Finite Elements in Analysis and Design 12 (1992) 313-318

313

Elsevier FINEL 273

A performance study of tetrahedral and hexahedral elements in 3-D finite element structural analysis
A . O . Cifuentes and A. Kalbag
IBM Research Di~,ision, P,O. Box 218, Yorktown Heights, NY 10598, USA

Received March 1992 Revised June 1992

Abstract. This study compares the performance of linear and quadratic tetrahedral elements and hexahedral

elements in various structural problems. The problems selected demonstrate different types of behavior, namely, bending, shear, torsional and axial deformations. It was observed that the results obtained with quadratic tetrahedral elements and hexahedral elements were equivalent in terms of both accuracy and CPU time.

Introduction

Finite element analysis has reached a state of maturity in which 3-D applications are commonplace. Most analysts, as well as most commercial codes ( M S C / N a s t r a n , Abaqus, etc.), use solid elements based on the isoparametric f o r m u l a t i o n - - o r variations of i t - - f o r 3-D analyses [1-4]. For simple geometries, or for applications in which it is possible to build a mesh " b y hand", analysts have relied heavily on the 8-node hexahedral e l e m e n t - - c o m m o n l y known as "brick" or " h e x a " [5]. For more complex geometries, however, the analyst must rely on an automatic (or semi-automatic) mesh generators. In general, automatic mesh generators produce meshes made of tetrahedral elements, rather than hexahedral elements. The reason is that a general 3-D domain cannot always be decomposed into an assembly of bricks. However, it can always be represented as a collection of tetrahedral elements. As the demand for analyses of more complex configurations has grown, coupled with the increasing popularity of automatic mesh generators, the need to understand better the relative merits of tetrahedral and hexahedral elements has become apparent. It is known, for example, that linear tetrahedral elements do not perform very w e l l - - a s e x p e c t e d - - b e c a u s e they are constant-strain elements; thus, too many elements are required to achieve a satisfactory accuracy. What remains unclear, however, is whether brick elements perform better or worse than quadratic tetrahedra, that is, tetrahedral elements including mid-side nodes. Specifically, for a given n u m b e r of nodes (or degrees of freedom), the analyst needs to know under what circumstances it is better to use bricks instead of quadratic tetrahedra. This amounts to investigating the accuracy and efficiency of such elements under a variety of problems characterized by different deformation patterns, such as, bending, shear, torsion and axial behavior. In addition, if a mesh made of linear tetrahedral elements does not yield a result within acceptable error, it is useful to know what strategy to follow: (a) decrease the size of the
Correspondence to: A.O. Cifuentes, IBM Research Division, P.O. Box 218, Yorktown Heights, NY 10598, USA.

0168-874X/92/$05.00 1992 - Elsevier Science Publishers B.V. All rights reserved

314

A.O. Cifuentes, A: Kalbag / Tetrahedral and hexahedral elements in 3-D analysis

elements while keeping them linear, or (b) make the elements quadratic by introducing additional (mid-side) nodes. Previous authors have proposed some useful benchmark tests for individual elements or simple arrays of elements [6-8]. However, no study comparing tetrahedra with hexahedra in a more general setting seems to be available. While it is difficult to give a final answer to all the issues involved, the aim of this study is to shed some light on this problem by investigating the performance of tetrahedral and hexahedral elements in a number of problems that have known analytical solutions. These findings are expected to be useful for finite elements analysts.

Method Four problems that have known solutions have been chosen for this study. Each problem is dominated by a different deformation pattern.

Problem 1--Bending
Consider a cantilever b e a m oriented in the y-direction and loaded in the z-direction (see Fig. 1). The b e a m has a rectangular cross-section and it deforms under the action of a load per unit of length equal to 0.01. The beam dimensions are as follows: L ( l e n g t h ) = 8, b (width) = 1 and h (height) --- 1. The material properties are: E (Young's modulus) = 1000 and u (Poisson's ratio) = 0.15. The analytical expression for the vertical displacement at the free end of the b e a m center-line, including both bending and shear deformations (although bending is the dominant effect in this case), yields a value of 0.0625 [9].

