S21 , S31 , and the tangent vector to the center-line of the deformed beam giving the axial stress component S11 . The lengths of the deformed base vectors are determined by the strains E11 , E22 , E33 and the deviation from orthogonality by the strains E13 , E23 , E12 . In the case of small strain the deformed base vectors still constitute a nearly orthogonal set of vectors of nearly unit length, and the second Piola–Kirchhoﬀ stress components are then nearly the ordinary stresses referred to this convected coordinate system. Thus, it turns out to be quite convenient that the stress components Sαβ refer to a plane with normal direction β in the initial conﬁguration with vector components α along the deformed base vectors.

158

Deformation and equilibrium of solids

6.2.2 Cauchy and Kirchhoﬀ stresses The Piola–Kirchhoﬀ stress refers to areas and directions in the reference state, and this gives an indirect relation to the current state in some situations involving large strains. It is therefore desirable to investigate the possibility of formulations based on the current state. Here the problem is posed in terms of small displacements du imposed on the current state, described by the coordinates x. The strains introduced by this motion will be the linear part of the Green strains, but with the current state as reference state. Thus these strains are ∂(duβ ) 1 ∂(duα ) dεαβ = . (6.31) + ∂xα 2 ∂xβ These strains may be related to the Green strain increments with components given by (6.16). When using the chain rule of diﬀerentiation on the derivatives of the displacements, the Green strain increments may be written as ! " ∂xλ ∂(duγ ) ∂xγ 1 ∂xγ ∂(duγ ) ∂xλ . (6.32) + dEαβ = ∂x0α ∂xλ ∂x0β 2 ∂x0α ∂xλ ∂x0β It is seen that the deformation gradient appears as a factor twice in each term, and thereby establishes the relation between the Green strain increment dE and the incremental linear strain dε as dE = FT dε F.

(6.33)

A similar relation applies to the variations δE and δε. The Cauchy stress tensor σ with components σαβ can now be introduced as conjugate to the linear current strains (6.31) when using the current volume and surface. This deﬁnition implies that the virtual work of the Cauchy stress through the virtual linear strain equals the virtual work of the second Piola–Kirchhoﬀ stress through the virtual Green strain,

δEαβ Sαβ dV0 = δεαβ σαβ dV. (6.34) V0

The relation between the two sets of stress components is obtained by introducing the relation (6.33) between the virtual strain components and by obtaining a relation between the reference volume element dV0 and the current volume element dV . The current volume element dV is related to the initial volume element dV0 by considering the sides ∂x/∂x0α of the deformed unit cube in Fig. 6.4.

6.2 Virtual work and stresses

159

The volume of the deformed cube is given by the triple product of the side vectors, ∂x ∂x ∂x ∂x α dV = (6.35) dV dV0 . × = det · 0 ∂x0β ∂x01 ∂x02 ∂x03 Introducing the Jacobian determinant J = det(F), the volume relation is dV = J dV0 = det(F) dV0 .

(6.36)

Introduction of this relation and the strain increment relation (6.33) into the left side of the internal virtual work equation (6.34) gives the following formula for the Cauchy stress: σ =

1 F S FT . J

(6.37)

The corresponding component relation is σαβ =

∂xβ 1 ∂xα Sγδ . 0 J ∂xγ ∂x0δ

(6.38)

It is seen that the role of the deformation gradient is to transform the reference of Sγδ components from the reference coordinates to the current coordinates. It follows from the polar decomposition (6.17) that in general this is not an orthogonal transformation, but involves deformation as well. The Cauchy stress σ conforms closely to the small deformation theory notion of stress with its use of current geometry. However, for processes involving large deformation the volume may change, and thereby the volume increment dV used in the principle of virtual work is not constant during the motion. This fact is accounted for by the so-called Kirchhoﬀ stress τ , that is introduced in a similar way as the Cauchy stress but using the original volume element dV0 . The reason behind the Kirchhoﬀ stress is that for actual processes the constitutive behavior of the material is associated with a certain mass of material. Thus, the volume integrals would actually be over mass elements dm = ρdV , where ρ is the mass density. The change of density with change of volume element would neutralize the apparent eﬀect of the changing volume increment dV . This eﬀect is often included by representing the volume element in the form J −1 dV . Thus, the Kirchhoﬀ stress components ταβ are introduced via the internal virtual work identity

δEαβ Sαβ dV0 = δεαβ ταβ J −1 dV. (6.39) V0

It follows directly by comparison with the Cauchy stress relation (6.34) that

160

Deformation and equilibrium of solids

the Kirchhoﬀ stress is τ = J σ = F S FT .

(6.40)

The similarity with the inverse relation (6.33) for the corresponding strain increments is evident.

6.2.3 Stress rates The Piola–Kirchhoﬀ stress components Sαβ describe tractions acting on material surfaces as described in Section 6.2.1. The change in these components can therefore be attributed to the constitutive behavior of the material. The Cauchy and Kirchhoﬀ stress components are diﬀerent in this respect, as they refer to tractions on surfaces ﬁxed in space. Thus, a solid body with a constant state of stress undergoing rigid body motion will not experience any change in the Piola–Kirchhoﬀ stress components. However, the changing direction of the body relative to the ﬁxed directions in space will in general imply that there is a change in the Cauchy and Kirchhoﬀ stress components. In this section the changes in the components of Cauchy and Kirchhoﬀ stresses that can be ascribed to changes in the state of the material are described. It is customary to imagine the associated motion as taking place in time, and the stress component changes are therefore called stress rates. The stress rates that can be attributed to changes of the state of the material are called objective stress rates. By implication the objective stress rate associated with a rigid body motion must vanish. This criterion is necessary, but not suﬃcient, to deﬁne an objective stress rate. Several objective stress rates have been deﬁned in the literature, see e.g. Holzapfel (2000), but only the two most common will be used here. A simple and direct approach to the derivation of objective stress rates starts out by observing that the Piola–Kirchhoﬀ stress S is deﬁned with reference to material surfaces and directions, and thus the rate of its components deﬁned as the time derivative S˙ is objective. This enables derivation of objective rates of other stress deﬁnitions directly from the relation of their original components to the Piola–Kirchhoﬀ stress. This approach leads to the so-called Truesdell stress rate. The simplest case is the Kirchhoﬀ stress τ . Inversion of the relation (6.40) gives S = F−1 τ F−T .

(6.41)

The time derivative of this relation is ˙ −1 τ F−T + F−1 τ F ˙ −T . S˙ = F−1 τ˙ F−T + F

(6.42)

6.2 Virtual work and stresses

161

Now, the basic argument is that the objective rate of the Kirchhoﬀ stress, ◦ denoted τ , should follow from the rate of the Piola–Kirchhoﬀ stress S˙ in the same way as the current components τ follow from S, namely by the relation (6.40). This deﬁnes the objective rate of the Kirchhoﬀ stress as ◦ τ = F S˙ FT .

(6.43)

Substitution of S˙ from (6.42) then gives ◦

˙ −1 τ + τ (F F ˙ −1 )T . τ = τ˙ + F F

(6.44)

It is seen that the eﬀect of convection of the material is included via a factor ˙ −1 and its transpose. FF The factors that represent the material convection in this formula can be given a precise interpretation. By deﬁnition of the time derivative, the deformation gradient has the components ∂ x˙ α . F˙αβ = ∂x0β

(6.45)

The time derivatives x˙ α are the components of the velocity of the particle. Thus (6.45) is the gradient of the particle velocity ﬁeld with respect to the reference coordinates x0β . The velocity gradient L with respect to the current coordinates is obtained via post-multiplication by F−1 , ∂x0 ˙ F−1 = ∂ x˙ α γ = ∂ x˙ α . L = F ∂x0γ ∂xβ ∂xβ

(6.46)

The velocity gradient tensor is deﬁned in terms of derivatives with respect to the current state, in contrast to the deformation and displacement gradients, where gradient refers to derivatives with respect to the coordinates of the reference state. ˙ −1 describing the convection in the objective stress rate The factor FF (6.44) is obtained by taking the time derivative of the identity FF−1 = I, whereby ˙ −1 = −F ˙ F−1 = −L. FF (6.47) The ﬁnal form of the Truesdell rate of the Kirchhoﬀ stress in terms of the velocity gradient is ◦

τ = τ˙ − L τ − τ LT .

(6.48)

Note that while τ˙ denotes the time derivative of the components of the Kirchhoﬀ stress tensor, constitutive relations must be formulated in terms ◦ of objective stress rates, such as the Truesdell rate τ .

162

Deformation and equilibrium of solids

The Truesdell rate of the Cauchy stress can be found via the time derivatives of the second Piola–Kirchhoﬀ stress tensor by a similar procedure. Inversion of (6.37) gives S = J F−1 σ F−T

(6.49)

and time diﬀerentiation leads to ˙ −1 σ F−T + J F−1 σ F ˙ −T . S˙ = F−1 (J σ˙ + J˙ σ)F−T + J F

(6.50)

This formula contains the time derivative J˙ of the Jacobian determinant. The Jacobian determinant V is the current volume of the material originally ˙ occupying unit volume. Thus, J/J is the current rate of volume change, which can be expressed in terms of the time derivatives of the strains, ˙ ˙ J/J = ε˙11 + ε˙22 + ε˙33 = tr(ε).

(6.51)

◦

Deﬁnition of the objective Cauchy stress rate σ from the original transformation formula (6.37) then leads to ◦

σ = σ˙ + tr(ε) ˙ σ − L σ − σ LT .

(6.52)

The term containing tr(ε) ˙ accounts for the change in the current volume element dV , an eﬀect that has been neutralized in the deﬁnition of the Kirchhoﬀ stress. Note that the circle symbol for the Truesdell stress rate has here been used to denote a particular procedure for deﬁning the objective rate, in which the stress – here Kirchhoﬀ or Cauchy stress – is transformed to a material frame of reference by a relation of the form (6.41) or (6.49). In the material frame a direct time derivative is objective and can be used to describe material behavior. The corresponding Truesdell rate of Kirchhoﬀ or Cauchy stress is therefore deﬁned by a similar transformation of the increments in the material frame. This procedure corresponds to the concept of a Lie derivative discussed e.g. by Marsden and Hughes (1983) and Holzapfel (2000). It is similar to the discussion in Section 4.2 of the virtual and actual changes of the local components of strain and curvature in the development of a large deformation beam theory. Example 6.4. Truesdell stress rate in constant volume shear. The constant volume shear problem introduced in Example 6.2 and illustrated in Fig. 6.5(b) is described by the deformation tensor, its inverse and its time derivative: 0 γ˙ 1 −γ 1 γ −1 ˙ . , F = F = , F = 0 0 0 1 0 1

6.2 Virtual work and stresses

163

The velocity gradient tensor is determined by (6.46) as 0 γ˙ −1 ˙ . L = FF = 0 0 The velocity gradient tensor determines the extra terms in the objective stress rate (6.48), ◦ τ12 + τ21 τ22 T . τ = τ˙ − L τ − τ L = τ˙ − γ˙ τ22 0 When this formula is used to express the time derivative of the Kirchhoﬀ stress, ◦ τ12 + τ21 τ22 τ˙ = τ + γ˙ , τ22 0 the last term represents the eﬀect of the convection of the material in the spatial frame of reference. It is seen that the convection contribution is independent of the stress component τ11 . Intuitively this stress may be associated with tension in an imagined ﬁber in the x1 -direction, which does not change during the motion. On the other hand, the current tension τ22 in the direction of the x2 -axis contributes γτ ˙ 22 to the time derivative of the shear strain, and a current shear stress τ12 = τ21 contributes 2γτ ˙ 12 to the time derivative of the normal stress in the x1 -direction. Intuitively, eﬀects of this type were to be expected. In the present problem there is no change of volume, and therefore the stress rate formulae also apply to the Cauchy stress.

Other objective stress rates have also found use in modeling large deformation of solids. Probably the most used is the Jaumann stress rate. It is based on the idea that the change of stress components in a ﬁxed spatial coordinate system consists of two parts: one part corresponding to a change of stress in the material, and another due to convection of the material relative to the spatial frame of reference. In the Jaumann stress rate the eﬀect of convection is described by reference to an instantaneous angular velocity ω. This angular velocity, called the spin, is identiﬁed by resolving the velocity gradient tensor into a symmetric and an anti-symmetric part. The symmetric part is the strain rate ε, ˙ while the anti-symmetric part is the spin tensor W: L = ε˙ + W

(6.53)

164

Deformation and equilibrium of solids

with strain rate and spin tensor deﬁned by ε˙ =

1 2 (L

+ LT ),

W =

1 2 (L

− LT ).

(6.54)

It follows from the discussion in Section 3.1 that any skew-symmetric tensor can be associated with a vector representing an inﬁnitesimal rotation or angular velocity. Thus, according to (3.7) the spin tensor W can be expressed in terms of the angular velocity ω as   0 −ω3 ω2 ˆ =  ω3 (6.55) W = ω 0 −ω1  . −ω2 ω1 0 The Jaumann stress rate is deﬁned by a relation similar to the Truesdell rate (6.48), but now using only the skew-symmetric part of the velocity gradient. This gives the Jaumann rate

τ = τ˙ − W τ − τ WT

(6.56)

of the Kirchhoﬀ stress. The Jaumann rate only accounts for the spin, but not for the strain rate. The volume rate does not enter into the deﬁnition of the Jaumann stress rate, and thus the formula for the Jaumann rate of the Cauchy stress is of the same form as (6.56). The Truesdell stress rates for the Kirchhoﬀ stress and the Cauchy stress are easily expressed in terms of the Jaumann rate by substitution of the decomposition (6.53) of the velocity gradient tensor. The Kirchhoﬀ stress rate formula follows from (6.48), ◦

τ = τ − ε˙ τ − τ ε˙

(6.57)

and the Cauchy stress rate formula from (6.52), ◦

˙ σ − ε˙ σ − σ ε. ˙ σ = σ + tr(ε)

(6.58)

It is seen that the diﬀerence between the Truesdell and the Jaumann stress rates is described by the current state of stress and the current rate of strain, but is independent of any rate of rotation. With a suitable constitutive relation for the material the strain rate can be expressed in terms of the current state of the material and the rate of the loads, and thus the choice between the Truesdell and Jaumann stress rates is largely a matter of taste and convenience. Example 6.5. Jaumann stress rate in constant volume shear. The Truesdell stress rate of the constant volume shear problem was considered

6.3 Total Lagrangian formulation

165

in Example 6.4. In the Jaumann stress rate the convection terms are represented by the spin tensor W, deﬁned as the symmetric part of the velocity gradient tensor L. The strain rate ε˙ and the spin tensor W are found as the symmetric and anti-symmetric parts of the velocity gradient tensor L, determined in Example 6.4. 1 0 12 γ˙ 0 0 γ˙ γ˙ 2 L == , ε˙ = 1 , W = . 0 0 − 12 γ˙ 0 ˙ 0 2γ The spin tensor W determines the extra terms in the objective stress rate (6.56), τ12 + τ21 τ22 − τ11 T 1 . τ = τ˙ − W τ − τ W = τ˙ − 2 γ˙ τ22 − τ11 −(τ12 + τ21 ) The time derivative of the Kirchhoﬀ stress in terms of the Jaumann stress rate is τ12 + τ21 τ22 − τ11 1 τ˙ = τ + 2 γ˙ . τ22 − τ11 −(τ12 + τ21 ) It is seen that the Jaumann stress rate formula involves all the current stress components, while the Truesdell rate is independent of τ11 in this motion.

6.3 Total Lagrangian formulation As discussed in Chapter 2, non-linear problems of solids and structures may be formulated either in the total or the updated Lagrangian format. In the total Lagrangian format the motion of the material points is described by the total displacement ﬁeld u(x0 ) with reference to the initial position x0 of the material point. This formulation leads naturally to deformations described in terms of the Green strain tensor E and stresses described in terms of the second Piola–Kirchhoﬀ stress tensor S. An advantage of this method is that the total stress appears directly in the equations, while a disadvantage is that strain increments become non-linear, as they refer to the initial state. Alternatively, the non-linear problem may be described in the updated Lagrangian format in terms of the displacement increments du(x) with reference to the current state x. This formulation leads to linearized incremental strains dε and Cauchy or Kirchhoﬀ stresses. This gives the advantage that the strain increments may be introduced in linear form. However, the stress increments must be represented by their non-linear objective rates, as discussed in the previous section. Thus, the choice between

166

Deformation and equilibrium of solids

the total and the updated Lagrangian formulations includes a choice of the speciﬁc form of the non-linearity in the equations to be solved. This section describes three main aspects of the total Lagrangian formulation: the formulation of equilibrium equations, the formation of the tangent stiﬀness matrix, and the implementation in the ﬁnite element format by use of shape functions. The formulation of these steps in the updated Lagrangian format is presented in Section 6.4. 6.3.1 Equilibrium and residual forces The total Lagrangian format makes use of the total displacement components uα (x0ξ ), considered as functions of the initial position x0ξ . This format makes use of the Green strain Eαβ (x0ξ ) corresponding to the conjugate Piola–Kirchhoﬀ stress components Sαβ (x0ξ ). In order to illustrate the general procedure the virtual work relation is obtained from the equilibrium equation, reversing the argument in Section 6.2.1. The equilibrium equation corresponding to the Piola–Kirchhoﬀ stress components was given in (6.26) as ∂xγ ∂ Sαβ + p0γ = 0 in V0 . (6.59) ∂x0β ∂x0α In this equation p0γ stresses and displacements are considered as functions of the initial position x0ξ . The corresponding virtual work equation is obtained by scalar multiplication of this equation with a virtual displacement ﬁeld δuγ (x0ξ ), followed by integration over the initial volume V0 :

∂xγ ∂ dV0 + δuγ δuα p0α dV0 = 0. (6.60) 0 ∂x0 Sαβ ∂x V0 V0 α β By use of the divergence theorem, expressed by (6.25), this relation may be written in the form of an equality of internal and external virtual work

0 δEαβ Sαβ dV0 = δuγ tγ dS0 + δuγ p0γ dV0 , (6.61) V0

where the virtual Green strain components δEαβ have been introduced from (6.16), ! " ∂(δuγ ) ∂xγ 1 ∂xγ ∂(δuγ ) δEαβ = (6.62) + 2 ∂x0α ∂x0β ∂x0α ∂x0β and t0γ is the surface traction vector, related to the stress tensor components by (6.27).

6.3 Total Lagrangian formulation

167

The right side of the virtual work equation (6.61) deﬁnes the external virtual work, i.e. the virtual work of the external loads,

0 δVext = δuγ p0γ dV0 δuγ tγ dS0 + (6.63) V0

while the left side deﬁnes the internal virtual work of the stresses,

δEαβ Sαβ dV0 . δVint =

(6.64)

Equilibrium is expressed by the statement δVint = δVext , valid for any displacement variation ﬁeld δxγ (x0ξ ). Thus, any lack of equilibrium may be considered equivalent to a distribution of residual forces rγ (x0ξ ), which accounts for the diﬀerence in the external and internal virtual work. This is expressed by the relation

δuγ rγ dV0 = δVext − δVint (6.65) V0

which, for any state of displacement uγ (xξ ) and stress Sαβ (xξ ), deﬁnes the residual force distribution rγ (x0ξ ). For a suitably discretized displacement ﬁeld this relation deﬁnes a discrete set of residual forces, which are eliminated in an iteration process that gradually establishes equilibrium. The ﬁnite element discretization is described in Section 6.3.3.

6.3.2 Tangent stiﬀness In the same way the virtual work equation for the total internal and external forces is used to check equilibrium and to deﬁne residual forces, an incremental form of the virtual work equation is established in order to obtain the tangent stiﬀness. Instead of considering the total loads t0γ , p0γ and stresses Sαβ , two neighboring conﬁgurations separated by the increments dt0γ , dp0γ and dSαβ are considered. In practice it is easier to obtain the incremental relation by diﬀerentiation of the virtual work (6.61) than by considering the two neighboring states individually. The key point is that while loads, stresses and displacements are incremented, the variation δuγ remains unaffected. This follows from the fact that δuγ is to be considered as an arbitrary ﬁeld, independent of the actual displacement ﬁeld. In the case of continua, where the displacement representation depends on rotations, this argument has to be modiﬁed as explained in Chapter 3. The tangent stiﬀness follows from the increment of the internal virtual

168

Deformation and equilibrium of solids

work δVint ,

d(δVint ) = d

δEαβ Sαβ dV0 .

