A geometrically exact curved beam theory and its nite element formulation/implementation

Jing Li

Thesis submitted to the faculty of the Virginia Polytechnic Institute and State University in partial fulllment of the requirements for the degree of

Master of Science in Aerospace Engineering

Rakesh K. Kapania, Chair Eric R. Johnson Raymond H. Plaut

December 28, 2000 Blacksburg, Virginia

Keywords: Curved beam, geometrically exact nonlinear, nite rotations, nite element.

c Copyright 2000, Jing Li

A geometrically exact curved beam theory and its nite element formulation/implementation
Jing Li Rakesh K. Kapania Aerospace and Ocean Engineering

(ABSRACT) This thesis presents a geometrically exact curved beam theory, with the assumption that the cross-section remains rigid, and its nite element formulation/implementation. The theory provides a theoretical view and an exact and ecient means to handle a large range of nonlinear beam problems. A geometrically exact curved/twisted beam theory, which assumes that the beam crosssection remains rigid, is re-examined and extended using orthonormal reference frames starting from a 3-D beam theory. The relevant engineering strain measures at any material point on the current beam cross-section with an initial curvature correction term, which are conjugate to the rst Piola-Kirchho stresses, are obtained through the deformation gradient tensor of the current beam conguration relative to the initially curved beam conguration. The Green strains and Eulerian strains are explicitly represented in terms of the engineering strain measures while other stresses, such as the Cauchy stresses and second Piola-Kirchho stresses, are explicitly represented in terms of the rst Piola-Kirchho stresses and engii

neering strains. The stress resultant and couple are dened in the classical sense and the reduced strains are obtained from the three-dimensional beam model, which are the same as obtained from the reduced dierential equations of motion. The reduced dierential equations of motion are also re-examined for the initially curved/twisted beams. The corresponding equations of motion include additional inertia terms as compared to previous studies. The linear and linearized nonlinear constitutive relations with couplings are considered for the engineering strain and stress conjugate pair at the three-dimensional beam level. The cross-section elasticity constants corresponding to the reduced constitutive relations are obtained with the initial curvature correction term. For the nite element formulation and implementation of the curved beam theory, some basic concepts associated with nite rotations and their parametrizations are rst summarized. In terms of a generalized vector-like parametrization of nite rotations under spatial descriptions (i.e., in spatial forms), a unied formulation is given for the virtual work equations that leads to the load residual and tangent stiness operators. With a proper explanation, the case of the non-vectorial parametrization can be recovered if the incremental rotation is parametrized using the incremental rotation vector. As an example for static problems, taking advantage of the simplicity in formulation and clear classical meanings of rotations and moments, the non-vectorial parametrization is applied to implement a four-noded 3-D curved beam element, in which the compound rotation is represented by the unit quaternion and the incremental rotation is parametrized using the incremental rotation vector. Conventional Lagrangian interpolation functions are adopted to approximate both the reference curve and incremental rotation of the deformed beam. Reduced integration is used to overcome locking problems. The nite element equations are developed for static structural analyses, including deformations, stress resultants/couples, and linearized/nonlinear bifurcation buckling, as well as post-buckling analyses of arches subjected to conservative and non-conservative loads. Several examples are used to test the formulation and the Fortran implementation of the element.

ii

ACKNOWLEDGEMENTS The support provided for this research by the grant DAAH04-95-1-0175 from the Army Research Oce with Dr. Gary Anderson as the grant monitor is greatly appreciated. I owe a great deal of thanks to Dr. Rakesh Kapania for his kind support and patient guidance during my graduate studies. I would like to thank Dr. B. Grossman, Department Head, Aerospace and Ocean Engineering, for providing considerable computational resources. I would also like to thank Dr. Raymond Plaut (Dept. of Civil and Environmental Engineering) and Dr. Eric Johnson (Dept. of Aerospace and Ocean Engineering) for their valuable discussions and participation as members of my committee during the course of my graduate studies.

iii

Table of Contents
Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Chapter 2: Geometric assumptions and the geometric representation of the curved/twisted beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1 Rigid cross-section assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2 Quasi-prismatic beam assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.3 The geometries of the reference beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 2.4 The geometry, elongation and shearing of the moving beam . . . . . . . . . . . . . . . . . . . . . . . . . .9 Chapter 3: Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3.1 Finite rotations by orthogonal transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3.2 Spatial and material descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3.3 Linearized increments and derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 3.4 Spin and rotational speed as a general term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 3.5 Corotated derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Chapter 4: Strains and stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 4.1 Curvatures and curvature changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 4.2 Deformation gradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 4.3 Stresses, stress resultants and stress couples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.4 Internal power: the strain measures conjugate to the rst Piola-Kirchho stresses . . . 35 Chapter 5: Balance equations and principle of virtual work . . . . . . . . . . . . . . . . . 42 5.1 Reduced balance equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 5.2 Reduced virtual work equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Chapter 6: Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 6.1 Linear constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 6.2 Material nonlinearity and its linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 iv

Chapter 7: A general formulation of geometrically exact nonlinear curved beam elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 7.1 External loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 7.2 Parametrization of nite rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 7.3 Admissible variations/linearizations with respect to deformation . . . . . . . . . . . . . . . . . . . . 68 7.4 Virtual work equation and its linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Chapter 8: A 3-D four-noded curved beam element implementation . . . . . . . . 87 8.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .87 8.2 Solution procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 8.3 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Chapter 9: Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Appendix: Fortran programs for a 3-D four-noded curved beam element: ARCHCODE Version 1.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

List of Illustrations
Fig. 1 Fig. 2 Fig. 3 Fig. 4 Fig. 5 Fig. 6(a) Fig. 6(b) Fig. 7(a) Fig. 7(b) Fig. 8 A 3-D curved beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 A dierential slice for the straight reference conguration . . . . . . . . . . . . . . . . . . . . . 43 Dierent load models considered . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 Unrolling and re-rolling a circular cantilever beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Clamped-hinged circular arch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 45-degree bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Non-uniform mesh of 45-degree bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Deformed shapes of a cantilever . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .100 Tip deections of a cantilever beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .101

Circular arch with two hinges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

vi

List of Tables
Table 1 Reduced strain measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Table 2 Applied external load densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Table 3 Displacement comparison of loaded point of unrolling a beam . . . . . . . . . . . . . . . . . . 94 Table 4 Buckling load comparison of clamped-hinged circular arch . . . . . . . . . . . . . . . . . . . . . 95 Table 5(a) Displacement comparison of loaded point of 45-degree bend . . . . . . . . . . . . . . . . . 97 Table 5(b) Distortion test of non-uniform meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Table 6 Buckling load comparison of a circular arch with two hinges . . . . . . . . . . . . . . . . . . 102

vii

Chapter 1: Introduction
Beams have found many applications in civil, mechanical and aerospace engineering. Exact and ecient nonlinear analysis of structures, built up from beam components, using robust numerical methods, e.g. nite element methods, should be based on proper nonlinear beam theories. Reissners nite strain beam theory (1972, 1973, 1981) is one of the simplest and most important ones. This theory is based on Timoshenkos plane cross-section assumption, which has been extended, given in detail, and used by many other authors (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; Cardona and Geradin, 1988; Iura and Atluri, 1988, 1989; Saje, 1991; Simo and Vu-Quoc, 1991; Jeleni and Saje, 1995; Simo et al, 1995; Ibrahimbegovi, c c 1995; etc.) for 2-D and 3-D cases for both static and dynamic problems. Most authors dealt with straight beams and used orthonormal reference frames. For curved/twisted beams, the slender beam or rod assumption is hidden in the beam theories (see e.g. Simo et al, 1995; Ibrahimbegovi, 1995; etc.). c For truly geometrically exact curved/twisted beam theory for moderate thick beams, Iura and Atluri (1988, 1989), etc., kept the rigid cross-section assumption and used curvilinear reference frames for the formulation. Certain authors considered shear and torsion warping by the introduction of a simple warping function (see e.g. Simo and Vu-Quoc, 1991; etc.) for straight beams. For computationally analyzing curved beams or arches, many authors prefer using straight beam elements based on straight beam theories (see e.g. Simo and Vu-Quoc, 1986; Cardona and Geradin, 1988; Ibrahimbegovi et al, 1995; Franchi and Montelaghi, 1996; etc.). This c is a simple and good approximation for slender curved beams or exible curved beams although more elements will be used to get a satisfactory accuracy. Others prefer using curved beam/arch elements to analyze curved beams or arches based on slender beam theories to 1

reduce the number of elements used (see e.g. Sandhu, Stevens and Davies, 1990; Saje, 1991; Simo et al, 1995; Jeleni and Saje, 1995; Ibrahimbegovi, 1995). c c However, for thick and moderately thick curved beams, an increase in the accuracy of the nite element solution by increasing the number of straight beam elements or curved beam elements based on the slender beam theories has its limit, especially when long-term dynamic responses as well as strains and stresses in three-dimensional level are needed for design purposes. In this case, more rened curved beam theories should be used. In this thesis, a geometrically exact nite-strain curved and twisted beam theory with large displacements/rotations is re-examined and extended using orthonormal frames and the rigid cross-section assumption. The corresponding nite element formulation/implementation is also dealt with. In Chapter 2, we summarize the geometric assumptions and geometric descriptions of the curved/twisted beam model. A straight reference beam conguration is introduced in addition to the initially curved/twisted beam and moving beam or current congurations. In Chapter 3, we, for the readers reference convenience, briey review some preliminary concepts associated with nite rotations, spatial and material descriptions as well as co-rotated derivatives. In Chapter 4, the strains and stresses for initially curved/twisted beams are reduced from the 3-D to 1-D beam level. The deformation gradient tensor of the initially curved/twisted beam is obtained with the help of the straight reference beam conguration, from which the strain measures, with an initial curvature correction term, at any material point on the beam cross-section are obtained through internal power conjugate analysis. The strain measures conjugate to the rst Piola-Kirchho stresses are addressed because of the convenience of numerical implementation. Other strains and stresses can be explicitly represented in terms of the rst Piola-Kirchho stresses. In Chapter 5, we summarize the Lagrangian equations of motion and the virtual work equation for initially curved/twisted beams for completeness though the nal results are similar 2

to those by Iura and Atluri (1988). In Chapter 6, the general, but time- and rate- independent, linear and nonlinear constitutive relations are considered for the rst Piola-Kirchho stress and engineering strain conjugate pair to reduce them from the three-dimensional beam level to the corresponding constitutive relations at the one-dimensional beam model. The geometrically exact nite-strained curved/twisted beam theory summarized here is different from that by Iura and Atluri (1988, 1989) in that (1) Iura and Atluri (1988, 1989) used mixed curvilinear and orthonormal (i.e., Cartesian) frames while we use only orthonormal (i.e., Cartesian) reference frames; (2) Iura and Atluri (1988, 1989) used the ber length on the beam reference curve as the reference to measure strains at any point on the beam crosssection while we use the local real ber length to measure strains; therefore, the stress/strain conjugate pair is dierent; and nally, (3) the resulting constitutive relations for the same kind of materials are dierent. Note that in this thesis, only orthonormal reference frames or rectangular coordinate systems are used and therefore, only vectors and second-order Cartesian tensors are used. Of the formulation and implementation of geometrically exact curved beam elements based on the geometrically exact curved and twisted beam theory as described above, the robustness of beam elements is considered in the sense that a beam element can be used to exactly and eciently predict the linear and nonlinear static responses of the structures built up of beam components no matter if they are straight or curved, slender or thick, as long as the rigid cross-section assumption is valid in practice. In Chapter 7, a general formulation of geometrically exact nonlinear curved beam elements is made. Three types of simple load models, self-weight, snow and pressure, are considered to illustrate the importance to identify conservative and non-conservative loads in a nite element formulation (see Section 7.1).

The concepts of nite rotations and their parametrizations are summarized in some detail in Section 7.2. The vector-like parametrization of nite rotations is generalized in terms of spatial descriptions. In Sections 7.3 and 7.4, the admissible variations/linearizations are derived and an element formulation is performed on the continuum setting through the generalized vector-like parametrization of nite rotations, which can recover the case of non-vectorial parametrizations with a proper explanation. From the nal formulae for internal/external load and tangential stiness operators, one may make choices of which are preferred in the beam element implementation from dierent points of view. In Chapter 8, a C 0 continuous four-noded 3-D curved beam element is implemented as an example for static problems. In the formulation, a non-vectorial parametrization of nite rotations is adopted taking the advantage of its simplicity and the clear and well-known classical meanings of rotations and moments. Though a non-symmetric tangential stiness matrix is obtained, it is not a problem if a non-conservative external loading is considered. The element has the following features: (i) can be used for 3-D natural curved/twisted slender and thick beams; (ii) has nite strains and nite rotations; (iii) can handle classical coupled cross-section material constants with an initial curvature correction term; and (iv) uses isoparametric Lagrangian interpolation with uniform reduced integration. Several examples for slender beam cases are presented in Section 8.3 to test the element formulation and its Fortran implementation.

Chapter 2: Geometric assumptions and the geometric representation of the curved/twisted beam
In addition to the geometric assumptions in general nonlinear continuum mechanics (see e.g. Ogden, 1997: 77-83), the geometrically exact beam theory is based on some assumptions. 2.1 Rigid cross-section assumption The beam theory is based on the rigid cross-section assumption: the plane cross-sections of the beam remain plane and do not undergo any shape-change during the deformation (see e.g. Iura and Atluri, 1988; Iura and Atluri, 1989; etc.). The rigid cross-section assumption is valid in practice for slender beams, thin beams, or rods, and in some special cases, also valid for moderately thick beams of many structural materials. 2.2 Quasi-prismatic beam assumption The contours or shapes of dierent beam cross-sections are not necessarily exactly the same but similar to each other in the sense of local ane mapping1 . They may vary smoothly (continuously) and slowly along the beam axis. This assumption is not really a restriction on, but a removal of, the previous uniform beam assumption (see e.g. Simo, 1985; Iura and Atluri, 1989; etc.) hidden in the developed beam theories, to address the fact that the beam theory can be used for varying cross-sections as long as the rigid cross-section assumption is valid in practice. As will be seen later, this assumption is not explicitly used in the formulation except that the local reference frames vary continuously along the beam reference curve and the lateral surface is piece-wise smooth. 2.3 The geometries of the reference beams The conguration of a physically unstrained/unstressed curved/twisted beam, simply called curved reference beam, is dened by a smooth and spatial xed
1 2

reference curve with its

Ane mapping consists of translation, rotation, stretch and shearing transformations. By spatial xed we mean it is xed in an arbitrarily chosen orthonormal frame ei (i = 1, 2, 3) that has

position vector 0 = 0 (1 ) R3
3

(1)

parametrized by its real arclength coordinate 1 s0 [0, L0 ] with L0 R1 as the total real arclength of the reference curve of the curved reference beam, and a family of spatial xed cross-sections connected by the reference curve through the geometry, mass, or elasticity centroids of the cross-sections, whichever is convenient. Hence, the reference curve is also called midcurve in the present paper. Consider the local orthonormal (i.e., Cartesian) frame t0i , which is rigidly attached onto the cross-section at 1 s0 [0, L0 ] with its origin at 0 (1 ) and the orthonormal base vectors: t0i = t0i (1 ) = t0ij ej R3 ; t0i t0j = ij (2)

referred to the arbitrarily chosen spatial frame ei spanned by the orthonormal base vectors ei R3 and ei ej = ij . In the above and later, the indices i, j = 1, 2, 3, the summation convention holds for repeated indices i and j; and ij denotes the Kronecker delta4 . Without losing generality, let the base vector t01 be normal to and base vectors t02 , t03 in the cross-section plane. Since the curved reference beam conguration is in an unstrained and unstressed state, it is conventionally assumed that the cross-section plane of the curved reference beam is normal to the unit tangent vector, 0,1 , of the midcurve at 1 s0 [0, L0 ]: 0,1 = t01 , and 0,1 t0i = 1i , where (),x = the paper. On the other hand, the position vector 0 = 0 (1 , 2 , 3 ) R3 of any material point (1 , 2 , 3 ) on the cross-section of the curved reference beam conguration can be described by 0 = 0 + 2 t02 + 3 t03 (3)
() x

for any variable x R1 throughout

no acceleration nor rotation in the 3-D inertial physical space. We also call ei the spatial frame. 3 n R denotes both the Euclidean point space and Euclidean vector space of dimension n for convenience. However, one should be careful of the dierence between them, especially the dierence between a position vector and a free-ended vector. 4 ij = 1 for i = j, ij = 0 for i = j

where 2 and 3 are the coordinates along base vectors t02 and t03 , respectively: (2 , 3 ) B and B = B(1 ) is the compact subset of R2 that gives the shape and size of the beam cross-section, and it may vary smoothly and slowly along the material point on the midcurve 1 s0 [0, L0 ] but invariant of any deformation and motion. Note that we have called (1 , 2 , 3 ) for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) the material point of the beam, which both labels a material point or particle and gives the coordinate values along the unstrained/unstressed curved reference beam midcurve and the unit base vectors t02 and t03 . The material point (1 , 2 , 3 ) for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) is independent of any deformation and motion for a given curved reference beam, though its position vector may change during deformation and motion. When a material point is on the midcurve of a beam, 2 = 3 = 0. For later convenience, we dene a spatial xed straight (untwisted) reference beam conguration whose midcurve is given by the position vector 00 = 1 e1 (4)

with its arclength coordinate s00 and total arclength L00 exactly the same as in the unstrained and unstressed state: s00 = 1 s0 ; and the local frame t00i of the cross-section is t00i Ei = ei (6) L00 L0 (5)

Hence, the position vector 00 = 00 (1 , 2 , 3 ) of any material point (1 , 2 , 3 ) for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) on the cross-section can be described by 00 = 00 + 2 t002 + 3 t003 = 1 E1 + 2 E2 + 3 E3 (7)

We call Ei the material frame in the current context, which is coincident with the spatial frame ei . A material point (1 , 2 , 3 ) referred to this material frame is the simplest choice because of the orthonormality condition. 7

Because both Ei and t0i are orthonormal frames, there exists an orthogonal tensor 0 = 0 (1 ) SO(3)5 relating t0i to Ei by t0i 0 Ei 0 t0i Ei = t0ij ej Ei = t0ji ei Ej = 0ij ei Ej (8)

where denotes the standard tensor product6 . One observes that the component form of the orthogonal tensor 0 , referred to base ei Ej , is

t011

t021 t022 t023

t031

[0 ]ei Ej = [0ij ]ei Ej = t012

t032

(9)

t013

t033

Therefore, considering t0ij = t0i ej , the directional cosine of t0i with respect to ej , the orthogonal tensor 0 determines the orientation of the cross-section of the curved reference beam conguration. For this reason, we may call the orthogonal tensor 0 = t0i Ei the orientation tensor of the curved reference beam cross-section. For a straight and untwisted beam, 0 is constant along the beam midcurve, while it varies for a general curved beam, leading to the major dierences between a straight beam and a curved/twisted beam. For the straight reference beam conguration, the corresponding orthogonal orientation tensor is 00 t00i Ei = Ei Ei = I3 = vi vi ; vi R 3 and vi vj = ij (10)

where I3 is the second-order identity tensor of dimension three.

