ON FINITE ELEMENT NONLINEAR ANALYSIS OF GENERAL SHELL STRUCTURES by SAID BOLOURCH B.Sc., Queen Mary College, London University (1973) S.M., Massachusetts Institute of Technology (1975) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHELOSOPHY at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY (ay 1979) Signature of Author... ... EOE ee Certified by Accepted BY ee Chairman, Department Comjttee on Graduate Students NSSAoMusE rE waitUTE JUL 20 1979ON FINITE ELEMENT NONLINEAR ANALYSIS GF GENERAL SHELL STRUCTURES by SAID BOLOURCHI Submitted to the Department of Mechanical Engineering on May 16, 1979 in partial fulfillment of the require- ments for the Degree of Doctor of Philosophy. ABSTRACT Continuum mechanics-based total Lagrangian and updated Lagran- gian formulations of structural elements are presented for large displace- ment and rotation analysis. It is shown that the two formulations yield identical element matrices, when they are applied to a straight Hermitian- based beam element, but the updated Lagrangian-based beam element is com- putationally more effective. A comparison of the straight beam element with an isoparametric bending element with independent nodal element displacement and rotation degrees-of-freedom shows that the isoparametric displacement/rotation nethod fs the more effective technique for fomulating general structural elements, A displacement-based isoparametric displacement/rotation shel] element is formulated and developed as a variable-number-nodes element. The element is described using linear, quadratic or cubic interpolating displacement fields, and can be a quadrilateral or a triangle. The ele- ment can also be employed as a fully compatible transition element to model shell to solid regions or shell intersections. Shell structures with eccentric and non-eccentric stiffeners can be modeled using the shell element and a compatible bending element. The shell element can be employed in analysis of general shell structures with linear and geometric and material nonlinear behavior. Various sample solutions of linear and highly nonlinear problems are presented to illustrate the effectiveness of the element in analyzing practical engineering problens. Thesis Supervisor: Klaus-Jlirgen Bathe Title: Associate Professor of Mechanical Engineering 2ACKNOWLEDGEMENTS I wish to express my deep appreciation to Professor K. J. Bathe for his supervision and guidance throughout this research. I am also grateful to the other members of my thesis committee: Professor T. H. H. Pian, Professor G, W. Strang and Professor J. H. Williams, Jr. T am grateful to S. Ramaswamy for some fruitful discussions. I also like to thank M. D. Snyder. I give special thanks to Prudence Young for typing this manuscript. The work reported in this report has been supported financi- ally by the ADINA users group. I acknowledge this support. I espectally thank my parents Aghdas and Asghar for their patience and support throughout my life.TABLE OF CONTENTS PAGE ABSTRACT 2 0 et 2 ACKNOWLEDGEMENTS © 0 ee eet 3 TABLE OF CONTENTS 22. - ee ee ee ee te tet 4 LIST OF TABLES . - 7 LIST OF FIGURES 0. ce ee ee eee te 8 NOTATION 0 cc ete uu 1. INTRODUCTION. 2 oe ee 20 2. INCREMENTAL CONTINUUM MECHANICS FORMULATION . ©... + + + = 27 2.1 Updated Lagrangian (U.L.) Formulation... - --- ++ 32 2.2 Total Lagrangian (T.L.) Formulation... .- +--+ +--+ 34 2.3 Finite Element Discretization.. 1... - +e eee 35 3. COMPARISON OF THE T.L. AND U.L. FORMULATION FOR A 37 STRUCTURAL ELEMENT So. eee eee ee eee 3.1 UL. Formulation of Bean Element... 2+ +++ += 39 3.2 T.L. Formulation of Beam Element... . +--+ +++ 41 3.3. Comparison of T.L. and U.L. Formulations... +--+ 43 4. COMPARISON OF HERMITIAN AND LAGRANGIAN BENDING ELEMENTS. . . 46 4.1. U.L. Formulation of Lagrangian Bending Element... - . - 48 4.2 T.L. Formulation of Lagrangian Bending Element... . ~~ 54 4.3. Comparison of Hermitian Beam and Lagrangian Bending Element... eee eee cree ert ctet 555. TABLE OF CONTENTS (Continued) FORMULATIONS OF THE NONLINEAR GENERAL SHELL ELEMENT... .. 5.1 U.L. Formulation of the Shell Element... 2.1... . 5.2 T.L. Formulation of the Shell Element... ......-. 5.3 Constitutive Relations... 1... 2.22 eee eee 5.4 Mass Matrices. 2... eee ee ee ee eee INTERPOLATION FUNCTIONS FOR SHELL ELEMENTS... 1... 6.1 Demonstrative One-Dimensional Element Example... . . 6.2 Variable-Number-Nodes Quadrilateral Shell Element . . . . 6.3 Variable-Number-Nodes Triangular Shell Element... . . 6.4 Variable-Number-Nodes Transition Element... ..... SOME IMPORTANT FEATURES OF THE SHELL ELEMENT... ..... 7.1 Element Description... .-.. 2... +2. ee eee 7.2 Numerical Integration. . 2... eee eee eee 7.3 Modeling of Shel] Connections in Engineering Problems . 7.3.1 Modeling of Shell Intersections. ........-- 7 .2 Modeling of Thin Shell to Solid Transition... . ~ 7.3.3. Modeling of Stiffened Shells... .....2005 7.4 Use of Global or Element Convected Coordinate System . . 7.5 Effects of Element Aspect Ratio... ..-..-.--- 7.6 Important Element Nodal Layouts... ....-.+-+-+- 7.7 Comparison of Element with Elements Based on Plate Theory... ee ee ee eee PAGE 58 62 ae 1 82 86 87 89 94 98 99 99 99 + 104 104 104 106 +108TABLE OF CONTENTS (Continued) 8. SAMPLE ANALYSES 2. 1. ee ee ee ee 8.1 Linear Analysis of Three Contilever...-...--- + 8.2 Linear Analysis of Folded Plates... ...-.--.-- 8.3 Linear Analysis of a Cylindrical Shell... ... 1... 8.4 Elastic-Plastic Dynamic Analysis of a Simply Supported Plate. 2 6. et ee ee ee ee ee 8.5 Large Displacement/Rotation Analysis of a Cantilever . . 8.6 Elastic-Plastic Instability Analysis of a Column... . 8.7 Large Deflection Analysis of a Diamond Structure... . 8.8 Large Displacement Analysis of an Elastic Simply- Supported Plate... 0... eee 8.9 Large Deflection Analysis of a Shallow Cylindrical Shell. ee ee 8.10 Large Deflection Analysis of a Spherical Shell 9. CONCLUSIONS... 1 ee ee ee ee eee bee ee REFERENCES » 2 1 1 ee ee ee ee BIOGRAPHICAL NOTE. 2 2. ee ee ee PAGE 113 13° 17 9 127 127 127 131 139 14. 143 174 179LIST OF TABLES PAGE UPDATED LAGRANGIAN FORMULATION . 147 TOTAL LAGRANGIAN FORMULATION... 1 ee ee ee ee ee 150 FINITE ELEMENT MATRICES IN ALL ANALYSIS... .. 2... 153 MATRICES USED IN STRAIGHT HERMITIAN-BASED BEAM ELEMENT FORMULATION. 2... 2 ee ee ee ee ee 155 MATRICES USED IN UPDATED LAGRANGIAN FORMULATION QF SHELL ELEMENT. 2... ee ee ee ee ee ee 163 MATRICES USED IN TOTAL a FORMULATION OF SHELL ELEMENT... . e See 166 3-D ELASTIC-PLASTIC INCREMENTAL CONSTITUTIVE MATRIX, VON MISES YIELD CONDITION, AND ISO- TROPIC HARDENING 5. 2. ee ee ee ee eee ee eee 170 SHELL ELEMENT INTERPOLATION MATRIX... 2... eee 171 INTERPOLATION FUNCTIONS FOR VARIABLE-NUMBER-NODES SHELL ELEMENT 2 2 1 ee et ee ee 172i FIGURES 21 3.1 3.2 4.1 4.2 5.1 5.2 5.3 5.4 5.5 5.6 5.7 6.1 6.2 6.3 6.4 6,5 LIST OF FIGURES Motion of Body in Cartesian Coordinate System... . Schematic View of the 3/D Beam Element Local Coordinate Axes... ee ee ee ee Motion of 3/D Beam Element and its Local Coordinate Axes Shown in Global Coordinate System . . Variable-Number-Nodes Bending Element. ....... Definition of Two-Node Bending Element ....... Variable-Number-Nodes Quadrilateral Shell Elements Variable-Number-Nodes Triangular Shell Elements . . . Some Derived Elements of Variable-Number-Nodes Transition Elements... 1... ee eee ee ee Definition of a General Shell Element... .. 2... Normal to Shell Mid-Surface.........-205 Definition of Vectors at Nodal Pointk ....... Definition of the shell Element Orthogonal Local Coordinate Axess... - . ee eee ee Schematic Construction of a Variable-Number-Nodes One Dimensional Element interpolation Functions . . . Node Numbering Scheme of a Variable-Number-Nodes , Quadrilateral Shell Element... .......--% Construction of Shell Element Interpolation Functions ©. 1 ee ee ee ee Interpolation Polynomials of Some of the Shell Elements... ... ee rey Node Numbering of a Variable-Number-Nodes Triangular Shell Element... 1... ee eee eee 28 38 40 a7 50 60 61 64 66 67 75FIGURE 71 7.2 7.3 74 8.1 8,2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11 8.12 8.13 8.14 LIST OF FIGURES (Continued) PAGE Triangular Integration Scheme. ....... aoe 102 Modeling of Shell Surfaces Intersections ...... 105 Modeling of Shel1-To-Solid Transitional Regions... 107 Recommended Shell Elements... ....-..--% 110 Effects of Aspect Ratio in Linear Analysis of a Uniform Cantilever... 2... ee ee eee 14 Linear Analysis of a Tapered Cantilever... . . us Linear Analysis of a Stiffened Plate... .... 116 Analysis of Folded Plates... .......0. 18 Pinched Cylindrical Shel]... 2.2... 