(6.66)

The stress components are symmetric, and the virtual strain can then be substituted from the ﬁrst term in (6.62). It then follows from diﬀerentiation that

∂(δuγ ) ∂(duγ ) (6.67) S dV + δEαβ dSαβ dV0 . d(δVint ) = 0 αβ 0 ∂x0β V0 V0 ∂xα Now, assume that there exists a linear constitutive relation between the stress increments dSαβ and the strain increments dEγδ in the form 0 dEγδ . dSαβ = Cαβγδ

(6.68)

Speciﬁc forms of this relation corresponding to elastic and elasto-plastic materials are discussed in Chapter 7. The increment of the internal virtual work then takes the form

∂(δuγ ) ∂(duγ ) 0 S dV + δEαβ Cαβγδ dEγδ dV0 . (6.69) d(δVint ) = 0 αβ 0 0 ∂x ∂x V0 V0 α β Due to the symmetry of the stress and strain components the elastic tensor Cαβγδ must be symmetric in the ﬁrst and last pair of subscripts, and the strain variation and strain increment can therefore be introduced in the non-symmetric form corresponding to the ﬁrst term in (6.62). With this, the increment of the internal virtual work is

∂(duγ ) ∂(δuγ ) d(δVint ) = Sαβ dV0 0 ∂x ∂x0β V0 α (6.70)

∂(δuκ ) ∂xκ 0 ∂xλ ∂(duλ ) + 0 Cαβγδ ∂x0 ∂x0 dV0 . 0 V0 ∂xα ∂xβ γ δ This form of the incremental virtual work is similar to the tangent stiﬀness relation (2.68) obtained for bar elements. It consists of a constitutive stiﬀness contribution and a geometric stiﬀness contribution, where the deformation gradients appear as a consequence of the ﬁnite change of geometry. In order to emphasize the simple structure of the relations they are also given in matrix form using the symbol : for contraction on two subscripts,

T (δD dD) : S dV0 + δET : C0 : dE dV0 . (6.71) d(δVint ) = V0

6.3 Total Lagrangian formulation

169

The parentheses indicate that the product δDT dD should be formed, leaving two free subscripts that are then contracted with S. The previous formulae may help in the detailed interpretation of the matrix products. Several special cases of the incremental virtual work (6.67) are of interest. If the initial stress contribution and the dependence on the current displacement are omitted, the traditional linear form is obtained. Note that material 0 non-linearities can still be included via the coeﬃcients Cγδαβ . In updated formulations and in linearized stability theory the initial stress terms are included, while the dependence on the current displacement is left out, see e.g. Washizu (1974). There is a close relation between the symmetry of the incremental virtual work and the existence of a strain energy function for the material. If the material has a strain energy density function ϕ(Eαβ ), the stresses are given by the derivatives Sαβ =

∂ϕ . ∂Eαβ

(6.72)

Then the stress increments follow by further diﬀerentiation, dSαβ =

∂2ϕ 0 dEγδ = Cαβγδ dEγδ . ∂Eαβ ∂Eγδ

(6.73)

0 as double derivatives of the strain This establishes the coeﬃcients Cαβγδ 0 0 = Cαβγδ follows energy density ϕ(Eαβ ), and the symmetry relation Cγδαβ from interchanging the order of diﬀerentiation. With this symmetry the incremental internal virtual work is symmetric with respect to interchange of du and δu. This symmetry in turn leads to a symmetric tangent stiﬀness relation. It should be noted that several material models do not lead to a strain energy density, see e.g. Ottosen and Ristinmaa (2005). Those materials must therefore be treated by incremental relations. An important example is the incremental theory of plasticity treated in the next chapter. The principle of virtual work may also be written as an equation expressing that the change in virtual work δV is zero for arbitrary variations of the displacement ﬁeld:

δV = Sαβ δEαβ dV0 − t0γ δuγ dS0 − p0γ δuγ dV0 . (6.74) V0

If the material has a strain energy density function, and if the loads are conservative, this may be considered as the variation of a potential energy functional Φ(u) found by integration. In the case of constant loads the

170

Deformation and equilibrium of solids

potential energy has the form

ϕ(Eγδ ) dV0 − Φ(u) =

t0γ uγ dS0 −

p0γ uγ dV0 .

(6.75)

In this formulation it is seen that the geometrical non-linearity is contained completely in the non-linear strain deﬁnition, while material non-linearity determines the form of the strain energy function. 6.3.3 Finite element implementation Within the ﬁnite element method the most common way of discretizing problems of solid mechanics consists in representing the displacement ﬁeld u(x0 ) in terms of the displacement un at selected nodes x0n , n = 1, 2, . . . This corresponds to a representation in the form u(x0 ) = hn (x0 ) un , (6.76) n

where hn (x0 ) is the shape function corresponding to a unit displacement at node n. Thus, the shape functions must satisfy the relation hn (x0m ) = δnm ,

(6.77)

where δnm is the Kronecker delta. This relation implies that hn (x0 ) is unity at the node x0n and zero at all other nodes. In practice the shape functions may not be explicitly given functions of the spatial coordinates x0 , but be given as a mapping from a normalized set of coordinates. In the formulation of the virtual work, the external virtual work can be evaluated directly from (6.63) by substituting a representation of the virtual displacement ﬁeld of the same form as (6.76), hn (x0 ) δun . (6.78) δu(x0 ) = n

The external virtual work can now be used to deﬁne equivalent external nodal forces fnext via the relation

$ T ext T hn (x0 ) t0 dS0 + hn (x0 ) p0 dV0 , (6.79) δVext = δun fn = δun S0

where summation over the nodes n = 1, 2, . . . is implied. The components of the virtual nodal displacements δun can be selected independently, and thus the scalar relation (6.79) deﬁnes the external nodal forces as

ext fn = hn (x0 ) t0 dS0 + hn (x0 ) p0 dV0 . (6.80) S0

6.3 Total Lagrangian formulation

171

Thus, the external nodal forces are obtained by integrating the surface traction and the volume force, weighted by the corresponding shape function hn (x0 ). In spite of the fact that the stresses Sαβ and virtual strains δEαβ are components of two-dimensional tensors, it is often computationally convenient to represent these components as a one-dimensional array. In the following a column format will be used. The symmetry of the stress and strain components leads to the six-component form of the virtual work δEαβ Sαβ = δE11 S11 + δE22 S22 + δE33 S33 + 2δE23 S23 + 2δE31 S31 + 2δE12 S12 .

(6.81)

This expression of virtual work may alternatively be formed by use of the one-dimensional arrays for the strain increments and stresses δE = [ δE11 , δE22 , δE33 , 2δE23 , 2δE31 , 2δE12 ]T

(6.82)

S = [ S11 , S22 , S33 , S23 , S31 , S12 ]T ,

(6.83)

and

where the factor two has been included in the virtual strains. The internal virtual work (6.64) involves the virtual strain δEαβ , given by (6.62) in terms of products of the deformation gradient F = [∂xγ /∂x0α ] and the displacement gradient δD = [∂(δuγ )/∂x0β ]. It is computationally desirable to arrange these tensors in a format that gives the incremental strain δE in a product format. This is accomplished by considering the columns of the component matrix of the deformation gradient F = [∂xα /∂x0β ] = [ f1 , f2 , f3 ]

(6.84)

and the virtual displacement gradient δD = [∂(δuα )/∂x0β ] = [ δd1 , δd2 , δd3 ].

(6.85)

It is seen that the summation in the deﬁnition of the virtual strain increment δEαβ is on the subscript denoting the row number, and thus the summation can be expressed directly by the boldface notation introduced by the columns in (6.84) and (6.85). The array format (6.82) of the virtual strain then takes

172

the form

Deformation and equilibrium of solids

    δE =    

f1T δd1 f2T δd2 f3T δd3 f2T δd3 + f3T δd2 f3T δd1 + f1T δd3 f1T δd2 + f2T δd1





       =       

f1T 0 0 0 f3T f2T

0 f2T 0 f3T 0 f1T

0 0 f3T f2T f1T 0

    δd1   δd2 .   δd3 

(6.86)

The column vectors δd in the last factor are expressed in terms of the node displacements and the shape functions by use of (6.78): δdβ =

∂hn ∂(δu) = δun = hn,β δun , 0 0 ∂xβ ∂xβ n n

(6.87)

where the notation hn,β has been used for the material derivatives of the shape functions. The relation (6.86) for the virtual strain can now be expressed in the compact form ˆ Bn δun = δE(x0 ) = B0n (x0 ) δun , (6.88) F n

ˆ is deﬁned by the ﬁrst factor in (6.86), and the material where the matrix F derivatives of the shape functions are contained in the matrices   hn,1 I

Bn =  hn,2 I .

(6.89)

hn,3 I

Most often the factors ∂xα /∂x0β and ∂hn /∂x0γ are given implicitly and the evaluation of the matrix B0n (x0 ) is carried out numerically at selected points, see e.g. Hughes (1987) and Zienkiewicz and Taylor (2000) for computational details. The internal virtual work is now used to deﬁne internal nodal forces fnint by substitution of the virtual strain ﬁeld into (6.64): #

$ δVint = δuTn fnint = δuTn (6.90) B0n (x0 )T S dV0 , V0

where summation over the nodes n = 1, 2, . . . is implied. By selecting nonvanishing nodal displacements one node at a time, the following deﬁnition of the internal nodal forces is obtained:

int B0n (x0 )T S dV0 . (6.91) fn = V0

The calculation of the internal nodal forces requires evaluation of a volume

6.3 Total Lagrangian formulation

173

integral of the product of the current Piola–Kirchhoﬀ stress and the matrix B0n (x0 ). The integral form of the residual force relation (6.65) can now be written in discretized form in terms of nodal residual forces rn , rn = fnext − fnint .

(6.92)

For a given conﬁguration x(x0 ) with Piola–Kirchhoﬀ stress distribution S(x0 ) the external and internal forces are calculated from (6.80) and (6.91), respectively. Any deviation from equilibrium is then expressed by the nodal residual forces rn , determined by (6.92). In the absence of equilibrium the residual forces are eliminated by iterations using the tangent stiﬀness matrix. The tangent stiﬀness is evaluated from the increment of the internal virtual work (6.70) by substitution of the displacement and the incremental strain representations (6.78) and (6.88), whereby

# ∂h $ ∂hm n T 0T 0 S dV + B C B dV dum , d(δVint ) = δun I 0 0 0 αβ n m 0 ∂x0β V0 ∂xα V0 (6.93) where I is the 3 × 3 unit tensor with components δαβ . The 6 × 6 matrix C0 is the incremental stiﬀness corresponding to the relation (6.68), when written in terms of the array notation (6.82) and (6.83) for strains and stresses, dS = C0 dE.

(6.94)

The incremental virtual work relation (6.93) is of the form d(δVint ) = δuTn Knm dum ,

(6.95)

where summation over n and m is implied, and Knm is the contribution to the tangent stiﬀness matrix from the combination of nodes n and m. It follows from (6.93) and (6.95) that the stiﬀness matrix contribution Knm is given by

∂hn ∂hm 0 Knm = I S dV0 + B0T (6.96) n C0 Bm dV0 . 0 αβ ∂x0 ∂x V0 V0 α β The global stiﬀness matrix is assembled as discussed e.g. in Section 2.5. The total Lagrangian formulation is complicated by the fact that the matrix B0n (x0 ) depends on the components of the deformation gradient F = ∂x/∂x0 . Simpliﬁcations arise if the current conﬁguration x coincides with the initial conﬁguration x0 . In that case ∂xα /∂x0β = δαβ , and the components of the matrix F are either unity or zero. Hereby the formulation

174

Deformation and equilibrium of solids

becomes similar to that of a linear ﬁnite element analysis, in which an initial stress term has been included. This simpliﬁcation of the computation of internal forces and the tangent stiﬀness matrix is a characteristic feature of the updated Lagrangian formulation discussed in the next section.

6.4 Updated Lagrangian formulation In the previous section it was demonstrated how the ﬁnite deformation primarily enters into the total Lagrangian formulation via the non-linear form of the strain increments. In this section it is described how the non-linearities may be transferred into the constitutive behavior by use of an updated Lagrangian formulation. In the updated Lagrangian formulation each increment uses the current conﬁguration as reference. This gives some simpliﬁcations in the kinematic description, but on the other side the change of stress must be formulated in an objective format to justify evaluation by a constitutive relation for the material properties. First, a direct transformation of the total format into updated form is demonstrated. Then, formulations derived directly from the principle of virtual work in the current state are discussed, and ﬁnally a brief description of implementation issues is given.

6.4.1 Transformation from total to updated format One way to obtain an updated Lagrangian formulation is by transformation of the total format presented above. The quantities needed are: the external virtual work, the internal virtual work, and the incremental virtual work representing the tangent stiﬀness. The external virtual work follows immediately from (6.63) by introducing t as the traction vector with reference to the current bounding surface element dS and p as the distributed load intensity with reference to the current volume element dV . The external virtual work then follows immediately from (6.63) in the form

δuγ tγ dS + δuγ pγ dV. (6.97) δVext = S

In a similar way the internal virtual work follows from (6.64) by using the relation (6.34), changing the format to linearized current strain, Cauchy stress and current volume:

δεαβ σαβ dV. (6.98) δVint = V

6.4 Updated Lagrangian formulation

175

These relations enable evaluation of residual forces representing any deviations from equilibrium. The incremental internal virtual work was given by (6.67). The ﬁrst term represents geometric eﬀects, while the second term is the constitutive relation representing the material behavior. The geometric contribution can be transformed into current conﬁguration by diﬀerentiation ‘through’ the current conﬁguration via the chain rule of diﬀerentiation. The result is

∂(δuγ ) ∂(duγ ) ∂(δuγ ) ∂xκ Sαβ ∂xλ ∂(duγ ) S dV = dV, 0 αβ 0 ∂xκ ∂x0α J ∂x0β ∂xλ ∂x0β V0 ∂xα V (6.99) where the volume element has also been changed to current volume by use of the relation dV = JdV0 . By using the relation (6.38) it is seen that the factor in parentheses is simply the Cauchy stress component matrix σκλ . The second term involves the stress increment dSαβ . The deﬁning property of the Truesdell rate of the Cauchy stress is that objective increments of Cauchy and Piola–Kirchhoﬀ stress are related by the same transformation as their total components. Thus, the incremental form of the virtual work relation (6.34) is

◦ δEαβ dSαβ dV0 = δεαβ Dσαβ dV, (6.100) V0

◦

where the notation D has been introduced to denote the increment corresponding to the Truesdell rate. The precise form follows from (6.52): ◦

Dσαβ = d(σαβ J) J −1 −

∂(dxβ ) ∂(dxα ) σγβ − σαγ , ∂xγ ∂xγ

(6.101)

where dσαβ represents the increments of the components σαβ . As discussed in Section 6.2.3, constitutive relations must be formulated in terms of objective stress rates. In the present context this means a relation between the Truesdell stress increment and the increment of the linearized stress of the form ◦

Dσαβ = Cαβγδ dεγδ .

(6.102)

When using the strain and stress transformation rules (6.33) and (6.37) it can be demonstrated that the new stiﬀness component matrix Cαβγδ is 0 connected to the previous stiﬀness component matrix Cαβγδ via the relation Cξηκλ

0 ∂xξ ∂xη Cαβγδ ∂xκ ∂xλ = . ∂x0α ∂x0β J ∂x0γ ∂x0δ

(6.103)

176

Deformation and equilibrium of solids

This relation can also be established by diﬀerentiating the ﬁrst and last factors of the last integral in (6.70) ‘through’ the current state variable xκ . Note that the initial coordinates x0 with index 0 are contracted by the indices of the stiﬀness coeﬃcients of C0 , also with index 0, while the stiﬀness coeﬃcients of the current state C correspond to the indices of the current conﬁguration x. The relation (6.103) corresponds to a transformation of the material properties from the initial to the current conﬁguration. Combination of the geometric contribution (6.99) and the constitutive contribution (6.100) leads to the incremental virtual work in the form

d(δVint ) = V

∂(duγ ) ∂(δuγ ) σαβ dV + ∂xα ∂xβ

δεαβ Cαβγδ dεγδ dV.

(6.104)

It is seen that the external and internal force relations (6.97) and (6.98) and the incremental virtual work relation (6.104) are obtained from their total Lagrangian equivalents (6.63), (6.64) and (6.70) by changing to Cauchy stress and linearized current strain and using current surface and volume elements. In this formulation the complications associated with the nonlinear Green strain interpolation matrix B0n have been absorbed into the constitutive relations for the material. However, this does not imply that the complications arising from the non-linear kinematics of the problem have vanished. As described in Chapter 7, the classical deﬁnition of (hyper)elastic materials as derivable from an elastic energy function implies that the elas0 tic coeﬃcients Cαβγδ from the material description can be constants, while the corresponding coeﬃcients Cαβγδ must depend on the deformation gradient Fξη . It is of interest to note that the present reformulation of the total Lagrangian equations into updated Lagrangian format establishes exact correspondence between the use of Green strain and Piola–Kirchhoﬀ stress in the total formulation with the use of linearized current strain and Cauchy stress combined with its Truesdell rate in the updated formulation. Alternatively, other stress rates may be introduced, when the updated Lagrangian formulation is derived directly from the equilibrium equation in the current conﬁguration.

6.4.2 Virtual work in the current conﬁguration The updated Lagrangian formulation may be obtained directly from the equation of virtual work formulated in the current conﬁguration. The start-

6.4 Updated Lagrangian formulation

177

ing point is the equilibrium equation ∂σαβ + pα = 0 ∂xβ

in V

(6.105)

expressed in terms of the components σαβ (x) of the Cauchy stress and the distributed force pα (x) per unit current volume. The virtual work equation is obtained by multiplying the equilibrium equation (6.105) with the virtual displacement ﬁeld δuα (x), followed by integration over the current volume. After use of the divergence theorem, the equation of virtual work takes the form

δuα pα dV, (6.106) δεαβ σαβ dV = δuα tα dS + V

where the virtual strain components are ∂(δuβ ) 1 ∂(δuα ) δεαβ = + 2 ∂xβ ∂xα

(6.107)

and the surface traction components tα are related to the stress components by tα = σαβ nβ .

(6.108)

In this formulation the distributed surface loads as well as the surface normal components nβ refer to the current surface area. It follows immediately from the virtual work equation (6.106) that the external virtual work in the current conﬁguration is

δVext = δuα tα dS + δuα pα dV, (6.109) S

where the loads are normalized with respect to the current surface and volume, respectively. The internal virtual work is

δεαβ σαβ dV. (6.110) δVint = V

These relations are identical to (6.97) and (6.98) derived from transformation of the total Lagrangian formulation. The tangent stiﬀness follows from the increment of the internal virtual work,

d(δεαβ ) σαβ dV + δεαβ d(σαβ J) J −1 dV. (6.111) d(δVint ) = V

The Jacobi determinant is included in the stress factor, because J −1 dV = dV0 is then invariant with respect to the increment. In order to derive the

178

Deformation and equilibrium of solids

stress increment from a constitutive material relation it must be expressed in objective form. It turns out to be convenient ﬁrst to use the Truesdell stress rate to establish the basic result. The corresponding results for other stress rates can then be obtained from this result by simple substitution. ◦

When the Truesdell stress increment Dσαβ deﬁned in (6.101) and the linearized virtual strain from (6.107) are introduced into the second integral, the following result is obtained:

δεαβ d(σαβ J) J

−1

◦

dV =

δεαβ Dσαβ dV

∂(δuβ ) ∂(duβ ) ∂(duα ) ∂(δuα ) 1 + 2 dV σγβ + σαγ ∂xγ ∂xβ ∂xα ∂xγ V

∂(δuβ ) ∂(duβ ) ∂(duα ) ∂(δuα ) 1 + 2 dV. σαγ + σγβ ∂xβ ∂xγ ∂xα ∂xγ V

(6.112)

The stress component matrix σαβ is symmetric, and therefore the two terms in each of the square brackets are identical. Thus, the relation may be written in the more compact form

δεαβ d(σαβ J) J V

+ V

−1

dV =

◦

δεαβ Dσαβ dV V

∂(duγ ) ∂(δuγ ) dV + σαβ ∂xβ ∂xα

∂(duγ ) ∂(δuα ) σαβ dV. ∂xγ ∂xβ

(6.113)

The ﬁrst two integrals on the right side correspond to the incremental virtual work as given in (6.104). It is now demonstrated that in the full form (6.111) of the incremental virtual work the ﬁrst integral containing the contribution from the increment of virtual strain exactly cancels the last integral in (6.113). The contribution to the incremental virtual work from the increment of the virtual strain is given by

d(δεαβ ) σαβ dV = V

∂(δu ) α σαβ dV, ∂xβ

d V

(6.114)

where the deﬁnition (6.107) of the linearized virtual strain and the symmetry of the strain component matrix have been used. The variation δuα is kept constant, and the increment arises because the position of a material point currently at x = x1 changes position to x = x1 + du. The inﬂuence of this on the partial derivatives in (6.114) is found by ﬁrst diﬀerentiating through

6.4 Updated Lagrangian formulation

179

the current state variables x1γ by the chain rule of diﬀerentiation, d

∂(δu ) ∂(δuα ) ∂x1γ α = . d ∂xβ ∂x1γ ∂xβ

(6.115)

The increment has now been isolated to the last factor, which is independent of any particular variational displacement δuα . It is evaluated by considering the identity ∂x1γ ∂xλ = δγβ . (6.116) ∂xλ ∂x1β This relation is simply a statement of the deﬁnition of partial diﬀerentiation of the variable x1γ with respect to the variable x1λ . For identical variables γ = λ the result is unity, while for diﬀerent variables γ = λ it is zero. When the variables xγ are ﬁxed, the increment of this relation is ∂x1 ∂x ∂x1 ∂x ∂x1γ ∂(dxλ ) γ γ λ λ = d + = 0. ∂xλ ∂x1β ∂xλ ∂x1β ∂xλ ∂x1β

(6.117)

At the current state xγ = x1γ , and thus ∂x1α /∂xβ = ∂xα /∂x1β = δαβ in this state. When using this ∂x1 ∂(dxγ ) γ d = − . (6.118) ∂xβ ∂x1β By use of this result in (6.115), the integral (6.114) takes the form

∂(duγ ) ∂(δuα ) d(δεαβ ) σαβ dV = − σαβ dV. (6.119) 1 ∂xγ ∂x1β V V In the current state x1α = xα , and it is seen that this integral is exactly equal to the last term in (6.113) with opposite sign. Thus, the ﬁnal form of the incremental virtual work derived directly from the virtual work equation in the current conﬁguration is

◦ ∂(δuγ ) ∂(duγ ) σαβ dV + δεαβ Dσαβ dV. (6.120) d(δVint ) = ∂xα ∂xβ V V This form of the incremental virtual work is identical to (6.104), when the linear relation (6.102) is introduced between the Truesdell stress increment and the incremental strains dεγδ . It is seen that this simple form of the incremental virtual work is connected with the use of the Truesdell stress increment. In this formulation the Truesdell stress increment is assumed to be provided by a constitutive relation representing the material behavior. A diﬀerent objective stress rate or increment can be used, but then it

180

Deformation and equilibrium of solids

must be introduced into the incremental virtual work equation in a consistent manner. In essence this amounts to substitution of the new objective stress increment into the incremental virtual work equation (6.120) via its relation to the Truesdell stress increment. As a result, additional terms will appear in the ﬁrst integral on the right side of (6.120), representing a modiﬁcation of the geometric stiﬀness following from a modiﬁed interpretation of convection of the current state of stress. In general this will lead to a non-symmetric tangent stiﬀness, as seen in the case of the Jaumann stress increment by substitution of the incremental form of (6.58). The lack of symmetry of the tangent stiﬀness for alternative stress rates is discussed in detail in connection with the rotated Green–Naghdi stress rate in section 7.3.2.2 of Simo and Hughes (1998).