5

SO(3) denotes the special group of proper orthogonal tensors of order two, whose component repre-

sentations, referred to a given orthonormal frame, are a set of proper three-by-three orthogonal matrices. Proper means that the determinant of the orthogonal tensor or matrix is 1. All the reference frames in this paper are orthonormal, whose mutually orthogonal unit base vectors follow the conventional right hand rule. Therefore, only proper orthogonal tensors are used. 6 For any four vectors v1 = v1j ej , v2 = v2j ej , v3 = v3j ej , v4 = v4j ej R3 , v1 v2 represents a

second-order tensor, whose component representation, referred to the base ei ej , is [v1 v2 ]ei ej = [v1i v2j ], a three-by-three matrix. Two important identities exist: (v1 v2 )v3 = (v2 v3 )v1 and (v1 v2 )(v3 v4 ) = (v2 v3 )(v1 v4 ).

2.4 The geometry, elongation and shearing of the moving beam Assume that the beam moves and deforms from the curved reference beam conguration at time t0 to the current beam conguration at time t. The position vector of the material point 1 s0 [0, L0 ] on the beam midcurve moves from 0 R3 at time t0 to R3 at time t through a translational displacement u = u(1 ) R3 , i.e., = (1 ) = 0 + u (11)

and during the same deformation and motion process, the local frame of the moving beam is rotated, along with the rigid cross-section, from t0i = t0i (1 ) R3 at time t0 to ti = ti (1 ) R3 at time t, which stays orthogonal and unitary (ti tj = ij ) through the so-called orthogonal rotation tensor r = r (1 ) SO(3) as follows: ti r t0i = r 0 Ei Ei r where = (1 ) r 0 (13) ti t0i ; ti Ei SO(3) (12)

Considering ti = tij ej , one observes that the component representation of the orthogonal tensor , referred to ei Ej , is

t11

t21 t22 t23

t31

[]ei Ej = [ij ]ei Ej = t12

t32

(14)

t13

t33

Therefore, the orthogonal tensor determines the orientation of the moving beam crosssection at time t. Similar to 0 , we may call the orthogonal tensor = ti Ei the orientation tensor of the moving or current beam cross-section. Note that the arc-element ds of the current beam midcurve corresponding to the material 9

point at 1 s0 [0, L0 ] on the curved reference beam midcurve is ds = J d1

1

(15)

where J = J (1 ) ||,1 || and the L2 -norm ||v|| = (v v) 2 for any vector v Rn . Then the elongation e or extension ratio of the midcurve of the moving beam at time t is dened by e = e(1 ) = ds 1=J 1 ds0 (16)

Thus, the unit tangent vector of the midcurve of the moving beam at time t corresponding to the material point 1 s0 [0, L0 ] on the beam midcurve is calculated as ,s = 1 1 ,1 = J 1 + e ,1 (17)

Generally, the unit normal vector t1 of the deformed beam cross-section does not coincide with the unit tangent vector ,s because of the shearing; the angle changes between the tangent vector of the midcurve and t1 and away from orthogonal to t2 and t3 are the angles of shearing, denoted by 1i and determined by (also see e.g. Iura and Atluri, 1989; etc.) ,s t1 = cos11 ,s t2 = cos( ,s t3 or equivalently, 1 t1 = cos11 1 + e ,1 1 t2 = sin12 1 + e ,1 1 t3 = sin13 1 + e ,1 12 ) = sin12 2 = cos( 13 ) = sin13 2

(18)

(19)

At time t0 , the moving beam coincides with the curved reference beam. Similarly, we may rewrite Eq. (19) for the curved reference beam though the corresponding elongation and shearing vanish: 1 t01 = cos011 1 + e0 0,1 10

1 t02 = sin012 1 + e0 0,1 1 t03 = sin013 1 + e0 0,1 e0 = 0 ; 01i = 0

(20)

Also, the position vector = (1 , 2 , 3 ) R3 of any material point (1 , 2 , 3 ) for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) on the moving beam cross-section at time t can be described by = + 2 t2 + 3 t3 (21)

For the reduced beam or one-dimensional beam, the moving beam conguration at time t can be completely determined by the position vector and orthogonal orientation tensor of the beam cross-section as (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; etc.) = (1 ) (, ) for 1 s0 [0, L0 ] at any time t, whose special cases are 0 = 0 (1 ) (0 , 0 ) for 1 s0 [0, L0 ] (23) (22)

at time t0 when it coincides with the spatial xed curved reference beam conguration, and 00 = 00 (1 ) (00 , 00 ) for 1 s0 [0, L0 ] (24)

at time t00 when it coincides with the spatial xed straight reference beam conguration. Note again that both the curved and straight reference beam congurations are spatially xed and independent of time though the moving beam conguration may coincide with these by respectively taking the pre- subscripts 0 and 00. Note that 1 s0 s00 . The three congurations of a spatially curved beam are shown in Fig. 1. The right end cross-section is cut through the midcurve at a material point 1 s0 = [0, L0 ] and the left end corresponds to 1 s0 = 0.

11

Translational motion:
The position vector changes from 0 to via displacement u 0 = 0 + u 0 t2 current moving beam ( , ) t1 local moving frame t3 Rotational motion: The orthogonal tensor changes from 0 to via orthogonal rotational tensor r = r 0 and following relations hold: t0i = 0 Ei 0 ti = r t0i = Ei t00i = Ei = ei

curved reference beam ( 0 , 0 ) t02 0 t03 0

t01 0 0

spatial fixed frame e2 t002 t001 (material frame) (E2 ) e1 e3 ( E1) t003 ( E3) straight reference beam ( 00 , 00) 00
Fig. 1 A 3-D curved beam

12

Chapter 3: Preliminaries
In this Chapter, we review and summarize the denitions of some basic concepts associated with nite rotations so that the geometrically exact curved beam theory under the rigid cross-section assumption can be easily handled. Mostly, we focus on the objects of the same material point of the moving beam, either on the midcurve which is parametrized by 1 s0 [0, L0 ], or any point on the cross-section (1 , 2 , 3 ) R3 for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ), whichever is applicable. 3.1 Finite rotations by orthogonal transformations In Eq. (12), we have dened the orthogonal rotation tensor r , which rotates the local frame of the moving beam from t0i at time t0 to ti at time t through an orthogonal transformation (see e.g. Argyris, 1982; Simo, 1985; Atluri and Cazzani, 1995; etc.) ti = r t0i . Though the orthogonal orientation tensors 0 , as dened in Eq. (8), and , as dened in Eqs. (12) and (13), determine the orientations of the local frames of the unstrained/unstressed curved reference beam and current beam congurations, respectively, we may also view them as rotation tensors that rotate the local frame of the moving beam from t00i = Ei at time t00 , to t0i at time t0 and to ti at time t, respectively. In this sense and also from Eq. (13) , we may consider = r 0 as a compound rotation, which is the result of the rotation 0 of the moving frame from t00i = Ei at the time t00 to t0i at time t0 and then a following rotation r of the moving beam from t0i at time t0 to ti at time t. It follows that if we consider a nite incremental rotation of the moving frame from ti at time t by the orthogonal incremental rotation tensor rw SO(3) to ti at time t , then we have ti = rw ti = rw r t0i = rw r 0 Ei r t0i Ei (25)

13

where r = rw r SO(3) (26)

is the updated compound rotation tensor that rotates the moving frame from t0i at time t0 to ti at time t as a compound result (in fact, r is also a compound rotation tensor); and the updated orthogonal orientation tensor = rw = rw r 0 SO(3) (27)

represents the orientation of the moving frame at time t or the rotation of the moving frame from t00i = Ei at time t00 to ti at time t , as a sequence of rotations, 0 , r and rw . For an updated rotation tensor, the reverse-ordered multiplicative procedure as in Eq. (26) should be used when we observe the rotations referred to the xed spatial frame ei (see e.g. Argyris, 1982; etc.). 3.2 Spatial and material descriptions The objects of a beam relating nite rotations are strongly orientation dependent because of the geometric constraint resulting from the rigid cross-section assumption. It helps to dene the spatial and material descriptions of an object. An arbitrary vector v R3 of the beam induced by the translation, rotation and deformation away from the straight reference beam to the moving beam at any time t is a spatial object (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; and the references therein), and may be referred to either the spatial frame ei or the moving frame ti as v = vj ej = vj tj (28)

which is called the spatial form of the spatial vector v. On the other hand, v can be further v v written as v = vj Ej = j Ej = with as the corresponding orientation tensor; thus, we have v = vj Ej = t v (29)

14

which is called the material form of the spatial vector v (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986). The superscript t of ()t means the transpose of a vector, tensor or matrix () throughout the thesis. The special cases of spatial vectors are the local base vectors, ti at any time t, of the moving beam, and its corresponding material form is i = t ti = Ei t (30)

For two arbitrary spatial vectors v R3 at any time t and v R3 at any time t of the same material point of the moving beam with and as their corresponding orientation tensors, respectively, if there exists a linear transformation relation given by v = Tv the corresponding transformation relation in the material form is given by v v = T (32) (31)

where v = ( )t v and v = ( )t v are the material forms of v and v , respectively; T and T are the second-order tensors in the spatial and material forms, respectively. It is easy to conrm that the following relations hold: T = T( )t ; T = ( )t T (33)

The special cases of spatial second-order tensors are the orthogonal tensors r and of the moving beam at any time t which dene the transformations ti = r t0i and ti = Ei , respectively, from Eq. (12). Their corresponding material forms are t i = r 0i = Ei ; t where r = t r 0 = I3 ; and i = Ei = Ei t (34)

and = t 00 = I3 ;

The above notations (upper bar for a material object and no upper bar for a spatial object) and relations are used throughout the paper. 15

Notice that the components of a material object relating the moving beam at any time referred to the material frame Ei are equal to those of its spatial object referred to its local frame (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986). The same holds for second-order tensors. Every beam conguration has its own material objects. It can be seen later that the advantages of using the material descriptions are: (1) arbitrary rigid-body rotations are removed; (2) the objectivity is kept for any observer transformation (see e.g. Simo, 1985; Ogden, 1997: 74, 130-139); (3) summation and subtraction may be performed directly for certain objects. 3.3 Linearized increments and derivatives We specify the notations for linearized increments and derivatives of an object. For an innitesimal change or increment of some parameter x R1 , denoted by x, the corresponding linearized or innitesimal increment of an object () will be denoted by x () (the variational operator x () for the same material point and the dierential operator d() for a variation of the material point in space are two special examples, and the same operation rules as partial derivatives or dierentiations hold); and the corresponding (partial) derivative by (),x = for y R1 . 3.4 Spin and rotational speed as a general term We consider the linearized increment x of the orthogonal tensor (which determines the orientation of the moving beam at any point of the midcurve at any time t) due to an innitesimal increment x. Using the chain rule of partial derivatives for linearized operations, we have x (t ) = x t + x t = x (I3 ) = 0 16 (35) () x () = x x and (),xy = [(),x ],y = y [ x () ] 2 () x = xy y

and dene x w x t x wt ; then x wt = x t = x w; (x w = x w)

t

(x w t x )

(36)

(37)

Therefore, (the spin of the moving beam relative to the straight reference beam in the linearized incremental form due to x) is a skew symmetric tensor (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; Iura and Atluri, 1989; etc.), x w so(3)7 (x w so(3)) in the spatial (material) form, associated with the spatial (material) axial vector x w R3 (x w R3 ) (the innitesimal or linearized incremental rotation vector of the moving beam relative to the straight reference beam). In the derivative form, x w,x x w ,x t = x t so(3); x ( x ,x so(3)) (38)

which is called spin (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; Iura and Atluri, 1989; etc.), as x varies, of the moving beam relative to the straight reference beam, associated with the spatial (material) axial vector x = xj tj w,x R3 ; ( x = xj Ej t x R3 ) (39)

which is the orientation change rate (angular or rotational speed) vector of the moving beam relative to the straight reference beam as the parameter x varies (with x as time parameter).
7

so(3) denotes the group of skew symmetric tensors. A second-order Cartesian tensor v is a skew symmet-

ric tensor if and only if v = t , whose associated axial vector v R3 is its only eigenvector corresponding v to zero eigenvalue vv = 0. For another arbitrary vector v1 R3 , an identity exists: v v1 = vv1 . The component representation of the skew symmetric tensor v corresponding to an axial vector v = vj ej , referred to ei ej , is 0 v3 0 v1 v2

[ ]ei ej = v3 v v2

v1 0

v An important identity exists along with the standard tensor product: v v1 = (v v1 )2 I3 + ( v1 )t . The above notation with an upper head for the skew symmetric tensor of a vector is valid throughout the paper.

17

Noticing = r 0 according to Eq. (13) and using the chain rule of partial derivatives, we have ,x = r,x 0 + r 0,x Thus, ,x t = r,x 0 t + r 0,x t = r,x t + r 0,x t t r 0 r (41) (40)

Therefore (for skew symmetric curvature case, see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; Ibrahimbegovi, 1995), c x = rx + r 0x t r where rx wr,x = r,x t so(3); r ( rx = t rx = t ,x t 0,x so(3)) 0 (43) (42)

is the incremental spin, as x varies, in the spatial (material) form of the moving beam at any time t relative to the curved reference beam associated with the spatial (material) axial vector rx R3 ( rx R3 ) while 0x w0,x = 0,x t so(3); 0 ( 0x = t 0x 0 = t 0,x so(3)) 0 0 (44)

is viewed as the spin, as x varies, in the spatial (material) form of the curved reference beam at any time t relative to the straight reference beam associated with the spatial (material) axial vector 0x R3 ( 0x R3 ). Eq. (42) implies that the spin of the current beam relative to the straight reference beam cannot be obtained by a simple addition of an incremental spin relative to the previous beam orientation in their spatial forms; it needs to align the spin of the previous orientation to the current orientation by applying the corresponding relative rigid-body rotation. The associated axial vector relation for Eq. (42) is x = rx + r 0x 18 (45)

However, one may easily conrm that a simple addition can be performed in their material forms: x = rx + 0x ; x = rx + 0x (46)

Note that when x t the physical time or conguration transformation due to deformation and motion, 0t 0x |x=t = 0; hence, t x |x=t = rt rx |x=t (48) ( 0t 0x |x=t = 0) (47)

because the curved reference beam is spatially xed and independent of the current time t! We use an upper dot to denote the true time derivative throughout the paper; thus, the time spin is w w,t = t = wr wr,t = r t ; r w0 w0,t = 0 t = 0 0 (49)

It should be mentioned that we have no denition for w nor for w; they always come along with the linearized operator x in front of them, and just denote an incremental rotation as x R1 varies. Conventionally, we let (x w),y = but (x w),y = x (w,y ); w,xy = w,yx (51) y (x w) ; y w,xy = (w,x ),y =
y [ x w ] x y

(50)

in general even though x and y are independent of each other because we have no denition for w nor for w! 19

3.5 Co-rotated derivatives v For any spatial vector v = vi ti = R3 of the moving beam at any time t, we have, using x = ,x t , (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986) v v,x = ,x v + ,x = (,x t )( ) + (t v),x = x v + (t v),x v and dene (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986) v,x vi,x ti (t v),x ,x v,x x v v,x x v v (53) (52)

as the co-rotated derivative of the spatial vector v with respect to x, with upper tilde to identify it throughout the paper. The meaning of the co-rotated derivative is that the derivative of a spatial object is taken by an observer xed in the moving frame, only on the components referred to the corresponding moving frame. To an observer who stays still in the xed spatial frame, he/she needs to pull-back the object to the material form v = t v to perform the usual derivative operation and then push-forward (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; and the references therein) to the spatial form ,x ; or equivalently, he/she needs to subtract the spin eect v x v from the usual derivative v,x (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986) to have the same objective observation as by the observer xed in the moving frame. A valuable result of application of the co-rotated derivative of a spatial vector to the current frame ti is i,x = ti,x x ti = (t ti ),x = (Ei ),x = 0 t i.e. a set of important formulae: ti,x = x ti = x ti t1,x = x3 t2 x2 t3 t2,x = x3 t1 + x1 t3 t3,x = x2 t1 x1 t2 20 (55) (54)

which may be called Frenet-Serret formula in a general sense. On the other hand, for any spatial second-order tensor T = Tij ti tj that denes a transformation, v = Tv between vectors v = vj tj R3 at any time t and v = vi ti R3 at any time t of the moving beam associated with the orientations = ti Ei SO(3) and = ti Ei SO(3) of the same material point of the beam and the corresponding material objects as in Eqs. (31) through (33), one may dene the corresponding co-rotated derivative as T,x Tij,x ti tj [( )t T ],x ( )t T,x ( )t T,x ( x T T x ) (56) and the chain rule holds for the co-rotated derivative operation: v v,x = T,x v + T ,x (57)

21

Chapter 4: Strains and stresses

4.1 Curvatures and curvature changes The elongation and shearing on the midcurve have been taken into account conceptually in a classical or engineering sense for a moving beam in Section 3.2. That the crosssection rotates away from the orthogonality with the tangent vector of the beam midcurve is considered as shearing, while the relative orientation change, as the material point 1 s0 [0, L0 ] varies along the beam midcurve, of the cross-sections at the same material point on the midcurve between the moving beam and the curved reference beam determines the curvature change of a curved beam. Taking the derivatives of local frames t0i and ti , respectively, with respect to the arclength coordinate 1 of the (unstrained and unstressed) curved reference beam midcurve and noticing Eq. (55) (also see e.g. Iura and Atluri, 1988; Iura and Atluri, 1989; etc.), we may obtain the Frenet-Serret formulae in their original sense t0i,1 = 0 t0i = 0 t0i t01,1 = 03 t02 02 t03 t02,1 = 03 t01 + 01 t03 t03,1 = 02 t01 01 t02 for the curved reference beam conguration, and the generalized Frenet-Serret formulae ti,1 = ti = ti t1,1 = 3 t2 2 t3 t2,1 = 3 t1 + 1 t3 t3,1 = 2 t1 1 t2 for the current beam; where, noticing Eq. (43) 0 01 = w0,1 = 0,1 t 0 22 (59) (58)