120 Finite Element Models of a Pinched Cylindrical Shel. eee ee ee ee a2 Numerical Value of We and Uy es 122 Displacements Convergence Study of a Pinched 124 Cylindrical Shell... ee ee Stress Distributions of a Pinched Cylindrical a Shel. Elastic and Elastic-Plastic Dynamic Response of a Simply Supported Square Plate... ...... 128 Large Deflection Response of a Cantilever Subjected to an End Moment... . 1... eee 129 Elastic-Plastic Instability Analysis of a Column . . 130 Load Deflection Curve for a Diamond Structure... . 132 Deformed Shapes of the Diamond Structure for pif rer a Values of the Loading Parameter 187/43 oloiolo 0G Glold G6 Olnidicio'o S000 133FIGURE 8.15 8,16 8.17 8.18 LIST OF FIGURES (Continued) Large Deflection Analysis of a Simply-Supported Square Plate Subjected to Pressure Loading... . . Membrane Stresses of the Simply-Supported Square Plate (Uniform In-Plane Edge Displacement)... . Central Deflection of a Hinged Shallow Cylindri- cal Shell... bee eee wee Shell... ee 140NOTATION All notation is defined in the text when used first. The following is only a list of some frequently used symbols. A bar (-) on a quantity denotes that the quantity is referred or measured in local coordinates. With regard to tensor and vector subscripts and superscripts, the following convention is employed: A left superscript denotes the time of the configuration in which the quantity occurs. A left subscript can have two different meanings. If the quantity considered is a derivative, the left subscript denotes the time of the configuration, in which is measured the coordinate with respect to which is differentiated. Otherwise the left subscript denotes the time of the configuration in which the quantity is measured. Right lower case subscripts denote the components of a tensor or vector. Components are referred to a fixed global Cartesian coordinate sys- tem; 1,J, ... = 1,2,3. Differentiation is denoted by a right lower case subscript following a comma, with the subscript indicating the coordinate with respect to which is differentiated » 1.e.5 Uy 5 = au/2*x,;- A local variable with a right lower case subscript following a comma de- notes the derivative of the local variable with respect to the local = .0- coordinate axis, i.e., 45 denotes gu ,/2 Xj t, that = time t and ttAt, before and after time increment At iWOy ty. trdty ry syt k —tyk —teatyk Es > ey > MANE, Oyk — tyk — thatyk sir Vsi > Voi = denoting "variation in" = volume of body in the configuration at time 0,t,ttat = area of body in the configuration at time 0 = fixed Cartesian coordinate in the con- figuration at time 0,t,ttat = local convected Cartesian coordinate in the configuration at time 0,t,t+it = Cartesian cuordinate of nodal point k in the configuration at time 0,t,ttat = derivative of coordinate in the con~ figuration at tine 0,t with respect * t, 0 to coordinate “xj, x, = natural element coordinates = component of unit vector in the di- rection t of nodal point k at time O,t, teat = Component of unit vector in the di- rection s of nodal point k at time O,t, teat 12tyk ni? OVE > ttt, kk si Vt V Da hy og tg? teat" § = Component of unit normal vector to shell mid-surface at the nodal point k at time 0,t,t+at = component of displacement vector from the initial configuration at time 0 to the configuration at time t, ttat = increment in displacement component, = ttt, t, use uu, > “uy = displacement component of nodal point k in the configuration at time t tk = increment in “u; = increment in unit normal vector ty, at nodal point k, i.e., VK, = ttityk tyk Yai ~ ‘ni = a, tyk k = increment in KS , “VE. = derivative of displacement component ‘in the configuration at time t, teat with respect to coordinate °K; = derivative of displacement increment with respect to coordinate ox, : t teat, J 13t, that e4y° €45 t. tat, oij’ Oij teat, ij orig? taj t, that, ij oj? tg og? tg = component of Almansi strain tensor in the configuration at time t,ttat referred to the configuration at time 0 = component of Green-Lagrange strain tensor in the configuration at time t,ttAt referred to the confjguration at time 0 = component of Green-Lagrange strain tensor in the configuration at time ttAt referred to the configuration at time t (i.2., using displacement increnients from the configuration at time t to time ttat) = component of Green-Lagrange strain increment tensor referred to the configuration at time 0,t = component of total infinitesimal strain tensor at time t, ttt = increment in *e,, = linear part of strain increment ofij ? tj = nonlinear part of strain increment oj? tig 14/ : Sisrs t, te Ossrs * tijrs tet, af ttt, = component of constant constitutive tensor = component of constitutive tensor at time t referred to the configuration at time 0,t = component of tangent constitutive tensor at time t referred to the configuration at time 0,t = specific mass of body in the configu- ration at time 0,t = component of body force vector per unit mass, at time t+At, referred to the configuration at time 0 = component of surface traction vector per unit area, at time t+At, referred to the configuration at time 0 = external virtual work expression corresponding to the configuration at time that defined in Eq. (2.2) = component of stress tensor at time t, calculated assuming infinitesimal displacement conditions < t = increment in oy 15ty | ttt that, ig og? t5ig component of Cauchy stress tensor in the configuration at time t,t+at component of 2nd Piola-Kirchhoff stress tensor in the configuration at time t,ttAt referred to the configu- ration at time 0 = component of 2nd Piola-Kirchhoff stress tensor in the configuration at time ttAt referred to the configuration at time t = component of 2nd Piola-Kirchhoff stress increment at time t = component of deviatoric stresses = finite element interpolation function associated with nodal point k = finite element Hermitian interpolation function in the local direction i associated with the kth degree of freedom of the beam element, and measured in the local configuration at time 0,t, respect- ‘ively. 16Matrices oh > eB c = component of the transformation matrix of the beam local axes at time 0,t with respect to the global axis = component of the transformation matrix of the beam local axes at time t,ttAt with respect to original beam local axes = linear strain-displacement matrix in the configuration at time t referred to the configuration at time 0,t = linear strain-displacement matrix assuming small displacements = nonlinear strain-displacement matrix in the configuration at time t re- ferred to the configuration at time 0,t = stress-strain material property matrix = tangent material property matrix at time t and referred to the configu- ration at time 0,t 7k tt ok? th t, t ofwe > thu iz trate tt ty , taty of tt? teat = vector of nodal point forces at time t = time independent linear elastic, small displacement stiffness matrix = linear strain stiffness matrix in the configuration at time t referred to the configuration at time 0,t = nonlinear strain stiffness matrix in the configuration at time t, referred to the configuration at time 0,t = tangent stiffness matrix at time t = mass matrix = vector of external loads at time t+At = 2nd Piola-Kirchhoff stress matrix and vector in the configuration at time t and referred to the configuration at time 0 = vector of element nodal point dis- Placements at time t,t+at = vector of element nodal point incre- mental displacements at time t and referred to the local axes at time 0, t, teat 18u ty = fe Op Ry tate vector of elenent nodal point incre- mental displacements at time t vector of element assemblage nodal point displacements at time t,ttat vector of elenent assemblage incre- mental nodal point displacements at time t = vector of stresses at time t in materially nonlinear analysis only Cauchy stress matrix and vector in the configuration at time t = deviatoric stress matrix in the configuration at time t = matrix transforming the beam nodal points displacements measured in the beam local axis at time 0,t,ttAt to the global axis = matrix transforming the straight beam nodal points’ displacements measured in the local axes at time t,t+\t to the beam local axes in the configuration at time 0 191, INTRODUCTION A very important element of a structure is in many cases a shell, since it is an efficient load carrying member. Shells are used in various structures such as spacecrafts, missiles, pressure vessels and so on. Thus reliable linear and nonlinear analysis of shell struc- tures are of much importance. For decades, a large amount of research effort has been spent to develop capabilities of general shell analysis. Fung and Sechler [1] have given a comprehensive survey of thin shell structural capabilities. In addition many governing equations of the be~ havior of thin shells have been derived, depending on the specific assump- tions made on the behavior or geometry of