Problem 2--Shear
Consider a short shear b e a m deforming under a unit distributed load (load per unit of length) as depicted in Fig. 2. The b e a m is oriented in the y-direction and loaded in the z-direction. The b e a m dimensions are: L = 1, b = 0.6 and h = 1. The material properties are: E = 1000 and u = 0.15. The vertical displacement at the free end of the b e a m center-line, considering both bending and shear deformations (which in this case are dominant) is 0.00538 [10].

Problem 3--Torsion
Consider a b e a m with a square cross-section oriented along the y-axis (see Fig. 3). The b e a m dimensions are: L = 16, b = 1 and h = 1. Material properties: E = 1000 and u = 0.15.

load per unit of length

load per unit

of length

...................
L

-}--~y

[]
b
cross

b
L ~

__ i b
cross section

longitudinal view

section

longitudinal view

Fig. l, Bending problem.

Fig. 2. Shear problem.

A.O. Cifuentes, A. Kalbag / Tetrahedral and hexahedral elements in 3-D analysis

315

constrained in the x and z direction torsional moment

iz

[] h
Y x ~L ..~

b
cross

longitudinal view

section

Fig. 3. Torsion problem

Displacements in the x- and z-directions are fixed at one end. At the other end, which is free, a rotation of 0.03 radians is applied (this corresponds to a 0.1146 torsional moment). The maximum value of the shear stress occurs at the mid-points of the cross-section sides. A solution based on a series expansion gives a value of 0.551 for the maximum shear stress [11]. This solution allows warping of the cross-section.
Problem 4 - - A x i a l behavior

Consider a short beam L = 4, b = 1 and h = 1. volume) = 1. The natural This problem was chosen
A n a lyses

clamped at both ends and oriented in the y-direction (see Fig. 4). In addition, E = 1000, v = 0.0 and p (mass density per unit of frequency corresponding to the first axial mode is 3.953 Hz [12]. because it involves a nonuniform axial displacement field.

The finite element analyses were performed using Abaqus, a general-purpose finite element code for structural analysis [13]. Three solid elements were tested: (a) C3D4, a 4-node tetrahedral element. This element was included only for comparison purposes; its performance was not expected to be good since it is a constant-strain element. One integration point is used. (b) C3D10, a second-order 10-node tetrahedral element. In this study, the "intermediate" nodes were located exactly halfway between the corner nodes. Four integration points are used. (c) C3D8, an 8-node isoparametric hexahedral element. This is a trilinear element. In this case "full" Gauss integration was employed in the stiffness matrix determination. This

.....................................

"

b L longitudinal view
cross

section

Fig. 4. Axial vibration problem.

316

A.O. Cifuentes, A. Kalbag / Tetrahedral and hexahedral elements in 3-D analysis

means that the Gauss scheme used integrates the stiffness matrix terms exactly if (i) the material properties are constant throughout the element and (ii) the Jacobian of the mapping from the isoparametric coordinates to the physical coordinates is constant and diagonal throughout the element. Each problem was solved using four different models (four different meshes), described as follows:
Mesh 1.

This is a regular mesh made of linear tetrahedral elements (C3D4).

Mesh 2. This is a regular mesh made of quadratic tetrahedra (C3D10) obtained by adding mid-side nodes to Mesh 1. This represents an attempt to improve the accuracy of the results obtained with the first mesh. Mesh 3. This mesh corresponds to another attempt to improve the results obtained with Mesh 1, but in this case decreasing the size of the linear tetrahedra (C3D4). This mesh obviously has more nodes than the mesh employed in the first model, but exactly the same number of nodes as Mesh 2. This is to make the second and third model comparable in terms of the size (same number of degrees of freedom) and therefore address the issue of what strategy is better if one wants to improve the accuracy of the results given by a mesh of linear tetrahedra (Mesh 1): to increase the order of the interpolation (Mesh 2) or to reduce the size of the elements (Mesh 3). Mesh 4. This is a regular mesh of brick elements. Again, the number of nodes is the same as in Mesh 2. This is to compare the performance of two meshes with the same number of degrees of freedom, one made with bricks and the other made with quadratic tetrahedra. (Notice that nodal coordinates in Mesh 2 coincide with those of Mesh 4).