 0 01 = t 0 0 = t 0,1 0 0 and r 1 = w,1 = ,1 t = r + r 0 t 1 = t = t ,1 = r + 0 as well as r r1 = wr,1 = r,1 t r r r1 = t r = t ,1 t 0,1 0

(60)

(61)

(62)

where, noticing Section 3.4 about spin as a skew symmetric tensor and rotational speed as the associated axial vector as well as the corresponding notations and operation rules, 0 , and r so(3) (0 , and r so(3)) are the skew symmetric tensors in their spatial forms (material forms) associated with the axial vectors 0 , and r R3 ( 0 , and r R3 ) , respectively, in their spatial forms (material forms). Note that 0 01 = w0,1 = 0j ej = 0j t0j 1 = w,1 = j ej = j tj r r1 = wr,1 = rj ej = rj tj for the spatial forms, and 0 01 = 0j Ej = t 0 0 1 = j Ej = t r r1 = rj Ej = t r = 0 for the material forms. 0 ( 0 ) is the curvature (vector) of the (unstrained/unstressed) curved reference beam conguration in the spatial (material) form, denoting the rotational speed or orientation change rate of the cross-section with respect to 1 as the material point 1 s0 [0, L0 ] varies along 23 (64) (63)

the beam midcurve. The component 01 is the twist rate around the tangent vector t01 , and 02 and 03 are the corresponding curvature components around t02 and t03 , respectively. Similarly, we also call ( ) the curvature (vector) of the current beam in the spatial (ma terial) form in a general term though it denotes the rotational speed or orientation change rate of the cross-section of the current beam with respect to 1 not s [0, L] as the material point 1 s0 [0, L0 ] varies along the curved reference beam midcurve. The component 1 is the twist rate around the normal vector t1 , and the components 2 and 3 are the curvatures around t2 and t3 of the current beam, respectively. Therefore, we further call r ( r ) the curvature change (vector) of the current beam relative to the curved reference beam in the spatial (material) form and it denotes the rotational speed of the current beam cross-section relative to the curved reference beam cross-section with respect to 1 as the material point 1 s0 [0, L0 ] varies along the beam reference midcurve. The component r1 is the twist rate change around the normal vector t1 , and the components r2 and r3 are the curvature changes around t2 and t3 of the current beam relative to the curved reference beam, respectively. On the other hand, 0 , and r (0 , and r ) are called the corresponding skew symmetric curvatures or curvature changes in their spatial (material) forms. So far, the elongation and shearing on the beam midcurve as well as the curvature change have been specied. 4.2 Deformation gradients The study of the deformation gradients helps determine the strain measures at any material point on the current beam cross-section. Simo (1985) (also see e.g. Simo and Vu-Quoc, 1986; etc.) easily obtained the deformation gradient tensor for initially straight beams simply using orthonormal reference frames by aligning the straight beam to coincide with the straight reference beam conguration. In the case of the initially curved/twisted beam, Iura 24

and Atluri (1989) used the covariant and contravariant reference frames to help obtain the strain measures, in which case the real ber length is not a direct consideration. In this section, we will indirectly obtain the deformation gradient tensor of the current beam conguration relative to the (unstrained/unstressed) curved reference beam conguration by rst obtaining the deformation gradient tensors of the two curved beam congurations relative to the straight reference beam conguration followed by a reference conguration change, to avoid using the covariant and contravariant reference frames or curvilinear coordinate systems. We repeat the same procedure as Simos (1985) for the latter two deformation gradient tensors for completeness. Noticing Eqs. (3) and (21) for the expressions of the position vectors 0 and of any point (1 , 2 , 3 ) for 1 [0, L0 ] and (2 , 3 ) B(1 ) in the curved reference beam and current beam congurations, as well as the Frenet-Serret formulae in Eqs. (58a) and (59a), respectively, we have 0,1 = 0,1 + 0 (2 t02 + 3 t03 ); for a point in the curved reference beam, and ,1 = ,1 + (2 t2 + 3 t3 ); for a point in the current beam 8 . Therefore, noticing 0 = t0i Ei and = ti Ei as given in Eqs. (8) and (12), respectively, as well as the spatial and material descriptions for vectors and second-order tensors in Section 4.2, the deformation gradient tensors, F0 and F, of the curved reference beam and current beam relative to the straight reference beam, respectively, are dened and determined by d 0 F0 d 00 F0 = 0,j Ej =
8

0,2 = t02 ;

0,3 = t03

(65)

,2 = t2 ;

,3 = t3

(66)

E1 + t0i Ei = 0i t0i E1 + t0i Ei =

E1 + 0 = 0 F0

0,i and ,i are used as natural base vectors for the curved reference beam and current beam, respec-

tively, by some authors (see e.g. Iura and Atluri, 1988; Iura and Atluri, 1989; etc.).

25

 F0 = 0 E1 + I3 = 0i Ei E1 + Ei Ei and d Fd 00 F = ,j Ej = E1 + ti Ei = i ti E1 + ti Ei = E1 + = F F = E1 + I3 = i Ei E1 + Ei Ei respectively, where

0

(67)

(68)

0j t0j = 0 + 0 (2 t02 + 3 t03 ); j tj = + (2 t02 + 3 t03 );

0 0j Ej = 0 + 0 (2 E2 + 3 E3 ) j Ej = + (2 E2 + 3 E3 )

(69) (70)

and 0 0j t0j = 0,1 t01 ; j t0j = ,1 t1 ; 0 0j Ej = t 0,1 E1 0 j Ej = t ,1 E1 (71)

Noticing Eqs. (20) and (19), the relations between the elongations and shearings for the curved beam reference beam and current beam congurations, we obtain the components of 0 = 0 (1 ) and = (1 ) referred to their local frames: 0j = 0,1 t0j t01 t0j = 0 01 = 0,1 t01 1 = (1 + e0 )cos011 1 02 = 0,1 t02 = (1 + e0 )sin012 03 = 0,1 t03 = (1 + e0 )sin013 e0 = 0; and j = ,1 tj t1 tj 1 = ,1 t1 1 = (1 + e)cos11 1 2 = ,1 t2 = (1 + e)sin12 3 = ,1 t3 = (1 + e)sin13 26 (73) 01i = 0 (72)

Then, the components of

0 (1 , 2 , 3 )

and

= (1 , 2 , 3 ), in Eqs. (69) and (70),

referred to their local frames will be, noticing Eqs. (58) and (59), 01 = 01 + (3 02 2 03 ) 02 = 02 + (3 01 ) 03 = 03 + (2 01 ) and 1 = 1 + (3 2 2 3 ) 2 = 2 + (3 1 ) 3 = 3 + (2 1 ) respectively. Note that the component representations of F0 and F in the spatial forms as well as F and F0 in the material forms can be easily identied from Eqs. (67) and (68) as [F0 ]t0i Ej = [F0 ]Ei Ej 1 + 01 0 0
=

(74)

(75)

02 03

1 0

(76)

0 1

and [F]ti Ej = [F]Ei Ej 1 + 1 0 0

=

2 3

1 0

(77)

0 1

respectively. After some simple elementary linear algebra, the determinants of F0 and F can be easily obtained as g0 detF0 = (det0 )(detF0 ) = detF0 = 1 + 01 = 1 + 01 + (3 02 2 03 ) 27 (78)

and g detF = (det)(detF) = detF = 1 + 1 = 1 + 1 + (3 2 2 3 ) respectively, and the inverses of F0 and F, respectively, are F1 = (0 F0 )1 0 1 1 1 = F1 t = ( 0 E1 + I3 )t = 0 t01 + t = t ( 0 0 0 0 0 g0 g0 g0
0

(79)

t01 + I3 ) (80)

and, similarly, 1 1 1 F1 = ( E1 + I3 )t = t1 + t = t ( t1 + I3 ) g g g (81)

Noticing the denitions of F0 and F in Eqs. (67) and (68) as well as (80) and (81), one may dene and obtain the deformation gradient tensor of the current beam relative to the (unstrained/unstressed) curved reference beam through a change of reference congurations (see e.g. Ogden, 1997: 120) as d Fr d 0 Fr = FF1 = 0 Fr = 1 g0
r

t01 + r =

1 ri ti t01 + ti t0i = Fr t 0 g0 (82)

1 1 r E1 + I3 = ri Ei E1 + Ei Ei g0 g0

where, noticing Eqs. (69) through (75), noticing r r , r rj Ej 0 = r + r (2 E2 + 3 E3 ) = r + (3 r2 2 r3 )E1 + (3 r1 )E2 + (2 r1 )E3
r

= r = r

= r + r (2 t2 + 3 t3 )

= r + (3 r2 2 r3 )t1 + (3 r1 )t2 + (2 r1 )t3 = rj tj rj j 0j = ,1 tj 0,1 t0j 28

r1 = r1 + (3 r2 2 r3 ) r2 = r2 + (3 r1 ) r3 = r3 + (2 r1 ) rj = j 0j = ,1 tj 0,1 t0j = ,1 tj 1j and r 0 = t ,1 t 0,1 = rj Ej 0 r = r = ,1 r 0,1 = r 0 = = ,1 t1 = rj tj (84) (83)

for 0j = 0j = 0, where the meanings of r and r are given in Table 1 at the end of this chapter. The determinant and inverse of Fr can be obtained as, using Eqs. (78) and (79), gr = detFr = det(FF1 ) = detFdetF1 = 0 0 and, using Eqs. (67) and (81), F1 = (FF1 )1 = F0 F1 = t ( r 0 r respectively. Note that
0

detF g r1 = =1+ detF0 g0 g0

(85)

1 g

t1 + I3 )

(86)

(0 ) and

() are, respectively, the strain vectors at any point on the cross-

section and 0 (0 ) and () are the strain vectors on the beam midcurve for the curved reference beam and current beam congurations relative to the straight reference beam conguration (see e.g. Simo, 1985). They determine the corresponding elongations and shearings relative to the straight reference beam as seen e.g. in Eqs. (72) and (73) for the beam midcurve. By comparison of the similarity between Eq. (82) and Eq. (68), one may intuitively choose
r /g0

(r /g0 ) as the right strain vector of the current beam conguration relative to the curved

reference beam conguration that is conjugate to the corresponding rst Piola-Kirchho stress vector (see Section 5.4). 29

Now we address the geometrical meaning of g0 , which is the same as that used by Iura and Atluri (1988, 1989). First, by its denition, g0 = detF0 , as given in Eq. (78), is the scalar between the dierential volumes of the curved reference beam and straight reference beam congurations at any material point (1 , 2 , 3 ) of the beam: dV0 = g0 dV00 = g0 d1 d2 d3 dV00 = d1 d2 d3 (87)

where V0 and V00 are the volume domains of the curved reference beam and straight reference beam congurations, respectively. Second, a unit-length ber parallel to E1 (the normal of the straight reference beam crosssection) at any material point on the cross-section is stretched to be g0 in the direction of t01 (the normal of the curved reference beam cross-section) if the moving beam moves and deforms from the straight reference beam conguration to the curved reference beam conguration. In fact, on use of Eq. (67) for the deformation gradient tensor F0 of the curved reference beam relative to the straight reference beam and Eq. (78) for the expression of g0 , we have (F0 E1 d1 ) t01 = (
0

+ t01 ) t01 d1 = (

t01 + t01 t01 )d1 = (01 + 1)d1 g0 d1 (88)

which is dependent on the curvatures of the curved reference beam conguration, but not on the twist for a given point on the cross-section, see Eq. (78). This is a result of the assumption that the beam cross-section remains undeformed during any motion and deformation. Third, Eq. (88) also implies that any cut slice of the curved reference beam through two cross-section planes with dierential length d1 , at any point on the midcurve, is linearly tapered and its thickness in direction t01 varies according to g0 d1 as the material point varies on the curved beam cross-section. Any ber parallel to t01 has a real length g0 d1 at (1 , 2 , 3 ) if a ber parallel to t01 has a real length d1 at 1 on the midcurve. 30

From the above explanations, we may be clear that the factor

1 g0

in front of

in the

deformation gradient tensor Fr of the current beam relative to the curved reference beam in Eq. (82) is due to the fact that it is taken with respect to the real undeformed ber length variations, which can also be observed from the denition of d Fr d 0 in the same set of equations, in which d 0 and d R3 are the spatial vectors of an oriented dierential ber with real length before and after deformation in the physical space with the orthonormal reference frame ei or ti . Because of its importance in the truly geometrically exact curved beam theory, we may call g0 the initial curvature correction term, whose eect may be signicant for thick and moderately thick curved beams and small for slender beams. Similar explanations may be made for g = detF as dened and given in Eq. (79) and gr = detFr as in Eq. (85). 4.3 Stresses, stress resultants and stress couples There are many choices of stress measures (see e.g. Ogden, 1997: 145-168; etc. for their denitions). We only illustrate the Cauchy stress and rst and second Piola-Kirchho stresses for the curved beam and then represent each of them in terms of the components of the rst Piola-Kirchho stress and its conjugate strain (which will be conrmed in the next section), along with the Cauchy stress for the conjugate strain analysis in the next section. Then, we dene the stress resultant and stress couple in the classical or engineering sense (see e.g. Reissner, 1972; Reissner, 1973; Reissner, 1981; Simo, 1985; Simo and Vu-Quoc, 1986; Iura and Atluri, 1989; etc.). Denote by the true and symmetric Cauchy stress tensor referred to a dierential volume of the current beam at any material point (1 , 2 , 3 ) for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ): j tj = ji ti tj = ji Ei Ej 31

 j ji ti j = ji Ei ij ji (89)

where j is the associated stress vector acting on the face and referred to the real area of the same face of the current beam with tj as unit normal vector.
1 Then, on use of Eq. (86) for the inverse of the deformation tensor, F1 = t ( g r r r t1 + I3 ),

and the relations gr = 1 +

rj g0

as given in Eq. (85), the unsymmetric rst Piola-Kirchho

stress tensor 0 referred to a dierential volume of the curved reference beam can be obtained as, according to its denition (see e.g. Ogden, 1997: 85, 153; etc.), 0 gr Ft = 0 t0j = ji ti t0j 0 r j 0 = ji Ei Ej 0 0 ji ti = gr j 0 j j 0 = ji Ei 0 ji = gr ji 0 0 Ft Fr ( 0 )t r 0 = 1 1 (90) rj 1i = ij 0 g0 rj 1 g0

where 0 is the corresponding stress vector acting on the deformed face in the current beam j corresponding to the reference face normal to t0j in the curved reference beam and referred to the real area of the same reference face in the curved reference beam. Inversely, we have, noticing Eq. (82) for Fr , 1 0 t 1 0 rj 0 1 0 ri 0 Fr = (ji + 1i )ti tj = (ij + 1j )ti tj gr gr g0 gr g0 (91)

Similarly, for later reference, the rst Piola-Kirchho stress tensor 00 referred to a dierential volume of the straight reference beam can be given by 00 gFt = g0 0 Ft = 00 t00i 0 i 32

 00 = 1 = 0 1 1 00 = g 2 2 1 = g0 0 02 0 2 2 1 00 = g 3 3 1 = g0 0 03 0 3 3 1 00 Ft F( 00 )t (92)

where 00 is the corresponding stress vector acting on the deformed face in the current beam i corresponding to the reference face normal to t00i = Ei in the straight reference beam and referred to the real area of the same reference face in the straight reference beam. Note that the rst Piola-Kirchho stress vector referred to the cross-section of any beam conguration is the same as the real (Cauchy) stress vector on the cross-section because the cross-section plane remains undeformed. Besides, the symmetric second Piola-Kirchho stress tensor (see e.g. Ogden, 1997: 159, 157; etc. for its denition), denoted by 0(2) , is given in terms of the rst Piola-Kirchho stress and Cauchy stress components as 0(2) gr F1 Ft F1 0 = j r r r
0(2) 0(2) = ji Ei Ej 0(2)

t0j = ji t0i t0j

0(2)

0(2)

F1 0 = F1 (ji ti ) = ji (F1 ti ) = ji t0i 0 0 r r j r

0(2)

0(2) 0(2) = ji Ei j

ij

0(2)

ri 0 rj 0 j1 = ij 0 gr g0 gr g0 i1 ri rj 1 ri rj = gr ij ( 1j + 1i ) + ( )( )11 g0 g0 gr g0 g0 = ji 0

(93)

where j

0(2)

is the stress vector acting on the deformed face in the current beam corresponding

to the reference face normal to t0j in the curved reference beam and referred to the real area of the same reference face in the curved reference beam ( 0 ) and then is contracted back to j the curved reference beam. It can be seen that the dierences among the Cauchy stress and rst and second PiolaKirchho stresses are obvious for nite strain problems, though the dierences tend to vanish 33

for small strain problems. For the reduced beam or one-dimensional beam model, it is convenient to dene the stress resultant (internal force vector acting on the current beam cross-section) and stress couple (internal moment vector acting on the current beam cross-section) in the classical or engineering sense. The stress resultant n = n(1 ) R3 is dened as n = ni =
B B B

0 t01 d2 d3 0 d2 d3 = 1 1i d2 d3 0
B

00 t001 d2 d3
B

t1 d2 d3

Fr 0(2) t01 d2 d3

00 d2 d3 = 1

1 d2 d3 = ni ti (94)

and the stress couple m = m(1 ) R3 , referred to the position vector on the current beam reference curve, is dened as m =
B B B