6.4.3 Finite element implementation The ﬁnite element implementation of each step of the updated Lagrangian formulation follows that of the total Lagrangian formulation closely, but contains three essential diﬀerences: the current geometry with current surfaces and volumes is used, the virtual and incremental strains are simpliﬁed because they refer to current geometry, and ﬁnally the stress components must be updated from the objective increment provided by the constitutive relation in order to refer to ﬁxed spatial coordinates. In the updated Lagrangian formulation the virtual and incremental displacements δu(x) and du(x) are considered as functions of the current conﬁguration x. This implies a representation of the virtual displacements in the form hn (x) δun (6.121) δu(x) = n

in terms of shape functions hn (x) and nodal values δun , n = 1, 2, . . . A similar representation is used for the incremental displacements in terms of their nodal values dun . The external virtual work is used to deﬁne equivalent external nodal forces fnext via the relation

$ T ext T hn (x) t dS + hn (x) p dV , (6.122) δVext = δun fn = δun S

where summation over the nodes n = 1, 2, . . . is implied, and loads refer to current surface and volume. By the standard procedure, in which each nodal component is selected as the only non-vanishing component, this scalar

6.4 Updated Lagrangian formulation

equation generates the external nodal forces

hn (x) t dS + hn (x) p dV. fnext = S

181

(6.123)

Thus, in the updated Lagrangian formulation the external nodal forces are obtained by integrating the surface traction and the volume force, weighted by the corresponding shape function hn (x). Also the Cauchy stress and the linear strain components satisfy component symmetry conditions that permit replacement of the nine-component tensor index format with a six-component format. The six-component form of the Cauchy stress is σ = [ σ11 , σ22 , σ33 , σ23 , σ31 , σ12 ]T ,

(6.124)

while the corresponding virtual strain components are δε = [ δε11 , δε22 , δε33 , 2δε23 , 2δε31 , 2δε12 ]T .

(6.125)

A similar formula holds for the strain increment dε. In the updated Lagrangian formulation the internal virtual work involves the virtual strain components δεαβ , given by (6.107). When substituting the virtual displacement representation (6.121) into the strain deﬁnition, the virtual strain is represented as δε(x) = Bn (x) δun , (6.126) n

where the matrix strain interpolation matrix Bn (x) corresponding to node n is given by           Bn =         

∂hn ∂x1

∂hn ∂x2

∂hn ∂x3

∂hn ∂x3 ∂hn ∂x2

0 ∂hn ∂x1

∂hn ∂x3 ∂hn ∂x2 ∂hn ∂x1 0

        .        

(6.127)

The strain interpolation matrix Bn (x) for the updated Lagrangian formulation is seen to be a special case of the strain interpolation matrix B0n (x0 ) for the total Lagrangian formulation. It corresponds to the choice of the

182

Deformation and equilibrium of solids

current conﬁguration x as initial conﬁguration x0 . This introduces the simpliﬁcation ∂xα /∂x0β = δαβ and removes the need for the superscript 0 in the formula (6.88) for B0n (x0 ). In spite of the apparent simplicity of the formula (6.127) for Bn (x), the shape functions hn (x) are usually only given as implicit functions of the current coordinates x and thus the matrix Bn (x) must be evaluated numerically at speciﬁc points. The internal virtual work deﬁnes internal nodal forces fnint by substitution of the virtual strain ﬁeld into (6.110): $ #

δVint = δuTn fnint = δuTn Bn (x)T σ dV , (6.128) V

where summation over the nodes n = 1, 2, . . . is implied. This scalar work equation generates the internal nodal forces

int Bn (x)T σ dV. (6.129) fn = V

The calculation of the internal nodal forces requires evaluation of a volume integral of the product of the current Cauchy stress and the strain representation matrix Bn (x). In the discretized problem equilibrium, or deviation from equilibrium, is expressed in terms of the nodal residual forces rn = fnext − fnint .

(6.130)

The residual forces are calculated via the external and internal nodal forces by integration of the current loads and estimate of the Cauchy stresses. If the residual forces are not negligible, they are eliminated by iterations using the tangent stiﬀness matrix. The tangent stiﬀness is evaluated from the increment of the internal virtual work (6.120) using the constitutive incremental relation (6.102) for the Truesdell stress increment. When substituting the virtual and incremental strains from representations of the form (6.121), the incremental internal virtual work is obtained in the form

$ # ∂h ∂hm n T d(δVint ) = δun I σαβ dV + BTn C Bm dV dum , (6.131) ∂xβ V ∂xα V where I is the two-dimensional unit tensor. The 6 × 6 matrix C is the incremental stiﬀness (6.102) for the Truesdell increment of the Cauchy stress, ◦

Dσ = C dε.

(6.132)

As for the total Lagrangian formulation, the incremental internal virtual

6.4 Updated Lagrangian formulation

183

work relation is of the form d(δVint ) = δuTn Knm dum ,

(6.133)

where summation over n and m is implied, and Knm is the contribution to the tangent stiﬀness matrix from the combination of nodes n and m. In the updated Lagrangian format the tangent stiﬀness matrix contribution Knm is given by

∂hm ∂hn Knm = I σαβ dV + BTn C Bm dV. (6.134) ∂xβ V ∂xα V When comparing this expression to the equivalent total Lagrangian formula (6.96) it is seen that in the ﬁrst integral, representing the geometric stiﬀness, the deformation gradient changes the gradient factors, leaving the Cauchy stress instead of the Piola–Kirchhoﬀ stress, and in the second integral, representing the constitutive relation of the material, the deformation gradient also changes the strain representation matrix to B and the constitutive stiﬀness to C for the Truesdell increment of the Cauchy stress. While the ﬁrst of these changes is a simple substitution, the second involves basic considerations if the coeﬃcient matrix C is to be derived directly from material behavior. The iterative procedure provides an estimate of the displacement increment dun at each of the nodes. This estimate gives the strain increment dε by the representation (6.126). In the updated Lagrangian formulation the strain increment is related to an objective stress increment, e.g. the Truesdell increment of the Cauchy stress shown in (6.132). The update of the stress components must then be made by use of the deﬁning relation for that particular objective stress increment. In the case of the Truesdell stress increment the component formula follows from (6.101) as ◦

dσαβ = Dσαβ − dεγγ σαβ +

∂(duβ ) ∂(duα ) σγβ + σαγ . ∂xγ ∂xγ

(6.135)

This formula can be expressed in a computationally more convenient form using the six-component stress format, ◦

dσ = Dσ − σ dεγγ + Tn dun . The increment of the volume strain is calculated from ∂h ∂h ∂h n n n dun dεγγ = ∂x1 ∂x2 ∂x3 n

(6.136)

(6.137)

184

Deformation and equilibrium of solids

while the matrix Tn is given by a formula similar to that of the strain representation matrix Bn , but with the terms weighted by the stress components, 



∂hn  2σ1γ ∂xγ

  0     0  Tn =    0   ∂hn   σ3γ ∂xγ   ∂hn σ2γ

∂xγ

0 ∂hn 2σ2γ ∂xγ 0 σ3γ

∂hn ∂xγ 0

σ1γ

∂hn ∂xγ

   0   ∂hn   2σ3γ ∂xγ  . ∂hn   σ2γ ∂xγ   ∂hn  σ1γ  ∂xγ  

(6.138)

In all these formulae, summation over the contributing nodes n is implied. The formula (6.135) is for inﬁnitesimal increments. Actual calculations work with ﬁnite increments. This suggests a reinterpretation of the stress update in which the strain increment ∆ε is considered as a linearized Green strain increment with the current state as reference state. This strain increment produces a Piola–Kirchhoﬀ stress increment ∆S, also with the current state as reference state. The current conﬁguration is denoted x1 , and the displacement increment ∆u leads to the new conﬁguration x2 = x1 + ∆u. The updated Cauchy stress then follows from the transformation formula (6.38) in the form 2 σαβ

∂x2 J1 ∂x2α 1 β = , σγδ + ∆Sγδ 1 J2 ∂xγ ∂x1δ

(6.139)

where the terms in parentheses represent the updated Piola–Kirchhoﬀ stress with x1 as basis conﬁguration. In summary, the implementation of the updated Lagrangian formulation using the Truesdell increment of the Cauchy stress is similar in structure to the total Lagrangian formulation but leads to a simpler strain representation matrix Bn . However, some of this simpliﬁcation is oﬀset by the need to update the stress components from their objective increments. This leads to a matrix fairly similar to the strain representation matrix B0n of the total Lagrangian formulation. The use of an alternative objective stress increment will introduce additional terms in the tangent stiﬀness matrix, that will then generally lose its symmetry, even for a symmetric constitutive matrix.

6.5 Summary of non-linear motion of solids

185

6.5 Summary of non-linear motion of solids Motion of a solid body is described either with reference to the individual particles of the body, the material or Lagrangian description, or with reference to the current positions in space, the spatial or Eulerian description. In the material formulation the properties of the local deformed state around a particle are described via the deformation gradient, which can be used to deﬁne several diﬀerent strain measures. A special role is taken by the Green strain, which is a quadratic function of the displacement gradient. While the Green strain may not be ideal for the formulation of constitutive relations for large deformation, its quadratic form leads to a particularly simple formulation of problems with large displacement and moderate strain, used e.g. in the formulation of momentum- and energy-conserving algorithms for dynamics in Chapter 9. The principle of virtual work serves as an important tool for the formulation of consistent theories of continuous deformable bodies. In its basic form it is a statement of equality between the external virtual work, performed by the external loads through a virtual displacement ﬁeld, and the internal virtual work, performed by the internal stresses through the corresponding virtual strain ﬁeld. In the case of the Green strain the postulated form of the internal virtual work leads to a clear interpretation of the associated Piola–Kirchhoﬀ stress, simply by reformulation of the involved integrals. The principle of virtual work also serves as a convenient means of connecting a material description in terms of e.g. Green strain and Piola–Kirchhoﬀ stress with an equivalent spatial formulation in terms of linearized small strain and Cauchy stress. When the kinematic relation between the strain increments has been established, the relation between the corresponding conjugate stresses again follows from simple reformulation of the integrals. The solution of speciﬁc problems relies on the constitutive relation between the stresses and strains in the body. The material properties, including the state of deformation, reside in the material frame of reference, and thus a spatial formulation must be based on proper representation of these material properties in the spatial frame. An important aspect is the relation between the change of stress as observed in the spatial frame, and as observed relative to a local material frame. This issue is dealt with by the so-called objective stress rate, i.e. a formulation in the spatial frame, that accounts for the convection eﬀects introduced by the motion of the material body. The basic form, corresponding to the Lie derivative from tensor analysis, is derived for the Cauchy stress. It is observed that the use of current volume in the deﬁnition of the Cauchy stress leads to an additional

186

Deformation and equilibrium of solids

term in the objective stress rate accounting for material dilation. A simpler formulation is obtained by re-normalizing the stress with respect to a constant amount of material, and this leads to the introduction of the Kirchhoﬀ stress. The Kirchhoﬀ stress plays a more direct role when accounting for internal properties such as internal energy, because it refers to a speciﬁc amount of material. The chapter concludes with a discussion of the formulation of ﬁnite deformation problems for solids by the total and the updated Lagrangian methods. The total Lagrangian formulation is obtained from the virtual work equation in terms of Green strain and Piola–Kirchhoﬀ stress and leads to the deﬁnition of residual forces and tangent stiﬀness in terms of a shape function representation of the displacement ﬁeld. In the updated Lagrangian formulation the current state is used as reference state. The corresponding formulation is obtained from the total formulation by transforming the stresses and strains. This transformation leads directly to the objective stress rate in terms of the Lie derivative. Speciﬁc constitutive relations for elastic and elasto-plastic materials are discussed in the following chapter. 6.6 Exercises Exercise 6.1 A uniform state of deformation can be expressed by the linear relation x = a + Ax0 . Consider the special two-dimensional case given by 0 x1 3.0 0.8 −0.2 x1 = + . 2.0 0.6 1.2 x2 x02 (a) Plot the initial and current base vectors e1 , e2 and the current location of the initial square with corner coordinates x0 = [±1, ±1]T . (b) Find the deformation gradient component matrix Fαβ . (c) Find the Green strain component matrix Eαβ . (d) Find the ratio of current area to original area via the determinant J. Exercise 6.2 In a rectangular two-dimensional bi-linear element the displacement ﬁeld is of the form u1 = a0 + a1 x01 + a2 x02 + a3 x01 x02 , u2 = b0 + b1 x01 + b2 x02 + b3 x01 x02 . (a) Find the deformation gradient component matrix Fαβ . (b) Find the Green strain component matrix Eαβ . 2 , E 2 , E 2 and (c) Determine the polynomial degree of the components E11 22 12 the necessary order of Gauss quadrature for exact integration of all terms in a linear elastic material.

6.6 Exercises

187

Exercise 6.3 The purpose of the objective stress rates discussed in this chapter is to isolate changes associated with the material frame, and thus they must be independent of any superimposed rigid body rotation. A rigid body rotation transforms the current coordinates x to xR = Rx. The corresponding transformed deformation gradient is FR = RF, while the transformed Kirchhoﬀ stress is τ R = Rτ RT . (a) Express the transformed velocity gradient LR in terms of the original ˙ velocity gradient L, the rigid body rotation R and its time derivative R. (b) Use the expression (6.48) for the Truesdell rate of Kirchhoﬀ stress to ◦ ◦ prove the transformation formula τ R = Rτ RT . Exercise 6.4 The symmetric matrix U deﬁned by the polar decomposition theorem (6.17) is called the (right) stretch tensor. Its principal directions deﬁne the directions of zero shear, and the corresponding eigenvalues λ1 , λ2 , λ3 are called the stretches. (a) Find the stretches λ1 , λ2 for the shear problem of Example 6.2 in terms of the parameter γ and illustrate the result graphically. (b) Introduce η by the variable transformation sinh(η) = 12 γ, and express the stretches λ1 and λ2 in terms of η. (c) Give an interpretation of η in terms of logarithmic strain, introduced in Chapter 2. Exercise 6.5 The polar decomposition theorem (6.17) provides a factorization of the deformation gradient tensor, F = RU. (a) Obtain an expression for the velocity gradient tensor L in terms of R ˙ T. and U, and identify the angular velocity tensor Ω = RR (b) Express the angular velocity tensor Ω for the shear problem of Example 6.2 in terms of ϕ. ˙ Exercise 6.6 In the shear problem considered in the examples of this chapter there is no volume change, and for a linear isotropic elastic material the stress rate is then proportional to the strain rate. In terms of the ˙ where µ is the Jaumann stress rate this amounts to the relation τ = 2µε, shear modulus. (a) Show, on the basis of the expressions in Example 6.5, that the mean stress vanishes identically, τ11 + τ22 = 0. (b) Show that the remaining elasticity equations can be expressed as τ˙11 − γ˙ τ12 = 0,

τ˙12 + γ˙ τ11 = µ γ. ˙

τ11 = −τ22 = µ (1 − cos γ).

188

Deformation and equilibrium of solids

(d) Plot and discuss the solution. This solution was given by Dienes (1979) as part of a general discussion of stress rates.

7 Elasto-plastic solids

A central feature of the mechanical behavior of solids and structures is the constitutive relation connecting stresses and strains. This chapter presents the basic theory of elastic and plastic solids, including the techniques needed to implement the models in the numerical framework described in previous chapters. The theoretical basis for the material models described here has developed over the last 50 years, from the early work of Ziegler (1963), over potentials with internal variables (Rice, 1971; Hill and Rice, 1973), to a fully developed theory including conjugate variables and potentials (Halphen and Son, 1975; Germain et al., 1983). A full account of this development is outside the present scope, and the chapter is limited to elastic and elasto-plastic solids. The presentation is deliberately simpliﬁed to purely mechanical effects, leaving out e.g. thermal eﬀects, but retaining the general structure of the formulation. First the notion of reversible elastic deformation is introduced in Section 7.1 in connection with an internal energy potential, and the concept of conjugate variables and potentials is described. The role of stress and strain invariants and some of their properties is described and used in the formulation of isotropic elasticity. The general theory of rate-independent elasto-plasticity with internal variables is then developed in Section 7.2. The key ingredients are the internal energy potential, expressing the recoverable energy in terms of elastic strains and the primary internal variables, and a yield potential, describing the limit of the region of reversible behavior. The equations of plasticity theory are then developed from the postulate of maximum rate of dissipation during plastic deformation. It is demonstrated by elementary arguments that this requirement leads to evolution conditions formulated in terms of the gradients with respect to the conjugate variables. The natural appearance of the conjugate variables has fundamental implications, explaining the diﬀerence between the classic notions of associated 189

190

Elasto-plastic solids

and non-associated plastic ﬂow in terms of properties of the internal energy function. The following two sections describe basic properties of elasto-plastic models and their numerical implementation on two levels. First, the basic properties like hardening and numerical integration are treated on a simple basis in connection with von Mises plasticity, and then the corresponding problems are treated in a general form. Section 7.5 illustrates the formulation of a non-associated plasticity theory for granular materials by using simple arguments on special stress states, and then ‘unfolding’ the theory for general stress states by more qualitative arguments. For simplicity the main presentation is based on small displacement theory. Extension of the results to ﬁnite displacements is treated in Section 7.6, where two diﬀerent deﬁnitions of ﬁnite elastic strain are identiﬁed and discussed. Some elastic materials, such as e.g. wood and ﬁber composites, have anisotropic properties, i.e. properties that depend on the orientation of loading relative to directions ﬁxed in the material, see e.g. Lekhnitskii (1963) and Cowin and Mehrabadi (1995). However, in the following the theory will be specialized to isotropic materials, i.e. materials whose properties are independent of the orientation of the material. This enables a fairly general, yet simple, identiﬁcation of two basic deformation mechanisms: change of volume and change of shape.

7.1 Elastic solids A material is called elastic when the current state of stress is a function of the current state of the conjugate strain. Here the basic considerations will be based on linear kinematics using the Cauchy stress σ with components σαβ and the conjugate linear strain ε with components εαβ . Thus, the property of elasticity is expressed as σαβ = σαβ (εγδ ).

(7.1)

This relation is illustrated in Fig. 7.1 as a single curve, as reversing the strain history leads to reversal of the stress history. During straining of the material the stresses perform work as deﬁned via the principle of virtual work. The accumulated work per unit volume is

εαβ σαβ (¯ εγδ ) d¯ εαβ . (7.2) ϕ(εαβ ) = 0

The integral is illustrated schematically in Fig. 7.1. In order for the accumulated work to be a unique function of the current state of strain, the

7.1 Elastic solids

191

Fig. 7.1. Elastic stress–strain curve with internal energy density ϕ(ε) and complementary energy density ψ(σ).

integrand must satisfy the conditions ∂σγδ ∂σαβ = . ∂εγδ ∂εαβ

(7.3)

Conversely, if these conditions are satisﬁed, the integral ϕ(ε) is only a function of the current state of strain and the stresses are given in terms of the internal energy density function as σαβ =

∂ϕ(ε) . ∂εαβ

(7.4)

Such a material is called hyper-elastic or Green-elastic. Hyper-elasticity implies that reversal to a previous state of strain will imply reversal to the energy at this previous state. Thus, the deformation of a hyper-elastic material is fully reversible. An elastic material without the integrability relations (7.3) is called Cauchy-elastic. The additional freedom in the formulation of Cauchy-elastic models is obtained at the cost of lack of energy reversibility for closed strain cycles, see e.g. Ottosen and Ristinmaa (2005). It is sometimes more convenient to use the complementary energy density ψ(σγδ ), deﬁned in terms of the strain energy density as ψ(σ) = σαβ εαβ − ϕ(ε).