the structure. One of these thin shell theories derived for geometrically linear and nonlinear gen- eral shell structural behavior is presented by Reissner [1]. More de~ tailed discussions of shell theory assumptions are given in [2]. The nonlinear partial differential equations governing the be- kavior of shell structures are extremely complex. Analytical closed form solutions to these equations hardly exist except for a few small deflection analyses of simple cases of geometry and loading. Some series solution techniques have been used to predict the approximate solution for simple nonlinear shel! problems. However, it is impossible to ex- tend these methods to the analysis of realistic engineering shell struc- tures with complex geometries (e.g. variable thickness, cut outs, branches, intersections and connection with other types of structural 20members). These difficulties are specially observed due to the complex loading of the engineering shell structures and the high degree of pre- cision which is needed in determining the safety of these structures. The short coming of the classical analytical techniques has forced the designers to search for numerical methods. Advances in numerical methods and the availability of high speed digital computers has created the new field of computational mechanics. In the past two decades a large number of papers have been devoted to the numerical analysis of shell structures. A detatled review of all the Papers on this topic is beyond the scope of this study. The two basic numerical techniques employed in the analyses of plate and shell struc- tures are, the finite element method [3-5], and the finite difference method [6-7]. In this study we concentrate on the finite element method because of its generalities and flexibilities. The finite element method has proved to be a powerful, effec~ tive and reliable approximate technique for solying many solid mechanics problems. Finite element analyses of thin shell structures may be divided ‘into two approaches as follows; firstly, what may be called a classical procedure, and secondly, an assumed displacement isoparametric element formulation, In the first approach a classical concept is employed in which a plate or shell theory is used as the starting point of the finite ele- ment formulation, This plate or shell theory has been developed from the three-dimensional field equations by incorporating various assumptions 21appropriate to the structural behavior. Using variational formulations based on these theories various finite element models have been developed; namely, displacement, hybrid and mixed formulations [8]. Atluri and Pian [9] have given a comprehensive review and systematic classification of finite element formulations for the linear analysis of general shells. In recent years, many elements have been developed based on the classical procedure by employing different plate or shel] assumptions to analyze shell structures. Many researchers have approximated the shell geometry using flat elements, Among the recent papers on this subject, [10-13] deal with linear analysis and [3, 14-16] deal with geometrically nonlinear analysis of shells of large deflections and rotations, using flat elements. Displacement type formulation were used in [11,13,14], and hybrid stress formulations are employed in [12,15,16]. ‘A number of authors haye concentrated on the analysis of shells of revo- lution, formulations for linear analysis are presented in [17,18] while [19] deals with nonlinear analysis of shells of revolution. A general shell theory has been employed in [20,21] to formulate shell elements for analyzing the linear behavior of general shell structures. A mixed iso~ parametric formulation was employed by Noor et al. [22] to study the large deflection behavior of general shells. Boland and Pian [23] used shallow shell theory in formulating shell elements to analyze the elas- tic behavior of general shells. In the second approach, isoparametric elements with indepen- 22dent rotational and displacement degrees of freedom are employed. This procedure was originally introduced by Ahmad et. al. [24] for the linear analysis of moderately thick and thin shells, and has been further de- veloped for the linear analysis of elastic thin shells [25-27]. A simi- lar approach has recently been applied to the geometrically nonlinear analysis of shells by Ranm [28]. Kr&keland [29] has extended this work to analyze also the materially nonlinear behavior of shells. In considering current capabilities for the computer analysis of shell structures, reference may also be made to the comparative study of the linear and nonlinear finite element shell solutions reported by Hartung and Ball [31], in which several specific problems solved by different members of the structures community have been compared. The comparison of the results shows good agreement for linear analysis. How- ever, the nonlinear results (geometric nonlinearity only) shows virtually No agreement and the reason for the differences could not be explained. The conceptual advantage of the displacement/rotation iso- parametric elements is their inherent generality, which is analogous to the generality of the isoparametric elements in the analysis of two- and three dimensional continuum problems [34,35]. In contrast to the classical approach, no specific classical plate or shell theory is em- ployed; instead, the geometry and the displacenent field of the structure is directly discretized and interpolated as in the analysis of continuum problems, This approach is equivalent to using a general shell theory and can be employed efficiently for the analysis of general structural 23configurations by using variable-number-nodes elements [32,33]. The basic ideas for the development of the displacement/ rotation isoparametric elements evolved from the difficulties that are encountered when using the usual displacement isoparametric elements in the analysis of plates and shells. Firstly, computational diffi- culties can result when these isoparametric elements are very thin, be- cause the stiffness coefficients corresponding to the transverse dis- placement degrees of freedom are considerably larger than those corres- ponding to the longitudinal displacements. Secondly, errors are intro- duced in the analysis because erroneous strain energy corresponding to the normal stresses in the thickness direction is included. These two difficulties are overcome in the approach introduced by Ahmad et.al. , because it is assumed that the normal to the shell surface remains straight and does not extend, and the normal stresses in the direction of the shell thickness are ignored in the element formulation. Scope of Present Study The objective in this investigation is to present a general vartable-number-nodes displacenent/rotation isoparametric shell ele- ment for geometric and material nonlinear analysis of general plate and shell structures. The element formulations are based on the con- sistance continuum mechanics principles. We first present the basic continuum mechanics virtual work theory and then define an updated Lagrangian (U.L.) and a total Lagran- gian (T.L.) formulation, Chapter 2, These formulations allow for very large displacements and rotations as well as the material nonlinear 24behavior of the continuum. The linearization of these formulations, and the finite element discretization are also presented in this chapter. In Chapter 3, the T.L. and the U.L. formulations are employed to formulate a straight beam element, in which Hermitian polynomials are employed to interpolate displacements within the element. It is shown in this chapter that both formulations are identical for this example. In Chapter 4, a variable-number-nodes Lagrangian bending ele- ment with independent displacement and rotation is presented. It is shown that it is more effective to employ the Lagrangian bending element than the Hermitian beam element to analyze curved structures. The general variable-nunber-nodes displacement/rotation iso- parametric shell element is presented in Chapter 5. This element can be employed to analyze general shell structures with geometric and material nonlinear behavior. In this chapter, we present the derivation of the element matrices based on the T.L. and the U.L. formulation. We also present the lumped and consistent mass matrices, elastic and elastic- plastic constitutive relations to be employed in the element formulation. In Chapter 6, the Lagrangian interpolation functions of the variable-number-nodes general shell element are presented. The tables of interpolation functions are presented for both the quadrilateral and the triangular elements. The shell element can possess 3 to 16 mid-surface nodes. Each