( ) [ 0 t01 ]d2 d3 ( ) [t1 ]d2 d3 ( ) 0 d2 d3 = 1

B B

( ) [ 00 t001 ]d2 d3

( ) [Fr 0(2) t01 ]d2 d3

B

( ) 00 d2 d3 = 1

( ) 1 d2 d3

= mi ti m1 = m2 = m3 =
B B B

(2 13 3 12 )d2 d3 0 0 (3 11 )d2 d3 0 (2 11 )d2 d3 0 (95)

in the spatial forms. For mi in the last set of equations, use has been made of = 2 t2 + 3 t3 from Eq. (21). The stress resultant and stress couple in the material forms are n = n(1 ) = t n = ni Ei and m = m(1 ) = t m = mi Ei 34 (97) (96)

respectively. In Eqs. (94) through (97), n1 is the normal force component in the cross-section normal direction t1 while n2 and n3 are the shear force components in the directions t2 and t3 , respectively. On the other hand, m1 is the torque component around the cross-section normal t1 while m2 and m3 are bending moment components around t2 and t3 , respectively. Note that if the second Piola-Kirchho stresses are used, one needs to stretch the stresses to the rst Piola-Kirchho stresses to obtain the stress resultant and couple in the classical or engineering sense if the strains are not small. 4.4 Internal power: the strain measures conjugate to the rst Piola-Kirchho stresses There are many choices of strain measures, each of which has its own conjugate stress though there are some exceptions (see e.g. Ogden, 1997: 159). The general power balance condition states that the external surface traction power plus the body force power is equal to the kinetic power plus the internal power (stress power) for a given reference volume domain of the continuum body using the Lagrangian descriptions if only the mechanical energy is concerned (for the exact mathematical statement, see e.g. Ogden, 1997: 156). In the dierential form, the internal power per unit reference volume of the continuum is pint = Trace[( 0 )t Fr ] in terms of the rst Piola-Kirchho stress tensor 0 and material time derivative Fr of the deformation gradient tensor Fr , which is an objective scalar, independent of the observer and reference frame at a given material point (see e.g. Ogden, 1997: 156). Like Simo (1985) and Simo and Vu-Quoc (1991), we may examine the objective internal power to determine the strain measures that are conjugate to the rst Piola-Kirchho stresses for the curved beam. We rst need to obtain the material time derivative of the deformation gradient tensor Fr

35

as follows, noticing Eqs. (82), (49) and (56): Fr = (Fr t ),t = wFr Fr w0 + Fr = wFr + Fr 0 where 1 Fr = Fr t = r t01 0 g0 1 Fr = r E1 g0 r = r + r (2 E2 + 3 E3 ) r = r + r (2 t2 + 3 t3 ) r = r r = r t (99) (100) (101) (102) (103) (104) (98)

Note that the upper tilde denotes the co-rotated derivative as designated in Section 3.5. The current beam internal power per unit volume of the curved reference beam at any material point (1 , 2 , 3 ) is, noticing that Trace(Tt T2 ) = Trace(T1 Tt ) 1 2 for any two second-order Cartesian tensors T1 and T2 , pint = pint (1 , 2 , 3 ) Trace[( 0 )t Fr ] = Trace( 0 Ft ) r = Trace[( 0 )t wFr ] + Trace[( 0 )t Fr ] = Trace[ 0 (wFr )t ] + Trace[( 0 )t Fr ] (105)

The rst term of the above equation is due to rigid-body rotation and should vanish. In fact, noticing Eq. (90) for the relation between the rst Piola-Kirchho stress tensor 0 and the Cauchy stress tensor as well as the symmetry of the Cauchy stress tensor and skew symmetry of w, we have (also see e.g. Simo and Vu-Quoc, 1991) t Trace[ 0 (wFr )t ] = Trace[gr Ft Ft w ] = gr Trace[ w] r r t = gr Trace[ t w] = gr Trace[ w ] = gr Trace[ w] = 0 36 (106)

The second term becomes 1 Trace[( 0 )t Fr ] = Trace[(t0j 0 )( r t01 )] j g0 1 = Trace[( 0 r )(t01 t01 )] 1 g0 1 0 = r g0 1

(107)

It follows that the current beam internal power per unit volume of the curved reference beam at any material point (1 , 2 , 3 ) is pint = Trace[( 0 )t Fr ] = Trace[( 0 )t Fr ] = Trace[( 0 )t Fr ] = 0 ( 1 Therefore,
1 g0 r

1 1 r ) = 0 ( r ) 1 g0 g0
1 g0 r

(108)

in the spatial form (or

in the material form) is the current beam strain

vector at a material point (1 , 2 , 3 ) on the beam cross-section conjugate to the rst Piola1 Kirchho stress vector 0 (or 0 = t 0 ). 1 1 This also means that we may dene the corresponding strain tensor E r (E r ) as, noticing Eq. (82), E r (Fr t0i r t0i ) t0i = Fr r 1 1 ri ti t01 = Erij ti t0j r t01 = g0 g0 t t (Fr 0i r 0i ) 0i = Fr I3 t = = 1 1 r E1 = ri Ei E1 = Erij Ei Ej g0 g0 (109)

Er

which is conjugate to the rst Piola-Kirchho stress tensor 0 = gr Ft with the component r form
r1

g [E r ]ti t0j = [E r ]Ei Ej = r2 0

r3 g0

g0

0 0

0 0

(110)

0 0

In fact, E r = Fr r = Fr (r ),t t = Fr (I3 ),t t Fr 0 0 37 (111)

and E r Fr Note that the geometrical meaning of the strain vector native denition: 1 g0
r 1 g0 r

(112) can be observed from the alter-

Fr t01 r t01

(113)

which is the stretching of an oriented unit-length ber t01 of the curved reference beam at any material point to Fr t01 with the rigidly-rotated reference part t1 = r t01 removed. The component and
1 g0 r3 1 g0 r1 1 g0 r2

along t1 may be called the extensional strain, and the components

along t2 and t3 , respectively, called the shear strains, which are much like the
1 g0 r

extensional and shear strains for small strain problems. In fact, for small strain problems, the three components of the strain vector become the extensional strain and shear
1 g0 r

strains in the classical or engineering sense. For example, the three components of on the beam midcurve become r1 = e; r2 = 12 ; r3 = 13

if the elongation e and angles of shearing 12 , 13 are small as seen in Eq. (73). On the other hand,
1 , 1 g0 r1 g0 r2

and

1 g0 r3

are really conjugate to the normal stress component 11 and shear 0

stress components 12 and 13 , respectively. Therefore, we may call those strain components 0 0 engineering strains dened by oriented ber deformation for nite strain problems. Many researchers may prefer to use the symmetric Green strain tensor EG EG 1 t (F Fr I3 ) = EGij t0i t0j 2 r = EGij Ei Ej r r r1 + 1 g0 g0 g0 2

1 r2 2 g0 1 r3 2 g0

1 r2 2 g0

1 r3 2 g0

[E G ]t0i t0j = [E G ]Ei Ej =

0 0

(114)

38

which is conjugate to the second Piola-Kirchho stress tensor (notice Eq. (93) for its relation with the rst Piola-Kirchho stress components) in the sense of the internal power density 1 0(2) 1 0 pint Trace[ 0(2) E G ] Trace[( 0 )t Fr ] = ij EGij = ij Erij = 0 ( r ) g0 However, the symmetric Eulerian strain tensor EE EE [E E ]ti tj 1 (I3 Ft F1 ) Ft E G F1 = EEij ti tj r r r r 2 = EEij Ei Ej r r r1 1 g1r g0 g0 1 r2 1 r3 g0 2 2 g0 2 g0 1 1 r2 = [E E ]Ei Ej = 0 0 2 g0 gr 1 r3 0 0 2 g0 (115)

(116)

does not have its conjugate stress tensor, nor does the symmetric Cauchy stress tensor have its conjugate strain tensor; the Green strain tensor and Cauchy stress tensor (noticing Eq. 91 for its relation with the rst Piola-Kirchho stress components) relate the objective internal power density by pint gr Trace[] Trace[( 0 )t Fr ] ij ij = 1 ( where Ft E G F1 ij t0i t0j r r
r1
g0

1 1 r ) = 0 ( r ) 1 g0 g0

(117)

[]t0i t0j =

1 gr

1 r2 2g 0
1 r3 2 g0

1 r2 2 g0

1 r3 2 g0

0 0

(118)

is the Eulerian strain rate tensor (see e.g. Ogden, 1997: 155), which cannot be obtained simply taking the material time derivative on E E nor on E E in Eq. (116). Note that the components of both the Green strain tensor E G and Eulerian strain tensor E E consist of those of the symmetric part of the engineering strain tensor E r . In terms of the components of E r , both the Green strain tensor and Eulerian strain tensor are nonlinear and consist of the quadratic term 1 r r 1 rj rj = 2 g0 g0 2 g0 g0 39

We should point out that Iura and Atluri (1989) chose g0 Erij = ,i tj 0,i t0j as their strain measures, which are conjugate to
1 0 . g0 ij

The only non-zero components of

g0 Erij are g0 Eri1 = ri , which are the strains that are measured with reference to the ber length on the midcurve. For the one-dimensional beam or reduced beam model, the current beam internal power per unit arclength of the curved reference beam midcurve is Pint = Pint (1 ) = = = [
B B

1 r )g0 d2 d3 g0 B B 0 [r + r (2 t2 + 3 t3 )]d2 d3 = 0 [r + r ( )]d2 d3 1 1 pint g0 d2 d3 = 0 ( 1

B

0 d2 d3 ] r + [ 1

( ) 0 d2 d3 ] r 1 (119)

= n r + m r = n r + m r

Therefore, r and r (r and r ) are the strain measures conjugate to the stress resultant n and stress couple m ( and m), respectively. n In the literature, the one-dimensional or reduced strain measures for the curved beams are available in the vector/tensor forms (see e.g. Simo and Tarnow, 1995; Ibrahimbegovi, 1995, c etc.). In the above, we have obtained them directly from the three-dimensional curved beam model, and now summarize them into Table 1 for reference convenience.

40

Table 1 Reduced strain measures Strain type Spatial form Material form

Translational strains

r = ,1 t1

r = t r

Bending strains

r = r1 r1 = r,1 t r

r = t r

In Table 1, the translational strain vector r = rj tj consists of the extensional strain com ponent r1 along t1 normal to the current beam cross-section, and shear strain components r2 and r3 along t2 and t3 , respectively. The bending strain (curvature change) vector r = rj tj consists of the twist rate change component r1 around t1 normal to the current beam cross-section, and curvature change or bending strain components r2 and r3 around t2 and t3 , respectively. Once the reduced strain vectors are determined, the strain vector
r

g0

at any material point

(1 , 2 , 3 ) R3 for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) on the current beam cross-section can be determined according to Eq. (83) for
r

and Eq. (78) for the initial curvature

correction term g0 = 1 + 3 02 2 03 . Other strain measures can also be obtained through their relations with
r

g0

, e.g. Eq. (114) for Green strains and Eq. (116) for Eulerian strains.

41

Chapter 5: Balance equations and principle of virtual work

5.1 Reduced balance equations The Lagrangian dierential equations of motion referred to the reference beam conguration are, see e.g. Simo (1985) and Ogden (1997), Div( 0 )t + 0 b = 0 ; 0 Ft = Fr ( 0 )t r (120)

for a material point = (1 , 2 , 3 ) R3 of the continuum body of the current conguration, 2 where = is the acceleration for translation; 0 = 0 (1 , 2 , 3 ) is the mass density per t2 unit volume of the reference conguration of the continuum body, dependent on the choice of the reference conguration; b = b(1 , 2 , 3 ) is the body force per unit mass of the continuum, which is independent of any reference conguration. Simo (1985) started with Eq. (120a) directly on the initial beam conguration. Because the initial beam conguration is also the straight reference beam conguration, he easily obtained, using the divergence theorem for a xed orthonormal frame and Eq. (120b), the reduced balance equations for uniform straight beams with the line of geometry centroids of the beam cross-sections as the beam reference curve, which may also be a good approximation for very slender curved beams. Equivalently, Iura and Atluri (1989, 1988) started with the general principle of virtual work for the reduced balance equations of the initially curved/twisted beams. Now we examine the case of an initially curved/twisted beam with (slowly) varying crosssection from the basic Lagrangian equations of motion using orthonormal frames only. In this case, it is not convenient to work directly on the initially curved/twisted beam because the divergence term in Eq. (120a) is inconvenient to expand in the varying local frame t0i along the midcurve, though taking directional derivatives may be an alternative choice. Therefore, we may also work on the straight reference beam conguration to obtain the equations of motion of the current conguration for the initially curved/twisted beam, but start with the 42

integral counterpart of Eq. (120) (according to the divergence theorem, see e.g. Ogden, 1997: 153): 00 00 dS00 + 00 bd1 d2 d3 = 00 d1 d2 d3 (121)

S00

V00

V00

for the linear momentum balance condition, and ( v) ( 00 00 )dS00 + 00 ( v) bd1 d2 d3 = 00 ( v) d1 d2 d3 (122) for the angular momentum balance condition, where S00 is the arbitrarily chosen surface domain, 00 the outward unit vector of the dierential surface dS00 , V00 the corresponding volume domain surrounded by S00 , 00 = g0 0 the mass density because of the mass conservation condition, and 00 the rst Piola-Kirchho stress tensor as given in Eq. (92); and v R3 is an arbitrarily spatially xed (position) vector. Applying Eqs. (121) and (122) to a parallel cut slice through the straight reference beam with the dierential length d1 and normal to the straight reference beam midcurve (on the E1 axis in fact) for the surface 9 and volume integration domains, and then using variable and domain changes, we can obtain the reduced balance equations of the moving beam referred
9

S00

V00

V00

We separate dS00 into the lateral surface dS00L and cut surface S00N = S00N S00N + for the straight

reference beam conguration, and similarly, dS0 into dS0L and S0N = S0N S0N + for the corresponding tapered cut slice of the curved reference beam conguration. See the Fig. 2 for the dS00 :

dS00L

S00Nd1

S00N+

Fig. 2

A dierential slice for the straight reference conguration

43

to the (initially unstrained/unstressed) curved (and twisted) reference beam conguration: n,1 + N = A0 + w S 0 + w [w S 0 ] for the translational motion (linear momentum balance condition), and m,1 + ,1 n + M = S 0 + I 0 w + w [I 0 w] (124) (123)

for rotation (angular momentum balance condition). In the above equations in the spatial forms, the stress resultant n and couple m have been given in Eqs. (94) and (95), re2 spectively; = t R3 , the acceleration vector of the current beam reference curve for 2 translation; w = t =
t w t

R3 the rotational speed vector of the current beam cross-

section; w = t t R3 the acceleration vector of the current beam cross-section for rotation; the (scalar) mass density A0 per unit arclength of the curved reference beam midcurve is A0 = g0 0 d2 d3 (125)

the rst mass moment density S 0 (vector) per unit arclength of the curved reference beam midcurve: S 0 = S0 3 = S0 2 = g0 0 (2 t2 + 3 t3 )d2 d3 = S0 3 t2 + S0 2 t3 g0 0 2 d2 d3 g0 0 3 d3 d3 (126)

B B B

and the rotational mass or second mass moment density I 0 (second-order tensor) per unit arclength of the curved reference beam midcurve: I 0 = = g0 0 [(2 t2 + 3 t3 ) ]d2 d3 g0 0 [( 2 t2 + 3 t3 )2 I3 (2 t2 + 3 t3 ) (2 t2 + 3 t3 )]d2 d3
2

B B

= I0 ij ti tj I0 11 = I0 22 + I0 33 I0 12 = I0 13 = I0 21 = I0 31 = 0 44

 I0 22 = I0 33 =

B B

g0 0 (3 )2 d2 d3 g0 0 (2 )2 d2 d3
B

I0 23 = I0 32 =

g0 0 2 3 d2 d3

(127)

while the reduced external force density per unit arclength of the curved reference beam midcurve is N = = = = 00 ( 00 dS00L ) + 00 bd2 d3
B

S00L S0L S0L CB

(g0 0 Ft )( 0

1 t F 0 dS0L ) + g0 0
B

g0 0 bd2 d3

0 0 dS0L +

g0 0 bd2 d3
B

g0 0 0j dCB + 0 0CB j

g0 0 bd2 d3

(128)

and the reduced external moment density per unit arclength of the undeformed beam reference curve is M = = = ( ) [ 00 ( 00 dS00L )] +
B

S00L S0L CB

00 ( ) bd2 d3

( ) ( 0 0 dS0L ) +

g0 0 ( ) bd2 d3
B

g0 ( ) ( 0 0j )dCB + j 0 0CB

g0 0 ( ) bd2 d3

(129)

Note that the above expressions for N and M include the load boundary conditions for the lateral surface tractions with outward unit vector 0 = 0j t0j of the curved reference beam conguration; dCB is the dierential element of the contour line CB of the cross-section domain B and 0CB = 2CB t02 + 3CB t03 the unit outward normal vector of CB in the cross section plane of the curved reference beam conguration; N and M also include the body forces. For an untwisted straight beam of homogeneous material (0 = constant) with no initial elongation of the beam midcurve (g0 = 1), the rst mass moment density S 0 given in Eq. (126) vanishes if the beam reference curve is chosen as the geometry centroid line of the beam cross-section. In this case, the underlined terms in Eqs. (123) and (124) vanish, and the 45

balance equations reduce to the original forms given by Simo (1985) and Simo and Vu-Quoc (1986). In addition, if the beam is also uniform, Eqs. (128) and (129) will also reduce to those given by Simo (1985). For an initially curved beam, g0 = 1, and if the beam reference curve is chosen as the geometry centroid line, S 0 does not vanish in general though its entries are small for slender beams. On the other hand, if one chooses the mass centroid line as the beam reference curve, S 0 also vanishes. However, as will be seen later for linear (linearized) constitutive relations, the elasticity centroid line is not the mass centroid line, nor the geometry centroid line all because of the initial curvature correction term g0 = 1. On the other hand, in contrast to Iura and Atluri (1988), we write the inertial terms or linear and angular momentum rates on the right-hand sides of Eqs. (123) and (124), respectively, in their nal forms to avoid users wrongly introducing or ignoring some terms, as we have addressed for Eq (122) that the reference position vector v for angular momentum is spatially xed and independent of time. The rest, for example, the formulae for inertia constants, are the same as in Iura and Atluri (1988). The material forms of the reduced equations of motion may also be given. However, it seems that the spatial forms are more natural for dynamic problems, which is in contrast to the cases of strains and stresses as well as the constitutive relations where the material forms are more natural. 5.2 Reduced virtual work equations For numerical implementations using the curved beam theory, e.g. displacement-based nite element methods, it is more convenient to use the principle of virtual work (weak formulation) to build equilibrium equations. In fact, Iura and Atluri (1989, also see the references therein) started with the general principle of virtual work to obtain the balance equations in the static case for the curved beam theory. On the other hand, the virtual work equation may also be

46

obtained from the reduced balance equations Eq. (123) and (124) (see e.g. Ibrahimbegovi, c 1995; and the refences therein). For completeness, we may repeat a similar procedure to obtain the proper virtual work equation along with the end boundary conditions. Using the static method (see e.g. Iura and Atluri, 1989; and the references therein), we may let N M = N (A0 + w S 0 + w [w S 0 ]) = M (S 0 + I 0 w + w [I 0 w]) (130)

be the generalized external force and moment densities to include the inertial forces. Taking the dot product with Eq. (123) using an arbitrary but kinematically admissible variation (virtual displacement), a = a u R3
10