(7.5)

The role of ψ as the complement of ϕ is illustrated schematically in Fig. 7.1. The complementary energy is considered as a function of current stress. The diﬀerential increment of (7.5) therefore takes the form ∂ϕ ∂ψ dσαβ = εαβ dσαβ + σαβ − dεαβ . ∂σαβ ∂εαβ

(7.6)

The terms in parentheses cancel due to (7.4), leaving the relation between

192

Elasto-plastic solids

the strain and the complementary energy density, εαβ =

∂ψ(σ) . ∂σαβ

(7.7)

This relation is a clear analogue to the relation (7.4) between stress and the strain energy density. The use of a relation like (7.5) to deﬁne a complementary potential is called a Legendre transformation and is a procedure widely used in mechanics, see e.g. Washizu (1974) and Sewell (1987). The purpose here is to interchange the roles of stress and strain relative to the original potential. In the following it is convenient to use boldface notation for the stress and strain tensors, and also to introduce the notion of a compact array notation, in which only the independent components appear. In this notation σ and ε are represented by the column arrays σ = [ σ11 , σ22 , σ33 , σ23 , σ31 , σ12 ]T ,

(7.8)

ε = [ ε11 , ε22 , ε33 , 2ε23 , 2ε31 , 2ε12 ]T .

(7.9)

As discussed in the previous chapter, this implies that the contribution to the internal virtual work is represented as σ T dε. This establishes a precise analogy with the boldface tensor notation for conjugate tensor quantities, when the transpose is indicated explicitly. A similar analogy applies to the energy potentials ϕ(ε) and ψ(σ), where the increments can be expressed as dϕ =

∂ϕ dε = σ T dε ∂ε

(7.10)

and ∂ψ dσ = εT dσ. (7.11) ∂σ It is seen that derivatives correspond to the transpose of the array. In the following the boldface notation will be used with the possibility of interpretation as tensors or as component arrays. In the few cases where ambiguities of interpretation can arise, these are resolved by indicating e.g. products in a more explicit form. dψ =

7.1.1 Stress invariants The discussion of material models is here limited to isotropic materials, i.e. materials whose properties are independent of the orientation in space. This gives the possibility of formulating the theory in terms of the so-called stress and strain invariants, i.e. special combinations of tensor components that

7.1 Elastic solids

193

are independent of the orientation. While a similar approach is also possible for anisotropic materials, the number of invariants increases considerably because for anisotropic materials additional invariants must be included to account for the orientation of the stress and strain components relative to special directions in the material. The properties of the stress tensor σ, and thereby the state of stress at a point, can be described via the eigenvalue problem σ nj = σj nj

(no sum).

(7.12)

In this equation nj is a unit vector, deﬁning a section through the point with stress tensor σ. The equation deﬁnes sections with stress vector σj in the direction of the normal nj . These stresses are called the principal stresses and the directions are similarly called the principal directions of the stress tensor. The principal stresses and directions are found by writing the eigenvalue problem in matrix form, ( σ − σj I ) nj = 0

(no sum).

(7.13)

A non-trivial solution for the direction vector nj requires the matrix to be singular, corresponding to the component determinant equation % % σ11 − σ σ12 σ13 % % σ21 σ22 − σ σ23 % % σ31 σ32 σ33 − σ

% % % % = 0. % %

(7.14)

The principal stresses σj are the solution to the corresponding cubic characteristic equation σ 3 − I1 σ 2 + I2 σ − I3 = 0,

(7.15)

in which the coeﬃcients I1 , I2 and I3 are the so-called principal stress invariants. The principal stresses σ1 , σ2 and σ3 are the roots of the characteristic equation, and it can therefore be expressed in the form σ 3 − (σ1 + σ2 + σ3 )σ 2 + (σ2 σ3 + σ3 σ1 + σ1 σ2 )σ − σ1 σ2 σ3 = 0.

(7.16)

The invariance of the parameters I1 , I2 and I3 follows from their identiﬁcation with the coeﬃcients in terms of principal stresses. They can be expressed in terms of the general stress components via the determinant

194

Elasto-plastic solids

equation (7.14) as I1 = tr(σ) = σγγ = σ1 + σ2 + σ3 , I2 =

2 1 2 [tr(σ)

− tr(σ 2 )] =

1 2 (σαα σββ

(7.17) − σαβ σβα )

= σ 2 σ3 + σ3 σ1 + σ1 σ2 , I3 = det(σ) = εαβγ σ1α σ2β σ3γ = σ1 σ2 σ3 .

(7.18) (7.19)

Here tr(σ) denotes the trace deﬁned as the sum of the diagonal terms and det(σ) denotes the determinant of the component matrix. It is an interesting fact that the stress tensor σ satisﬁes its own characteristic equation, when interpreted in tensor format. The proof is quite simple and consists in pre-multiplication of the eigenvalue equation (7.13) with factors containing the two remaining principal stress components. The order of the factors can be interchanged, and thus the resulting equation can be written in the form ( σ − σ1 I )( σ − σ2 I )( σ − σ3 I ) nj = 0.

(7.20)

By changing the order of the factors to place the factor with subscript j last, the equation is seen to be satisﬁed for all three directions n1 , n2 and n3 . Thus, the product of the three factors must be the zero tensor, and multiplication gives the cubic tensor equation σ 3 − (σ1 + σ2 + σ3 )σ 2 + (σ2 σ3 + σ3 σ1 + σ1 σ2 )σ − σ1 σ2 σ3 I = 0. (7.21) This is the Cayley–Hamilton equation, stating that the stress tensor satisﬁes its own characteristic equation. The equation implies that σ 3 can be expressed as a linear combination of σ 2 , σ and I with coeﬃcients in terms of the stress invariants I1 , I2 and I3 . If the equation is multiplied with σ the fourth power σ 4 can be reduced to lower powers, and so on. By continuing this process it can be concluded that a power series in the stress tensor σ can be expressed in terms of a basis consisting of the three members I, σ and σ 2 with coeﬃcients in terms of the stress invariants I1 , I2 and I3 . Scalar functions of the stress tensor σ must be formed by invariants of the series expansions, and thus must be functions of the three stress invariants. The implications may be illustrated with reference to the complementary energy ψ(σ). It follows from the argument above that it can be expressed as a function of the three stress invariants, i.e. in the form ψ(I1 , I2 , I3 ). The strain was found by the stress derivatives as shown in (7.7). This can be

7.1 Elastic solids

195

expressed using the invariant format as ε =

∂ψ ∂ψ ∂I1 ∂ψ ∂I2 ∂ψ ∂I3 = + + , ∂σ T ∂I1 ∂σ T ∂I2 ∂σ T ∂I3 ∂σ T

(7.22)

where the transpose has been introduced to make the notation consistent with the condensed component format. It is seen from this relation that the derivatives of the invariants with respect to σ serve as a three-component basis for the strain tensor. The derivatives of the invariants follow from their deﬁning relations (7.17)– (7.19). They are most easily derived in component form. The derivative of I1 follows directly from the deﬁnition as the trace, ∂I1 = δαβ . ∂σαβ

(7.23)

Similarly, the derivative of the second invariant I2 is found as ∂I2 = σγγ δαβ − σαβ . ∂σαβ

(7.24)

Finally, the derivatives of the invariant I3 follow from the observation that the deﬁnition as the stress tensor determinant can be expanded as products of components and co-factors along the row or column containing the component σαβ . In this expansion the selected component σαβ only appears once, namely as factor to the corresponding co-factor. Thus, the derivative is the co-factor corresponding to σαβ . The co-factor can be expressed in component form as ∂I3 = 12 εαγδ εβκλ σγκ σδλ . (7.25) ∂σαβ The factor 12 compensates for the inclusion of two permutations that each contribute to the full result. It is often convenient to use a diﬀerent set of stress invariants that correspond more directly to the properties of materials. In this connection the mean stress σm =

1 3 tr(σ)

1 3 σγγ

1 3 I1

(7.26)

plays a central role. The idea is to decompose the total stress into an isotropic stress state deﬁned by the mean stress and the remaining stress representing the deviation from an isotropic state of stress, and therefore called the deviatoric stress. The deviatoric stress is deﬁned as σ = σ − 13 tr(σ)I = σ − σm I

(7.27)

196

Elasto-plastic solids

corresponding to the component form = σαβ − 13 σγγ δαβ = σαβ − σm δαβ . σαβ

(7.28)

The mean stress state is described by the principal stress invariant I1 , and two new invariants that do not contain the mean stress are now introduced to characterize the deviatoric stress state. The deviatoric stress invariants are deﬁned by the corresponding powers of the deviatoric stress tensor as J1 = tr(σ ) = σγγ = 0,

(7.29)

J2 =

2 1 2 tr(σ )

1 2 σαβ σβα ,

(7.30)

J3 =

3 1 3 tr(σ )

1 3 σαβ σβγ σγα .

(7.31)

The fact that the mean deviatoric stress vanishes provides a relation between the components. This leads to two alternative formats for the expressions of the deviatoric invariants in terms of principal stresses: J2 =

2 1 2 (σ1

+ σ22 + σ32 ) = −(σ2 σ3 + σ3 σ1 + σ1 σ2 ),

(7.32)

J3 =

3 1 3 (σ1

+ σ23 + σ33 ) = σ1 σ2 σ3 ,

(7.33)

where the ﬁrst expression follows directly from the deﬁnition, while the second can be demonstrated by use of the zero mean deviatoric stress property. An important use of the deviatoric invariants is in the expression of the complementary energy density in the form ψ(I1 , J2 , J3 ). When expressed in terms of these variables, the strain relation (7.7) takes the form ε =

∂ψ ∂ψ ∂I1 ∂ψ ∂J2 ∂ψ ∂J3 = + + . T T T ∂σ ∂I1 ∂σ ∂J2 ∂σ ∂J3 ∂σ T

(7.34)

In this format the basis consists of the derivatives ∂I1 /∂σ T , J2 /∂σ T and ∂J3 /∂σ T . The ﬁrst of these is the unit tensor I as shown in (7.23). The two remaining terms are evaluated by diﬀerentiation through the deviatoric stress tensor σ . The components of this diﬀerentiation follow from (7.27) as ∂σγδ = δαγ δβδ − 13 δγδ δαβ . (7.35) ∂σαβ The derivative of the second deviatoric stress invariant J2 then follows from (7.30) as ∂J2 = σαβ . (7.36) ∂σαβ

7.1 Elastic solids

197

Clearly, this component is independent of the mean stress σm , in contrast to the derivative of I2 given in (7.24). This also follows from the property ∂J2 /∂σγγ = 0. The derivative of the third deviatoric stress invariant J3 follows from (7.31) as ∂J3 = σαγ σγβ − 23 J2 δαβ . ∂σαβ

(7.37)

Also for this derivative the contracted form vanishes, ∂J3 /∂σγγ = 0, indicating the independence of the mean stress. Example 7.1. Graphic representation of stresses. A state of stress σ is conveniently represented graphically in terms of its principal components σ1 , σ2 , σ3 . In particular, this representation permits an illustration of the decomposition of the stress state into the sum of the mean stress state Iσm and the deviatoric stress state σ , as shown in Fig. 7.2. The geometry of this stress space is central to the development of elasto-plastic material models and is discussed in some detail, e.g. by Chen and Han (1988).

Fig. 7.2. Principal stress space with deviatoric plane.

The mean stress is represented by the vector (1, 1, 1)σm in the direction of the principal diagonal. The deviatoric components are represented by the vector (σ1 , σ2 , σ3 ) that is orthogonal to the mean state vector. Thus, the deviatoric stress lies in the deviatoric plane illustrated by the triangle that intersects the axes at 3σm . The location of the stress in the principal stress space is conveniently determined by the stress invariants I1 , J2 of the√deviatoric plane is described by its distance and J3 . The location √ from the origin 3σm = I1 / 3. The length of the deviatoric stress vector √ (σ1 , σ2 , σ3 ) is 2J2 , and the direction in the deviatoric plane is determined by a combination of J3 and J2 as discussed in Section 7.5.

198

Elasto-plastic solids

7.1.2 Strain invariants and small strain elasticity The properties relating to principal components, invariants, etc. have been described relating to the stress tensor σ. Similar results apply to the strain tensor ε. Within small strain analysis it is customary to use the linear volumetric strain εv = tr(ε) = εγγ

(7.38)

instead of the mean strain. This gives the deviatoric strain components in the form εαβ = εαβ − 13 εγγ δαβ = εαβ − 13 εv δαβ .

(7.39)

The principal strain invariants I¯1 , I¯2 and I¯3 follow from (7.17)–(7.19), when replacing the stress symbol σ with the strain symbol ε. As in the case of stresses, the ﬁrst principal strain invariant I¯1 = tr(ε) = εγγ = ε1 + ε2 + ε3

(7.40)

can be used together with the second and third deviatoric strain invariants, deﬁned in the same way as the deviatoric stress invariants, J¯2 =

2 1 2 tr(ε )

1 2 εαβ εβα ,

(7.41)

J¯3 =

3 1 3 tr(ε )

1 3 εαβ εβγ εγα .

(7.42)

The property of zero mean value of the deviatoric strain leads to two alternative forms of the formulae for the principal component forms, exactly like (7.32) and (7.33) for the deviatoric stress invariants. For isotropic materials the internal energy can be expressed in terms of the deviatoric strain invariants in the form ϕ(I¯1 , J¯2 , J¯3 ). The stress relation (7.10) then takes the form ∂ϕ ∂ I¯1 ∂ϕ ∂ J¯2 ∂ϕ ∂ J¯3 ∂ϕ = ¯ + ¯ + ¯ . (7.43) T T T ∂ε ∂ I1 ∂ε ∂ J2 ∂ε ∂ J3 ∂εT In this format the basis consists of the derivatives ∂ I¯1 /∂εT , J¯2 /∂εT and ∂ J¯3 /∂εT . The components follow from the similar stress basis given in (7.23), (7.36) and (7.37) as ∂ϕ ∂ϕ ∂ϕ 2 ¯ ∂ϕ σ = I + ¯ ε + ¯ ε ε . − (7.44) J 3 2 ¯ ¯ ∂ I1 ∂ J3 ∂ J2 ∂ J3 σ =

It follows from this relation that a linear relation between conjugate stress and strain tensors implies an internal energy function of the form ϕ(I¯1 , J¯2 ), because dependence on J¯3 would lead to a term with the quadratic tensor product ε ε .

7.1 Elastic solids

199

The linear isotropic elastic model follows from (7.44), when the internal energy is a hom*ogeneous function of degree two in the strain components as ϕ(I¯1 , J¯2 ) =

1 ¯2 2 k I1

+ 2µJ¯2 =

2 1 2 k(εγγ )

+ µεαβ εβα .

(7.45)

This energy function is described by two additive terms: a volumetric term multiplied by the bulk modulus k, and a deviatoric term multiplied by the shear modulus µ. The corresponding stresses follow by diﬀerentiation with respect to the strain components as σαβ =

∂ϕ = kεγγ δαβ + 2µεαβ = (k − 23 µ)εγγ δαβ + 2µεαβ . ∂εαβ

(7.46)

The ﬁrst form shows the split into mean stress and deviatoric stress: σm =

1 3 σγγ

= kεγγ ,

σαβ = 2µεαβ .

(7.47)

These two equations describe the two basic deformation mechanisms of linear isotropic materials – change of volume and change of shape. The change of shape is associated with zero-mean stress and strain states and is sometimes called generalized shear. The relations are illustrated in Fig. 7.3.

Fig. 7.3. Linear isotropic elasticity in dilation and generalized shear.

The linear elastic relation (7.46) can be written in boldface format as σ = C ε,

ε = C−1 σ,

(7.48)

where the relations can be interpreted either in full tensor format or in the reduced vector format as need may be. The stiﬀness tensor components follow from (7.46) as Cαβγδ = (k − 23 µ)δαβ δγδ + µ(δαγ δβδ + δαδ δβγ ).

(7.49)

In the vector format C is a 6 × 6 stiﬀness matrix, and its inverse C−1 a 6 × 6 ﬂexibility matrix. For hyper-elastic materials the existence of an energy function implies the symmetry relations Cαβγδ = Cγδαβ .

(7.50)

200

Elasto-plastic solids

These relations are equivalent to symmetry of the 6 × 6 stiﬀness matrix C and its inverse. For linear isotropic materials the elastic parameters are often represented by Young’s modulus E and Poisson’s ratio ν, describing the uniaxial tension stiﬀness and transverse contraction, respectively. The somewhat more basic parameters k and µ are expressed in terms of these as k =

E , 3(1 − 2ν)

The elastic ﬂexibility matrix  C−1 =

1 −ν  1  −ν

E 

−ν 1 −ν

E . 2(1 + ν)

µ =

(7.51)



−ν −ν 1

    

2(1 + ν) 2(1 + ν)

(7.52)

2(1 + ν)

and the corresponding stiﬀness matrix 

1−ν ν ν ν 1−ν ν   ν ν 1−ν 

E C = (1+ν)(1−2ν)  

 1 2 −ν

1 2 −ν

    

(7.53)

1 2 −ν

follow from (7.49), when using the vector representation (7.8) for stresses and (7.9) for strains.

7.1.3 Isotropic elasticity at ﬁnite strain In practice the isotropic linear elastic relation developed in the previous section is limited to small strains. The theory demonstrates the important distinction between change of volume and change of shape. In the small strain theory the change of volume is represented by the linear volume strain, deﬁned in (7.38) as εv = εγγ . However, the interpretation of εv as a volume strain only holds within linearized displacement theory. In the case of ﬁnite motion and ﬁnite strain, volume changes must be described by a precise measure of the change of volume, e.g. the Jacobian J, introduced in (6.36) as the ratio between a volume element in the current state and in the reference state, J = dV /dV0 . A general discussion of ﬁnite strain models in elasticity can be found in e.g. Ogden (1984) and Holzapfel (2000). In the context of ﬁnite deformation it is often convenient to describe the current conﬁguration relative to the original in a factored format without

7.1 Elastic solids

201

explicit use of strains. For this purpose the right Green deformation tensor is introduced by the deﬁnition C = FT F.

(7.54)

In terms of this tensor the Green strain tensor deﬁnition (6.12) takes the form E =

1 2 (C

− I).

(7.55)

In the development of constitutive relations, volume changes play an important role and will often be represented directly in the model, e.g. via the Jacobian determinant of the deformation gradient tensor F, J = det(F) = det(C)1/2 .

(7.56)

The dilatational part of the motion, described by J, must be supplemented by a tensor measure of the change of shape, the so-called isochoric or volumepreserving deformation. A split of the inﬁnitesimal motion into a volumetric and an isochoric part can be accomplished by introducing a factored format, in which the local dilation appears as a separate factor (Simo et al., 1985; Simo, 1988a,b). This is expressed by the factored format F = Fvol Fiso = Fiso Fvol .

(7.57)

The volumetric part Fvol represents a simple isotropic scaling with the appropriate power of the Jacobian determinant J: Fvol = J 1/3 I,

Fiso = J −1/3 F.

(7.58)

Hereby the motion represented by Fiso is volume-preserving, as demonstrated by the property det(Fiso ) = 1. The factored format of the deformation gradient leads to the following factored format for the corresponding right Green deformation tensor: C = J 2/3 Ciso ,

Ciso = FTiso Fiso .

(7.59)

The deformation tensor Ciso is then used to describe the isochoric part of the motion by assuming an internal energy density composed of two additive parts, whereby ϕ(C) = ϕvol (J) + ϕiso (Ciso ),

(7.60)

representing the dilatational and the isochoric part of the motion, respectively. The Piola–Kirchhoﬀ components of the stress follow from diﬀerentiation of the energy density function with respect to the ﬁnite Green strain E. When

202

Elasto-plastic solids

using the relation (7.55) the diﬀerentiation is changed to the deformation tensor C, S =

∂ϕ ∂ϕvol ∂J ∂ϕiso ∂Ciso ∂ϕ = 2 = 2 +2 . ∂E ∂C ∂J ∂C ∂Ciso ∂C

(7.61)

The ﬁrst term describes the dilatational behavior. The potential ϕvol (J) should exhibit monotonic dependence on J on both sides of a minimum at J = 1. The potential ϕvol (J) =

2 1 1 2 k [ 2 (J

− 1) − ln(J) ]

(7.62)

was introduced by Simo (1988a,b) and Simo and Miehe (1992). Other speciﬁc forms can be used, e.g. ϕvol (J) = 12 k(ln J)2 from Belytschko et al. (2000). The derivative with respect to the deformation tensor C is evaluated in two steps. First, the derivative of the particular function ϕvol (J) with respect to J, ∂ϕvol = 12 k ( J − J −1 ), (7.63) ∂J followed by the derivative of J. The derivative of J follows from a general formula for the derivative of a determinant with respect to its components. The contribution to a determinant from any particular component can be expressed as the product of this component with the corresponding co-factor. The co-factor is deﬁned by the determinant of the matrix, where the row and column containing the component have been deleted. Thus, diﬀerentiation with respect to the component gives the co-factor. This result can be expressed for the matrix C in the compact notation ∂ det(C) = cof(C) = det(C) C−T . (7.64) ∂C The last result follows from the expression of the inverse in terms of cofactors. When using the expression (7.56) for J and the symmetry of C, the formula takes the form ∂J = 12 J C−1 . (7.65) ∂C This completes the evaluation of the dilatational part, ∂ϕvol ∂J = 21 k (J 2 − 1) C−1 . ∂J ∂C For small displacements the leading term of this relation is Svol = 2

Svol

2 1 2 k (J

− 1) I =

1 2 k [ det(C)

− 1 ] I k Eγγ I.

(7.66)

(7.67)

7.2 General plasticity theory

203

This corresponds to the dilation relation (7.47a) for the linearized theory. The isochoric motion is described in terms of the scaled deformation tensor Ciso deﬁned in (7.59), ϕiso (Ciso ) =

1 2 µ [ tr(Ciso )

− 3 ].