Results

The four problems described before were solved using the four different meshes. The analyses were performed on an IBM RISC/6000 workstation using Abaqus. The models were setup with Foxi, a geometric modeler that has a parametric solid object representation and is integrated with an automatic mesh generator and an Abaqus preprocessor [14]. Regular meshes were employed in all cases. In each case, the error was computed by comparing the result given by the finite element model against the analytical solution. Tables 1-4 summarize the results. The nomenclature is as follows: N is the number of nodes in the mesh (including mid-side nodes when quadratic tetrahedra are used); E is the

Table 1 Results for Problem 1 (bending) N Mesh Mesh Mesh Mesh 1 2 3 4 225 1225 1225 1225 E 576 576 4608 768 T 51 105 290 106 Ax 0.5 0.5 0.25 0.25 Ay 0.333 0.333 0.1666 0.1666 Az 0.5 0.5 0.25 0.25 Ra x 10 -2 3.822 6.210 5.334 6.264 Error (%) 38.9 0.7 14.7 0.2

a R represents the vertical displacement at the end of the beam center line. The analytical solution gives R = 6.254 X 10 -2.

A.O. Cifuentes, A. Kalbag / Tetrahedral and hexahedral elements in 3-D analysis

317

Table 2 Results for Problem 2 (shear) N Mesh 1 Mesh 2 Mesh 3 Mesh 4 768 4805 4805 4805 E 2700 2700 21600 3600 T 176 687 1840 589 Ax 0.3 0.3 0.15 0.15 Ay 0.0666 0.0666 0.0333 0.0333 Az 0,0666 0,0666 0.0333 0.0333 Ra x l 0 -3 4.687 4.808 4.739 4.756 Error (%) 12.8 9.2 11.8 11.5

a R represents the vertical displacement at the end of the beam center line. The analytical solution gives R = 5.375 10- 3.

Table 3 Results for Problem 3 (torsion) N Mesh 1 Mesh 2 Mesh 3 Mesh 4 153 825 825 825 E 384 384 3072 512 T 46 79 294 80 Ax 0.5 0.5 0.25 0.25 Ay 1.0 1.0 0.5 0.5 Az 0.5 0.5 0.25 0.25 Ra 0.3859 0.5799 0.4736 0.5300 Error (%) 30.0 5.2 14.1 3.8

R represents the maximum shear stress on the cross section of the beam. The analytical solution gives R = 0.5511.

Table 4 Results for Problem 4 (axial vibration) N Mesh 1 Mesh 2 Mesh 3 Mesh 4
a

E 1536 1536 12288 2048

T 153 595 1488 499

nx 0.25 0.25 0.125 0.125

Ay 0.25 0.25 0.125 0.125

Az 0.25 0.25 0.125 0.125

Ra 3.947 3.945 3.951 3.951

Error (%) 0.2 0.2 0.1 0.1

425 2673 2673 2673

R represents the fundamental frequency (Hz) for axial vibrations. The analytical solution gives R = 3.953.

n u m b e r of e l e m e n t s in the mesh; A x , A y , Az d e n o t e the n o d e spacing in the x-, y- a n d z - d i r e c t i o n (in m e s h e s m a d e of q u a d r a t i c t e t r a h e d r a , the spacing is d e t e r m i n e d by the distance b e t w e e n c o r n e r nodes); a n d T d e n o t e s the total C P U time in seconds.

Discussion
T h e m a i n goal of this analysis was to investigate the p e r f o r m a n c e of h e x a h e d r a l e l e m e n t s versus q u a d r a t i c t e t r a h e d r a u n d e r similar conditions. This has b e e n achieved by c o m p a r i n g the results given by M e s h 2 a n d M e s h 4. T h e location of the n o d e s is identical in both meshes. T h u s , the n u m b e r of active d e g r e e s of f r e e d o m is exactly the same. This is necessary to m a k e a m e a n i n g f u l c o m p a r i s o n . I n addition, the e l e m e n t aspect ratio in b o t h m e s h e s is e q u i v a l e n t (the ratio b e t w e e n the n o d e spacing in the x-, y- a n d z - d i r e c t i o n is the same in b o t h meshes). It c a n be observed that the resuls o b t a i n e d with bricks a n d q u a d r a t i c t e t r a h e d r a , in t e r m s of b o t h accuracy a n d C P U time, are roughly equivalent. This is significant b e c a u s e it indicates that analysts who rely o n a u t o m a t i c m e s h g e n e r a t o r s (which in g e n e r a l g e n e r a t e m e s h e s m a d e of t e t r a h e d r a l e l e m e n t s ) do n o t have a d i s a d v a n t a g e c o m p a r e d to those analysts who use