, of the position vector , and taking

the dot product with Eq. (124) using an arbitrary but kinematically admissible incremental rotation a w = a wr R3 associated with the skew symmetric tensor a w = a t = a r t so(3) (virtual incremental rotation), and summing up the two scalar virtual work r equations for the tapered dierential slice of the beam, and then taking integration over the whole beam, one may obtain that, after rearranging the terms, (also see e.g. Simo, 1985) [a (n,1 + N ) + a w (m,1 + ,1 n + M )]d1 = 0 (131)

L0

or [a n,1 + a w m,1 + a w (,1 n)]d1 + (a N + a w M )d1 = 0 (132) For the above equation, taking integration by parts for the n,1 and m,1 terms, and noticing a w (,1 n) = (a w ,1 ) n, one may easily obtain that
L0

L0

L0

[(a ,1 a w ,1 ) n + (a w),1 m]d1

10

a denotes the variation of the deformation of the current beam conguration; note that the initial beam

conguration is spatially xed and the variation of any objects associated with the initial beam conguration vanishes, e.g., a t0i = 0, a 0 = 0 and a w0 = 0

47

[(a n)|L0 + (a w m)|L0 ] 0 0 = 0

L0

[a N + a w M ]d1 (133)

It follows that the virtual work equation for the beam in the current conguration can be written as (also see e.g. Simo and Vu-Quoc, 1986; Ibrahimbegovi, 1995) c a Wint a Wext = 0 where a Wext is the external virtual work a Wext =
11

(134)

along with the end boundary conditions, (135)

12

L0

(a N + a w M )d1 + (a n)|L0 + (a w m)|L0 0 0

a Wint is the internal virtual work, the variational form of the reduced internal power given in Eq. (119): a Wint = (a r n + a r m)d1 = (a r n + a r m)d1

as

(136)

L0

L0

in which, noticing the corotated variation notation a , the variational form of the co-rotated derivatives as summarized in Section 3.5, a r = a ,1 a w ,1 = a ,1 = a ,1 1 = a (,1 t1 ) t a r = t a r a r = (a w),1 = (a wr ),1 a r = t a r (137)

To this end, it should be mentioned that Ibrahimbegovi (1995) used the reduced strain c vector (co-rotated) variations to obtain the reduced strain measures for the curved beam
11

In the static method, the external virtual work includes the virtual kinetic-energy-lost because the

generalized external loads include the inertia terms as in Eq. (130). We may separate the virtual kinetic energy from the external virtual work in Eq. (134) if necessary for the dynamic problems. 12 In the more general term, the power (energy) balance equation in its incremental or variational form becomes the virtual work equation while the internal power term becomes the internal virtual work. Therefore, the time derivative terms in Section 5.4 may be replaced by the corresponding variational terms with the operator a .

48

theory as demonstrated for r in Eq. (137) with the consideration of the vanishing ,1 t1 as the pure rigid-body motion. In order to use the solution procedures of Newton type, one needs the linearized equilibrium or state equation, which can be achieved through the linearization of the principle of virtual work in its continuum form. The linearized virtual work equation can be written as13 (also see e.g. Ibrahimbegovi, 1995) c a Wint a Wext + a a Wint a a Wext + a Wint a Wext = 0
13

(138)

Note again that the symbol a () denotes the linearized increment of the quantity () and a () the ad-

missible variation of the quantity () at the same point of the beam midcurve (the same arc-length coordinate 1 ), due to deformation only while both the loading factor and initial conguration of the structure remain constant; the subscript a is used to identify deformation. Also, () denotes the linearized increment of the quantity () due to a change in the loading factor while both the initial and deformed congurations of the structure remain constant. For dynamic problems, the loading factor may be replaced by the time parameter t.

49

Chapter 6: Constitutive relations

For the nite strain beam theories, a hyperelastic, isotropic and homogeneous material was usually assumed (for straight beams, see e.g. Simo and Vu-Quoc, 1986; etc.; for curved beams, see e.g. Iura and Atluri, 1989; etc.). Therefore, the reduced constitutive relations are very simple. In fact, it is easy to extend the constitutive relations to more general cases (see e.g. Simo, 1985). In what follows, we consider the general linear and nonlinear constitutive equations in more explicit forms for the convenience of performing computations. 6.1 Linear constitutive relations Let us rst assume that the beam material is hyperelastic but not necessarily isotropic nor necessarily homogeneous. The linearity is kept between the rst Piola-Kirchho stress vector and its conjugate strain vector at any material point (1 , 2 , 3 ) R3 for 1 s0 [0, L0 ] and (2 , 3 ) B(1 ) on the current beam cross-section and given in the component forms as rj 1i = Cij ; 0 g0 Cij = Cij ; 0 = (1 , 2 , 3 ) (139)

where Cij = Cji are the general material elasticity constants for a given material point but 0 0 may vary over dierent material points; Cij = Cji are arbitrarily chosen reference material constants and do not vary over dierent material points; is the corresponding scalar 0 between Cij and Cij , and may vary over material points according to Cij . Then, the linear constitutive relation for a material point on the current beam cross-section may be described in terms of vectors and tensors in the spatial and material form as 0 = 1 1 1 1 C r ; 0 = Cr g0 g0 C = Cij ti tj ; C = Cij Ei Ej

(140)

Substituting Eq. (139) into the formulae for the components ni and mi of the stress resultant (n) and couple (m) vectors in Eqs. (94) and (95), respectively, also using the formulae for 50

the components rj of the strain vector

without the initial curvature correction term in

Eq. (83), we may nally obtain the general reduced linear constitutive relations for the curved/twisted beam: n = Cnn r + Cnm r m = Cmn r + Cmm r n = Cnn r + Cnm r m = Cmn r + Cmm r

Cpq = Cpqij ti tj Cpq = Cpqij Ei Ej Cpq = Cpq t for p, q = m, n

0 Cnnij = Cij A 0 0 Cnmi1 = Ci3 S3 Ci2 S2 0 0 Cmn1j = C3j S3 C2j S2 0 Cnmi2 = Ci1 S2 0 Cmn2j = C1j S2 0 Cnmi3 = Ci1 S3 0 Cmn3j = C1j S3 0 0 0 0 Cmm11 = C22 I22 + C33 I33 (C23 I23 + C32 I32 ) 0 Cmm22 = C11 I22 0 Cmm33 = C11 I33 0 0 Cmm12 = C31 I32 C21 I22 0 0 Cmm21 = C13 I32 C12 I22 51

 0 0 Cmm13 = C21 I23 C31 I33 0 0 Cmm31 = C12 I23 C13 I33 0 Cmm23 = C11 I23 0 Cmm32 = C11 I32

A = S2 = S3 = I22 = I33 = I23 =

d2 d3 B g0 3 d2 d3 B g0 2 d2 d3 B g0 (3 )2 d2 d3 B g0 (2 )2 d2 d3 B g0 I32 = 2 3 d2 d3 B g0

(141)

Note that the reduced elasticity constants have an overall symmetry for any hyperelastic material because of Cij = Cji . For other materials, the symmetry may not hold because of non-existence of the strain energy functional. In general, the couplings exist, such as stretch-bending coupling, stretch-torsion coupling and torsion-bending coupling, etc. On the other hand, we may align the beam reference curve so that S2 = S3 = 0. This means that the cross-section elasticity centroid line is chosen as the beam reference curve. However, the elasticity centroid line does not coincide with the mass centroid line for a general curved beam even though the beam material is homogeneous because the initial curvature correction term g0 appears as the numerator in the integrals for the inertia constants while it appears as the denominator for the elasticity constants. The simplest case of the cross-section elasticity constants is when the beam material is isotropic and homogeneous, 11 0 = E r1 g0

52

12 0 13 0

= = i.e.

r2 g0 r3 G g0 G

C11 C22 Cij

= = = and

E C33 = G 0, otherwise

(142)

where E is the Youngs modulus and G the shearing modulus in the conventional sense. Then, the cross-section elasticity constants become EA [Cnn ]ti tj = [Cnn ]Ei Ej = 0 0
GS2

0 Gks A 0

0 0

Gks A E S2 E S3 0 0 0 E I22 0 0 0

[Cnm ]ti tj = [Cnm ]Ei Ej =

G S3 Gkt I11 [Cmm ]ti tj = [Cmm ]Ei Ej =

0 0

E I23

[Cmn ]ti tj = [Cmn ]Ei Ej =

0 E S2 E S3

E I32 E I33 GS2 GS3 0 0

0

A = S2 = S3 = I22 =

1 d2 d3 g0 1 3 d2 d3 g0 1 2 d2 d3 g0 1 (3 )2 d2 d3 g0 53

 I33 = I11

1 (2 )2 d2 d3 B g0 22 + I33 = I
B

I23 = I32 =

1 2 3 d2 d3 g0
1 g0

(143) by g0 , and ks

which would be the same as given by Iura and Atluri (1989) if we replaced and kt are the correction factors for shearing and torsion, respectively.

For slender curved beams or straight beams, we may let g0 = 1. If the beams are further assumed to be built of isotropic, homogeneous and linear elastic material, we may take the beam cross-section geometry centroid line as the beam reference curve and align t2 and t3 to coincide with the cross-section principal axes. Then, [Cnn ] and [Cmm ] become diagonal, and Cnm and Cmn vanish. The simple constitutive form, used by Simo (1985) and Simo and Vu-Quoc (1986), Simo et al (1995), etc., is recovered. 6.2 Material nonlinearity and its linearization For nite strain problems, the linearity for the constitutive relations may not be applicable. The problems also involve material nonlinearities. Assume that the nonlinear stress and strain relations exist, which are, for example, time-and-rate independent, 0 1i = 1i ( 0 then, we may obtain the tangential coecients as Cij = g0 (1i ),rj 0 which are dependent on the strain state
rj g0

rj ) g0

(144)

(145)

and may vary over dierent material points.

In this case, the linearized constitutive relations may be used for the solution procedure of Newton type, i.e., 1 a 0 = 1 a 0 1 C a r g0 1 Ca r = g0 54

 1 1 a 0 = a 0 a
r

= a r

(146)

in the three-dimensional beam level, and (also see e.g. Simo, 1985) a n = Cnn a r + Cnm a r a m = Cmn a r + Cmm a r a n = Cnn a r + Cnm a r a m = Cmn a r + Cmm a r a n = a n = a n a w n a m = a m = a m a w m a r = a r = a ,1 a w ,1 a r = a r = (a w),1 (147)

in the one-dimensional beam level, where a () is the co-rotated derivative in the linearized incremental form (see Section 4.5). C (C) and Cpg (Cpg ) will be understood as the tangential material coecient tensors, dependent on deformation. The rest is the same as the linear case. However, attention should be paid that the stresses in the three-dimensional level should be updated according to the nonlinear relations between the stresses and strains. Then, Eqs. (94) and (95) should be used to update the stress resultants and couples. Note that Eq. (147) is used to linearize the internal virtual work in Eq. (138) for both linear and nonlinear constitutive relations. For dynamic problems, the (co-rotated) linearized increment operators in Eq. (147) may be replaced by the corresponding (co-rotated) time rate operators.

55

Chapter 7: A general formulation of geometrically exact nonlinear curved beam elements

7.1 External loads External loads applied to a structure are very complex in practice. The interaction between the structure and its environment must be considered. In the present study, only simplied distributed loads are considered in the formulation, while the concentrated loads can be treated with no additional eort but will not appear in the formulation. Here, three types of simplied distributed loads are considered to model self-weight, snow, and pressure loads, each of which is given in the form of some load density. The relevant denitions are in Table 2 and the descriptions following it. Table 2 Applied external load densities Load type Force density Moment density

I: Self-weight

N g (1 ) = Ngj (1 )ej

Mg (1 ) = Mgj (1 )ej

II: Snow

N d (sd ) = Ndj (sd )ej

Md (sd ) = Mdj (sd )ej

III: Pressure

N p (1 ) = Npj (1 )tj

Mp (1 ) = Mpj (1 )tj

Note that all the components of the applied load densities in Table 2 are constant for given 1 in load types I and III and sd in load type II. Type I: The applied load density is given by per unit unstressed arc-length of the beam 56

midcurve and referred to the global xed frame ei , invariant with respect to the deformation of the beam. Self-weight may be dened by this type of load. Therefore, this type of load may be called self-weight type of load. The dierential force, dfg , and moment, dmg , exerted on the arc-element d1 are calculated as follows: dfg = N g (1 )d1 dmg = Mg (1 )d1 (148) (149)

where is the proportional loading factor. This type of load is deformation invariant and usually conservative. Type II: The applied load density is given as constant in space in the sense that the load acting on unit projection length dsd of the deformed arc-element ds corresponding to the undeformed arc-element d1 at a material point 1 on the midcurve onto any plane with N normal dN = ||N d || R3 is constant, given by d

dfd =

N d dsd d

(150)

where N d (therefore, dN ) is constant with respect to both space and the beam itself; however, dsd depends on the deformation and motion of the beam. dsd relates the deformation and undeformed arc-element d1 as well as the direction of N d by dsd = [(dN ,s ) dN ] ,1 d1 = [(dN )2 ,s ] ,1 d1 (151) Assuming that the extension or elongation of the beam midcurve is small and can be ignored, i.e. ds = d1 , the above equation for dsd can be simplied as dsd = [(dN )2 ,1 ] ,1 d1 57 (152)

Therefore, Eq. (150) can be rewritten as

dfd =

N d {[(dN )2 ,1 ] ,1 d1 }d
0

= Nd

{[(dN )2 ,1 ] ,1 }dd1 (153)

= cN Nd d1 where cN = 1
0

[(dN )2 ,1 ] ,1 d

(154)

Similarly, we may dene the dierential moment dmd , exerted on the arc-element d1 of the beam midcurve, as dmd = cM Md d1 where cM = dM 1 2 [(dM ) ,1 ] ,1 d 0 Md = R3 ||Md || (156) Note that both dfd and dmd are dependent on deformation. The force density N d may be a good approximation for snow loads acting on a uniform beam when the loading process of snowing is addressed and the snow itself may cause relatively large displacements/rotations if the slipping of snow is not considered. Therefore, when large displacements/rotations are considered, the snow loads can be non-conservative. On the other hand, when displacements/rotations due to snow loading are small and other loads dominate, all the snow can be considered to add onto the structure before deformation and then move along with the structure if the slipping of snow is not considered. In this case, the coecients for the dierential force dfd , and moment, dmd , exerted on the arc-element d1 may be calculated as cN 0 = [(dN )2 0,1 ] 0,1 58 (157) (155)

 cM 0 = [(dM )2 0,1 ] 0,1 which are really independent of deformation and may be conservative.

(158)

Type III: The applied load density is given by per unit unstressed arc-length of the beam midcurve referred to the local moving frame ti , invariant with respect to the deformation of the beam. Pressure or hydrostatic (wind) loads belong to this type if both the shear deformation and cross-section variation along the beam axis are small and may be ignored. The dierential force, dfp , and moment, dmp , exerted on the arc-length d1 are calculated as follows: dfp = N p d1 = N p d1 dmp = Mp d1 = Mp d1 (159) (160)

where N p = Npj ej and Mp = Mpj ej are in the material form. This type of load depends on rotational displacements and therefore is non-conservative. See Fig. 3 for the demonstration of the three types of loads.
d fg
ocess

d fd d md ds
dsd

d mg ds

d fp d mp ds d1

Defo

rmat ion p r

d1

d1

d fg = Ng ds d mg = Mg ds Type I

d fd
d

= Nd dsd = Md dsd

d fp = Np ds d mp = Mp ds Type III

d md
d

Type II

Fig. 3 Now dene

Dierent load models considered

N = [N g + cN N d + N p + ]

(161)

59

as the force density, and M = [Mg + cM Md + Mp + ] as the moment density along the beam midcurve at current loading in general. 7.2 Parametrization of nite rotations In the previous context, the so-called rotation tensor r SO(3) has been frequently encountered, e.g., Eqs. (12) and (13), Table 1, and wherever the orthogonal tensor SO(3) or the base vectors ti of the local moving frame are involved, which is in contrast to small rotation problems. The proper orthogonal rotation tensor r consists of nine parameters, of which only three are independent because of r t = I3 , the orthogonality condition. r Therefore, it seems necessary to parametrize the rotation tensor using three independent parameters (see e.g. Iura and Atluri, 1989; etc.), which represent the proper three degrees of freedom of the rotation, so that the three independent rotation parameters and the associated rotation tensor as well as their linearized increments and admissible variations are well dened (one-to-one mapping) and updated, which is the key to represent in detail the virtual work equation (equilibrium equation) and its linearization (to obtain the tangential stiness matrix for the solution procedures of Newton type). In what follows, therefore, a brief summary of the parametrizations of nite rotations is given and discussed. In the available literature on nite rotations, the dominant methods to parametrize the rotation tensor r are to use Euler angles (the three angles rotated in a certain sequence about three reference axes attached to the rotated rigid body), the rotation vectors (to be dened soon), as well as quaternion parameters (e.g. see Argyris, 1982; Simo and Vu-Quoc, 1986; Cardona and Geradin, 1988; Ibrahimbegovi et al, 1995; Betsch et al, 1998; and references c therein), of which there are total parametrizations and incremental parametrizations as well as the combinations, leading to multiplicative or additive updating procedures and symmetric or non-symmetric tangential stiness matrices. Eorts have been made to overcome the singularities (rank decient matrices) among those parametrizations. In what follows, 60 (162)

only the rotation vectors (including the incremental rotation vector in the spatial form) and quaternion parametrizations are addressed. Rotation tensor, rotation vector and compound rotation When a rigid body rotates from one orientation to another with a nite angle |||| about a spatially xed axis (e.g. see Argyris, 1982) represented by a unit vector i R3 , the rotation will carry a vector v R3 , which is rigidly attached to the rigid body, to be v R3 on a cone around axis i . This rotation can be described mathematically using an orthogonal rotation tensor r SO(3), via the (exact) linear transformation (mapping) (e.g. see Argyris, 1982) v = r v R3 while the so-called rotation vector is dened as (e.g. see Argyris, 1982) = i R3 (164) (163)

Argyris (1982) used a simple geometric approach to conrm the famous Rodrigues formula: r () = I3 + sin 1 cos 2 () + ( )2 (165)

where so(3) is the skew symmetric tensor associated with the axial vector . i i Using the Taylor series for sin and cos and the relations ( )2k1 = (1)k1 , ( )2k = i (1)k1 ( )2 for k = 1, 2, , and i = / , it can be easily veried that (e.g. see Argyris, i 1982): r () = I3 + ()2 + + = exp[] 1! 2! (166)