(7.68)

Diﬀerentiation of this energy potential gives Siso = 2

∂ϕiso ∂Ciso = µ J −2/3 [ I − 13 tr(C)C−1 ], ∂Ciso ∂C

(7.69)

where Ciso is represented by (7.59). In total the two contributions give the Piola–Kirchhoﬀ stress S =

2 1 2 k (J

− 1) C−1 + µ J −2/3 [ I − 13 tr(C)C−1 ].

(7.70)

The physical content of this relation is more easily seen when expressed in terms of the spatial Kirchhoﬀ stress. The Kirchhoﬀ stress is obtained by the transformation (6.40). The transformation cancels the factors C−1 , and leads to the left Green deformation tensor B together with its scaled isochoric counterpart Biso , B = F FT ,

Biso = Fiso FTiso = J −2/3 B.

(7.71)

In terms of these tensors the Kirchhoﬀ stress is found as τ =

2 1 2 k (J

− 1) I + µ[ Biso − 13 tr(Biso ) I ].

(7.72)

This formula brings out the similarity with the linear stress–strain relations (7.47) for dilatational and isochoric motion.

7.2 General plasticity theory Plasticity theory is used to denote constitutive relations in which irreversible processes take part, when the stress or strain state exceeds a certain limit, described by the yield condition. Thus, plasticity theory is concerned with constitutive relations which combine an elastic state of deformation with additional irreversible eﬀects, typically represented by additional straining. The discussion in this chapter is limited to rate independent forms of plasticity, i.e. forms of plasticity theory in which the additional straining is independent of time. Theories in which the irreversible deformation is associated with a time scale are called visco-plasticity, see e.g. Lemaitre and Chaboche (1990) or Ottosen and Ristinmaa (2005). In this section the structure of rate-independent plasticity theory is presented by using the modern approach of potentials, but without including non-mechanical eﬀects like

204

Elasto-plastic solids

temperature. Thus, the theory takes the simplest form that illustrates the interplay between internal energy and the limit of reversible deformation, described by a yield surface. This section presents the general framework and proceeds to discuss the special case of associated plasticity theory, often used e.g. for metals. The implications of the more general non-associated plasticity theory are discussed in Section 7.4, and application of the general theory to the area of geotechnical problems is presented in Section 7.5.

7.2.1 Reversible deformation Consider a material point in a state of deformation described by the elastic strain εe and a set of internal variables κ. The subscript e denotes elastic and refers to reversible elastic deformation. The internal variables κ are parameters describing irreversible processes in the material. They are chosen such that they do not change during reversible elastic deformation. An example is the speciﬁc pore volume in a porous material that will contain a part due to the elastic deformation, but may also develop an irreversible contribution due to constitutive changes, e.g. in a granular material. The internal parameters are here selected such that they represent only the irreversible part, while any reversible contribution is included in the elastic properties. The material is assumed to have an internal energy density described by the function ϕ(εe , κ). This form includes two extensions relative to the elastic deformation described in Section 7.1: the elastic strain εe may be only part of the total straining of the material, and the internal variables κ may contribute to the energy density. Conjugate variables are deﬁned via the partial derivatives of the energy function, σ =

∂ϕ = ∇ε ϕ(εe , κ), ∂εTe

ζ =

∂ϕ = ∇κ ϕ(εe , κ). ∂κT

(7.73)

Here σ is the stress tensor, and ζ is the set of internal variables, conjugate to κ. It is here convenient to introduce the special notation for partial derivatives in terms of the symbol ∇ with an appropriate subscript indicating the variable with respect to which the function is diﬀerentiated. The system is assumed to change with time, and the time derivatives are denoted by a dot over the symbol. The theory to be developed is rate independent, and thus time appears merely as a parameter, and can be considered as ‘pseudo time’. The time derivatives of the conjugate variables

7.2 General plasticity theory

205

in (7.73) follow by diﬀerentiation through εe and κ, ˙ σ˙ = (∇εε ϕ) ε˙ e + (∇εκ ϕ) κ,

(7.74)

˙ ζ˙ = (∇κε ϕ) ε˙ e + (∇κκ ϕ) κ.

(7.75)

The role of the internal parameters κ as describing irreversible changes of material conﬁguration implies that κ˙ = 0 in the elastic state. Thus, the reversible development of the conjugate external variables σ and internal variables ζ is governed by the relations σ˙ = (∇εε ϕ) ε˙ e ,

ζ˙ = (∇κε ϕ) ε˙ e .

(7.76)

The ﬁrst relation identiﬁes ∇εε ϕ as the reversible elastic incremental stiﬀness tensor, C(εe , κ) = ∇εε ϕ(εe , κ).

(7.77)

The second relation demonstrates that for a material in which the external and internal variables are coupled, the conjugate internal variables ζ will change under reversible elastic deformation, while the primary internal variables κ remain constant.

Fig. 7.4. Domain of reversible elastic deformation. Reversible deformation only takes place when the primary variables εe and κ remain within a limited domain, described by a condition of the form F (εe , κ) ≤ 0.

(7.78)

This relation and the role of the yield function F (εe , κ) deﬁning the domain of reversible deformation are illustrated in Fig. 7.4. While the material point is inside the domain of reversible deformation the strains εe and thereby the stresses σ may change, while the internal variables κ remain constant. Thus,

206

Elasto-plastic solids

the yield function can be expressed in terms of the stress tensor σ and κ. For convenience this is expressed here without change of symbol as F (σ, κ) ≤ 0.

(7.79)

The stress gradient of the yield function F (σ, κ) is obtained by using the incremental relation between the stress σ and the elastic strain εe for constant value of the internal parameters κ: ∇σ F = (∇εε ϕ)−1 ∇ε F = C−1 ∇ε F.

(7.80)

In any reversible elastic process the internal variables κ remain constant, and thus in this particular state the limiting surface, also called the yield surface, can be described as a function of stress F (σ, κ) with the internal variables κ acting as internal parameters. This, the most commonly used form, is illustrated in Fig. 7.5.

Fig. 7.5. Domain of reversible elastic deformation in stress space. It is important to note that while the yield surface can be expressed either by using the reversible elastic strain εe or the stress σ, there may be restrictions on the representation in terms of the internal variables. The original internal variables κ were introduced in such a way that they remain constant during reversible elastic deformation. Thus, the internal variables κ appear as constant parameters for processes inside the current yield surface. In contrast, the conjugate internal variables ζ will change during reversible processes according to (7.76b), if not all the mixed derivative components ∇κε ϕ vanish. When all mixed derivatives vanish, the conjugate variables ζ remain constant during reversible processes. In that case the yield condition can be expressed in terms of the conjugate internal variables ζ instead of the original internal variables κ. However, in the general case the original internal variables κ, that remain constant during reversible deformation, must be used.

7.2 General plasticity theory

207

In the following it will be assumed that the yield function F is convex in its arguments. This corresponds to the statement that for any two states within the yield surface any weighted linear average of the state variables will also be inside the yield surface. This condition is important for the uniqueness of the irreversible processes to be described next.

7.2.2 Maximum plastic dissipation rate In the case of reversible deformation the internal variables κ remain constant. Thus, it follows from (7.73a) that the work of the stresses σ through the strain rate ε˙ is equal to the rate of change of the internal energy ϕ. It is noted that in reversible deformation the elastic strain constitutes the total strain, and therefore (7.73a) can be written in time rate form as σ T ε˙ = ε˙ Te ∇ε ϕ(εe , κ) = ϕ(ε ˙ e , κ),

(7.81)

where the superscript T denotes transpose, and has been introduced to make the notation cover tensor and matrix notation. In essence this is a statement that the total work of the stresses is stored in the internal energy ϕ, and thus is recoverable. In the case of irreversible deformation dissipation will occur, and the rate of dissipation is deﬁned by ˙ e , κ) ≥ 0. D = σ T ε˙ − ϕ(ε

(7.82)

˙ while The ﬁrst term is the work of the stresses σ through the strain rate ε, the second term is the rate of recoverable work stored in the internal energy function ϕ(εe , κ). The energy function ϕ is a function of the current state of the recoverable strain εe and the internal variables κ, but does not depend explicitly on time. Thus, the rate of change of the internal energy is given by ϕ(ε ˙ e , κ) = ε˙ Te ∇ε ϕ + κ˙ T ∇κ ϕ.

(7.83)

The gradients in this expression are the conjugate variables σ and ζ deﬁned in (7.73). Thus, the internal rate of energy can be written in terms of the rates of the primary variables, multiplied by the corresponding conjugate variables: ˙ ϕ(ε ˙ e , κ) = σ T ε˙ e + ζ T κ.

(7.84)

When this is introduced into (7.82), the rate of dissipation takes the form ˙ D = σ T (ε˙ − ε˙ e ) − ζ T κ.

(7.85)

208

Elasto-plastic solids

The plastic strain rate ε˙ p is now deﬁned as the diﬀerence of the total and the elastic strain rates, appearing in the ﬁrst term. Hereby the total strain rate is the sum of the elastic and the plastic strain rates, i.e. ε˙ = ε˙ e + ε˙ p .

(7.86)

The dissipation relation then takes the simpler form ˙ D = σ T ε˙ p − ζ T κ,

(7.87)

where it is clear that only the plastic part of the strain rate contributes to the dissipation, together with a contribution from the internal variables. The constitutive relations of the material follow from a postulate of maximum dissipation rate under the condition that the variables are bounded by the yield condition deﬁned by the inequality (7.78). The basic idea goes back to von Mises (1928) and Hill (1948), and has since been used by many authors, typically in relation to convex analysis, see e.g. Eve et al. (1990) for a survey. Here, a simple and direct formulation is given, following the approach of optimization theory. The maximum dissipation problem can be stated as a constrained extremum problem of the form maximize

D = σ T ε˙ p − ζ T κ˙

subject to

F (εe , κ) ≤ 0.

σ, ζ

(7.88)

Here the original form of the yield function F (εe , κ) has been retained to stress its role as the bounding surface of the domain of reversible deformation, described by the variables εe and κ. This form helps clarify the role of the internal variables in determining the rules for the development of plastic strains. The maximum rate of dissipation is found by considering the rates ε˙ p and κ˙ as ﬁxed, and then searching for the maximum by considering the conjugate variables σ and ζ as subject to variation. The yield function is given in terms of the original variables εe and κ, and therefore derivatives with respect to the conjugate variables must be obtained by using the deﬁning relations (7.73). If the maximum is attained at a point inside the domain deﬁned by the yield condition, the derivatives of D with respect to the independent variables σ, ζ must vanish. This gives the equations ε˙ p = 0,

κ˙ = 0.

(7.89)

These equations state that there is no development of plastic strain and no development of the internal variables κ. These are precisely the conditions characterizing reversible deformation, as described in Section 7.2.1.

7.2 General plasticity theory

209

Now, assume that the maximum rate of dissipation is attained for a set of variables σ, ζ corresponding to a point on the yield surface. According to (7.88a), surfaces corresponding to equal dissipation rate are hyperplanes ˙ The point of maximum dissipation rate must with normal vector (ε˙ p , −κ). therefore be located on the yield surface and have the same normal vector in σ, ζ space. If this were not the case, a larger value of the dissipation rate D could be found in the neighborhood of the assumed intersection of the dissipation contour and the yield surface. It is convenient to formulate this condition in terms of a potential function G(σ, ζ) that describes the yield surface in terms of the variables σ, ζ: G(σ, ζ) = F (εe , κ),

(7.90)

where the variables are transformed according to (7.73). The partial derivatives of the two potentials F (ε, κ) and G(σ, ζ) are related via the chain rule of diﬀerentiation: ∂σ %%T ∂ζ %%T ∇ε F = % ∇σ G + % ∇ζ G, ∂εe κ ∂εe κ (7.91) ∂ζ %%T ∂σ %%T ∇κ F = % ∇ζ G. % ∇σ G + ∂κ ε ∂κ ε When the conjugate variables σ, ζ are introduced by their deﬁnition (7.73), the relation between the partial derivatives of the potentials takes the form ∇σ G ∇εε ϕ ∇εκ ϕ ∇ε F = . (7.92) ∇κ F ∇κε ϕ ∇κκ ϕ ∇ζ G This provides the relation of the partial derivatives of the yield function F in terms of the original variables εe , κ and the yield function G in terms of the conjugate variables σ, ζ. The solution to the constrained maximization problem (7.88) can now be stated in concise form. In the σ, ζ space the condition of co-directional normals of the dissipation rate function D and the yield potential G takes the form G ∇ ∇σ D σ = λ˙ , (7.93) ∇ζ D ∇ζ G where λ˙ ≥ 0 is a factor of proportionality. This factor is conveniently combined with the yield potential in the product condition λ˙ F (εe , κ) = 0.

(7.94)

This condition covers both of the following two cases. Inside the yield surface

210

Elasto-plastic solids

λ˙ = 0, and the processes are fully reversible with D = 0, as discussed in connection with (7.89). Conversely, points on the yield surface have F = 0 and thus permit λ˙ > 0, corresponding to dissipative processes. These are the Kuhn–Tucker conditions, here expressed in terms of conjugate variables (Kuhn and Tucker, 1951). When substituting the rate of dissipation from (7.85), the evolution equations are obtained as ε˙

= ε˙ e + λ˙ ∇σ G, (7.95)

κ˙ =

− λ˙ ∇ζ G.

For λ˙ = 0 the elastic equations (7.89) corresponding to reversible deformation are recovered. The internal variables κ have been chosen to represent the deviation from the quasi-stationary equilibrium in the elastic state, and as a consequence changes are limited to the plastic state. The plastic strain rate and the rate of the internal variables κ˙ are given by the gradients of the potential function G, and this is therefore often called the ﬂow potential. A similar set of equations was obtained by Lemaitre and Chaboche (1990) by a diﬀerent approach, using special properties of the Legendre transformation of hom*ogeneous functions of degree one instead of the maximum dissipation condition, see e.g. Sewell (1987) for the mathematical background. The plastic strain rate and the gradient of the potential G are illustrated in Fig. 7.6. It is seen that in the case of coupling between external and internal variables the plastic strain increment does not appear as orthogonal to the yield surface in stress space.

Fig. 7.6. Yield surface with coupled plastic ﬂow.

The solution (7.95) for the rates of plastic strain and the internal variables leads to a simple form of the rate of dissipation D in terms of gradients of

7.2 General plasticity theory

the ﬂow potential G. Substitution of ε˙ p and κ˙ e into (7.85) gives D = λ˙ σ T ∇σ G + ζ T ∇ζ G .

211

(7.96)

The dissipation rate is here given entirely as a function of the stress σ and the conjugate internal variable ζ. Thus, this expression may be used as a theoretical basis for derivation of the ﬂow potential G(σ, ζ) from the physics of the dissipation process, see e.g. Schoﬁeld and Wroth (1968) and Krenk (1998) for the case of granular friction materials, discussed in Section 7.5. Much work on plasticity theory and many speciﬁc plasticity models have been based on the uncoupled case, in which ∇εκ ϕ = ∇κε ϕ = 0, corresponding to an additive format of the internal energy: ϕ(ε, κ) = ϕε (ε) + ϕκ (κ).

(7.97)

For this class of plasticity problems the conjugate variables are deﬁned by the uncoupled relations σ = ∇ε ϕε (εe ),

ζ = ∇κ ϕκ (κ).

(7.98)

This implies that the transformation between the variable εe and its conjugate variable σ is independent of the transformation between the internal variable κ and its conjugate internal variable ζ. As a consequence, the gradients of the two yield function representations F and G are given by the uncoupled form of (7.92) as ∇ε F = (∇εε ϕ)∇σ G,

∇κ F = (∇κκ ϕ)∇ζ G

(7.99)

for the class of uncoupled materials. For this class of plastic materials the plastic strain rate is proportional to the stress gradient of the yield function F (σ, κ) as illustrated in Fig. 7.7.

Fig. 7.7. Yield surface with uncoupled plastic ﬂow.

212

Elasto-plastic solids

In the literature, plasticity theory with internal variables that are uncoupled from the elastic strain (or the stress) is often called associated plasticity, because the gradient property used in the ﬂow rule follows directly from the yield function F . In contrast, the general coupled theory is called non-associated plasticity theory, because the direction of plastic ﬂow is not directly given by the original yield function F (σ, κ). In the present context it is probably more descriptive to classify these two types of models as uncoupled and coupled, respectively.

7.2.3 Evolution equations Plastic ﬂow requires λ˙ > 0, and according to the Kuhn–Tucker criterion (7.94) this leads to the condition F = G = 0. This identity must remain satisﬁed during plastic ﬂow, a property often called the consistency condi˙ tion. This provides the extra equation needed to determine the multiplier λ, thereby closing the system of evolution equations. The most elegant general formulation is obtained by expressing the consistency condition in terms of the yield function F (εe , κ). Time diﬀerentiation of the yield condition in terms of this potential gives the consistency condition in the form F˙ (εe , κ) = ε˙ Te ∇ε F + κ˙ T ∇κ F = 0.

(7.100)

The time derivatives ε˙ e and κ˙ are substituted from the ﬂow equations (7.95), whereby

F˙ = ε˙ T ∇ε F − λ˙ (∇σ G)T (∇ε F ) + (∇ζ G)T (∇κ F ) = 0. (7.101) This gives an expression for the plastic multiplier λ˙ in terms of the total ˙ strain rate ε, λ˙ =

(∇ε F )T ε˙ . (∇σ G)T (∇ε F ) + (∇ζ G)T (∇κ F )

(7.102)

It is to be noted that the products of the gradients of the potentials F and G are formed directly by products of their components. This property is a consequence of the gradients being formed in terms of conjugate variables. The evolution equations for elasto-plastic ﬂow are now found from (7.95) ˙ In the ﬁrst of these equations the rate of elastic strain by substitution of λ. ε˙ e is replaced by the stress rate σ by multiplication with the elastic stiﬀness according to (7.76a): σ˙ = (∇εε ϕ) I −

(∇σ G) ⊗ (∇ε F )T ˙ ε, (∇σ G)T (∇ε F ) + (∇ζ G)T (∇κ F )

(7.103)

7.2 General plasticity theory

213

where I is the appropriate unit tensor, and the symbol ⊗ is introduced to indicate that the components of two factors are not contracted. The corresponding evolution law for the internal variables κ follows from the second equation (7.95b) as κ˙ = −

(∇σ G) ⊗ (∇ε F )T ˙ ε. (∇σ G)T (∇ε F ) + (∇ζ G)T (∇κ F )

(7.104)

This is the most compact explicit formulation of the evolution equations, making use of the dual yield and ﬂow potentials F (εe , κ) and G(σ, ζ). In practice, plasticity problems are integrated by forming ﬁnite increments, and thus the gradients appearing in the evolution equations typically change value over the time increment. The format of the evolution equations (7.103)–(7.104) is well suited for explicit integration schemes, in which the non-linear gradients are evaluated using the current value of the variables. However, explicit integration schemes have limitations regarding accuracy and numerical stability, and therefore implicit schemes are often preferred, see e.g. the analyses of Krieg and Krieg (1977), Ortiz and Popov (1985), Ortiz and Simo (1986). In implicit schemes the functions occurring in the evolution formulae are evaluated at an intermediate or the ﬁnal time. This requires iterations, and the evolution equations must therefore be given a format that is suitable for this. In the context of plasticity the incremental step will most often start with a predicted value of the total strain increment ∆ε, and the objective of the numerical scheme will then be to ﬁnd corresponding increments of the variables ∆σ, ∆κ and ∆λ. Here the evolution equations will be reformulated using the time derivatives, while the ﬁnite increment form will be discussed in connection with numerical integration schemes later. The idea is to cast the set of the two evolution equations (7.95a,b) and the consistency equation (7.100) in the form of three equations for the rates ˙ First the strain increment equation (7.95a) is written in the ˙ κ˙ and λ. σ, form (7.105) ε˙ = C−1 σ˙ + λ˙ ∇σ G, where the elastic strain rate has been replaced by the stress rate by use of the elastic stiﬀness tensor C = ∇εε ϕ from (7.77). The equation (7.95b) for the internal variable κ is written in hom*ogeneous form as 0 = κ˙ + λ˙ ∇ζ G.

(7.106)

In the ﬁnal equation, the consistency equation (7.100), the elastic strain is replaced as independent variable by the stress, yielding the equation F˙ = σ˙ T ∇σ F + κ˙ T ∇κ F = 0.

(7.107)

214

Elasto-plastic solids

These three equations are conveniently collected in block matrix form as      0 (∇σ G) C−1 σ˙ ε˙      0 I (∇ζ G)   κ˙  , (7.108) 0 =  T T ˙ ˙ F λ (∇σ F ) (∇κ F ) 0 where F˙ has been indicated explicitly for clarity. It is noted that for materials, where the internal variables are coupled with the elastic strains, these equations are non-symmetric. However, for materials with uncoupled internal parameters symmetry can be obtained by multiplication of the second row with the symmetric matrix (∇κκ ϕ). By (7.99b) this replaces the gradient ∇ζ G with ∇κ F , and the equations then take the symmetric form      0 (∇σ F ) C−1 σ˙ ε˙      (∇κ F )   κ˙  . 0 (∇κκ ϕ) (7.109) 0 =  T T ˙ ˙ F λ (∇σ F ) (∇κ F ) 0 The integrated forms, in which the time derivatives are replaced by ﬁnite increments, are suitable for numerical plasticity algorithms, when the entries in the matrix are recognized as representative mean values, typically evaluated iteratively to correspond to the ﬁnal time. It is sometimes convenient to use a reduced format, in which the eﬀect of the internal parameters is combined into a single parameter H, and the update of the internal parameters κ is considered separately. This format follows, when the time rate of internal variables κ˙ is eliminated from the consistency condition (7.107) by use of the evolution equation (7.106). This gives the consistency equation in the form F˙ = σ˙ T ∇σ F − H λ˙ = 0,

(7.110)

where the contribution from the internal parameters has been combined in the hardening modulus H = (∇ζ G)T (∇κ F ).