mid-surface node has three displacement degrees of freedom and two rotational degrees-of-freedom. The interpolation function can be linear, quadratic, or cubic polynomial. The quadri- 25lateral element can have 1 or 4 internal nodes. The 9-nodes element (including one internal node), and 16-nodes element (including four in- ternal nodes) represent quadratic and cubic Lagrangian element, respec- tively. In this chapter, the interpolation functions of transition ele- ment are also presented. This element can have mid-surface nodes and top- and bottom- surface nodes. This element can be used for connecting the shell element to solid continuum elements. Each top- or bottom- surface node has three displacement degrees-of-freedom. In Chapter 7, we describe some important aspects pertaining to the implementation and the usage of the element such as; geomerty discription, spatial integration over quadrilateral and triangular space, connectivity of the element with other continuum elements, mode- ling of sharp edges and branches by using constraint equations, and the effect of the element aspect ratio on the response prediction. Some linear and highly nonlinear numerical examples are pre- sented in Chapter 8. These examples demonstrate some of the capabili- ties and the effectiveness of the element. The numerical results are com- pared to analytical or experimental results when available. Finally the conclusions of this study and recommendations for further studies are given. 262. INCREMENTAL CONTINUUM MECHANICS FORMULATION Consider the large displacement motion of a general body as a function of time and assume that the solutions for the static and kine- matic variables are known for the discrete time points, 0, At, 2At, sest. The motion of the body is considered in a fixed Cartesian coor- dinate system, Fig. 2.1. The basic aim of the analysis is to establish an equation of virtual work from which the unknown static and kinematic variables in the configuration at time t+At can be solved. Since the dis- placement-based finite element procedure shall be employed for numerical solution, we use the principle of virtual displacements to express the equilibrium of the body. Assuming that the direction and magnitude of the body and sur- face loading are independent of the body configuration, the principle of virtual displacements is: tit, 5 tty, - trite Tig Strat ?ag (2.1) treaty trat, = ; 4, where “tz, = Cartesian components of Cauchy stress tensor at time ttAt and measured in the configuration at time t+At [Physically equal to force per unit area of the deformed body]; teat®ig ~ Cartesian components of infinitesimal strain tensor in- crement measured in the configuration at time t+At ‘Bu. ou, swhich are defined to be ssaitsy “2( ae + “t-) a x. 27ttat, tat. t+at Pe Tee, Ks) pita, txas! CONFIGURATION AT TIME trot CONFIGURATION PO x2 xp? AT TIME t ie Xan oxo, iu xa, CONFIGURATION AT TIME O Ong, tng IFAs FIGURE 2.1 MOTION OF BODY IN CARTESIAN COORDINATE SYSTEM 28and, **4tR is the total external virtual work expression due to the sur- tH face tractions with components ‘ts, and body forces with components ttt, Of? ttt, ttt, 0 R J ity Ou, Oda + 0, ‘A teat, 0, (2.2) ork Sl av v It should be noted that Eq- (2.1) is identical to the virtual work principle for linear analysis, except thet in the linear formulation the integral is performed over the original volume. Equation (2.1) is an integral equaticn over the unknown deformed volume of the body at time ttt. The quantities in this equation can be veferred back to any of the known configurations at time O,dt, 2Mt, ...t, i.e. the integral can be evaluated over the known volume v, “ty, 2Aty, Here we show how to refer the Eq. (2.1) to the configuration at time t (where t could be 0,At, 2At, ...t). Define a mathematical stress measure, 2nd Piola-Kirchhoff stress", [36, pp. 220- 224] tats 2 t ttt t ‘2 t, ij” Bat, teat®i,s Top teat’ jyr (2.3) Farthough some pictorial representation of the 2nd Piola-Kirchhoff stress are given (e.g. see Fung [-37] page 437, Malvern [36] page 221), there is no physical representation of this stress measure. This stress tensor is merely used for mathematical convenience. 29where *, St, - density at time t, trat + ttl - es i ttat, 4 A atx, t= ttt i,s at tat, 5 Inverting Eqn. (2.3), tet, teat, tl tet, teat, 3 = 8 tes t5rs 8% is (2.4) t Substituting Eqn. (2.4) into Eqn. (2.1) we obtains teat, trag, tat, tH, ttt, ttte . | 2 Ses ivr ys Steae’ay wv (2-5) Jt ty Introducing the increment in the Green-Lagrange strain measure neon tht, tte. rs Pir Bs ,s teat®ij (2.6) and the total Green-Lagrange strain at time ttit is tay = Feng = (2.7) i, fers * ps 30where fe,, = component of Green-Lagrange strain tensor at tine t measured in the configuration at time t. Using Eq. (2.7), the variation in the Green-Lagrange strain becomes ; ste soe (2.8) Using Eqs. (2.5) to (2.8) and since from the conservation of mass teat, that, . & Fay (2.9) the virtual work principle becomes tet, _ [ttt tat. ¢ R= $Syg 6 Feng “dV (2.10) It should be noted that Eqs. (2.1) and (2.10) are equivalent. It should also be noted that a large number of stress and strain measures could have been used to transform Eq. (2.1) into a form other than the one in Eq. (2.10). However, the well known 2nd Piola-Kirchhoff stress tensor is used in this study, because it is a simple and symmetric tensor (see Eq. (2.3)), which results in a symmetric stiffness matrix. In order to solve for the static and kinematic variables of the body at time ttit, in essence, two different formulations can be employed [34]. In the total Lagrangian (T.L.) formulation all static and kinematic variables are referred to the initial configuration at time 0. The up- 31dated Lagrangian (U.L.) formulation is based on the same procedures that are used in the T.L. formulation, but in the solution all static and kine- matic variables are referred to the last calculated configuration at time t. Various applications of both formulations in the analysis of continuum problems are presented in [ 35,38 ] where it was shown, in particular that both the T.L. and U.L. formulations include all nonlinear effects due to large displacements, large strains and material nonlinearities. The only advantage of using one formulation rather than the other is its better numerically efficiency. 2.1 Updated Lagrangian Formulation In the U.L. formulation the virtual work principle expressing the equilibrium and compatibility requirements of the body at time tat referred to the configuration at time t, is teat, tet, ty, teat, I t4j 6 tea OF R (2.11) Vv tat, Kirchhoff stress tensor and Green-Lagrange strain tensor both referred to ttt ttt, where the +54, j and the 4g are the components of the 2nd Piola- the configuration at time t respectively; “te is the external virtual work defined by Eq. (2.2). The above relation represents a nonlinear equation in the unknown static and kinematic variables of the body at time ttAt. The first step is to obtain an approximate solution to Eq. (2.11) by linearizing the equation about the last calculated configuration 32at time t. This approximate solution can then be improved by equilibrium iteration until Eq. (2.11) is satisfied to a required tolerance. Using our usual notation, Table 1 summarizes the linearization and solution of Eq. (2.11) by the modified Newton-Raphson iteration. The objective is to perform static and dyn>mic analysis, where in dynamic analysis the applied body forces include inertia forces. Assuming that the mass of the body is preserved, the inertia forces are conservative and can, in both the T.L. and U.L. formulations, be evalu- oO. ttt. 0 ou, ated using oy dy. Corresponding to the linearized equation given in the table the equations of motion to be solved in the U.L. formu- lation by finite element discretization are: QO ttt. 0 t | - ii, Su, Pav +f sir rs 88g (2.12) 0 t v where the definition for tte is still given in Eq. (2.2), but it should be noted that the body forces no longer include inertia forces, and C5 in. is the incremental material property tensor, referred to the configuration t at time t. The derivation of t%iirs is presented in Section 5.3. The Tj 33are given Cauchy stresses acting in the configuration at time t, and #ig> tag 2Pe the Tinear and nonlinear incremental strains referred to the configuration at time t. 2.2 Total Lagrangian Formulation In the T.L. formulation, the same basic concepts as in the U.L. formulation are employed, but the initial configuration at time 0 is used as reference. Thus, the governing virtual work principle is trit, —gttdt, 0, teat, f §555 Oey Cav = te (2.13) oy where the vars and vate are the components of the 2nd Piola-Kirch- hoff stress tensor and Green-Lagrange strain tensor, both referred to the initial configuration at time 0. Equation (2.13) represents a nonlinear equation in the unknown static and kinematic variables of the body at time t+At. Using our usual notation, Table 2 summarizes the linearization and solution of Eq. (2.13) by the modified Newton-Raphson iteration. Separating the inertia forces from the body forces in the linearized equations of motion given in the table leads to the equations to be solved ‘in the T.L. formulation by finite element discretization: 0, ttt, 60 9, fr ii, Suydv + foesin ofrs 8855 dv (2.14)where gC; jpg 18 the incremental material property tensor referred to the configuration at time 0, 45 are given 2nd Piola-Kirchhoff stresses acting at time t, and 0°. are the linear and nonlinear incremental ig? O45 strains referred to the configuration at time 0. 