Therefore, the rotation tensor r is an exponential mapping (e.g. see Simo and Vu-Quoc, 1986; etc.). According to Eulers theorem, when a rigid body rotates from one orientation to another, which may be the result of a series of rotations with one rotation superposed onto the previous one, the rotation will be equivalent to a single compound rotation (e.g. see Argyris, 1982) 61

around a spatial xed axis. In fact, one can easily conrm, by post-multiplication of both sides of Eq. (165) with and noticing the identity = 0 for a skew symmetric tensor, that the rotation vector is the eigenvector of the rotation tensor r with 1 as the eigenvalue (see e.g. Argyris, 1982): (r I3 ) = 0 (167)

Hence, the rotation tensor r carries its corresponding rotation vector to itself or does not aect its corresponding rotation vector. Therefore, in the above context and later, the rotation vector can and will be understood as a compound rotation, which globally or totally parametrizes the compound rotation tensor r via the Rodrigues formula (Eq. 165). In general, if the compound rotation tensor r can be uniquely represented by some real parameters pk R1 , (k = 1, 2, 3, ), i.e. r = r (p1 , p2 , p3 , ), then this kind of parametrization of the nite rotations can be called total parametrization. Therefore, the use of the rotation vector = j ej to represent the rotation tensor as in Eq. (165) is a total parametrization. The use of the nine components ij of the rotation tensor r = ij ei ej is also a total parametrization. In the case of the use of the compound rotation vector to parametrize the compound rotation tensor r as in Eq. (165), for given , r can be uniquely determined, but for given r , cannot be uniquely determined without knowledge of the number of full cycles because of the periodical nature of the sine and cosine functions. Incremental rotation and multiplicative updating rule Following the previous compound rotation, e.g. parametrized by the rotation vector as r = r () = exp[], if there is an incremental rotation with an angle a w about a spatially xed axis represented by a unit vector iw R3 , the rotation will carry the previously rotated vector (on the rigid body) v to be v R3 on a cone around axis iw . If, similar to dening the rotation vector previously, the incremental rotation is (directly) described by the incremental rotation vector a w dened as a w = a w iw 62 (168)

whose corresponding rotation tensor can be determined using the Rodrigues formula (Eq. 165) or its equivalent exponential mapping form rw = r (a w) = exp[a w] then v = rw v = rw r v = r v R3 where r = rw r (171) (170) (169)

is the updated compound rotation tensor corresponding to the previous compound rotation r followed by the superposed incremental rotation rw that is, the reverse-ordered multiplicative rule applies to obtaining the compound rotation tensor of a series of rotations about a series of spatially xed axes with one rotation following the previous compound one. It should be noted that Eq. (171) is, in fact, independent of any particular parametrizations of the compound rotation tensor r and incremental rotation tensor rw . However, similar to the use of the compound rotation vector to parametrize the compound rotation, the use of the incremental rotation vector a w to parametrize the incremental rotational tensor rw is a more natural and convenient choice. Consider the new compound rotation vector = + a (172)

which parametrizes the new compound rotation tensor r as in Eq. (171), with a the increment of the rotation vector due to the incremental rotation described by the incremental rotation vector a w. Now rewrite Eq. (171) in two parts in the form with a scalar parameter as follows: r = r (a w)r () (173)

63

and r (a w) = r ()r ( + a ) (174)

Taking the derivative of Eq. (173) with respective to and then setting = 0, one can, using the Rodrigues formula Eq. (165), obtain the formula for the linearized increment of the rotation tensor: a r = [r (a w)r ()]=0 = a wr (175)

or the admissible variation of the rotation tensor (see e.g. Simo, 1985; Simo and Vu-Quoc, 1986; Ibrahimbegovi, 1995; etc.): c a r = a wr (176)

One can observe that, because of a w being skew symmetric, the linearized increment or admissible variation of the rotation tensor, a r , is no longer orthogonal. In fact, a w belongs to the tangential space of the rotation tensor r SO(3) (see. e.g. Simo and VuQuoc, 1986; Ibrahimbegovi and Frey, 1995; etc.). Therefore, Eq. (171) is used for updating c the rotation tensor r while Eqs. (175) and (176) are used for the linearized increment and admissible variation operations, respectively, if the incremental rotation is parametrized using the incremental rotation vector. Similar to Eq. (175), taking the derivative of both sides of Eq. (174) with respect to the scalar parameter and setting = 0, one can, repeatedly using the Rodrigues formula (Eq. 165), obtain the linearized relation between the incremental rotation vector, a w, and the increment of the rotation vector, a , as (see e.g. Ibrahimbegovi, 1995; Ibrahimbegovi et c c al, 1995;) a w = Tw a where Tw = Tw () = and the determinant of Tw etc.) det Tw = 2(1 cos ) ( )2 64 (179) sin sin 1 cos + (178) I3 + 2 3 is (e.g. see Ibrahimbegovi, 1995; Ibrahimbegovi et al, 1995; c c (177)

Therefore, a = a w for general 3-D rotations unless a w and are coaxial. Besides, when = 2k for k = 1, 2, 3, , Tw () becomes rank decient, which may lead the tangential stiness tensor to be rank decient too. In this case, iterative solution procedures of Newton type do not converge and give spurious bifurcation information because the determinant of the tangential stiness matrix vanishes. Apart from the above singular points, the following inverse transformation is well dened (e.g. see Ibrahimbegovi, 1995; c Ibrahimbegovi et al, 1995; etc.): c a = Tw a w where Tw = Tw () = T1 = w 1 /2 1 /2 I3 + 1 tan( /2) 2 ( )2 tan( /2) (181) (180)

To avoid the singularity due to the use of the rotation vector to parametrize the rotation tensor, a rescaling remedy is available (e.g. see Ibrahimbegovi et al.) as follows when c > is identied: = 2ni (182)

where n = int[( + )/2], the number of full-cycle rotations. This remedy makes sure [, ], and therefore overcomes the singularity. It should be addressed that either Eq. (177) or (180) is a linearized relation between a and a w in the general 3D case. For a 2D problem, however, one can conrm that a w = a as mentioned previously though Tw = I3 or Tw = I3 in general. Quaternion parametrization and quaternion multiplicative updating rule Another way to avoid a singularity is to use Euler parameters or quaternion parameters to parametrize the rotation tensor (see e.g. Argyris, 1982; Simo and Vu-Quoc, 1986; Ibrahimbegovi, 1995; Betsch et al, 1998; and the relevant references therein). The unit quaternion c parameters {q0 , q} associated with the rotation vector are dened as (see e.g. Argyris,

65

1982; Simo and Vu-Quoc, 1986; Betsch et al, 1998; and the relevant references therein) {q0 , q} = {cos , sin } 2 2 (183)

and similarly, the unit quaternion representing the rotation tensor that corresponds to the incremental rotation, a w, is given as (e.g. see Simo and Vu-Quoc, 1986; Ibrahimbegovi, c 1995) {q0w , qw } = {cos a w a w a w , sin } 2 a w 2 (184)

Note that the normality conditions (q0 )2 + q q = 1 and (q0w )2 + qw qw = 1 have to be satised. In order to establish the relations between the unit quaternions and the associated rotation tensors, one can obtain the alternative form of the Rodrigues formula (Eq. 165) using = i and the standard relation ()2 = ( )2 I3 as follows (see e.g. Betsch et al, 1998): r () = r ( i ) = cos I3 + (1 cos )i i + sin i Using the standard relations cos = 2(cos 2 ) 1 2 2 ) 2 (185)

1 cos = 2(sin and sin = 2 sin

cos 2 2

as well as the denition of the unit quaternion in Eq. (183), it can be easily conrmed that (see e.g. Argyris, 1982; Ibrahimbegovi, 1995; Betsch et al, 1998; and the relevant references c therein) r = r (q0 , q) = [2(q0 )2 1]I3 + 2q0 q + 2q q and therefore, rw = r (q0w , qw ) = [2(q0w )2 1]I3 + 2q0w qw + 2qw qw 66 (187) (186)

Substituting the above two equations into Eq. (171), one can verify that the new compound rotation tensor can be written as r = r (q0 , q ) = r (q0w , qw )r (q0 , q) = [2(q0 )2 1]I3 + 2q0 q + 2q q where q0 = q0 q0w q qw and q = q0 qw + q0w q qqw (190) (189) (188)

are the unit quaternion parameters for the new compound rotation following the quaternion multiplicative rule (see e.g. Argyris, 1982; Ibrahimbegovi, 1995; Betsch et al, 1998; etc.) c A general form of vector-like parametrizations of nite rotations Assume that there exists a vector p = pj ej R3 with three independent components pi , which can uniquely determine the rotation tensor, i.e. r = r (p) One expects to establish the additive update rule p = p + a p (192) (191)

but has to determine the linearized transformation relation between a p and incremental rotation vector a w a p = Tpw (p)a w and its inverse linearized transformation a w = Twp (p)a p (194) (193)

because the incremental rotation vector a w has a clear geometrical and physical meaning. Usually, a p = a w or Twp = I3 , and the relation between p and a w is a nonlinear dierential manifold (Simo and Vu-Quoc, 1986; etc.). Equation (193) serves as projecting 67

a w onto the linear vector space to which p belongs, so that the simple additive update rule Eq. (192) can be used. The parametrizations of the compound rotation using p through Eq. (191) and incremental rotation vector a w using a p through Eq. (194) can be called vector-like parametrizations (see e.g. Ibrahimbegovi et al, 1995) or consistent linear c parametrization (see e.g. Pacoste and Eriksson, 1997). These, in fact, belong to the class of total Lagrangian descriptions (see e.g. Iura and Atluri, 1989). Within the class of vector-like parametrizations, the bending strain vector has an alternative vector form r = r (p) = Twp (p)p,1 for the compound bending strains, and rw = r (pw ) = Twp (pw )pw,1 (196) (195)

for the bending strains due to the incremental rotation vector a w, in which pw is the parameter p due to incremental rotation superposed onto the conguration with vanishing rotation. For example, when p is chosen to be the compound rotation vector , therefore, pw = a w, then r = r () = Tw () ,1 and rw = r (a w) = Tw (a w)a w,1 (198) (197)

which can be found in the papers by Ibrahimbegovi (1995), Ibrahimbegovi et al, 1995, etc. c c 7.3 Admissible variations/linearizations with respect to deformation Variations of translational displacement and position vector It is well known that a position vector and its displacement vector are in the same vector

68

space. The vector additive rule applies to them. From Eq. (11), one can readily obtain that a u = a Variations of orthogonal tensors One may repeat the procedures for Eq. (175) to obtain the admissible variation of the rotation tensor with respect to deformation: a r = [r (a w)r ()]=0 = a wr (200) (199)

Therefore, the variation of the orientation tensor is a = Variations of strains Since the variations of the orthogonal tensors have been determined, the other variations can be obtained using the chain rule of partial derivatives. In what follows, one needs to recall the strain measures given in Table 1. Considering r = ,1 t1 and noticing Eq. (200) and t1 = r t01 as from Eq. (12), one has the usual variation operation: a r = a (,1 t1 ) = a (,1 r t01 ) = a ,1 a (r t01 ) = a ,1 a r t01 = a ,1 a wr t01 = a ,1 a wt1 From r = t r , noticing a t = t a w according to Eq. (201), one obtains that a r = a (t r ) 69 (202) =
=0

(r 0 )

= a wr 0 = a w
=0

(201)

= a t r + t a r = t a w(,1 t1 ) + t (a ,1 a wt1 ) = t [a w(,1 t1 ) + a ,1 a wt1 ] = t (a ,1 a w,1 ) = t (a ,1 + ,1 a w) Therefore, one has the co-rotated variation: a r = a r = a ,1 + ,1 a w (204) (203)

which is equivalent to that given in Eq. (147). Similarly, considering r = r1 = r,1 t r from Table 1 and a t = t a w from Eq. (200), one has r r a r = a (r,1 t ) r = a r,1 t + r,1 a t r r = (a r ),1 t + r,1 (t a w) r r = (a wr ),1 t r,1 t a w r r = (a w,1 r + a wr,1 )t r,1 t a w r r = a w,1 r t + a wr,1 t r,1 t a w r r r = a w,1 + a wr r a w (205)

After some direct operations on the component form, one can show that for any two vectors v1 , v2 R3 and v = v1 v2 , v v1 v2 v2 v1 which leads Eq. (205) to its axial vector form: a r = a w,1 + a w r = a w,1 r a w From r = t r , one obtains that a r = a (t r ) 70 (207) (206)

= a t r + t a r = t a wr + t (a w,1 r a w) = t a w,1 and therefore, one obtains the co-rotated variation: a r = a r = a w,1 which is also equivalent to that given in Eq. (147). If one chooses a vector-like parametrization for nite rotations, it is convenient, noticing the identity a w,1 = a r + r a w from Eq. (207), to rewrite Eqs. (208) and (209) as a r = t (a r + r a w) and a r = a w,1 = a r + r a w respectively. Considering the transformation relation as given in Eq. (194) and the bending strain vector as given in Eq. (195) for a general case of the vector-like parametrizations, Eqs. (203), (204), (210), and (211) can be further written as a r = t (a ,1 + ,1 Twp a p) a r = a ,1 + ,1 Twp a p a r = t [a (Twp p,1 ) + r Twp a p] = t [a Twp p,1 + Twp a p,1 + r Twp a p] = t [(Twp,p a p)p,1 + Twp a p,1 + r Twp a p] = t [(Twp p,1 ),p a p + Twp a p,1 + r Twp a p] = t {[(Twp p,1 ),p + r Twp ]a p + Twp a p,1 } = t [Aa p + Twp a p,1 ] (214) 71 (212) (213) (211) (210) (209) (208)

and a r = a w,1 = a (Twp p,1 ) + Twp a p = Aa p + Twp a p,1 (215) respectively, where the second-order tensor A and the two directional derivatives, A1 and A2 of At along m with respect to p and p,1 , respectively, are calculated as follows: A = R + r Twp (216)

A1 = (At m),p = [Rt m + Tt mTwp p,1 ],p wp = (Rt m),p + (Tt mTwp p,1 ),p + Tt m(Twp p,1 ),p wp wp = H + (m) + Tt mR wp (217)

A2 = (At m),p,1 = (Rt m),p,1 + Tt mTwp wp = (m) + Tt mTwp wp (218) where R = (Twp p,1 ),p H = (Rt m),p (v) = (Tt v),p wp 72 (219) (220) (221)

and (m) = (Rt m),p,1 (= t (m)) (222)

In Eqs. (214) through (222), brief notations have been used: for any second-order tensor T = T(x), the third-order tensor T,x is the derivative of T with respect to the vector x; the second-order tensor (Tv),x = T,x v is the derivative of the vector Tv with respect to the vector x while the vector v remains constant, which is a convenient way to calculate directional derivatives. The closed-form representations for those tensors can be obtained after some elaborate operations for certain vector-like parametrizations. For the vector-like parametrization using the rotation vector in the same spatial form, those tensors have been obtained (see e.g. Ibrahimbegovi et al, 1995). Here, clear denitions of them have been c presented for readers convenience to reproduce the same results. Cardona and Geradin (1988), and then Pacoste and Eriksson (1997), gave dierent versions of similar tensors (the material form for incremental rotation vector was used) but the second-order derivatives of Twp were neglected. Variations of stress resultants/couples Recalling the linearized relations between stress resultants/couples and strain measures in Eq. (147) and considering the variations of strains in Eqs. (202) through (209), and denoting Cn = Cnn and Cm = Cmm , one obtains that a n = Cn a r + Cnm a r = Cn t (a ,1 + ,1 a w) + Cnm t a w,1 (223) and hence (by pull-back/push-forward operations) a n = a t n = a n = Cn t (a ,1 + ,1 a w) + Cnm t a w,1 73

 = Cn (a ,1 + ,1 a w) + Cnm a w,1 (224) for stress resultants, and similarly, a m = Cmn t (a ,1 + ,1 a w) + Cm t a w,1 (225) and hence, a m = Cmn (a ,1 + ,1 a w) + Cm a w,1 (226) for stress couples. For the conventional variation of spatial stress resultant, noticing the other form of the relation for the co-rotated variation a n = a n a wn, one may easily obtain that a n = a n + a wn = a n na w = Cn (a ,1 + ,1 a w) + Cnm a w,1 na w = Cn a ,1 + (Cn ,1 n)a w + Cnm a w,1 (227) Similarly, a m = Cmn a ,1 + (Cmn ,1 m)a w + Cm a w,1 (228) for the conventional variation of stress couple, which will be used in the derivation of Eq. (231). For a vector-like parametrization, one can substitute a w = Twp a p and a w,1 = Aa p + Twp a p,1 into Eqs. (223) through (228) for further representations. 74

7.4 Virtual work equation and its linearization The virtual work equation and its linearization, which produces the equilibrium or state equation, have been introduced in Eqs. (134) through (138). Now one may expand them as follows in the form more convenient to a C 1 continuous nite element formulation, though a C 0 continuous curved beam element will be implemented in the next chapter. Substituting Eqs. (203), (208), (204), and (209) into Eq. (136), one has the internal virtual work as (noticing the equivalence between the material and spatial representations) a Wint = =
L

(a r n + a r m)d1 {[t (a ,1 + ,1 a w)] n + (t a w,1 ) m}d1 [(a ,1 + ,1 a w) ( ) + a w,1 (m)]d1 n [(a ,1 + ,1 a w) n + a w,1 m]d1 (a r n + a r m)d1
t

=
L

=
L

=
L

a ,1 L w a

n d1 ,1 n

(229)

a w,1 which gives rise to internal loading.