318

A.O. Cifuentes, A. Kalbag / "Fetrahedral and hexahedral elements in 3-D analysis

bricks. In o t h e r words, the t r i l i n e a r brick e l e m e n t - - a l o n g - t i m e favorite o f m a n y finite e l e m e n t p r a c t i c i o n e r s - - a p p e a r s n o t to have a s u b s t a n c i a l a d v a n t a g e c o m p a r e d to t h e q u a d r a t i c tetrahedron. A s e c o n d conclusion is c o n c e r n e d with w h a t is t h e best a p p r o a c h to t a k e if a m o d e l m a d e o f l i n e a r t e t r a h e d r a d o e s not give satisfactory results ( M e s h 1). T h e s e analyses ( M e s h 2 versus M e s h 3) suggest that, in g e n e r a l , it s e e m s b e t t e r to i n c r e a s e t h e o r d e r of the e l e m e n t s r a t h e r t h a n refining t h e m e s h with s m a l l e r l i n e a r e l e m e n t s . E x c e p t for P r o b l e m 4, in which M e s h 2 a n d M e s h 3 give a p p r o x i m a t e l y the s a m e result, t h e q u a d r a t i c t e t r a h e d r a do b e t t e r t h a n the l i n e a r t e t r a h e d r a , for t h e s a m e n u m b e r o f nodes. In t e r m s of C P U time, t h e a d v a n t a g e of q u a d r a t i c t e t r a h e d r a is m o r e c l e a r - - t h e r e is a t h r e e f o l d penalty, in all cases, for using l i n e a r t e t r a h e d r a . T h i s is b e c a u s e M e s h 3 i n c l u d e s m a n y m o r e e l e m e n t s t h a n M e s h 2 a n d conseq u e n t l y the C P U t i m e r e q u i r e d to g e n e r a t e the stiffness m a t r i x a n d m a s s m a t r i c e s increases, as d o e s the t i m e for solving t h e r e s u l t i n g l i n e a r system o f e q u a t i o n s .

References
[1] B.M. IRONS, "Engineering applications of numerical integration in stiffness methods", J. Am. Inst. Aeronaut. Astronaut. 4 (ll) pp. 2035-2037, 1966. [2] I.C. TAIG, "Structural analysis by the matrix displacement method", English Electric Aviation, Report SO-17, 1961. 13] J. ERGATOUDIS, B. IRONS and O.C. ZIENKIEWICZ,"Curved isoparametric quadrilateral elements for finite element analysis", Int. J. Solids Struct. 4, pp. 31-42, 1968. [4] R.D. COOK, Concepts and Applications of Finite Element Analysis, Wiley, New York, 2nd edn., Chap. 5, 1981. [5] T.J.R. HUGHES, The Finite Element Method, Prentice-Hall, Englewood Cliffs, NJ, Section 3.5, 1987. [6] R.L. TAYLOR,J.C. SIMo, O.C. ZIENKIEW1CZand A.C. CHAN, "The patch test: A condition for assessing finite element convergence", Int. J. Numer. Methods Eng. 22 (l) pp. 39-62, 1986. [7] J. ROBINSON, "Single element test for aspect ratio sensitivity of solids", Finite Element News, pp. 26-32, February 1986. [8] G.P. BAZELEY, Y.K. CHEUNG, B.M. IRONS and O.C. ZIENKIEWICZ,"Triangular elements in plate bending --Conforming and non-conforming solutions", Proc. 1st Conf. on Matrix Methods in Structural Mechanics, Wright-Patterson Air Force Base, Ohio, pp. 547-576. 1968. [9] J.P. DEN HARTOG, Strength of Materials, Dover Publications, New York, 1961. [10] S. TIMOSHENKOand D.H. YOUNG, Elements of Strength of Materials, Van Nostrand, New York, 1968. [11] S. TIMOSHENKOand J.N. GOODIER, Theory of Elasticity, McGraw-Hill, New York, 1970. [12] R.D. Bt.Evms, Formulas for Natural Frequency and Mode Shape, Van Nostrand Reinhold, New York, 1979. [13] Abaqus Users' Manual, Version 4.8, Hibbit, Karlsson and Sorensen Inc., Providence, Rhode Island, 1989. [14] L.R. NACKMAN,unpublished IBM Research internal report, 1992.