2.3 Finite Element Discretization The basic aim is to establish and solve the equations of motion of a complete structure. Referring to the standard procedures for assem- bling the structure matrices, in the derivation of the required finite element matrices, attention need only be given to the calculation of the matrices corresponding to a single element [33 ]. In the U.L, formulation, using Eq. (2.12), the following matrix equations are obtained, treaty (ts a tu) 2 = tity te (2.15) In the T.L. formulation, using Eq. (2.14), the following matrix equations are obtained in the finite element derivations, ttt, t t, _ teat, t, wn (Sq + ou) as MEA oe (2.16) where M = mass matrix Gk, > {ky = linear strain incremental stiffness matricess of thy. » Eky, = nonlinear strain (geonetric or initial stress) ‘incremental stiffness matrices; 35= vector of externally applied element nodal loads at time teats i. tr = vectors of nodal point forces equivalent to the element stresses at time ts u_* vector of incremental nodal displacements, assembled from uf 5 tty = vector of nodal displacements at time ttt, assembled tron PBI tthty = vector of nodal point accelerations at time teat. Using our usual notation, Table 3 summarizes the verious inte- grals considered and the corresponding matrix evaluations, 363. GOMPARISON OF THE T.L. AND U.L. FORMULATION FOR A STRUCTURAL ELEMENT The total and the updated Lagrangian formulations were employed successfully in formulating continuum elements [32,39]. In this study we apply these formulations in formulating bending elements. It is important to show that both formulations yield identical element matrices, and the only difference in these formulations is their relative computa- tional efficiency . In this chapter it is shown that the two formulations are identical when they are applied to a straight 2-node beam element. The general three-dimensional straight beam element is formula- ted based on the T.L. and the U.L. formulations [40,41]. The element has two nodes with 6 degrees of freedom per node, and can transmit an axial force, two shear forces, two bending moments and a torque. Figure 3.1 shows a typical beam element. The element is assumed to be straight and of constant cross- section. It is assumed that plane sections of the beam element normal to the neutral axis remain plane and normal during defornation. The element can undergo large deflections and rotations, but small strains conditions are assumed. Thus, the cross-sectional area and the length of the beam element do not change during deformation. The principal moment of inertia axes of the beam element define the local courdinate system r,s,t as shown in Fig. 3.1. The two end nodes of the element, 1 and 2, plus a third auxiliary node, 3, are used to define these axes, where it should be noted that in the computations the r-s plane is defined by nodes 1, 2, and 3. 37BEAM ELEMENT END NODES AUXILIARY NODE o Ox, xg FIGURE 3.1 SCHEMATIC VIEW OF THE 3/0 BEAM ELEMENT LOCAL COORDINATE AXES 38Considering a typical straight beam element it is more effective to first evaluate the finite element matrices corresponding to the local principal axes "X, of the element (see Fig. 3.2), and then transform the resulting matrices to correspond to the global Cartesian coordinates prior to the element assemblage process [33 ]. The finite element matrices corresponding to the axes “x, are simply obtained by measuring all static and kinematic quantities in this coordinate systen. These new quantities are denoted by a bar (-) placed over them. 3.1 U.L. Formulation of Beam Element To describe the motion of a beam element the incremental dis- placement field within the element as a function of the incremental nodal point displacement components is required, 12 a) = i ak 3. ee Dat where the ,hl are the interpolation functions corresponding to the Tocal axis *i,, and the ,i* are the nodal point displacenent increments measured in the local axis at time t, see Fig. 3.2. The interpolation functions in Eq, (3.1) are Hermitian functions where bending displacements varying cubically and axial and torsional dis- placements varying linearly, Table 4.A. The kinenatic assumptions used in defining these interpolation functions hold for small strains, smal rigid body incremental rotations in each solution step, but any size translational displacements. These assumptions are appropriate for the up- 39CONFIGURATION AT TIME t+at CONFIGURATION AT TIME t 1% CONFIGURATION AT TIME O FIGURE 3.2 MOTION OF 3/D BEAM ELEMENT AND ITS LOCAL COORDINATE AXIS SHOWN IN GLOBAL COORDINATE SYSTEM 40dated Lagrangian formulation of beams, because the kinematic variables are linearized about the last-known position. A typical derivative required in the calculation of strain-dis- placement transformation matrices is (04i4/2°%;). This derivative is cal- culated using Eq. (3.1) and noting that the axes 4G = 1,2,3) corres- pond to convected coordinates axes r,s,t at tine t (i.e. “i; = rset (j.= 1,2,3)). Table 4.B summarizes the calculation of the matrices fB and tBu, that are required to evaluate the tangent stiffness matrix and nodal point force vector of an element corresponding to coordinate axes 5, (1=1,2,3). The element matrices have to be transformed to the global coordinate system prior to their assemblage into a system of beam elements. 3.2 TL. Formulation of Beam Element The displacement increments within the element at time t measured in the local axes at time 0 are related to the nodal point displacement increments of the element in its local axes using, see Figs. 3.1 and 3.2, 12 : ick S ok of (3.2) kel where the oi are the element nodal point displacement increments at time t, but measured in the °x, (1=1,2,3) coordinate system. The func~ tions gh{ (1=1,2,3) are the interpolation functions corresponding to the convected axes r,s,t and measured in the coordinate system %, (5=1,2,3). The interpolation functions hj are obtained using 4A 3 12 (3.3) hi = 2 2, te am ty 3.3) ok” 4 & im tin “nk » which transforms dis- where "Rin is the element (i,m) of the matrix *® placements measured in the coordinate system i, (i=1,2,3) to displace- ments measured in the system x, (i=1,2,3). This matrix is incrementally i calculated at each time step [41]. A typical derivative required in the calculation of the strain- 12 ‘ se (aa fate.) « (a.ntyade oak displacement transformation matrix is (39i;/3 «) . s (agh,/2 Xs )gil a kel To evaluate these derivatives it should be noted that the axes °%, (j=1,2,3) correspond to the convected axes r,s.t at time 0 (i.e., °% =, = 5, %, = t). Using Eq. (3.3) we have _ Ww 3 2 ” anu. 3. tr gk fH > DD ttn Tay of (3.4) a, kel om=l nel ae, Therefore double transformations are needed at each integration point for the strain calculation in the T.L. formulation. In comparison, the U.L. Formulation does not require the above transformation. In order to evaluate the strain increments it is also necessary to calculate the derivatives of the total displacements. The kinematics of the rigid body rotations of the beam give; ti Uy = - 6, i=1,2,3 (3.5) ating = By 7 Fig (433) where the "Rj, are the direction cosine of the ti, axis with respect to the 5 axis, and 44, 1s the Kronecker delta. 