The external virtual work, recalling Eqs. (148) through (160), can be further written as a Wext = [a (dfg + dfd + dfp ) + a w (dmg + dmd + dmp )]
L

{a [N g + cN N d + N p ]

+a w [Mg + cM Md + Mp ]}d1
a ,1 L w a

Mg

N g + cN N d + N p 0

d1 p + cM Md + M

a w,1

0 (230) 75

which gives rise to external loading. Using Eqs. (227) and (228) in their linearized incremental forms, one can obtain the linearized increment of the internal virtual work:
a ,1 L w a t

a a Wint =

n d1 a ,1 n

a ,1 L w a

a w,1 a t

m 0

a n d1 a ,1 n ,1 a n

a ,1 L w a

a w,1 a t

a m 0

Cn a ,1 + (Cn ,1 n)a w + Cnm a w,1 d1 a ,1 n ,1 [Cn a ,1 + (Cn ,1 n)a w + Cnm a w,1 ]

a ,1 L w a

a w,1 a t 0
0 ( 0

Cmn a ,1 + (Cmn ,1 m)a w + Cm a w,1 0 0 0 Cn ,1 Cn 0

Cn ,1 ,1 Cn ,1 Cmn ,1
d1

,1 Cnm

Cnm

0 + 0

a w,1 0 0 0

0 0 n

Cmn
a ,1 ) a w

Cm

n ,1 n 0 m

0 0

(231)

a w,1

in which the matrix with subscript M corresponds to the material part of the tangential stiness operator and the matrix with subscript G corresponds to the geometric part of the tangential stiness operator, and both of the matrices are due to the internal loading. The symmetry of the material part of the tangential stiness is obvious, while the geometric part is not symmetric in general. The geometric part of the tangential operator can recover symmetry at the equilibrium state if the external loads are conservative; if external loads 76

are nonconservative, it will be non-symmetric (see e.g. Simo and Vu-Quoc, 1986). One can also obtain the linearized increment of the external virtual work from Eq. (230) as follows:
a ,1 L w a t

a a Wext =

a Mg

N g + cN N d + N p 0

d1 + cM Md + Mp

a ,1 L w a

a w,1 0 t a N d a cN + a N p

0 d1 Md a cM + a Mp

a ,1 L w a

a w,1 0 t a N d (cN a ,1 ) + a wN p
Md (cM

d1 p a ,1 ) + a wM

a ,1 L w a

a w,1 0 t a (N d cN )a ,1 + a wN p
(Md

d1 cM )a ,1 + a wMp

a ,1 L w a

a w,1 0 t a (N d cN )a ,1 N p a w
(Md

d1 cM )a ,1 Mp a w

a ,1 L w a

a w,1 a t 0
0 0

0 N d cN 0 Md cM 0 Np 0 Mp 0 0

d1 0 a w

0 a ,1 0 a w,1

a w,1

77

a ,1 L w a

0 ( 0

N d cN 0 Md cM

0 0

0 0 0 0 0 0

0 + 0

a w,1 0 0 Np 0 0

0 a

0 Mp 0

0 a ,1 ) d1 0 a w

0 0

a w,1 (232)

where the vectors cN and cM are dened as a cN = cN a ,1 and a cM = cM a ,1 respectively, and given as cN = and cM = 2
0

(233)

(234)

(dN )2 ,1 d

(235)

(dM )2 ,1 d

(236)

for the deformation-dependent loading of type II, or cN = 0 and cM = 0 (238) (237)

for the deformation-independent loading of type II. In Eq. (232), the matrix with subscript d corresponds to the part of the tangential stiness matrix operator contributed by the external loading of type II, and the matrix with subscript p by the external loading of type III - the pressure type of loading, both of which are non-symmetric. 78

The above formulation has been performed for parametrizing incremental rotations with the incremental rotation vector a w independent of any particular parametrizations of the compound rotation tensor. One may also formulate with a vector-like parametrization as follows. Referred to Eqs. (213) through (222) , the internal virtual work can be written as a Wint = =
L

{(r n + a r m)d1 {[a ,1 + ,1 Twp a p] n

+[Aa p + Twp a p,1 ] m}d1 =

L

[a ,1 n + (,1 Twp a p) n

+(Aa p) m + (Twp a p,1 ) m]d1 =

L

{a ,1 n a p [Tt ,1 n] wp

+a p [At m] + a p,1 [Tt m]}d1 wp

a ,1 L p a

n d1 t t A m Twp ,1 n

(239)

a p,1

Tt m wp

which gives rise to internal loading, and the external virtual work as a Wext = {a [N g + cN N d + N p ]

+a w [Mg + cM Md + Mp ]}d1 =
L

{a [N g + cN N d + N p ]

+(Twp a p) [Mg + cM Md + Mp ]}d1 =

L

{a [N g + cN N d + N p ]

+a p [Tt Mg + Tt cM Md + Tt Mp ]}d1 wp wp wp

79

a ,1 L p a

t Twp [Mg

N g + cN N d + N p 0

d1 p] + cM Md + M

a p,1

0 (240)

which gives rise to external loading. Using Eqs. (227) and (228) in their linearized incremental forms, and considering a w = Twp a p and a w,1 = Aa p + Twp a p,1 , as well as the derivative tensor notations in Eqs. (216) through (222), one can obtain the linearized increment of the internal virtual work as
a ,1 L p a t

a a Wint =

n d1 a t t A m Twp ,1 n

a ,1 L p a

a p,1 a t

a n d1 a (At m Tt ,1 n) wp

Tt m wp 0

a p,1

a n t a a ,1 d1 a At m + At a m L p a Tt n Tt n a wp ,1 wp a ,1 a p,1 t Twp ,1 a n

a (Tt m) wp 0

a T t m + T t a m wp wp

80

a n t a a ,1 (At m),p a p + (At m),p, a p,1 1 L p a +At m (Tt n) p a ,p a wp ,1 a p,1 Tt n Tt n wp a ,1 wp ,1 a

d1

a n t a a ,1 t A1 a p + A2 a p,1 + A a m d1 L p a ( n) p + Tt n a ,1 ,1 a wp a p,1 t Twp ,1 a n

(Tt m),p a p + Tt a m wp wp 0

a n t a t a ,1 Twp na ,1 d1 L p +[A ( n)] p ,1 a 1 a a p,1 +A2 a p,1 +At a m Tt ,1 a n wp

(m)a p + Tt a m wp 0

(m)a p + Tt a m wp

81

0 t a a ,1 0 ( L p a a p,1 0

Cn

Cn ,1 Twp +Cnm A

Tt ,1 Cn wp +At Cmn

Tt ,1 Cn ,1 Twp wp +At Cm A +At Cmn ,1 Twp Tt ,1 Cnm A wp Tt Cmn ,1 Twp wp +Tt Cm A wp 0

0 A2

Tt Cmn wp 0

Cnm Twp t Twp ,1 Cnm Twp t +A Cm Twp Tt Cm Twp wp

M

0 + 0

Twp n

Tt n wp

Tt ,1 nTwp wp +A1 (,1 n) At mTwp (m) Tt mTwp wp

a ,1 ) a p

d1

a p,1

(241) in which the matrix with subscript M corresponds to the material part of the tangential stiness operator and the matrix with subscript G corresponds to the geometric part of the tangential stiness operator, both of which are due to the internal loading. The symmetry of the material part of the tangential stiness operator is obvious, while the symmetry of the geometric part is not transparent. However, one may reason that the geometric part is always symmetric as long as the beam material works in the elastic range and the second-order derivatives of the transformation tensors are continuous for a vector-like parametrization. Therefore, in Eq. (222), the last term is enforced. 82

For the tangential stiness operator due to the extential loading, one can obtain the linearized increment of the external virtual work, using the results in Eq. (232), as follows:
a ,1 L p a

a a Wext =

a t Twp [Mg

N g + cN N d + N p 0

d1 + cM Md + Mp ]

a p,1

t a 0 a ,1 d1 t L p T [(M c ) a d M a ,1 Mp a w] wp + Tt [M + c M + M ] p a p,1 a wp g M d

0 (N d cN )a ,1 N p a w

t a 0 a ,1 d1 p Twp a p] L p Tt [(M c ) a wp d M a ,1 M p )a p a p,1 +(Mg + cM Md + M

0 (N d cN )a ,1 N p Twp a p

83

0 t a a ,1 (Mg )a p d1 t L p +Twp (Md cM )a ,1 a a p,1 (cM Md )a p t +[Twp Mp Twp (Mp )]a p

(N d cN )a ,1 + N p Twp a p

a ,1 L p a

0 0 0 0 0 0 0

0 ( 0

0 (Mg ) 0 0 0 0 0 0

d

a p,1 0 0

0 0

0 + 0

N d cN

Tt (Md cM ) cM (Md ) 0 wp 0 N p Twp 0 0 0

0 + 0

0 0 0 0

0 a ,1 ) d1 t p Twp (Mp ) 0 a p 0 Twp M

0 0

a p,1 (242)

where the matrices with subscripts g, d, and p correspond to the parts of the tangential stiness operator contributed by the external loadings of type I, II, and III, respectivelyall of which are non-symmetric. Therefore, when the deformation-dependent external loading is involved, the overall tangential stiness matrix is non-symmetric.

84

It is interesting to note that if one denes a p a w and Tww Twp (243)

which means using the incremental rotation vector a w to parametrize incremental rotation, while p has no meaning (noticing a w = Twp a p), a w Tww a w therefore, Twp = Tww = I3 (245) (244)

and let the directional derivative tensors = A = A1 = A2 = 0, then the formulae for a vector-like parametrization of nite rotations as obtained in Eqs. (239) through (242) recover those for the parametrization of the incremental rotation using the incremental vector a w, as given in Eqs. (229) through (232). Now, therefore, one has a unied formulation for both vector-like parametrizations and pure incremental parametrization without considering how the compound rotation is parametrized. Also note that the moment contributed by Mg , as in load type I, is deformation invariant. If the incremental rotation is parametrized using the incremental rotation vector a w, and Mg does not contribute any stiness to the beam, as seen in Eq. (232), then is conservative; however, if a vector-like parametrization of nite rotations is used, Mg does contribute stiness to the beam, as seen in Eq. (242)! It should be addressed that the benets of vector-like parametrizations of nite rotations have been said to be: (i) less secondary storage is needed; (ii) a simple additive update rule is applied; (iii) a symmetric element tangential stiness matrix may be deduced under conservative loading. However, much eort of vector-like parametrizations is on pure mathematical treatments; therefore, clear physical meanings may be lost. If the external loading is non-conservative, the symmetry of tangential stiness is destroyed, and because of some extra calculations needed to obtain the second-order tensors Twp , A, A1 , A2 , and the 85

advantage of vector-like parametrizations of nite rotations is not obvious. Therefore, in general, a vector-like parametrization is not necessarily a better choice. On the other hand, parametrizing the incremental rotation using the incremental rotation vector a w results in the simplest and cleanest formulation. With the help of quaternion representations for compound rotations and some results of vector-like parametrizations, it can still be a good choice, especially when the external loading is not conservative and a precise bifurcation analysis is needed.

86

Chapter 8: A 3-D four-noded curved beam element implementation

8.1 Formulation Assume that the whole curved beam is discretized into Nelm nite elements, and the e-th
e element has Nnode nodes (the total number of nodes in the curved beam is Nnode ). As previ-

ously implied, the non-vectorial parametrization of nite rotations is used (because at least the pressure type of loading will be considered): the incremental rotation is parametrized using the incremental rotation vector a w, while the unit quaternion parameters are used to help update the compound rotations and reduce secondary storage. Here, the conventional Lagrangian interpolation is used for the displaced position vector of the e-th element, e (therefore, e , a e , and a e , etc.), and the incremental rotation 0 vector a w (therefore, a w). That is, e = NI e I a e = NI a e I
e a we = NI a wI

(246) (247) (248)

for example, where NI = NI () is the Lagrangian interpolation function of the node I and e , a e , a e , etc. are the corresponding nodal values. Note that in the present paper, I I I
e the subscripts I, J = 1, 2, , Nnode , and the summation convention holds.

Substituting Eqs. (247) and (248) into Eq. (229), one obtains the internal virtual work in the discretized form
e e a Wint = Pe a e + Me a wI = qe a ae I intI I intI intI

(249)

where Pe = intI Me = intI

Le Le

NI,1 ne d1

(250) (251) (252)

, (NI,1 me NI e 1 ne )d1

qe = intI

Pe intI Me intI

87

and

a ae = I

a e I a we I

(253)

The qe is interpreted as the nodal internal load vector at node I contributed by element intI e, and ae the nodal variational dispacement vector. Similarly, the external virtual work becomes
e e e a Wext = Pe a e + Me a wI = qe ext0I I ext0I ext0I a aI

(254)

where Pe ext0I = = e NI [N e + ce N e + e N p ]d1 N g d NI [N e + ce N e + N e ]d1 N g d p

e

Le Le

(255)

Me ext0I = = and

Le Le

NI [Me + ce Me + e Mp ]d1 M g d NI [Me + ce Me + Me ]d1 M g d p

(256)

qe ext0I =

Pe ext0I Me ext0I

(257)

The qe ext0I is interpreted as the nodal external load vector at node I contributed by the distributed loads acting on element e for unit loading factor. The element tangential stiness matrix is obtained by the linearization of the nodal internal and external loads of the corresponding element, which gives the same result as from Eqs. (231) and (232), the continuum form. By the linearization of the nodal internal load vector (Eq. 252), one obtains that a qe = Kce a ae intI IJ J (258)

The tangential stiness matrix Kce (corresponding to internal loads) consists of two parts: IJ Kce = Kce + Kce IJ mIJ gIJ 88 (259)

The material part is

e L N N e e I J,1 ,1 Cn

NI,1 NJ,1 Ce n

NI,1 NJ Ce e 1 + NI,1 NJ,1 Ce n , nm , n , NI,1 NJ,1 Ce NI NJ e 1 Ce e 1 m , nm +NI,1 NJ Ce e 1 NI NJ,1 e 1 Ce mn ,

d1

Kce = mIJ

+ NI,1 NJ,1 Ce mn

(260)

which is always symmetric. The geometric part is

Kce = gIJ

Le

0 NI NJ,1 n
e

NI,1 NJ ne , NI NJ e 1 ne NI,1 NJ m
e

d1

(261)

which is not necessarily symmetric (see Simo and Vu-Quoc, 1986). By the linearization of the nodal external load vector (Eq. 254), one obtains that a qe = (Kde + Kpe )a ae IJ extI IJ J The tangential stiness Kde turns out to be IJ

Le

(262)

Kde = IJ

NI NJ,1 N e cet d N NI NJ,1 Me cet d M

0 0

d1

(263)

The tangential stiness matrix Kpe (called the pressure stiness matrix) is IJ Kpe IJ 0 = e NI NJ N p e 0 NI NJ Mp
d1

Le

(264)

Note that both Kde and Kpe can be neglected for small displacements/rotations but not for IJ IJ large displacements/rotations, especially when an exact bifurcation analysis is needed. The nodal displacement vector in the incremental form for element e is

a ae 1 a ae 2 . . . a ae node N

a ae =

(265)

The global displacement vector in the incremental form is grouped to be a with consideration of the compatibility conditions at nodes. 89

The nodal internal and external load vectors for element e are

qe int1 qe int2 . . . qe node intN

qe ext01 qe ext02 . . . qe ext0Nnode

qe = int

and qe = ext0

(266)

respectively. The global nodal internal and external load vectors are summed up at the corresponding nodes to be qint and qext0 , respectively. The virtual work equation in the discretized form becomes at R = 0 or, equivalently, one obtains the equilibrium or state equation at nodes: R = qint qext0 = 0 where R is the nodal load vector residual. The tangential stiness matrix of element e is Ke = Kce + Kce + Kpe + Kde T m g where the material part is Kce = [ Kce ] m mIJ (270) (269) (268) (267)

and similarly for the geometric part Kce and the parts contributed by external loads, Kpe g and Kde . Assemblying Ke , one obtains the global tangential stiness matrix KT = R,a . Therefore, T the state equation in linearized incremental form is R + R,a a + R, = 0 (271)

90

or equivalently, KT a = qext0 R (272)

It should be pointed out that the formulae for both the nodal internal load and tangential stiness matrix for an element are, in fact, identical to those by Simo and Vu-Quoc (1986) and Ibrahimbegovi (1995) except the parts related to the dierent treatments of locking c phenomena and external loads in addition to the dierent arrangements of the order of the nodal translational and rotational displacements. Simo and Vu-Quoc (1986) and the present study apply the uniform reduced integration technique to the nodal load vectors and tangential stiness matrix to remove the lockings, while Ibrahimbegovi (1995) uses the c so-called hierarchical terms for the same purpose. The reason for using reduced integration in the present study is based on the paper by Min and Kim (1996), which shows that for
e C 0 -continuous beam elements under (even) the non-uniform mapping, using (Nnode 1)e pointed Gaussian quadrature can produce (enough) (Nnode 1) constraints with no spurious e constraints (energy) when Nnode -noded (curved) beam element is used. This conclusion makes

it possible to elevate the degree of the curved beam element with C 0 -continuity to any order following the same procedure. Therefore, the current formulation of the curved beam element with large displacements/rotations is uniformly valid for any number of nodes in an element, though a four-noded element may be a wise choice for a general spatially curved beam to have a better representation of the initial geometry of the beam and keep a relative small band-width of the tangential stiness matrix. It should be addressed again that the use of the incremental rotational vector a w to parametrize the incremental rotation makes the expressions for the nodal internal/external load vectors and tangential stiness matrix concise and explicit, as opposed to using the increment of the compound rotation vector a in the vector-like parametrization (see e.g. Ibrahimbegovi, 1995). This seems more ecient c and robust for computations and more convenient to programming.

91

8.2 Solution procedures The update procedures for large displacements/rotations are the exact same as suggested by Ibrahimbegovi (1995) for the same class of parametrizations, and therefore are omitted c here. The full Newton-Raphson iterative method (see e.g. Simo and Vu-Quoc, 1986) and Riks arc-length control technique (see Criseld, 1981) are used to nd the nonlinear solutions (deformations, stress resultants/couples, nonlinear bifurcation buckling point, limit point, and post-buckling analyses of arches) subjected to the three types of distributed loads as dened in Table 2. For the details of those, one can refer to relevant references. 8.3 Numerical examples While the formulation in the present study is suitable for any number (three or more) of nodes in an element, only a four-noded curved beam element is implemented in the Fortran code. Several geometrically nonlinear test problems are solved using the four-noded 3-D curved beam element. Some notations are rst given here. E represents Youngs modulus; G shear modulus; Poissons ratio; A and A cross-sectional areas; I area moment of the cross-section corresponding to bending; M bending moment; P point load; p distributed load; U , V , and W displacement components in X, Y , and Z directions, respectively; R radius; L length; and Nelm total number of the four-noded curved beam elements used. In all the test examples, the grid points of the nite element mesh are uniformly distributed along the beam midcurve for convenience. Whether or not a non-uniform mesh would induce signicant mesh-distortion problems will be tested in the 45-degree bend example. Note also that all examples are run with the slender beam assumption, in which the initial curvature correction term has trivial eects on the numerical solutions. Example 1. Unrolling and Re-rolling of a Circular Cantilever Beam This example represents the geometrically-nonlinear analysis of a circular cantilever beam 92

with a bending moment applied at the free end. See Fig. 4 for the relevant data (Ibrahimbegovi, 1995:). c
M = 40 M = 60

E = 1200 =0 square cross-section of unit area R = 5/ M = 20

R M=0

The exact solution can be computed with classical Euler formula = M/EI

Fig. 4

Unrolling and re-rolling a circular cantilever beam

For the chosen data, the values of the free-end bending moments that turn the reference circular shape into the corresponding semi-circle and straight line are M = 10 and M = 20, respectively. If the moment is increased after the straight line is reached, for example, M = 30, M = 40, the semi-circle and complete circle can be formed in the other side. Moreover, another smaller circle is formed at M = 60. Eight elements are used to unroll and re-roll the beam as shown in the gure, which shows that the deformed shapes of the beam t the expected results every well. Dierent numbers of the 4-noded elements were tried. It turns out that four elements are enough to get good results for the unrolling case. In this case, two load steps are needed to unroll the beam to a straight line. Eighteen iterations are needed within each load step. Table 3 shows the displacement comparison.