(7.111)

In some cases the hardening modulus may be prescribed directly in the model, and the equations (7.108) can then be used in the reduced form C−1 (∇σ G) σ˙ ε˙ = (7.112) T ˙ F λ˙ (∇σ F ) −H supplemented with the evolution equation for the internal parameters, κ˙ = −λ˙ ∇ζ G.

(7.113)

7.2 General plasticity theory

215

This format can be used for a combined implicit–explicit iteration scheme ˙ ˙ κ˙ and λ. for the ﬁnite increments corresponding to σ, If the internal variables are uncoupled from the elastic strain, these equations can be recast in a symmetric form, where the potential G is eliminated. In the deﬁnition of the hardening modulus the gradient ∇ζ G is replaced by the gradient ∇κ F by use of (7.99b), H = (∇κ F )T (∇κκ ϕ)−1 (∇κ F ).

(7.114)

It follows from (7.99a) that ∇σ G = ∇σ F , and thus the block matrix equation takes the symmetric form C−1 (∇σ F ) σ˙ ε˙ = . (7.115) T ˙ F λ˙ (∇σ F ) −H Finally, the evolution equation for the internal parameters is expressed in terms of the gradient ∇κ F by use of (7.99b) as κ˙ = −λ˙ (∇κκ ϕ)−1 ∇κ F.

(7.116)

This simple format is often found in the literature, but in many cases without the identiﬁcation of the coeﬃcient matrix (∇κκ ϕ) as part of the internal energy function. To round oﬀ the general discussion of evolution equations the rate equations (7.102)–(7.104) are given in terms of stress derivatives, the elastic stiﬀness tensor C and the hardening modulus H: λ˙ = σ˙ =

(∇σ F )T C ε˙ , (∇σ G)T C (∇σ F ) + H

C−

κ˙ = −

C (∇σ G) ⊗ (∇σ F )T C ˙ ε, (∇σ G)T C (∇σ F ) + H

(∇σ G) ⊗ (∇σ F )T ˙ C ε. (∇σ G)T C (∇σ F ) + H

(7.117)

(7.118)

(7.119)

For materials with uncoupled internal variables, i.e. for ∇εκ ϕ = 0, the stress gradients are identical, ∇σ G = ∇σ F , and the hardening modulus is given by the symmetric relation (7.114). For materials with internal variables that are coupled to the elastic strains, i.e. for ∇εκ ϕ = 0, the relation between the gradients of the potentials F (εe , κ) and G(σ e , ζ) is non-trivial, and the directions of the two corresponding stress gradients are diﬀerent. This is generally called non-associated ﬂow. It should be noted that, according to the present formulation based on

216

Elasto-plastic solids

potentials and maximum energy dissipation, the stress gradients of the yield potential F (εe , κ) and its dual, the ﬂow potential G(σ, ζ), are not independent, but related through the energy function. In the literature the two potentials are often considered as independent potentials. These theories may be consistent in the sense of dual potentials, if they can be related via a complete internal energy function containing the internal variables. This issue will be considered further in Sections 7.4 and 7.5.

7.2.4 Isotropic and kinematic hardening As seen in the previous section, the plastic deformation of a material is governed by the internal energy function ϕ(ε, κ) and the yield function F (σ, κ). Thus, the initial shape of the yield surface and its possible development during plastic deformation are important for the representation of elastoplastic material models. Elasto-plastic materials whose yield surface does not change with plastic deformation are called perfectly plastic. For these materials the yield function F (σ) does not depend on internal parameters, and the hardening modulus H is identically zero, as follows from (7.110). This group has been intensively studied by analytical methods, see e.g. Hill (1950), Chen and Hahn (1988) and Chakrabarty (2000). In the present context perfect plasticity arises as a special case, in which the yield function F (σ) is independent of internal variables. Thus, perfect plasticity will not be given special attention here, but merely be considered as a limiting case. At any particular instant the current yield surface can be described by a relation of the form f (σ) = σ0 ,

(7.120)

where σ0 represents the yield stress under speciﬁc conditions, e.g. uniaxial loading. This corresponds to a yield function F (σ) of the form F (σ) = f (σ) − σ0 .

(7.121)

In general the yield function F depends on internal variables κ. Often the details of this dependence are not known in terms of ﬁrst principles, and two simple assumptions are often used, either separately or in combination. The ﬁrst is the so-called isotropic hardening, in which only the yield stress σ0 = σ0 (κ) is assumed to depend on the internal variables. This gives the functional form F (σ, κ) = f (σ) − σ0 (κ).

(7.122)

7.2 General plasticity theory

217

Due to the particular functional form, the yield surfaces constitute a set of surfaces typically enclosing each other as illustrated in Fig. 7.8. The name arises from the fact that the expansion is independent of precisely which stress path generated the change of size of the yield surface. As a consequence, there are inﬁnitely many stress paths that create the same sequence of yield surfaces, and thus the ﬁnal state gives only limited information on the process leading to the ﬁnal state.

Fig. 7.8. Isotropic hardening: expanding yield surface. In the isotropic hardening model the yield surface expands in all directions. Thus, the yield stress experienced after a load reversal will also be increased. This is not in accordance with the so-called Bauschinger eﬀect by which an increase in the yield stress for loading in one direction is typically accompanied by a change of the reverse yield limit in the same direction. A simple hardening model that includes the Bauschinger eﬀect is the socalled kinematic hardening, in which the yield surface retains its shape but translates during plastic loading, (Fig. 7.9). A translating yield surface with initial yield stress σ0 can be written in

Fig. 7.9. Kinematic hardening: translating yield surface.

218

Elasto-plastic solids

terms of the yield function F (σ, κ) = f (σ − κ) − σ0 .

(7.123)

The parameters κ represent the current position of the initial center of the yield surface. The direct interpretation of the parameters σ in terms of stresses has led to names like ‘pseudo stress’ and ‘back stress’ for these parameters. In the kinematic hardening model the translation is governed by a set ˙ of relations of the form (7.112) for the rate of the internal parameters κ. Commonly used forms are the rule of Melan (1938) and Prager (1955), deﬁning the translation by the plastic strain rate κ˙ = a(κ) ε˙ p

(7.124)

and the rule of Ziegler (1959), using a radial direction from an assumed center κ˙ = λ˙ a(κ) (σ − κ)

(7.125)

corresponding to radial motion controlled by the single scalar function a(κ) for calibration. These rules coincide and follow from the potential formulation if the yield function is a quadratic function of the shifted stress variables σ − κ, as discussed in Section 7.4.1.

7.3 Von Mises plasticity models It has been found that the theory of plasticity can provide accurate representation of deformation of materials like metals. These materials typically exhibit plastic strains that are orthogonal to the yield surface described by the yield function F , and thereby suggest the use of uncoupled plasticity theory, without need for the extra ﬂow potential G. It also appears that in the absence of voids in the material, the plastic strains are purely deviatoric, i.e. the plastic contribution to the volumetric strain vanishes. It then follows from the ﬂow rule (7.95a) for plastic strains that the yield function is independent of the mean stress and only depends on the deviatoric stress components. For isotropic materials this implies that the yield surface may be generated via its contour in the deviatoric stress plane illustrated in Fig. 7.2. The central model for this class of materials is the model of von Mises (1928), in which the yield surface is a circular cylinder. This model leads to several particularly simple theoretical relations between stresses, strains and hardening parameters, and also has a number of very attrac-

7.3 Von Mises plasticity models

219

tive features with respect to numerical integration algorithms. This section contains a brief presentation of some of these features. 7.3.1 Yield surface and ﬂow potential The von Mises yield surface constitutes a circular cylinder in principal stress space with the principal diagonal as center axis. The radius is expressed in terms of the so-called equivalent stress σe , deﬁned via the second deviatoric stress invariant as σe2 = 3J2 = 32 σαβ (7.126) σαβ = 23 σ1 σ1 + σ2 σ2 + σ3 σ3 . The numerical factor has been chosen such that a uniaxial stress σ corresponds to the equivalent stress σe = |σ|. The cylindrical von Mises yield surface is illustrated in principal stress space in Fig. 7.10.

(a)

(b)

Fig. 7.10. Von Mises yield surface in principal stress space.

The yield function is conveniently expressed in terms of the equivalent stress as F (σ, σ0 ) = σe (σ) − σ0 .

(7.127)

In this format σ0 represents the uniaxial yield stress. Note that the yield function is formulated with the dimension of stress. This leads to considerable algorithmic advantages, as discussed later. The yield stress σ0 is considered as an internal parameter that may change during plastic deformation. Extension to kinematic hardening is discussed later. The strains are derived from the yield function F (σ, σ0 ) with respect to the stresses. This requires the derivatives of the equivalent stress σe , conveniently calculated by the component relations ∂σe 1 ∂σe2 3 σαβ = = . ∂σαβ 2σe ∂σαβ 2 σe

(7.128)

220

Elasto-plastic solids

This result may be expressed in the more compact form ∇σ σe =

3 2

σe−1 σ .

(7.129)

It is noted that the stress gradient of the equivalent stress σe is purely deviatoric and therefore has mean value zero. In geometric terms this corresponds to the normals of the cylinder shown in Fig. 7.10(a) being orthogonal to the axis of the cylinder. The plastic strain increments are evaluated according to the rate equation (7.95a), ε˙ p = λ˙ ∇σ σe = 3 λ˙ σ −1 σ . (7.130) 2

It follows from the deﬁnition of the deviatoric stress that the volumetric plastic strain rate vanishes, ε˙pv = 0. It is convenient to deﬁne an equivalent plastic strain rate ε˙p , similar to the equivalent stress σe . For incompressible strain increments the deﬁnition is ε˙2p =

2 p p 3 ε˙αβ ε˙αβ .

(7.131)

In pure extension in the principal direction 1 with ε˙p1 = 0 the incompressibility condition gives ε˙p2 = ε˙p3 = − 12 ε˙p1 and thereby ε˙p = |ε˙p1 |. Thus, the equivalent strain rate ε˙p corresponds to the strain rate for uniaxial stress loading. For the von Mises model the equivalent strain rate follows from substitution of the plastic strain rate tensor (7.130) into the deﬁnition of the equivalent plastic strain rate (7.131): 2 ε˙2p = 23 32 λ˙ σe−1 σαβ σαβ = λ˙ 2 , (7.132) and thus the rate multiplier λ˙ is equal to the equivalent strain rate in this model: λ˙ = ε˙p . (7.133) ˙ This leads to some simple illustrative interpretations of e.g. the multiplier λ. Example 7.2. Hardening in von Mises plasticity. Let the hardening be described in terms of the accumulated plastic strain εp . The consistency condition is now expressed as the time derivative of the yield function in the form dσ0 ˙ σ σe − F˙ (σ, σ0 ) = σ∇ ε˙p = 0. (7.134) dεp Comparison with the form (7.110) identiﬁes the hardening modulus in the present model as dσ0 H = , (7.135) dεp

7.3 Von Mises plasticity models

221

i.e. the slope of the uniaxial stress–strain relation corresponding to yielding of the material.

Fig. 7.11. Uniaxial stress–strain curve.

Figure 7.11 shows a uniaxial stress strain curve and identiﬁes the initial modulus of elasticity E and the tangent modulus Et . The representation of the total strain rate as the sum of the elastic and plastic strain rates can be written as σ˙ σ˙ σ˙ + , (7.136) = Et E H from which the plastic hardening modulus H is obtained as 1 1 1 = − . H Et E

(7.137)

H has a range from zero for perfect plasticity with Et = 0 to inﬁnity for vanishing plastic strain. If the tangent modulus Et is negative the material is said to be softening, and the corresponding value of the hardening modulus H is negative. This phenomenon typically leads to localization of the strains in smaller regions and requires special computational measures. The tangent stiﬀness Cep of elasto-plastic deformation is deﬁned via the stress–strain relation ˙ σ˙ = Cep ε,

(7.138)

where the speciﬁc form of the elasto-plastic tangent stiﬀness follows from (7.118), when specialized to associated form: Cep = C −

C (∇σ F ) ⊗ (∇σ F )T C . (∇σ F )T C (∇σ F ) + H

(7.139)

In the present case the elastic properties are assumed to be isotropic linear elastic, and the stiﬀness tensor is then given in component form by (7.49). In von Mises plasticity with yield function (7.127) the stress gradient of the yield function F is equal to the gradient of the equivalent stress σe ,

222

Elasto-plastic solids

∇F = ∇σe . This gradient has already been calculated in (7.129), where it was demonstrated to be purely deviatoric. Therefore, multiplication with the isotropic elastic stiﬀness tensor C simply corresponds to multiplication with the scalar factor 2µ: C (∇σ σe ) = 3µ σe−1 σ .

(7.140)

The product in the denominator then follows from the deﬁnition (7.126) of the equivalent stress as (∇σ σe )T C (∇σ σe ) = 3µ.

(7.141)

The elasto-plastic tangent stiﬀness now follows directly by substitution of these expressions into (7.139) as Cep = C −

9µ2 σ σ T ⊗ . H + 3µ σe σe

(7.142)

With the transpose sign the expression is valid both when interpreted as a four-dimensional tensor and when interpreted in association with the vector representation (7.8)–(7.9) of stresses and strains as a 6×6 matrix. In neither formulation are the two last factors contracted. The problem of integrating the constitutive relations of plasticity is due to the fact that the elasto-plastic stiﬀness Cep depends on the current stresses – and possibly on hardening parameters as well. Thus, generally approximate numerical methods must be used. These can be either explicit, in which case the needed parameters are evaluated at the initial time of the increment, or implicit, which leaves a number of possibilities for representation of the development of the parameters over the increment. The following two subsections give a brief presentation of explicit and implicit algorithms, illustrated speciﬁcally with reference to von Mises plasticity.

7.3.2 Explicit integration In elasto-plastic analysis the solution is advanced by increments of the form σ n+1 = σ n + ∆σ and εn+1 = εn + ∆ε. Typically, the elasto-plastic stiﬀness is used in connection with a speciﬁed load increment to predict the corresponding strain increment ∆ε at the individual material points – typically element Gauss points – of the body to be analyzed. This leads to a local analysis at the individual material points, in which the strain increment ∆ε is imposed, and the corresponding stress increment ∆σ needs to be calculated. Once the stresses are known, they can be used to carry out an equilibrium check via residual forces, or in the case of explicit methods the

7.3 Von Mises plasticity models

223

stresses can be accepted and used as a basis for the next step. In either case it is necessary to solve the problem of determining the stress increment ∆σ resulting from an imposed strain increment ∆ε. This limited task, carried out one or more times at each material point of the model within each load increment, is discussed in terms of explicit methods in this subsection and in terms of implicit methods in the next subsection. Explicit integration, often termed the forward Euler procedure, consists in calculating the stress increment ∆σ directly from the rate equation (7.138) by suitable representation of the elasto-plastic stiﬀness Cep in terms of parameter values that have already been obtained. In its simplest form the stress and strain rates are replaced by ﬁnite increments, and the variables deﬁning the elasto-plastic stiﬀness are evaluated corresponding to the initial state. Thus, the incremental relation (7.138) for inﬁnitesimal increments is replaced by ∆σ = Cep (σ 0 , κ0 ) ∆ε,

(7.143)

where the notation κ is retained for internal variables, and superscript 0 denotes initial state. A similar representation can be used to update λ and κ from (7.117) and (7.119) if necessary. In the forward Euler scheme the new stress state may not satisfy the yield condition, and the errors tend to accumulate. Thus, reﬁned methods are needed. The simplest reﬁnement of the forward Euler method follows from the observation that the error in (7.143) depends on the square of the size of the increments. Therefore, the error can be reduced by dividing the original strain increment ∆ε into sub-increments, e.g. m equal sub-increments ∆ε/m. This ‘sub-incremental’ method is illustrated in Fig. 7.12. The corresponding formulae are δσ k = Cep (σ k−1 , κk−1 ) ∆ε/m, σ k = σ k−1 + δσ k ,

(7.144)

supplemented with similar formulae for λ and κ if necessary. Various methods have been proposed to determine a suitable number of sub-increments, see e.g. Crisﬁeld (1991). Although the use of sub-increments reduces the error, there is still no check on the yield condition, and thus f (σ k , κk ) may drift away from the zero value required by consistency. A special two-step procedure has been suggested by Zienkiewicz and Taylor (2000). In this procedure a mid-point stress σ 1/2 is ﬁrst estimated by use of the forward Euler formula ∆σ 1/2 = Cep (σ 0 , κ0 ) 12 ∆ε.

(7.145)

224

Elasto-plastic solids

Fig. 7.12. The sub-incremental method.

If necessary ∆κ1/2 is also determined. The increment ∆σ 1/2 is now used to determine the mid-point stress σ 1/2 , and the full increment ∆σ is then calculated as ∆σ = Cep (σ 1/2 , κ1/2 ) ∆ε.

(7.146)

The two-step procedure provides an error estimate by comparing stress increments based on the initial point and the mid-point, respectively. Thus, from (7.145) and (7.146) the error estimate is 0 ∆σ − 2∆σ 1/2 = C1/2 (7.147) ep − Cep ∆ε. This error measure may be used to control the increment size. Example 7.3. Explicit integration. The explicit integration methods are illustrated by a simple example with the von Mises yield criterion in a two-dimensional principal stress space σ = [σ1 , σ2 ]T , with the corresponding strain components ε = [ε1 , ε2 ]T . The elastic modulus is E and Poisson’s ratio is assumed to be zero, ν = 0, giving the simple diagonal elastic stiﬀness matrix C = EI = 2µI. The yield function F is given by (7.127), with equivalent stress σe : σe2 = σ12 + σ22 − σ1 σ2 . The material is assumed to be non-hardening, i.e. H = 0. The material is loaded in uniaxial stress to ﬁrst yield at the stress σ = [σ0 , 0]T and the strain ε = [ε0 , 0]T , where ε0 = σ0 /E is the uniaxial yield strain. This state is represented by the point A in Fig. 7.13. After ﬁrst yield at A an additional strain increment ∆ε = [ε0 , 0]T is imposed, i.e. an elongation with ∆ε2 = 0. The stress increment ∆σ from this imposed strain increment is now determined by the three explicit methods described above: the single increment, division into two strain sub-increments, and the mid-point procedure. The ﬁnal stress state is indicated with B, C and D, respectively in Fig. 7.13, and the numerical values given in Table 7.1. Comparison of the results of the ﬁrst two methods illustrates that half

7.3 Von Mises plasticity models

225

Fig. 7.13. Explicit stress increments.

Table 7.1. Explicit plastic stress integration.

σ T /σ0 σe /σ0

Single increment

Sub-increments

Mid-point

(1.200, 0.400) 1.058

(1.155, 0.356) 1.024

(1.109, 0.312) 0.991

the increment gives one quarter error, but to be added from the two steps, resulting in an overall result with half the original error. Comparison between the two last methods demonstrates that given that two computations are going to be used, an extra factor of three is gained on the error by using the estimated mid-point for calculating the total step.

7.3.3 Radial return algorithm The explicit methods have the inherent weakness that the stresses may leave the yield surface and thereby violate the consistency condition. This problem is addressed in the implicit methods, in which the consistency condition is satisﬁed, often by iterative procedures. The implicit methods can be developed from two diﬀerent points of view: either by recasting the combined elasto-plastic strain increment into algorithmic form, or by solving the non-linear matrix evolution equations developed in Section 7.2.3 for ﬁnite increments by an iterative numerical method. The direct algorithmic formulations are often called ‘return algorithms’ because they typically consist of an elastic prediction of the stress, followed by a correction due to plastic eﬀects, see e.g. Ortiz and Popov (1985), Ortiz and Simo (1986). The return algorithms have particular advantages in connection with special yield surfaces like cylinders or cones. The main idea of the radial return algorithm is

226

Elasto-plastic solids

illustrated in this section with reference to the von Mises plasticity model. The general integration problem is dealt with in the following subsections in connection with the discussion of internal variables and non-associated ﬂow. The basic idea of the return algorithms is to decompose the stress increment into two additive parts, ∆σ = C (∆ε − ∆εp ) = ∆σ e − ∆σ p .

(7.148)

Here ∆σ e = C∆ε is an equivalent elastic stress that would result in the absence of plastic strains, while ∆σ p = C∆εp represents a plastic correction. This decomposition is illustrated in Fig. 7.14.

Fig. 7.14. Elastic predictor and plastic corrector.

If the point B is inside the yield surface the increment is fully elastic and ∆σ p = 0. If B is outside the yield surface the plastic corrector −∆σ p returns the stress to the yield surface. The plastic strain rate is given by (7.95) and a representation in ﬁnite increment form is ∆σ p = C ∆εp = ∆λ C ∇σ G.

(7.149)

Thus, the direction of the plastic corrector is determined by the gradient of the ﬂow potential ∇σ G, while the length depends on ∆λ. The increment ∆λ is determined such that the point C is on the yield surface after hardening. Taylor expansion of the yield function F (σ, κ) around (σ, κ)B gives F (σ C ) F (σ B ) − ∆σ Tp ∇σ F + (∂F/∂λ) ∆λ = 0.