423.3 Comparison of T.L. and U.L. Formulations The total and updated Lagrangian formulations are compared for the straight beam element presented in the previous sections. The reference coordinate system used in the U.L. formulation is defined by the principal axes of the bean element in the position at time t, ie, *, i =1,2,3). Therefore, the local stiffness matrix and ‘the nodal point force vector are referred to this coordinate system. These matrices are transformed to the global coordinate system using tk = tal ty ty + a " = tpl tj tl (3.6) T = RR where *R is the transformation matrix relating the incremental-nodal point 9 12,3), and R is the transformation displacements measured in *%, to %%, ( matrix that expresses the nodal point displacements measured in the beam local coordinate system % (i=1,2,3) in terms of the global nodal point displacement In the T.L. formulation the reference coordinate system used is given by the element principal axes of inertia in the configuration 0; at time 0, a (i=1,2,3). Therefore, the complete stiffness matrix (including the linear and nonlinear strain stiffness matrices), the nodal point force vector, and the local displacement increments are 43referred to this coordinate system and must be transformed to the global coordinate system °x, (i=1,2,3)5 = OpT tz, = RT £% (3.7) The principal difference between the U.L. and the T.L. formu- lations is that in the T.L. formulation the transformation on the inter- polation functions in Eq. (3.3) is carried out to refer the displacement interpolations to the original configuration, and the [B,, matrix is included in the calculations. The transformations on the interpolation functions and the use of the (6, matrix in the T.L. formulation to- gether are equivalent to the additional transformation matrix *R that is employed on Eq. (3.6) in the U.L. formulation. Indeed, as we have shown in detail [40], using Eqs. (3.1) to (3.5), Table 4, and some algebra we obtain, tr t t t = ty th ‘k= [tho * oul = of (3.8) Substituting the above reiation into Table 3, when the 4%, (j=1,2,3) are used as the reference coordinate system to evaluate the linear strain stiffness matrices, we obtain ter fe the te (3.9)Therefore the two formulations lead to identical linear-strain stiffness matrices corresponding to the global coordinate system. The components of the Cauchy stress tensor referred to the 8, axes are numerically equal to the components of the 2nd Piola-Kirchhoff stress tensor referred to the %, axes [52], it follows from Eq. (3.8), Table 3, and the above fact, that the nodal point force vectors corresponding to the global axes are equal in both formulations. Bathe and Bolourchi [40] have shown that the two formulations lead to identical nonlinear strain stiffness matrices corresponding to the global coordinate system; (3.10) Although the same final element stiftness matrices and nodal point force vectors are generated in the two formulations, it is noted that using numerical integration the transformation on the interpolation functions in Eq. (3.3) and the evaluation of the &, matrix is carried out at each integration point. Therefore, the U.L. formulation is com- putationally more effective. 454. COMPARISON OF HERMITIAN AND LAGRANGIAN BENDING ELEMENTS Hermitian polynomials to interpolate nodal displacements were employed in the previous chapter to develop a straight beam element. The element was used to evaluate an updated and a total Lagrangian formulation. In the formulations of the straight beam element displacements and slopes considered as dependent variables, and the Hermitian functions used to interpolate displacements were referred to the beam convected coordinate axes. A specific difficulty using this approach lies in that the differ- ent displacement components are not interpolated alike; namely, the use of polynomials of different order for the bending, torsional and axial displacements establishes an interpolation directionality that requires special attention. This difficulty is clearly shown in the need for double transformations required in the evaluation of the convected inter- polation functions which are employed in the T.L. formulation, Eq. (3.3). An alternative method is to formulate a general nonlinear bean bending element in which the transverse displacements, axial displace- ments and the rotations are interpolated independently using the same polynomials, Fig. 4.1, Since the same functions are employed for all these kinematic variables, the problem of interpolation directionality under large rotations does not arise. In this chapter we first present the updated and total Lagrangian formulations of a 2 to 4 variable-nunber-nodes Lagrangian beam bending element, Fig. 4.1. The straight beam formulation presented in the previous chapter is then compared to this formulation. 464-NODE BENDING ELEMENT (N=4) ‘FIGURE 4.1 VARIABLE-NUMBER-NODES BENDING ELEMENT 474.1 U.L. Formulation of Lagrangian Bending Element The virtual work principle at time t+At, and its linearization is given in Table 1, Although in the table the general three-dimensional continuum formulation is presented, the stress matrices and stain-displacement transformation matrices in the evaluation of element stiffness matrix and force vector only corresponding to the three nonzero convected beam ele~ ment stress components. Geometry and Displacement Interpolation In the derivation of the bending element below , we follow the work presented for continuum element [38], but use displacement and rotational degrees-of-freedom [42]. Considering a rectangular cross-section and the notation defined in Fig. 4.2 the geometry of the bending element at tine t is interpolated using N N N ; t kas tyk we SD oe oF De hte >, Ve; (4.0) PI kel where *x, = Cartesian coordinate of any point on the element at time t hy(r) = Lagrangian isoparametric interpolation functions, defined in section 6.1. #4 = cartesian coordinate of nodal point k at time t a = thickness of element in t-direction at nodal point k 48ty ty Also, in Et which can using Eq. k Ve by = thickness of element in s-direction at nodal point k Ks = component i of unit vector in t-direction , ‘vk at nodal point k at time t k si ~ Component i of unit vector in s-direction, tyk at Ss nodal point k at time t. iq. (4.1) the variable N is equal to the number of nodes used, be 2, 3 or 4 as shown in Fig. 4.1. tat, = Ky > The displacenient increments u; (4.1), 4 are evaluated kel ko trAtyk tk Vea Ver > Vea (4.3) ko, ttAtyk tyk si = sa VS For the finite element solution we express the components and VK; in tems of rotation increments. To do so we use the unit vectors "VE and *yk corresponding to the configuration at time t which is known (the initi jal vectors ok and ok corresponding to the configuration at time 0 are defined and input to the analysis). Then we have approximately, for small incremental changes in ‘vE and "yf , 49FIGURE 4.2 DEFINITION OF TWO-NODE BENDING ELENENT 50(4.4) =s where 0X is a vector listing three small increments in rotations at node ty and ‘x, axes, respectively. k about the ‘x, , Substituting from Eq. (4.4) into Eq. (4.2) the incremental displace- ments in the beam are obtained in terms of incremental nodal point dis- placements and rotations. The finite element solution gives these nodal point quantities, which can then be employed to evaluate the vectors tratyk and trtyk vk accurately, thatyk v tt (4.5) tratyk tyk ~ ack x yk 0 1f ok is snail the integrations can be carried out to suffi- cient accuracy using the Euler method, which corresponds to the assump- tion in Eq. (4.4). Calculation of Element Matrices Considering a single bending element, the linear strain-dis- placement transformation matrix, {By » the nonlinear straiu-displacement matric ffiy., the stress matrix, “tare required. 51To evaluate the strain-displacement matrices we obtain from Eqs. (4.2) and (4.4) ' k Ue ner fl Mark, Mok, orgy a]f uf une l= Sm fo tosky Fosrk tesrkalf ot i, > Kk 98)14 $98)95 98) 35 (4.6) ky of trotyk. trgyk. tran yk k “bt hy [0 “(ats “(at)p, —“(9t) 34]}] 0 where we use the notation tyk k 0 Vga E tyk tyk v' 0 -4¥ (4.7) traesk 2] 83 sl (as)* = 3, tyk —tyk “Vs2Vs1 e tyk k o V3 ‘vee el jk tyk tank “2% 4K, oo ve (4.8) tyk tyk 0 “eo Ver and tik = steasik, + tk atrk, (4.9) ij A ij 52To obtain the displacement derivatives corresponding to the axes *x,, i=1,2,3 we employ the Jacobian transformation [33] a. tyl 3 HC: (4.10) where the Jacobian matrix, *J, contains the derivatives of the current coordinates *x,, i=1,2,3 with respect to the isoparametric coordinates r,s and t. Substituting from Eq. (4.6) into Eq. (4.10) we obtain k k ng yk au esr f6Di1 (624) 46341] | uf Boxy N 0k au k k k EY eS fate ele (62042 el6802 ) | PP 1) ak ok 2 buy h cenk, yleark, oak, | | Ey thk,3 lV, 4G2);3 £86974 ox. where i. (ty1 tk ty-1 tyggyk. 4 tye) tygeyk tome (‘3h (ot) Mr * (va (95)qi * Yn ooh,) (4.12) With the displacement derivatives defined in Eq. (4.11) we can now directly assenble the strain-displacenent matrices 8, and ‘By . Namely, Eq. (4.11) is employed to establish the global strain components (corvesponding to the *;, 1 = 1,2,3, axes) which are then employed to calculate using a second order tensor transformation the local strain 53components and st corresponding to the r,s and t axes. ter > t’rs Thus, we obtain torr tts} = 8 u (4.13) te 5 st The elements in {8 are defined by the elements in Eq. (4.11) and the direction cosines of the r, s and t axes with respect to the global Cartesian axes *x; , 1 = 1,2,3. The elements in the non] inear strain-displacenent matrix {By are defined analogously. 