93

Table 3 Displacement comparison of loaded point of unrolling a beam Model M = 10 Present (4-noded, curved) 4 6 8 Ibrahimbegovi (1995) c (3-noded, curved) Reference solution M = 20 Present (4-noded, curved) 4 6 8 Ibrahimbegovi (1995) c (3-noded, curved) Reference solution 10.000 0.000 10 10.009 10.002 10.001 9.998 0.026 0.007 0.002 0.017 0.000 6.366 10 0.009 0.002 0.001 -0.004 6.374 6.368 6.367 6.364 Nelm Horiz. displ Vert. displ

Example 2. Post-buckling analysis of a clamped-hinged circular arch This well-known example presents the nonlinear analysis of the pre- and post-buckling deformation of a circular arch, hinged at one end and clamped at the other end, under a vertical force applied at the apex. Material and geometric data for the arch are shown in Fig. 5(a).

94

load factor

EA = GA = 100 EI EI = 106 R = 100o Y,V = 145 Buckling Load at Limit Point P = 0 Classical Solution: P =179 P = 359 Pcr = 897 Present Result: Pcr = 897.27 P = 538
R

1 0.9 0.8

Plimit =897.27 limit points V U

X,U

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

P = 718 P = 897

Starting point
-200 -150 -100 -50 0

Clamped (a)

Hinged

U, V
(b)

Fig. 5

Clamped-hinged circular arch

When twelve 4-noded elements are used, the limit point with value 897.30 is found automatically using the full Newton-Raphson iterative method. One load step with 44 iterations is enough to converge to the limit point value if it is prescribed as a load increment. Dierent numbers of the 4-noded elements were tried, and only six elements were enough to obtain good results: see Table 4 ( the classical solution was contributed by DaDeppo and Schmidt, 1975).

Table 4 Buckling load comparison of clamped-hinged circular arch Model Present (4-noded, curved) Nelm 6 12 Limit Buckling Load 897.27 897.30

Ibrahimbegovi (1995) c (3-noded, curved) Simo and Vu-Quoc (1986) (2-noded, straight) Classical solution

20

897.3

40

905.28

897

Fig. 5(a) shows, using ve equal-load increments, the deformation process of the arch (six 95

elements) up to the load when the limit point is reached. Finally, the pre- and post-buckling analysis was done using six elements. The relations between the displacements of the loaded point and the load are shown in Fig. 5(b). Note that the second limit point is reached at Plimit = 73.60 here while it is 77.07 by Simo and Vu-Quoc (1986). A total of 198 load steps are used automatically for Fig. 5(b). Obviously, the arc-length control technique used here is quite robust. Example 3. 45-degree bend This example serves to test the performance of the curved beam element in a 3-D deformation problem. The curved cantilever beam with the reference conguration given as a 45-degree circular segment of radius R = 100 in the horizontal plane is shown in Fig. 6(a), from which one sees that the cantilever beam is undergoing a 3-D deformation.
E = 107 =0 square cross-section of unit area R = 100 P = 600 P = 450
Z

P = 300
X

45o
R P = 150

P=0

Fig. 6(a)

45-degree bend

Dierent numbers of 4-noded elements were tried. The tip displacement components are given in Table 5(a) for comparison. It can be seen that one element is enough to obtain converged results14 .
14

The present results in Table 5(a) were obtained using torsion constant J = 0.140614 , while J = 1/314

was used in by Kapania and Li (1998)

96

Table 5(a) Displacement comparison of loaded point of 45-degree bend Load level Model Present (4-noded, curved) P = 300 Bathe and Bolourchi (1979) (2-noded, straight) Simo and Vu-Quoc (1986) (2-noded, straight) Cardona and Geradin (1988) (2-noded, straight) Present(4-noded) 1 2 P = 600 Ibrahimbegovi (1995) c (3-noded, curved) Bathe and Bolourchi (1988) Simo and Vu-Quoc (1986) (2-noded, straight) Cardona and Geradin (1988) (2-noded, straight) 13.75 -23.67 53.50 8 13.51 -23.48 53.37 13.40 -23.51 53.40 8 8 13.72 -23.76 53.51 13.73 -23.82 53.61 13.73 -23.82 53.61 13.73 -23.81 53.61 7.16 -12.07 40.35 6.8 -11.51 39.50 Nelm 1 2 8 U 7.18 7.18 7.18 6.97 V W

-12.17 40.46 -12.18 40.48 -12.17 40.48 -11.87 40.08

In case of a non-uniform mesh, we tested whether or not the four-noded curved beam has any signicant mesh-distortion problems through the 45-degree bend example under the point load P = 600. A non-uniform mesh was created using geometric progression: the p-th arc-length, numbered from the clamped end, of two adjacent nodes of the mesh is given by sp = rp1 s1 and s1 = L(1 r)/(1 rNnode ). Figure 6(b) shows how the mesh was distorted for the case of one element.

97

The results are given in Table 5(b) with dierent values of ratio r. The results show that the four-noded curved beam element is, for all practical purposes, free from mesh-distortion errors.
Ratio r = 0.25 with grid points 1, 2, 3, and 4 Ratio r = 0.5 Ratio r = 1.0 (uniform) Ratio r = 2.0 Ratio r = 4.0 with grid points i, ii, iii, and iv
X

45
i 1 ii iii 2 3 R

4 iv

Fig. 6(b)

Non-uniform mesh of 45-degree bend

98

Table 5(b) Distortion test of non-uniform meshes Nelm Ratio r 0.25 0.5 0.75 1 1.0 1.5 2.0 4.0 0.25 0.5 0.75 2 1.0 1.5 2.0 4.0 0.25 0.5 0.75 8 1.0 1.5 2.0 4.0 12.97 -23.25 13.69 -23.78 13.73 -23.82 13.71 -23.82 13.55 -23.81 12.60 -23.59 52.72 53.53 53.61 53.59 53.50 53.34 U V W 51.34 52.43 53.16 53.51 53.55 53.38 53.34 51.86 52.69 53.46 53.61 53.59 53.49 53.34

11.14 -21.81 12.69 -23.04 13.44 -23.55 13.72 -23.76 13.67 -23.82 13.41 -23.76 12.54 -23.59 10.87 -22.14 12.94 -23.22 13.65 -23.55 13.73 -23.82 13.70 -23.82 13.53 -23.80 12.59 -23.59

Example 4. Cantilever beam subjected to conservative and non-conservative distributed loads This example is used to illustrate the three types of distributed loads as dened previously (Table 2) through a uniform elastic cantilever beam: Type I self weight type of load, whose 99

density is given per unit arc-length; this type of load is conservative, being identied as Ic. Type II the density is given along the direction d, being coincident with the X axis in this example; for the case that the load is carried by the beam, it is conservative, identied as IIc; for the case that the load is xed in space, it is non-conservative, identied as IIn . Type III pressure type of load, which is non-conservative, identied as IIIn. Only force densities are involved in this example. Because the beam is straight, cases Ic and IIc are identical. See Fig. 7(a) for the illustration. Except for case IIn, the other two have been used by Kondoh and Atluri (1987) and Jiang and Olson (1994). Figure 7(a) also shows the deformed shapes for the four cases of loads at P L2 /EI = 25. Figure 7(b) shows the tip deections corresponding to the four cases of loads. The continuous curves are obtained by the present study. The scattered data in Fig. 7(b) were scanned from the paper by Kondoh and Atluri (1987). Results by Jiang and Olson (1994) are also very close to the current ones, though they are not shown here.
Uniformly distributed load p = P/L

0.1

-0.1

-0.2

III - n

Y
-0.3

II - n
-0.4

V U

I, II - c

-0.5

-0.6

-0.2

-0.1

0.1

0.2

0.3

0.4

0.5

Fig. 7(a)

Deformed shapes of a cantilever beam at P L2 /EI = 25

100

25

II - n

20

Normalized load, PL2/EI

I, II - c
15

I, II - c

III - n
10

III - n

Kondoh and Atluri (1987)

5

V
0 -0.4 -0.2 0 0.2 0.4 0.6 0.8

U
1

(L + U)/L, -V/L

Fig. 7(b)

Tip deections of a cantilever beam

Four 4-noded curved beam elements are used for the data in Fig. 7(a-b), though two such elements are enough to obtain the results close to exact ones. The full Newton-Raphson method is used for all the cases except for case II-n, in which Riks arc-length control technique is used because of the over-hardening phenomena. Note that for a small loading factor (therefore, small displacements/rotations), the deections are almost the same for both the conservative and non-conservative loadings, which can be seen in Fig. 7(b). However, the deections are very dierent from each other for a large loading factor or large displacements/rotations. Therefore, it is necessary to identify nonconservative loads for exible structures so that a better prediction of the structural responses can be achieved. Example 5. Bifurcation buckling of a circular arch with two hinges This example, as shown in Fig. 8, is used to test the performance of the 4-noded curved beam element in a bifurcation buckling problem under pressure load.

101

Uniform Pressure: p

Fig. 8

Circular arch with two hinges

Both linearized and nonlinear analyses were used. The results using 4 elements are shown in Table 6 (the buckling load is given in terms of the factor in the equation pcr = EI/L3 ), as a comparison with the classical solution available in the book by Timoshenko and Gere (1961).

Table 6 Buckling load comparison of a circular arch with two hinges H/L 0.1 0.2 0.3 0.4 0.5 Classical solution Linearized analysis 28.4 39.3 40.9 32.8 24.0 28.4 42.1 40.9 32.8 24.0 Nonlinear analysis 28.2 42.2 40.9 32.8 24.0

It turns out that the results from the linearized analysis are very close to the classical solution. The same holds for the nonlinear analysis except that when the arch tends to be shallower, the dierence increases.

102

Chapter 9: Summary and Conclusions

In the present study, we deal with a truly geometrically exact curved and twisted beam theory without placing emphasis on the computational aspects. Except for the rigid cross-section assumption, no further approximations are made in the formulation. All reference frames are orthonormal and therefore the geometric aspects may become more direct. The formulation starts with the three-dimensional theory and ends up with the one-dimensional beam model. The deformation gradient tensor of the current beam conguration relative to the initially curved/twisted beam conguration is explicitly obtained through the introduction of a straight reference beam. The rst Piola-Kirchho stress and its conjugate strain with an initial curvature correction term at any material point on the current beam cross-section are addressed, in terms of which other stresses and strains can be explicitly expressed. The stress resultant and couple are dened in a classical sense and the reduced strains are obtained from the three-dimensional beam model, which are the same as obtained from the reduced dierential equations of motion. The reduced dierential equations of motion are also re-examined for the initially curved/twisted beams. The corresponding equations of motion in general include additional inertial terms. The linear and linearized non-linear constitutive relations with couplings are considered for the engineering strain and stress conjugate pair in the three-dimensional beam level. The cross-section elasticity constants corresponding to the reduced constitutive relations are obtained with the initial curvature correction term. The geometry, mass and elasticity centroid lines do not coincide with each other in general. 103

It is believed that the initial curvature correction term may be signicant for thick and moderately thick curved beams, especially when long-term dynamic responses are concerned as well as when the strains and stresses in three-dimensional level are needed for strength checking. The current formulation method by the introduction of a straight reference beam can be used to extend the straight beam theory that considers simple shear-torsion warping eect (see e.g. Simo and Vu-Quoc, 1991; etc.) to the curved beam theory. Some eort has been made to discuss nite rotations using rotation vectors and unit quaternions, including total and incremental parametrizations. The so-called vector-like parametrizations, which may lead to an additive update rule and symmetric element tangential stiness matrix, are addressed and generalized in contrast to the non-vectorial parametrizations, which leads to a multiplicative update rule and non-symmetric tangential stiness matrix, when external loading is not a consideration. In terms of a general vector-like parametrization, a unied formulation is given for the linearized virtual work equations, which will lead to the nodal residual vector and tangential stiness matrix in the nite element formulation, which can recover the case of the non-vectorial parametrization if the incremental rotation is parametrized using the incremental rotation vector. It is addressed that when complex external loading is considered, the overall tangential stiness is non-symmetric in general. Therefore, the vector-like parametrization of nite rotations is not necessarily a better choice in general. As an example, taking advantage of the simplicity in formulation and clear classical meanings of rotations and moments, the previously suggested non-vectorial parametrization is applied to implement a four-noded 3-D curved beam element using Reissners slender beam theory, in which the compound rotation is parametrized using the unit quaternion and the incremental rotation is parametrized using the incremental rotation vector. The element has three displacement and three rotational degrees of freedom at each node. Conventional La-

104

grangian interpolation functions are adopted to independently model the displacements and rotations. Reduced integration is used in order to overcome locking problems. The nite element equations are developed for static structural analyses, including deformations, stress resultants/couples, and linearized/nonlinear bifurcation buckling, as well as post-buckling analyses of arches subjected to conservative and non-conservative loads. Illustrative examples show that the consideration of the initial curvature of a curved beam or arch and the elevation of the degree of interpolation polynomials to be cubic can increase the precision and eciency in the analysis of arches. Besides, the element is mesh-distortion-free from the practical view-point if a non-uniform mesh is used, which is especially important for shape optimization of arches.

105

Appendix Fortran programs for a 3-D four-noded curved beam element: ARCHCODE version 1.0 (See the separate PDF le: 2appendix.pdf)

106

References
Argyris, J. 1982: An excursion into large rotations. Computer Methods in Applied Mechanics and Engineering 49, 85-155 Atluri, S.N.; Cazzani, A. 1995: Rotations in computational solid mechanics. Archives Comput. Methods Engrg, 2(1), 49-138 Cardona, A.; Geradin, M. 1988: A beam nite element non-linear theory with nite rotations. International Journal for Numerical Methods in Engineering 26, 2403-2438 Franchi, C.G.; Montelaghi, F. 1996: A weak-weak formulation for large displacements beam statics: a nite volumes approximation. International Journal for Numerical Methods in Engineering 39, 585-604 Ibrahimbegovi, A. 1995: On nite element implementation of geometrically nonlinear c Reissners beam theory: three dimensional curved beam elements. Computer Methods in Applied Mechanics and Engineering 122, 11-26 Ibrahimbegovi, A.; Frey, F. 1994: Stress resultant geometrically nonlinear shell theory c with drilling rotations Part II. Computational aspects. Computer Methods in Applied Mechanics and Engineering 118, 285-308 Ibrahimbegovi, A.; Frey, F.; Koar, I. 1995: Computational aspects of vector-like c z parametrization of three-dimensional nite rotations. International Journal for Numerical Methods in Engineering 38, 3653-3673 Iura, M.; Atluri, S.N. 1988: Dynamic analysis of nitely stretched and rotated threedimensional space-curved beams. Computers & Structures 29, 875-889 Iura, M.; Atluri, S.N. 1989: On a consistent theory, and variational formulation of nitely stretched and rotated 3-D space-curved beams. Computational Mechanics 4, 73-88 Jeleni, G.; Saje, M. 1995: A kinematically exact space nite strain beam model nite c element formulation by generalized virtual work principle, Computer Methods in Applied Mechanics and Engineering 120, 131-161 Kapania, R.K. and Li, J. 1998: A Four-Noded 3-D geometrically nonlinear curved beam element with large displacements/rotations, in Modeling and Simulation Based Engineering, 107

S.N. Atluri and P.E. ODonoghue (ed.), Technology Science Press, 679-684 Ogden, R.W. 1997: Non-linear Elastic Deformations, Dover Publications, Inc., New York Reissner, E. 1972: On one-dimensional nite-strain beam theory: the plane problem. Journal of Applied Mathematics and Physics 23, 795-804 Reissner, E. 1973: On one-dimensional large displacement nite beam theory. Studies in Applied Mechanics 52, 87-95 Reissner, E. 1981: On nite deformations of space curved beams. Journal of Applied Mathematics and Physics 32, 734-744 Saje, M. 1991: Finite element formulation of nite planar deformation of curved elastic beams. Computers & Structures 19, 327-337 Sandhu, J.S.; Stevens, K.A.; Davies, G.A.O. 1990: A 3-D, co-rotational, curved and twisted beam element. Computers & Structures 35, 69-79 Simo, J.C. 1985: A nite strain beam formulation: the three-dimensional dynamics. Part I. Computer Methods in Applied Mechanics and Engineering 49, 55-70 Simo, J.C.; Tarnow, N.; and Doblare, M. 1995: Non-linear dynamics of three-dimensional rods: exact energy and momentum conserving algorithms. International Journal for Numerical Methods in Engineering 38, 1431-1473 Simo, J.C.; Vu-Quoc, L. 1986: A three dimensional nite-strain rod model. Part II: computational aspects, Computer Methods in Applied Mechanics and Engineering 58, 79-116 Simo, J.C.; Vu-Quoc, L. 1991: A geometrically-exact rod model incorporating shear and torsion-warping deformation, International Journal of Solids and Structures 27, 371-393

108

Vita Jing Li was born in Dalian, Liaoning, P.R. China in 1961. He received the B.S. and M.S. degrees in Aircraft Engineering from Northwestern Polytechnical University at Xian, China, in 1984, and 1987, respectively. From 1987 to 1995, he worked as a mechanical/structural engineer in the Department of Structures and Strength at Shenyang Aircraft Research Institute, Shenyang, China. In 1995, he joined in the Department of Aerospace and Ocean Engineering, Virginia Polytechnical Institute and State University. As a graduate research assistant, he is mainly involved in researches in the areas of geometrically nonlinear curved beam theory and its nite element implementation/simulation.

109