(7.150)

Comparison of the last term with the diﬀerential form (7.110) of the consistency equation identiﬁes the derivative as the hardening modulus, ∂F/∂λ = −H. Substitution of ∆σ p from (7.149) gives ∆λ =

F (σ B ) T (∇σ G) C (∇σ F )

(7.151)

It is instructive to compare this formula with (7.117) for the inﬁnitesimal

7.3 Von Mises plasticity models

227

increment dλ. The only change is that the numerator (∇σ F )T C ε˙ has been replaced with F (σ B ). The derivation leaves a number of issues. In the general case the gradients ∇σ G and ∇σ F vary along the return path from B to C. Then an iterative scheme must be used to determine the point C. Two diﬀerent approaches have been proposed: an explicit and an implicit, see Ortiz and Simo (1986), Simo and Hughes (1987). In the explicit scheme a sequence of linearized problems are solved, starting at the point B. In the implicit method the direction of the full plastic stress increment is determined by the gradient at the initially unknown point C. Each step consists of evaluating the gradient ∇σ G at the most recent estimate of point C, and then solving the consistency condition for ∆λ by a Newton-type iterative procedure. This corresponds closely to the Newton-type iteration procedure presented in Section 7.4.3.

Fig. 7.15. Radial return for perfect von Mises plasticity.

In the special case of a von Mises material the return algorithm takes on a particularly simple and robust form. In this case the gradient of the ﬂow potential is equal to the gradient of the yield function ∇σ G = ∇σ F . Furthermore, the particular choice (7.127) of the yield function as a linear function of the equivalent stress σe (σ) leads to the property that the gradient of F is constant along the radial line connecting B and C. This is illustrated in Fig. 7.15 for von Mises plasticity with isotropic hardening. As seen from the explicit expression (7.129), the gradient ∇σ σe represents a radial vector in the deviatoric stress plane pointing through the point D. This is the so-called radial return algorithm (Krieg and Krieg, 1977). Example 7.4. Radial return algorithm. The radial return algorithm is illustrated for a von Mises material in three-dimensional principal stress space σ = [σ1 , σ2 , σ3 ]T . As in Example 7.3 the elastic modulus is E and Poisson’s ratio is assumed to be zero, ν = 0, whereby the elastic stiﬀness tensor is C = E I = 2µ I. The yield function F is given by (7.127), with

228

Elasto-plastic solids

equivalent stress σe2 = σ12 + σ22 + σ32 − σ2 σ3 − σ3 σ1 − σ1 σ2 and the stress gradient from (7.129) or directly by diﬀerentiation as ∇σ σe =

1 −1 2 σe [2σ1

− σ2 − σ3 , 2σ2 − σ3 − σ1 , 2σ3 − σ1 − σ2 ]T .

The material is assumed to be non-hardening, i.e. H = 0, and loaded in uniaxial stress to ﬁrst yield at the stress σ A = [σ0 , 0, 0]T with strain εA = [ε0 , 0, 0]T , ε0 = σ0 /E. A strain increment ∆ε = [ε0 , −ε0 , 0]T is now imposed and the new stress state is to be determined. First the equivalent elastic stress increment is determined, leading to the stress σ B = σ0 [2, −1, 0]T . This corresponds to σe,B = 2.6458 σ0 , and thereby F (σ B ) = 1.6458 σ0 . The multiplier increment ∆λ is determined from (7.151), ∆λ = F (σ B )/3µ = 1.6458 σ0 /µ = 1.0972 σ0 /E. The gradient at B is ∇σ F |B =

σ0 [5, −4, −1]T . 2σe,B

The plastic stress increment ∆σ then follows from (7.149), and then the ﬁnal stress state σ C = σ B − ∆σ p =

2 −1 σ0 − 0

1.0367 −0.8293 σ0 = −0.2073

0.9633 −0.1707 σ0 . 0.2073

It is easily veriﬁed that σe,C = 1.000 σ0 , and thereby point C has been returned to the yield surface. In the case of isotropic hardening the yield stress is simply updated to σe,C . It is instructive to compare the radial return algorithm with explicit integration by use of the tangent stiﬀness at A. Here the gradient is σ0 ∇σ F |A = [2, −1, −1]T = [1, −0.5, −0.5]T . 2σe,A The increment ∆λ is determined from (8.21) as ∆λ = ∇σ F |A ∆σ e /3µ = σ0 /E. It is seen that this estimate of ∆λ deviates 10% from that of the radial return algorithm. The plastic strain increment is now evaluated by use of the gradient at A as σ C = σ B − ∆σ p =

2 −1 σ0 − 0

1 −0.5 σ0 = −0.5

1 −0.5 σ0 . 0.5

7.4 General aspects of plasticity models

229

This stress state deviates considerably from the yield surface, σe,C = 1.323 σ0 . The example illustrates how the radial return algorithm modiﬁes the direct tangent method in two ways: the direction of the plastic stress increment is determined from the gradient corresponding to the ﬁnal stress state, and the length of the increment is calculated via the yield function to give exact return to the yield surface. In the case of plane stress the von Mises yield function is constituted by a section of the cylindrical yield function shown in Fig. 7.10 in one of the principal stress planes. This leads to an elliptic surface, where the normal return procedure described above must be modiﬁed. However, the basically simple shape of the yield surface can also be used to construct eﬃcient return mapping schemes in this case, see Jetteur (1986) and Simo and Taylor (1986).

7.4 General aspects of plasticity models In this section three aspects of more general plasticity models are presented. The ﬁrst is the use of combined isotropic and kinematic hardening, which is of central importance for the use of plasticity models for reversed and cyclic load histories. The second is a simple demonstration of the use of the dual potentials F and G in connection with internal variables and their conjugates to introduce non-associated ﬂow. This problem is important in connection with many non-metallic materials, notably granular media, discussed in more detail in Section 7.5. Finally, the integration of the general plastic evolution equations derived in diﬀerential form in Section 7.2.3 is discussed in Section 7.4.3. The models considered in this section are mainly intended to illustrate basic properties of the plastic deformation, and the elastic properties are therefore assumed to be linear isotropic as discussed in Section 7.1. In isotropic material models the deformation consists of two mechanisms: change of volume and change of shape. The volumetric behavior is governed by mean stress and mean strain, represented by the scalar variables σm =

1 3 σγγ ,

εv = εγγ .

(7.152)

The change of shape is governed by the corresponding deviatoric tensor components with magnitudes σe2 =

3 2 σαβ σαβ ,

ε2e =

2 3 εαβ εαβ .

(7.153)

230

Elasto-plastic solids

The elastic energy is the sum of contributions from volumetric and deviatoric strain ϕ(εe , εv ) =

2 1 2 k(εγγ )

+ µεαβ εβα =

2 1 2 kεv

+ 32 µε2e .

(7.154)

In the models discussed below the internal energy is obtained by introducing suitable internal variable eﬀects into the elastic strain energy, and then introducing an appropriate ﬂow potential G in terms of the stresses and the conjugate internal variables. 7.4.1 Combined isotropic and kinematic hardening In modeling of cyclic loading there is typically a need to include the so-called Bauschinger eﬀect, by which an increase of the yield stress in positive loading leads to a shift in the yield stress experienced upon reversal of the loading. This involves a translation of the yield surface as discussed in general terms in the form of kinematic hardening in Section 7.2.4. Kinematic hardening requires an internal tensor variable κ to keep track of the translation. In addition, a scalar internal variable κ0 is included in the model here to illustrate the combined eﬀect. Often the translation of the yield surface in e.g. metals does not lead to noticeable deviations from normality of the plastic strain increment from the normal to the yield surface. Thus, the model will be formulated to represent associated plasticity, and therefore the internal variables will appear in the form of additive terms according to (7.97): ϕ(εe , εv , κ) =

2 1 2 kεv

+ 32 µε2e + ϕ1 (κe ) + ϕ0 (κ0 ).

(7.155)

The scalar invariant κe is here introduced in terms of the deviatoric components καβ in analogy with the equivalent stress σe , but without the numerical factor: κ2e = καβ καβ .

(7.156)

The stress is deﬁned as the conjugate variable to the strain, whereby σαβ =

∂ϕ = kδαβ εv + 2µ εαβ . ∂εαβ

(7.157)

The conjugate internal tensor variable follows by diﬀerentiation of the potential ϕ1 through the scalar argument κe as ζαβ

κ ∂ϕ1 ∂κe αβ = = ϕ1 , κe ∂κe ∂καβ

(7.158)

where ϕ1 denotes the derivative with respect to the argument κe . It is noted that the conjugate internal tensor variable ζαβ is purely deviatoric. Its

7.4 General aspects of plasticity models

231

magnitude is characterized by the scalar invariant ζe , deﬁned by a relation similar to (7.156) for κe . Contraction of the components on each side of (7.158) with themselves then gives the relation ζe = ϕ1 (κe ).

(7.159)

This relation establishes the internal variable ζe as conjugate to the primary internal variable κe . The conjugate scalar internal variable ζ0 is deﬁned by the similar relation ζ0 = ϕ0 (κ0 ).

(7.160)

In general these relations may be non-linear, but the linear case, arising from quadratic functions ϕ1 and ϕ0 , leads to a particularly simple formulation. The Jacobian of the transformation between the primary internal tensor follows from diﬀerentiation of (7.158) as variable καβ and it conjugate ζαβ καβ κγδ 1 ∂ζαβ ∂ϕ(κe ) 1 ϕ1 − ϕ1 . = = δαγ δβδ ϕ1 + ∂κγδ ∂καβ ∂κγδ κe κe κe κe

(7.161)

It is seen that the ﬁrst term is isotropic, while the second term is anisotropic – depending on the current direction of the internal variable tensor καβ . Now assume that κe = 0 corresponds to ζe = 0. It then follows from (7.159) that a Taylor expansion of the potential ϕ1 around this value must have the form ϕ1 (κe ) =

2 1 2 α κe

+ O(κ3e ).

(7.162)

It is convenient to select internal variables κα , κ0 with dimension of strain, whereby the parameter α gets the dimension of stress, similar to the shear modulus µ and the bulk modulus k. Substitution of this representation into (7.161) demonstrates that if the potential ϕ1 is represented by the quadratic term alone, then the second – anisotropic – term in the relation (7.161) vanishes, leaving the simple isotropic component relation ζαβ = α καβ

(7.163)

between the primary and conjugate internal variables. In contrast, any form of ϕ1 that is not simply quadratic in κe leads to a relation with anisotropy, induced via the internal parameter components καβ . In the present model the energy consists of additive contributions from elastic strains and internal variables. This implies that the relation between the internal variables καβ , κ0 and their conjugate variables ζαβ , ζ0 is independent of the state of strain. As discussed in Section 7.2.2, this implies that the yield potential F is equal to the ﬂow potential G and can be expressed in terms of stress or strain in connection with either the primary internal

232

Elasto-plastic solids

variables καβ , κ0 or their conjugate internal variables ζαβ , ζ0 . In the present context it is most direct to express the potentials F and G in terms of the stress and the conjugate internal variables. In kinematic hardening the stresses are ‘centered’ by the internal variable ζ. When using the von Mises yield surface this corresponds to introduction of the stresses via the translated equivalent stress σ ¯e2 (σ − ζ) =

3 2 (σαβ

). − ζαβ )(σαβ − ζαβ

(7.164)

The ﬂow potential can then be written in the form γ G(σ, ζ, ζ0 ) = σ (7.165) ¯e (σ − ζ) + 12 ζ : ζ − (σ0 + ζ0 ). α The quadratic term in ζ was introduced by Armstrong and Frederick (1966) and leads to ﬁnite translation of the yield surface, see e.g. Chaboche (1986) and Ottosen and Ristinmaa (2005). The evolution equations follow from the general format (7.108), when it is observed that also the isotropic hardening component κ0 must be included among the internal variables. It turns out to be advantageous to consider the eﬀect of this internal variable ﬁrst. The isotropic internal variable κ0 has the evolution equation κ˙ 0 = −λ˙

∂G ˙ = λ. ∂ζ0

(7.166)

Thus, the isotropic hardening variable κ0 can be identiﬁed with the plastic multiplier λ. In the following the plastic multiplier is replaced by the isotropic hardening parameter κ0 . Note that this representation remains valid also in the special case without isotropic hardening. In this case the internal energy is independent of κ0 , and thus ϕ0 ≡ 0 and ζ0 ≡ 0. Therefore the strain-like hardening variable κ0 is retained in the equations, and not its stress-like conjugate variable ζ0 . When λ˙ is eliminated from the general evolution equation (7.108), the following system of equations is obtained:      C−1 0 (∇σ G) ε˙ σ˙      0 I (∇ζ G)   κ˙  . (7.167) 0 =  T T ˙ κ˙ 0 F (∇σ F ) (∇κ F ) ∂F/∂κ0 Here it is observed that F ≡ G, and the two last derivatives in the last column are expressed as −1 ∂κ T ∇κ F = ∇κκ ϕ1 ∇κ F, (7.168) ∇ζ G = ∂ζ

7.4 General aspects of plasticity models

∂F ∂ζ0 ∂F = = −ϕ0 (κ0 ). ∂κ0 ∂ζ0 ∂κ0 Multiplication of the middle equation in ﬁnal form of the evolution equations,    0 C−1 ε˙    0 (∇κκ ϕ1 ) 0 =  T ˙ F (∇σ F ) (∇κ F )T

233

(7.169)

(7.167) by ∇κκ ϕ1 then gives the   (∇σ F ) σ˙   (∇κ F )   κ˙  . κ˙ 0 −ϕ0

(7.170)

Comparison with the general form of the evolution equations (7.109), including all hardening parameters, shows that elimination of the multiplier λ˙ leads to an identical system of equations apart from the isotropic hardening function −ϕ0 (κ0 ), now appearing in the lower right position. This is the same position, and the same role, as the hardening modulus in the reduced format (7.115). The double derivative ∇κκ ϕ1 in the center term has already been expressed in component form in (7.161). The format (7.170) of the evolution equations is general for associated plasticity theory with isotropic hardening described by an additive term ζ0 with kinematic conjugate variable κ0 . For the particular model with modiﬁed kinematic hardening described by the ﬂow potential (7.165), the equations are completed by introducing the gradients with respect to the stress and the kinematic hardening parameters: ∇σ G =

3 2

σ − ζ , σ ¯e

∇ζ G = − 23

γ σ − ζ + ζ . σ ¯e α

(7.171)

The ﬁrst of the evolution equations (7.167) gives the plastic strain rate as σ − ζ ε˙ p = λ˙ ∇σ G = λ˙ 32 . σ ¯e

(7.172)

Formation of the fully contracted product of the components on each side of this equation with themselves gives the equivalent plastic strain rate as 1/2 ˙ = λ. (7.173) ε˙pe = 32 εpαβ εpαβ If the kinematic hardening potential ϕ1 is assumed to be quadratic, then ζ = ακ and the evolution equation for κ follows from the second equation in (7.167) as κ˙ = −λ˙ ∇ζ G = ε˙ p − γ ε˙pe κ.

(7.174)

For monotonic proportional loading the kinematic hardening tensor κ exhibits a decaying exponential approach to the limit determined by κ˙ = 0.

234

Elasto-plastic solids

The limit is conveniently expressed in terms of the scalar invariant κe , deﬁned in (7.156): 3 1/2 −1 κ∞ γ , e = (2)

ζe∞ = ( 23 )1/2

α , γ

(7.175)

where the numerical factor is due to the diﬀerence in normalization between the kinematic hardening scalars and the stress/strain scalars in order to facilitate the discussion of non-linear kinematic hardening. A detailed discussion of the behavior of combined kinematic and isotropic hardening has been given by Ottosen and Ristinmaa (2005). The numerical integration of this type of equation is discussed in Section 7.4.3. Extensions of kinematic hardening models to large deformation kinematics have been presented, e.g. by Simo (1988a,b) and Wallin et al. (2003). The main points of this extension are brieﬂy indicated in Section 7.6.

7.4.2 Internal variables and non-associated ﬂow If the contribution of the internal variables κ to the internal energy ϕ is in the form of an additive term, the mixed derivatives ∇εκ ϕ will vanish. It was demonstrated in Section 7.2.2 that if this condition is satisﬁed the stress gradients ∇σ F and ∇σ G are identical, and the corresponding plasticity is called associated. Conversely, if there are coupling terms in the form of nonvanishing mixed derivatives ∇εκ ϕ, the stress gradients of the yield potential F and the ﬂow potential G will exhibit a systematic diﬀerence, governed by the coupling between stresses and internal variables in the internal energy. This eﬀect will here be illustrated in relation to a generalized Drucker– Prager model, in which the yield surface and ﬂow potential are represented by cones in principal stress space (Drucker and Prager, 1952). The basic model is probably the simplest plasticity model combining the mean stress and the deviatoric stress. More detailed models are discussed in Section 7.5. In this subsection the main point is the coupling between volumetric strain eﬀects in the internal energy and the direction of plastic ﬂow, notably the volumetric plastic strain. The discussion is therefore limited to a two-component format consisting of the mean stress and the equivalent deviatoric stress (σm , σe ) and the conjugate strain variables (εv , εe ), deﬁned in (7.152)–(7.153). The yield surface, represented by the yield function F (σe , σm ) = σe + α σm − σF ,

(7.176)

is shown in full line in Fig. 7.16. The inclination of the line is given via the non-dimensional coeﬃcient α. It is seen that the yield surface expands

7.4 General aspects of plasticity models

235

with increasing mean compression, corresponding to decreasing mean stress σm . This general behavior is typical of granular media, soils and rock. It is generally found that ﬂow strains based on the normal to the yield function F will predict a too large dilation. This suggests a ﬂow potential G with a smaller slope, as indicated by the dashed line in the ﬁgure: G(σe , σm ) = σe + β σm − σG .

(7.177)

The idea here is to illustrate how the experimentally observed behavior β < α can be generated via a physically plausible assumption about the internal energy ϕ.

Fig. 7.16. Drucker–Prager yield function F (σe , σm ) and ﬂow potential G(σe , σm ).

Materials represented via a Drucker–Prager yield surface of the form (7.176) often have, or develop, internal porosity. This implies that not all the externally observable volume strain εv reﬂects actual straining of the material. This can be represented, e.g. by an internal energy of the form ϕ(εe , εv , κ) =

1 2 k(εv

− κ)2 + 32 µε2e ,

(7.178)

where κ represents a stress-free volumetric strain. The internal energy associated with the deviatoric strain εe is unaﬀected by the internal variable κ. The stresses follow from the internal energy by diﬀerentiation as σm =

∂ϕ = k(εv − κ), ∂εv

σe =

∂ϕ = 3µ εe . ∂εe

(7.179)

The conjugate internal variable ζ follows similarly from diﬀerentiation with respect to the ‘porosity variable’ κ as ∂ϕ = − k(εv − κ). (7.180) ∂κ In this relation the strain εv can be eliminated by use of (7.179a), whereby ζ =

ζ = − σm .

(7.181)

This relation permits transformation between the ﬂow potential G(σe , σm , ζ) and the yield potential F (σe , σm , κ) by direct substitution.

236

Elasto-plastic solids

The inclination β of the ﬂow potential is assumed to be independent of the internal variable, and the ﬂow potential is therefore assumed of the following form in the stresses and the complementary internal variable ζ: G(σe , σm , ζ) = σe + β σm − (σ0 + γζ),

(7.182)

where γ is a positive constant. The dissipation rate follows from (7.96) in the form ∂G ∂G ∂G ˙ D = λ˙ σe = λ(G + σ0 ) = λ˙ σ0 . (7.183) +ζ + σm ∂σm ∂ζ ∂σe The result follows from the property of G as a hom*ogeneous function of degree one in σe , σm and ζ. This relation identiﬁes λ as an equivalent irreversible strain.

Fig. 7.17. Drucker–Prager ﬂow potential G(σe , σm ) with plastic strain rates (ε˙pv , ε˙pe ).

The plastic strain rate vector (ε˙pv , ε˙pe ) is normal to the ﬂow potential G(σe , σm ) as shown in Fig. 7.17. The corresponding yield function follows from the ﬂow potential by elimination of the internal variable ζ by use of (7.181): F (σe , σm ) = σe + (β + γ)σm − σ0 = 0.

(7.184)

It is seen that the introduction of an internal volumetric strain variable κ leads to a change of the slope β of the ﬂow potential G to α = β +γ

(7.185)

in the yield potential F . This illustrates that a stress-free volumetric strain component κ in the internal energy leads to a smaller slope of the ﬂow potential than the yield surface, and thereby to less plastic dilation than predicted by a corresponding associated version of the theory. Hardening can be introduced into the model by including a suitable additive function ϕ0 (σ0 ) in the internal energy.

7.4 General aspects of plasticity models

237

7.4.3 General computational procedure In the analysis of elasto-plastic behavior in solids and structures the incremental structure of the evolution equations leads to an extra level in the analysis procedure. Basically the analysis is performed by a sequence of load increments. Within the individual load step the predicted deformation may be accepted, perhaps with a check on unbalance, or equilibrium iterations may be included. In either case the load step leads to a predicted motion and thereby to a predicted strain increment ∆ε at each material point used in the model. Due to the special structure of the evolution equations of plasticity, the ﬁnite increment of the total strain must be split into a plastic part and an elastic part, from which the stress is determined. In the implicit methods this is typically done in an extra iterative loop at each individual material point. The input to this loop is the total strain increment ∆ε, and the result is the corresponding ﬁnite increments of elastic and plastic strains as well as stress and any internal parameters. When using local material models this step does not involve coupling between the diﬀerent material points, and thus this step in the analysis can be carried out sequentially at the material point level. The key point is to set up a consistent extension of the diﬀerential evolution equations to ﬁnite increments, and to obtain the necessary matrices for iterative solution of these ﬁnite increment equations. In the local analysis the total strain increment ∆ε has been predicted, and therefore remains ﬁxed, while the increments of the stress σ, the hardening parameters κ and the plastic multiplier λ are to be computed. The procedure consists in iterative elimination of the residuals deﬁned by the discretized form of the strain rate equation (7.95a), rε = ∆ε − ∆εe − ∆λ ∇σ G(σ, ζ)

(7.186)

and the discretized rate equation (7.95b) for the hardening parameters, rκ = − ∆κ − ∆λ ∇ζ G(σ, ζ).