4.2 T.L. Formulation of Lagrangian Bending Element The virtual work principle at time ttat and its linearization in the case of the total Lagrangian formulation is presented in Table 2. In this table the general three-dimensional continuum formulation is given, however, in the element implementation specialized such that the stress matrix and strain-displacement transformation matrices and force vec- tor only correspond to the three non-zero stress components. The main difference to the U.L. formulation (see Table 1) is that in the T.L. formulation an initial displacement effect is present in the calculation of the linear strain-displacement transformation matrix. This initia displacement effect is taken into account using Eq. (4.1) at time t and time 0. Tren, because ‘uy = x; - xj, we have 54t, tyk _Oyk 44 ah Vea Vea) (4.14) and using Eqs. (4.10) and (4.14) corresponding to the coordinates Oxys 1£1,2,3 we obtain tks tyk Ok t tuk (bt + $56 Ver > Vea) to aC Vey - N 1 0-1 tyk k +h > 253 ah (VES - vk) (4.15) ‘ ices t t The strain-displacement transformation matrices of. and eu. can now be obtained using Eq. (4.11) corresponding to the coordinates a i=1,2,3 and Eq. (4.15). The details of the evaluation are as in the U.L. formulation. 4.3. Comparison of Hermitian Beam and Lagrangian Bending Elenent Comparing the straight beam element formulation given in the previous chapter, and the displacement /rotation , isoparametric bending element formulation presented in this chapter the following differences are identified: (a) The geometry of the straight beam element is assumed to always 55(b) (c ) remain straight. In the Lagrangian bending element, however, the curved geometry of the original or deformed structure can be modeled with a higher order of accuracy because more nodes are used along the element. Also, it should be noted that the geo- metry of the element is interpolated to the same order as dis~ placements, Eqs. (4.1), (4.2). This difference in geometry interpolation and representation is very important for modelling of curved structures (especially when it is expanded for shell analyses). In the formulations of the straight beam element Hermitian inter- polation polynomials, referred to the beam convected coordinate axes, were employed. A specific difficulty of this approach lies in determining the relative contributions of nodal displacements, and/or angles to the bending and/or axial displacements. This establishes the interpolation directionality which requires extra transformations as shown in Eq. (3.3). The above difficulty is not present in the displacement/rotation isoparametric formu- lation with Lagrangian interpolation functions» as was demonstrated in the bending element formulation. The other advantage of the Lagrangian bending element is that the displacements and rotations are interpolated independently using the same functions. As the rotations are assumed to be independent variables, the shear deformation energy is approximately included in the formulation. In comparison, the inclusion of shear de- 56formation effects requires special attention for straight beams with Hermitian polynomial interpolation. The above differences indicate that the use of isoparametric displacement/rotation elements with Lagrangian interpolaticn functions is the more effective approach to expand for the analysis of two-dimensional bending structures, especially for shell structures where the accurate definition of the geometry is of importance. 575. FORMULATIONS OF THE NONLINEAR GENERAL SHELL ELEMENT In Section 4.3, we showed the advantages of employing indepen- dent nodal point rotation and displacement degrees of freedom. We also pointed out the importance of interpolating the geometry and the dis- placements and rotations using the same function. This, in effect, is the displacement/rotation isoparametric element formulation. In the formulations of the displacement/rotation isoparametric shel] elements no specific shell theory is assumed. This is in contrast to the classical approach of formulating shell elements. The geometry and displacement field of the structure are discretized and interpolated as in the analysis of continuum problems [32]. The other advantage of the displacement/rotation isoparametric shell element is its generality which is analogous to the well known isoparametric two- and three-dimen- sional continuum elements [4]. The objective in this chapter is to present the formulations of a general 3 to 32 variable-number-nodes displacement/rotation isopara- metric element for geometric and material nonlinear analysis of general shells, Figs. 5.1 to 5.3. The interpolation functions are presented in the next chapter, The element can have nodes on the middle surface only, or have nodes on the top and bottom surfaces and mid-surface, or on the top and bottom surfaces, see Figs. 5.1 to 5.3. The first elements are used for modelling thin shell structures, and elements with nodes on top, bottom and middle surfaces are employed for the transition regions between 5816-NODE CUBIC LAGRANGIAN ELEMENT (ALL NODES ARE ON THE SHELL MID-SURFACE) SOME DERIVED ELEMENTS: FIGURE 5.1 VARIABLE-NUNBER-NODES QUADRILATERAL SHELL ELEMENTS 59FIGURE 5.3 SOME DERIVED ELEMENTS OF VARIABLE-NUMBER-NODES TRANSITION ELEMENTS 61the thin shell section and sections of the structure modelled by continuum elements. The importance of each of the above elements are discussed in the next two chapters. Any top-or bottom-surface node has, in general, three displacement degrees-of-freedom. A mid-surface node, in general, has five degrees of freedom, three displacements and two rotations, see Fig. 5.4. The following constraints, corresponding to some of the usual shell theory assumptions, are introduced in formulating the element: (a) the strain energy corresponding to the stresses along the normal to the mid-surface is ignored by modifying the three-dimensional material law, and (b) the normal to the shel] mid-surface renains straight (not necess- arily norma?) and does not extend during deformation. 5.1 U.L. Formulation of the Shell Element The virtual work principle expressing the equilibrium and com- patibility requirement of the body at time t#at, in the U.L. formulation, is given in Eq. (2.11). This equation represents a nonlinear expression ‘in the unknown static and kinematic variables of the body at time trit. Using our usual notation, Table 1 summarizes the linearization and solu- tion of Eq. (2.11) by the modified Newton-Raphson iteration. Geometry and Displacement Interpolations In the derivation of an effective shell element, based on the U.L. formulation in Table 1, the displacements and rotational degrees-of- freedom are used, Referring to Fig. 5.4, the geometry of the variable- number-nodes shell element at time t is interpolated using 62tyk > eet he (5.1) where = Cartesian coordinate of any point on the shel1 at time t. h, = isoparametric interpolation functions, see Chapter 6. hy is a function of r and s only if k is a mid- surface node Figs. 5.1, 5.2 and, hy is a function of r, s and a linear function of t if k is a top-or botton-surface node, Fig. 5.3 *K = cartesian coordinate of node? point k at tine t 0 if node k is a top-or bottom-surface node nae 1 if node k is a mid-surface node a, = thickness of shell (in t direction) at nodal point k Wk; = component i of unit normal vector, ‘vy’, to the surface of the shell at nodal point k and time t. Also in Eq. (5.1), the isoparametric element coordinates are r,s and t,and N is the number of nodes of the element. To obtain an expression for uj» we use ttat, t, 44 xy > ky (5.2) and substitute from Eq. (5.1), 63CONFIGURATION AT TIME t SHELL MID- SURFACE MID-SURFACE NODAL POINT k TOP SURFACE NODAL POINT & COORDINATE tx , xg , 'x§, cooroinare ‘xf, ‘xf, 'xf norma. — ‘vi DEGREES- OF- FREEDOM: THICKNESS a, of uf, uf DEGREES - OF - FREEDOM: kook yk gk ok Uy Uz 43 a8 FIGURE 5.4 DEFINITION OF A GENERAL SHELL ELEMENT 64N N = k t k wD meh + BD mah wh (5.3) where tyke (5.4) Tt should be noted that the calculation of increments in the normal vector, VK, or new nomal directions are only performed at mid-surface nodes. For the Finite element solution we express the components V‘, in terms of rotations. To do so we use the vector "yk corresponding to the configuration at tine t which is known. The vector *vK does not pertain to an element but to mid- surface nodal point k. Thus, all shell elenents meeting at mid-surface nodal point k have this same vector as mid-surface normal as illustrated in Fig. 5.5 (For modeling ‘of intersecting shells see Chapter 7). Also, to start the solution, VK corresponding to the