(7.187)

In these relations the symbol ∆ denotes the total increment relative to the last established state of equilibrium. An iterative sequence of stress increments is illustrated in Fig. 7.18. The iteration follows the Newton–Raphson scheme, in which the nonlinear equations are expanded in terms of current derivatives and linearized sub-increments, denoted by δ. The linearized form of the equations (7.186)

238

Elasto-plastic solids

Fig. 7.18. Sequence of iterative total stress increments ∆σ k .

and (7.187) is δεe + δλ ∇σ G + ∆λ δ(∇σ G) = rε ,

(7.188)

δκ + δλ ∇ζ G + ∆λ δ(∇ζ G) = rκ .

(7.189)

The most systematic form of these equations is obtained by using the conjugate variables σ and ζ. Thus, the terms δεe and δκ must be recast in terms of their conjugate counterparts δσ and δζ. A systematic approach is developed by use of the complementary energy density deﬁned by ψ(σ, ζ) = σαβ εeαβ + ζαβ καβ − ϕ(εe , κ).

(7.190)

This extends the deﬁnition (7.5) to the case including internal variables. When using the complementary energy density ψ(σ, ζ), the roles of the primary variables εe , κ and their conjugate variables σ, ζ are interchanged relative to the discussion in Section 7.2.1. Thus, diﬀerentiation of the complementary energy density gives εe =

∂ψ = ∇σ ψ(σ, ζ), ∂σ T

κ =

∂ψ = ∇ζ ψ(σ, ζ), ∂ζ T

(7.191)

while further diﬀerentiation with respect to time gives ˙ ε˙ e = (∇σσ ψ) σ˙ + (∇σζ ψ) ζ,

(7.192)

˙ κ˙ = (∇ζσ ψ) σ˙ + (∇ζζ ψ) ζ.

(7.193)

These are the desired relations. They can also be obtained, although less elegantly, from the double derivatives of the internal energy ϕ(εe , κ) by inversion of (7.74)–(7.75). It is now observed that the terms δ(∇σ G) and δ(∇ζ G) lead to a similar format in terms of the double derivatives of the ﬂow potential G(σ, ζ). The equations (7.188)–(7.189) therefore take a particularly simple and compact

7.4 General aspects of plasticity models

239

form, if an algorithmic complementary energy is deﬁned as Ψ(σ, ζ) = ψ(σ, ζ) + ∆λ G(σ, ζ).

(7.194)

In terms of this potential function the equations (7.188)–(7.189) take the compact form ∇σσ Ψ ∇σζ Ψ δσ ∇σ G rε + δλ . (7.195) = δζ ∇ζ G rκ ∇ζσ Ψ ∇ζζ Ψ These equations must be supplemented by the consistency condition. According to their deﬁnition in connection with the maximum dissipation formulation the yield function F (εe , κ) and the ﬂow potential G(σ, ζ) are the same function, but expressed in terms of conjugate sets of variables. In the present context the iteration variables are σ, ζ, and it is therefore convenient to represent the yield condition in terms of the ﬂow potential. When the yield condition is expressed via the residual rG = − G(σ, ζ)

(7.196)

the corresponding linearized equation takes the form (∇σ G)T δσ + (∇ζ G)T δζ = rG .

(7.197)

When this equation is combined with (7.195) the resulting system of equations for the iterative increments is      ∇σσ Ψ ∇σζ Ψ ∇σ G δσ rε      (7.198) ∇ζζ Ψ ∇ζ G   δζ  =  rκ  .  ∇ζσ Ψ T T δλ rG (∇σ G) (∇ζ G) 0 It is remarkable that this local system of equations is symmetric, also for non-associated plasticity. In the case of associated plasticity the internal and external variables uncouple, and the equations can be expressed in terms of the sub-increments δσ, δκ together with the yield potential F (σ, κ). The key to the transformation is the incremental relation ∂ζ T δζ = δκ = (∇κκ ϕ)δκ (7.199) ∂κ and the corresponding diﬀerentiation rule ∇κ F =

∂ζ T ∇ζ G = (∇κκ ϕ)∇ζ G. ∂κ

(7.200)

The equation system (7.198) is transformed by introducing δζ from (7.199),

240

Elasto-plastic solids

followed by pre-multiplication of the second row by (∇κκ ϕ) (continues below).      ∇σσ Ψ ∇σκ Ψ ∇σ F δσ rε      (7.201) ∇κκ Ψ ∇κ F   δκ  =  rζ  ,  ∇κσ Ψ T T δλ rF (∇σ F ) (∇κ F ) 0 where the notation rF = rG has been introduced for consistency. In this formulation the residual rκ in the second row has been replaced by rζ = (∇κκ ϕ) rκ − ∆ζ − ∆λ ∇κ F (σ, κ).

(7.202)

The last approximation is given as the form, in which the ﬁrst term is represented by the linear increment ∆ζ instead of the product (∇κκ ϕ)∆κ. This is a slight modiﬁcation of the initial formulation of the discretization, but equally valid and more consistent with the formulation in terms of the yield potential F (σ, κ). Associated plasticity is a consequence of uncoupling of internal and external variables in the internal energy, i.e. ∇εκ ϕ = 0 and ∇σζ ψ = 0. This uncoupling in terms of energy does not necessarily imply uncoupling of the yield potential F (σ, κ) and the ﬂow potential G(σ, ζ). However, if the ﬂow potential function satisﬁes the uncoupling condition ∇σζ G = 0, then this implies that ∇σκ Ψ = 0 and leads to the often-used form      (∇εε ϕ)−1 + ∆λ∇σσ F 0 ∇σ F δσ rε      0 (∇κκ ϕ) + ∆λ∇σσ F ∇κ F   δκ  =  rζ  .  δλ rF (∇σ F )T (∇κ F )T 0 (7.203) This formula is quite similar to the rate equation (7.109), when the two diagonal terms are represented by their corresponding ‘algorithmic value’, representing the ﬁnite increment via a contribution proportional to ∆λ. The algorithmic tangent stiﬀness matrix Ca is deﬁned for associated plasticity theory via the top diagonal term as −1 C−1 + ∆λ ∇σσ F = C−1 + ∆λ ∇σσ F. a = (∇εε ϕ)

(7.204)

A similar algorithmic matrix for the internal variable is deﬁned by the center term. Consistent tangent matrices for incremental changes around a predicted state with ﬁnite multiplier increment ∆λ were discussed in relation to the return format by Simo and Taylor (1985). The concept arises naturally in connection with the Newton–Raphson scheme. The consistent tangent matrices play a role on two levels in the computation. They are generated on

7.5 Models for granular materials

241

a node-by-node basis within the stress iteration loop and improve the rate of convergence here. When the stresses have been determined at all Gauss points the corresponding residual forces may require an additional global determination of the deformation, strain, etc. In that connection changes on the sub-increment level are also related via the consistent tangent stiﬀness. The analysis can be simpliﬁed, if the changes of the internal variables are condensed into the hardening modulus H and determined separately. However, general plasticity models with coupled internal and external variables in both energy and ﬂow potential lead to full coupling of the local incremental equations as illustrated in (7.198).

7.5 Models for granular materials Granular materials like soils, grain, etc. represent an interesting case, in which basic mechanics principles can be used to formulate plasticity models that reﬂect several special features of the stress–strain behavior with remarkable accuracy. One of the basic features is observed in a triaxial test, in which a uniform isostatic compression is overlaid by an increasing uniaxial compression. In the beginning of such a test, where the uniaxial component is small relative to the isostatic component, the material will experience compaction – i.e. negative dilation. Just before failure, characterized by rapidly increasing strains, the behavior changes and becomes dilational. It turns out that the change from compaction to dilation is a characteristic of any particular granular material, described by the ratio of the stress diﬀerence to the mean stress. A mathematical theory incorporating this transition was developed by Schoﬁeld and Wroth (1968) under the name of Cam-Clay theory and has given name to the general concept of ‘critical state soil mechanics’. A simpliﬁed modern formulation using the concepts of plasticity theory can be found e.g. in Wood (1990), and later Collins and Houlsby (1997) have recast this theory into potential form, using the extremum properties of the energy dissipation rate. The original work on these theories was formulated in terms of two stress and two strain components – typically a mean value and the component diﬀerence. The following describes a fully triaxial plasticity theory for granular materials, in which the triaxial format arises as a natural extension in terms of invariants of a basic behavior in a hypothetical two-component generalized shear test (Krenk, 1998, 2000). Alternative models have been proposed e.g. by Lade and Kim (1995). The theory applies to granular materials without adhesion, and therefore is restricted to compressive stresses. In this section it has therefore been found convenient to use a notation in which stresses are positive in compression, and

242

Elasto-plastic solids

strains are positive in contraction. The ﬁnal equations are easily converted by changing the sign on all stress and strain components. 7.5.1 Flow potential and yield surface A typical yield surface of a granular material is shown in Fig. 7.19(a) in principal stress space. A suitable mathematical format should account for the smooth triangular shape in the deviatoric sections and the closed rounded shape with increasing mean stress. This is illustrated by the two generating contours in Fig. 7.19(b). It is convenient ﬁrst to identify a suitable format for the deviatoric contour and subsequently determine the dependence of the size of the contour on the mean stress, see e.g. Collins (2003).

(a)

(b)

Fig. 7.19. (a) Yield surface in principal stress space, (b) generating curves.

In the formulation of the model it is convenient to use the mean stress σm , deﬁned by (7.26) and (7.28). The and the deviatoric stress components σαβ simplest mathematical form of a smooth triangular contour that satisﬁes the symmetry requirements with respect to the three principal deviatoric components (σ1 , σ2 , σ3 ) is the cubic polynomial equation (σ1 + d)(σ2 + d)(σ3 + d) = η d3 ,

(7.205)

where d determines the size of the circ*mscribing triangle, and η is a nondimensional shape parameter (Krenk, 1996). It may be argued theoretically that the triangle should be the intersection of the deviatoric plane with the coordinate planes in principal stress space, whereby d = σm .

(7.206)

7.5 Models for granular materials

243

This is supported by experimental evidence, e.g. by Lade and Duncan (1975). The implication is that in any particular deviatoric plane the smooth triangular contour is deﬁned in terms of a single parameter η. Thus, the full surface is determined if this parameter is prescribed as a function of the mean stress, η = η(σm ). A set of deviatoric stress contours is shown in Fig. 7.20.

Fig. 7.20. Smooth triangular contours in deviatoric plane.

The surface is easily reformulated in full component format by using the stress invariants discussed in Section 7.1.1. The particular choice d = σm implies that the left side of (7.205) is simply the product of the three principal stress components, and it then follows from the deﬁnition of the third stress invariant I3 by (7.19) that the deviatoric contour is given by 3 I3 = η(σm ) σm .

(7.207)

Several other formats for the deviatoric contour have been used in the lit3 replaced by erature, e.g. the Matsuoka and Nakai (1985) format with σm σm J2 . However, not all of these formats are convex in their full parameter range. In the present formulation the deviatoric contour can be expressed in terms of the deviatoric stress invariants J2 and J3 from (7.30) and (7.31) as 3 J3 − σm J2 + (1 − η(σm ))σm = 0.

(7.208)

This latter format in terms of deviatoric invariants is particularly useful for identiﬁcation of suitable functions η(σm ) for the yield surface and the ﬂow potential. Most triaxial experiments on granular materials have been made with two of the principal stresses equal, either as ‘triaxial compression’ with e.g. σ1 > σ2 = σ3 , or as ‘triaxial tension’ with e.g. σ1 < σ2 = σ3 . These

244

Elasto-plastic solids

particular conditions lie on the lines connecting the center of the triangle in Fig. 7.20 with either a corner (full line) or the mid-point of a side (dotted line) respectively. Theoretically it is more convenient to refer to a condition of ‘generalized shear’, in which an isostatic state of compression is superposed by a state of pure shear. These conditions correspond to σ1 > σ3 ,

σ2 =

1 2 (σ1

+ σ3 )

(7.209)

or similar states obtained by index permutation. This state of stress is illustrated by the dashed line in Fig. 7.20. It has the interesting property that σ2 = 0, and J2 = 12 (σ1 − σ3 ) = τ, J3 = 0 (7.210) where τ is the maximum value of the shear stress. When these expressions are substituted into equation (7.207), the following formula is obtained: τ /σm = 1 − η(σm ) = γ(σm ). (7.211) According to the friction hypothesis of Coulomb the ratio τ /σm is equal to sin ϕ, where ϕ is the friction angle. In the Coulomb theory the friction angle ϕ is a constant. However, it has been argued by Krenk (1998) that when plastic ﬂow takes place the yield condition must be satisﬁed, and thus an eﬀective angle of friction ϕ∗ can be determined from the Coulomb condition. When combined with a physically argued dilation condition this leads to an analytical determination of the ﬂow potential G(σm , τ ) for the state of generalized shear. However, the result involves the combination of the sine and the ln function, and it is therefore of interest to develop a similar but simpler form of the ﬂow potential from the expression for the rate of dissipation. The following argument is based on a state of generalized shear stress as deﬁned in (7.209). The maximum and minimum stress are combined into mean stress and maximum shear stress: σm =

1 2 (σ1

+ σ3 ),

τ =

1 2 (σ1

− σ3 ).

(7.212)

Corresponding conjugate plastic strain rates are deﬁned by ε˙p = ε˙p1 + ε˙p3 ,

γ˙ p = ε˙p1 − ε˙p3 .

(7.213)

It is assumed that the dissipation from the intermediate stress and strain components can be neglected. The dissipation rate from (7.96) then takes the form D = σ1 ε˙p1 + σ3 ε˙p3 = σm ε˙p + τ γ˙ p .

(7.214)

7.5 Models for granular materials

245

After division by σm γ˙ p this equation takes the non-dimensional form ε˙p D τ + = . σm γ˙ p σm γ˙ p

(7.215)

The plastic strain rate vector (ε˙p , γ˙ p ) is along the gradient of the ﬂow potential G(σm , τ ), and thus satisﬁes the equation dσm ε˙p + dτ γ˙ p = 0,

(7.216)

where (dσm , dτ ) is an increment along an equipotential curve of G(σm , τ ). Substitution of this result into (7.215) leads to the following diﬀerential equation for the function γ(σm ) = τ /σm : σm

D dγ = . dσm σm γ˙ p

(7.217)

This function determines the shape of the surface as a function of the mean stress σm , and the physics enters the theory via an assumed expression for the dissipation rate D in terms of the stresses and plastic strain rates. In the original Cam-Clay theory it was assumed that the rate of dissipation was due to a friction-like mechanism and expressed via the product of the mean stress σm and the shear strain rate γ˙ p and a friction coeﬃcient n. Hereby the right side of the equation (7.217) takes the constant value n and is easily integrated to give σ0 τ . (7.218) = γ(σm ) = n ln σm σm In this formula σ0 represents the value of σ0 where the curve intersects the isostatic axis, i.e. the intersection of the surface as shown in Fig. 7.19 with the principal diagonal, and thereby constitutes a direct measure of the size of the current surface. The surface described by this function has an apex at the intersection with the principal diagonal. It has later been modiﬁed to a smooth shape in an associated plasticity theory for granular materials, see e.g. Wood (1990). However, the real problem with this form is not the apex, but the fact that the surface encloses regions with tension due to the rather simple assumption regarding the friction mechanism. This can be remedied in a fairly simple way by changing the assumption regarding the dissipation D. If the representative stress in the friction assumption is changed from the mean stress σm to the minimum stress σmin = σm − τ , the curve is found to be σ n τ m = γ(σm ) = 1 − . (7.219) σ0 σm

246

Elasto-plastic solids

This generalized shear contour is used to deﬁne the plastic ﬂow potential G(σ) according to the surface format (7.207). Direct substitution gives 3 G(σ) = − I3 + ηG (σm ) σm

(7.220)

with the contour described by the function

ηG (σm ) = 1 − γ(σm )2 = (σm /σG )n 1 − (σm /σG )n .

(7.221)

The corresponding surface is illustrated in Fig. 7.21, clearly showing the apex at σm = σG . This format was used in the implementation of Ahadi and Krenk (2000, 2003). However, from the point of numerical analysis it would be desirable to have the mathematical format of the surface in a form in which the stresses in the J2 and J3 terms are of degree one.

(a)

(b)

Fig. 7.21. (a) Flow potential in principal stress space, (b) generating curves.

It would now be desirable to introduce the internal elastic energy as described in Section 7.2 with suitable internal kinematic parameters, and then to derive the resulting yield surface F . However, no complete theory in terms of conjugate internal and external variables appears to be available at the present time. The yield surface is therefore introduced by retaining the function format 3 F (σ) = − I3 + ηF (σm ) σm ,

(7.222)

whereby the rounded triangular contour family is retained. The dependence on the mean stress is represented by the function ηF (σm ) = (σm /σF )m .

(7.223)

7.5 Models for granular materials

247

This surface is rounded at the intersection σm = σF with the principal diagonal. It was used for the illustration in Fig. 7.19.

7.5.2 Elasticity and hardening The elastic stress–strain relations are chosen in accordance with the original Cam-Clay theory, in which the bulk modulus is proportional to the mean stress σm and the shear modulus is constant: ε˙ev =

κ σ˙ m , σm

ε˙e αβ =

1 σ˙ , 2µ αβ

(7.224)

where the volumetric ﬂexibility is described by the non-dimensional parameter κ. In the elasto-plastic regime the total volumetric strain rate is given by the relation χ σ˙ m , (7.225) ε˙v = ε˙ev + ε˙pv = σm where χ > κ represents the increased volumetric ﬂexibility in the elastoplastic range. When subtracting the elastic part, the plastic strain rate is found as σm p ε˙ . (7.226) σ˙ m = χ−κ v In the case of isostatic loading σ˙ m = σ˙ F , and thus the relation may be considered as the evolution equation for the stress σF determining the current size of the yield surface. This is the hardening evolution rule adopted in the original Cam-Clay theory (Schoﬁeld and Wroth, 1968). In that theory the yield surface expansion was controlled by the rate of work of the mean stress σm ε˙pv . Thus hardening stops, and unlimited plastic deformation can take place, once the critical state of zero plastic dilation is reached. This does not reﬂect the experimentally observed behavior in typical tests, where the material passes into a dilative state accompanied by large plastic strains. This can be incorporated into the model by adding a weighted contribution from the deviatoric components to the plastic work, whereby the hardening equation becomes σ˙ F =

1 σm ε˙pv + w σαβ ε˙pαβ . χ−κ

(7.227)

In this equation w is a weight factor leading to a modest contribution to the hardening from the plastic work from the deviatoric components. The plastic hardening is characterized via the hardening modulus H as

248

Elasto-plastic solids

deﬁned via F˙ in (7.110), H = −

∂F ∂σF = H 1 H2 . ∂σF ∂λ

(7.228)

The factor H1 gives the dependence of the yield function on the size parameter σF . It follows from (7.222) by diﬀerentiation, H1 = −

m+1 ∂F 2 . σm /σF = m σm ∂σF

(7.229)

The factor H2 describes the plastic hardening via the hardening rule (7.227). After substitution of the plastic strain rates in terms of the ﬂow potential G, the following result is obtained: 1 ∂σF ∂G ∂G H2 = = σm . (7.230) + w σαβ ∂σm ∂σαβ ∂λ χ−κ The diﬀerentiations are most easily carried out by using the format (7.208) for G in terms of mean and deviator stress components, 1 4 dηG (1 − w) 3J3 − 2σm J2 + σm . (7.231) H2 = dσm χ−κ This completes the formulation of the model. It contains a total of six parameters – three parameters µ, κ and χ relating to stiﬀness, and three non-dimensional parameters n, m and w specifying the potential functions in the theory. Parameter calibration and numerical implementation are discussed by Ahadi and Krenk (2000, 2003).

(a)

(b)

Fig. 7.22. Stress ratio and volumetric strain for loose Baskarp sand.

Figure 7.22 shows the ability of the model to represent the development of stress and volumetric strain in a typical triaxial cylinder test on sand

7.6 Finite strain plasticity

249

by Borup and Hedegaard (1995). The development of volumetric strain in Fig. 7.22(b) clearly shows the initial contraction phase, followed by dilation. The parameter p0 refers to the uniform compressive stress at the beginning of the test.

(a)

(b)

Fig. 7.23. Constant volume tests on Sacramento sand: (a) loose, (b) dense.

Figure 7.23 illustrates the ability of the model to represent the results from so-called undrained tests on sand by Lee and Seed (1967). In the undrained test the sample is saturated with water and sealed, whereby the total volume is kept approximately constant. The ﬁgure shows a vertical plane containing the principal diagonal in stress space. The tests are started by loading the specimen to an isosta

Non-linear Modeling and Analysis of Solids and Structures - PDF Free Download (2024)

References