initial configuration is defined prior to the incremental analysis. To express the components VK, in terms of rotation angles, we first define two unit vectors ‘yt and *vk that are orthogonal to *y* (see Fig. 5.6). tyk o tyk T tyk: Wee Oey x KO) SYK) (5.5) tyk tyk where we set *vK equal tox, if ‘vy! is paraltel to x,5 and (5.6) 65JOVAUNS-GIW TIEHS OL IVNYON $*S 3uNDI4 ‘ ey Be texy 8 O SWIL LY NOILVENDI4NOO BOVAUNS-OIN T13HS 4 LNIOd TVGON / + 3WIL iv NOILVYNDIANOD 66FIGURE 5.6 DEFINITION OF VECTORS AT NODAL POINT k 67Let ak and aX be the rotations of the normal vector about the vectors "yf and yk from the configuration at tine t to the configuration at time t+At. then approximately (for small incremental angles a and af) Koo. tyk ok tyk ok v= 7 bak + fyi! (5.7) Substituting from Eq. (5.7) into Eq. (5.3) we thus obtain the incremental internal element displacements in terms of the nodal point ‘incremental displacements and rotations N N kt tyk gk, ty gk > nuts $ 2 magn evay © VI Od (5.8) kel . The finite element solution will yield the nodal point vari- ables uf if node k is a top-or botton-surface node, or uf, a€ and Bk if node k is a mid-surface node. The displacements and rotation of mid- ttt, surface nodes can then be employed to evaluate accurately **4tyk, +1 k - ttatyk 2 tyk y My = tn k ak ate B dak + ‘Kt apt (5.9) 1f of and B* are small, the integration in Eq. (5.9) can be carried out to sufficient accuracy using the Euler method [33], (5.10) 68which corresponds to the assumption employed in Eq. (5.7). Calculation of Element Matrices Considering a single she11 element, as summarized in Table 1 and 3, the linear strain-displacenent transfomation matrix, $8, , the stress matrix, 'r, and the stress vector, *?, are required. To evaluate the strain-displacement matrices we obtain from Eq. (5.8) tok uk u, ber OT tata Hf 4 ir N ued = tok iss 2 hs 0 tay ae (8.11) htgk “it Met eats tka, where we use the notation tk 2d k 914 FER V2 ee) k “ii al Saq 2 MK*k To obtain the displacement derivatives corresponding to the axes ‘es » 1=1,2,3, we now employ the inverse Jacobian transformation (5.13) isle 69where the Jacobian matrix, ty, contains the derivatives of the current. coordinates ',, i=1,2,3 of Eq. (5.1) with respect to the isoparametric coordinates r,s and t, ty t, t, (5.14) Substituting from Eq. (5.11) into Eq. (5.13) we obtain i tk k tik k k y ther 14 StS 4 *y y Ou, 1 |= hy tk gk tk gk k a ky 2 tke 914 82924 t% « (5.15) a h, k gk k gk k = t t, atx, tks 974 483924 483 5 where 2 tal t)-1 t,-1 Ct er ko arty ty-1 ty-1 18h = tC hg + TOFD eg) + Gg hy (5.16) 70and i is the element (i,j) of the matrix *y”! in Eq. (5.13). With the displacement derivatives defined in Eq. (5.15) we can now directly assemble the strain-displacenent matrices £8, and Table 5 gives these matrices and defines also the corresponding stress matrix t. and stress vector *e, 5.2 T.L. Formulation of the Shell Element In the T.L. formulation, the virtual work principle expressing the equilibrium and compatibility requirement of the body at time ttat is given by Eq. (2.13). This equation represents a nonlinear equation in the unknown static and kinematic variables of the body. Using our usual notation, Table 2 summarizes the linearization and solution of Eq. (2.13) by the modified Newton-Raphson iteration. Geometry and Displacement Interpolations In the T.L. formulation of a shell element, not only the geo~ metry and displacement increments are required to be interpolated, but also we need to interpolate the total displacements, see Tables 2 and 6. The geometry at time t, and the displacement increments of the variable-number-nodes shell element are interpolated using Eq. (5.1) and Eq. (5.3), respectively. The evaluation of the normal vector increments and calculation of the normal vectors at time ttAt are as given in the updated Lagrangian formulation presented in the previous section, The initial displacement effect (which is present in the calcu- 1lation of the linear strain-displacement transformation matrix) is taken t, -t, 0 into account using Eq. (5.1) at time t and 0. Then because Uy = Ky ky we have - yk ) (5.17) Calculation of Element Matrices Consider a single shell element, as summarized in Tables 2 and 3, the linear and nonlinear strain-displacement transformation matrices, t t, ax ti or, te OB and oBy» and the stress matrix,o5, and stress vector, §§ . are required. To obtain the derivates with respect to the axes %., i=1,2,3, we employ the inverse Jacobian transformation °y"!, fe th (5.18) ax = where the Jacobian matrix, J, is defined: 0, 9, Myr 2p *3yr (5.19) 0, is %2,5 3,5 0. 0. rt %2,t Xa.t Substituting from Eq, (5.11) into Eq. (5.18) we obtain 72u iy tok gk tok gk k 2x, cry 34 0% 921 oft | | Yi N ou tok gk tk gk sat] => foe 4 82 924 of (8.20) axel Ou. H or a k yy. oka 14 03 924 of3 J LP 3 where = Oy1 Oy 05-1 oft 7 iT Mr * 42 Meas * 743 Met k ° (5.21) ge Oy 05-1 -1 OF = UT Me * Ii2 Mkys) * 94s Mk and %4;} is the etenent (4,3) of °F”! tn Eq. (5.16). Using Eqs. (5.17), (5.18) and (5.21), the derivatives of the initial displacements corresponding to the coordinates °x,, i=1,2,3 are obtained: atu, . tyk 7 0-1 k k i. - ra Dd os +h D ee (tes +53 ny) - OK) Sl k=l (5.22) With the displacement derivatives in Eqn. (5.20) and initial displacement derivatives in Eq. (5.22) we can directly assemble the strain 73displacement matrices, {8, and $84). Table 6 gives these matrices and de~ fines also the corresponding stress matrix [S and stress vector fs 5.3 Constitutive Relations In the previous sections we assumed that appropriate constitutive relations are used, Tables 1 and 2. The constitutive tensor, in the U.L. formulation relates Cauchy stresses and Almansi strains, and in the T.L. formulation relates 2nd Piola-Kircihoff stresses to Green-Lagrange strains. We should note that in the formulations given so far all the six stress and strain components in the Cartesian coordinate directions are calculated, and the constitutive tensor corresponding to these com- ponents is used in Tables 1, 2 and 3. To obtain this tensor constrained by the shell assumption that the stress normal to the shell is zero, we use the usual three-dimensional constitutive relations corresponding to the local orthogonal directions Fig. 5.7, impose the condition that tne nomal stress in t-direction) is negligible and then transform the modified constitutive relations to correspond to the global Cartesian coordinate axes. Linear Elastic Material Model Young's moduli, E, and Poisson's ratio, v, are employed to de- fine the usual three-dimensional isotropic constitutive tensor [33]. The con- straint of zero stress along the normal to.the mid-surface is. satisfied: by 74t CONFIGURATION AT TIME t = UNIT VECTOR TANGENT TO r-AXIS = UNIT VECTOR ALONG t-AXIS FIGURE 5.7 DEFINITION OF THE SHELL ELEMENT ORTHOGONAL LOCAL COORDINATE AXESstatically condensing out the appropriate row and column of the constitu- tive tensor, This leads to an anisotropic constitutive relation, in the linear analysis, and in the T.L. formulation, lv 0 aao 1 0 a a0 t E t= o 1? ao oao lv (5.23) an symmetric Ip o The anisotropic constitutive tensor, measured in the *%, co- ordinate system at time t, employed in the U.L. formulation C, is numeri- cally identical to (C of Eq. (5.23). The above constitutive tensors are transformed to global co- ordinate systems using; fe = te | (r= 0,t) (5.24) % = % % (t = 0,t) (5.25) where the elements of the matric O t= 0,(t) are the direction cosines of the orthogonal local axes iy with respect to the fixed global axes, 76in linear or T.L. (UsL,) formulation. The direction cosines of the unit vectors °%, and °%, are obtained from the normalized values of the Jaco- bian matrix 9, Eq. (5.19), i.0.5 % and 9%, are along the natural coor- dinate r and t axes at time 0 respectively. The third unit vector is calculated as °%, = 5 x fy, Fig. 5.7, The directions of the unit gy (used in U.L. formulation) are obtained similarly by employing Eq. (5.14). It should be noted that the advantage of employing the local orthogonal axes "X; as a reference system rather than natural coordinates r,s,t is especially seen when triangular or distorted quadrilateral she11 elements are formulated where the axes r and s are non-orthogonal. It should also be mentioned that change of density is ignored in £q. (5.24) because of the small strain assumption. The constitutive relations in Eq. (5.24) are used in the evaluation of the element stress matrices and stress vectors, i.e., we calculate from total Green-Lagrange or Almansi strains directly the total 2nd Piola-Kirchhoff or Cauchy stresses, respectively. However, in the calculation of the linear strain stiffness matrices at time t, we need the following tangent material property tensors 77