Professional Documents
Culture Documents
Abstract
The ®nite element software ABAQUS offers several algorithms for dynamic analysis. The direct integration methods include the implicit
and the explicit methods which can be used for linear and nonlinear problems. The performance of these two methods are compared for
several dynamic problems including the impact of an elastic bar and a cylindrical disk on a rigid wall. The advantages of the implicit
method for small wavefront problems and the explicit method for short transient problems are veri®ed. # 2000 Elsevier Science S.A. All
rights reserved.
Keywords: Implicit ®nite element method; Explicit ®nite element method; Dynamic problems
*
Corresponding author. Tel.: 65-874-2205; fax: 65-779-1459. The implicit procedure uses an automatic increment
E-mail address: mpeleehp@nus.edu.sg (H.P. Lee). strategy based on the success rate of a full Newton iterative
0924-0136/00/$ ± see front matter # 2000 Elsevier Science S.A. All rights reserved.
PII: S 0 9 2 4 - 0 1 3 6 ( 0 0 ) 0 0 5 8 0 - X
J.S. Sun et al. / Journal of Materials Processing Technology 105 (2000) 110±118 111
and
3. Example problems
u_
i1 u_
i Dt
1 ÿ g
u
i g
u
i1 (4)
3.1. Impact of an elastic bar against a rigid wall
with
b 14
1 ÿ a2 ; g 12 ÿ a; 1
3 a0 (5) As the ®rst example, the impact of an elastic bar against a
rigid wall is presented. This model is used to compare the
a ÿ0:05 is chosen by default in ABAQUS as a small results by ABAQUS and the reported results by Zhong [6].
damping term to quickly remove the high frequency noise The dimensions of the bar are 1 1 10 m3 . Young's
without having a signi®cant effect on the meaningful, lower modulus is given as E100 kPa. Possion's ratio is n 0, and
frequency response. the mass density is r 0:001 kg mÿ3 .
The explicit procedure is based on the implementation of Eighty C3D8 and C3D8R (eight-node bricks) elements
an explicit integration rule along with the use of diagonal are used for the implicit and explicit methods, respectively.
element mass matrices. The equation of motion for the body An initial velocity of 1 m sÿ1 is prescribed to the bar striking
is integrated using an explicit central difference integration the rigid wall. The elements of the rigid wall are R3D4. The
rule: mesh is shown in Fig. 1.
The impact time can be simply estimated by
u
i1 u
i Dt
i1 u_
i1 ;
u_
i1=2 u_
iÿ1=2 12
Dt
i1 Dt
i
u
i 2L
Dt
c
the acceleration, and i and i ÿ 12,
where u_ is the velocity, u where L is the length of the bar, and c the wave speed for this
1
i 2 refer to the increment number and mid-increment kind
numbers. p of material, which is determined by
E=r 2:0 10ÿ3 m sÿ1 .
For the implicit method, the half-step residual tolerance
i Mÿ1
F
i ÿ I
i
u (6)
HAFTOL is a very important parameter to control the
where M is the mass ``lumped'' matrix, F the applied load computational accuracy [1]. A large value of 10 is set
vector and I the internal force vector. initially. In Fig. 2a, the contact reaction forces are compared.
It only takes 19 steps and less than 1 s CPU time for the analyze and predict failure. Comparing the stresses by these
explicit method, while it takes 106 s of CPU time for the two methods is helpful for failure diagnosis.
implicit method. However, under this condition, the accu- A three-dimensional example is needed for the above
racy of the results by the explicit method is better than that purpose. This example is shown in Fig. 3. A cylindrical disk
by the implicit method. Fig. 2b shows the stress component is formed by two cylinders. The radius of the bottom
S11. The values of S11 by the implicit method are found to cylinder is 1 m, and the radius of the top cylinder is
oscillate much more than the corresponding results by the 0.25 m. The bottom cylinder is of 0.5 m height, and the
explicit method. height of the top one is 0.1 m. The distance between the
From Fig. 2a, the reaction force is found to be about centers of these two cylinders is 0.31 m. There are 548 eight-
10 N. In order to secure more accurate results, 10% of the node brick elements and 810 nodes for the mesh. Elements
reaction force, 1.0, is set as the revised HAFTOL value. The 173 and 409 are chosen for the purpose of stress comparison.
reaction force and S11 are shown in Figs. 2c and d. The Nodes 311 and 709 are one of the eight nodes for elements
values of S11 by the implicit methods still oscillate more 173 and 409, respectively.
than those by the explicit method, but both the stresses and The material is ductile steel with a Young's modulus of
reaction force oscillate less than the corresponding results E 200 GPa and density of r 7833 kg mÿ3 . Poisson's
for the previous case with HAFTOL10. The CPU time for ration is taken to be n 0 in order that the stresses can
the implicit method signi®cantly increases, from 106 to be simply calculated by
524 s.
E @ui @uj
The periods of the FEM deviate from that of the exact si;j ; i; j 1; 3 (7)
solution. As mentioned above, a is set to ÿ0:05 for the 2 @xj @xi
implicit method to induce arti®cial damping. The numerical
dissipation leads to an amplitude decay of 2px [5], where x is
the algorithmic damping ratio. This is illustrated in Figs. 2a
and c.
Fig. 5. (a) The contact reaction force RF2; (b) the displacement component U1 of node 311; (c) the displacement component U2 of node 311; (d) the stress
component S11 of element 173; (e) the stress component S22 of element 173; (f) the stress component S33 of element 173.
J.S. Sun et al. / Journal of Materials Processing Technology 105 (2000) 110±118 115
Fig. 5. (Continued ).
116 J.S. Sun et al. / Journal of Materials Processing Technology 105 (2000) 110±118
in which si;j are the six independent components of stress, ui 3.3. Slow contact between an elastic cylindrical disk and a
are the three components of displacement, and xj are the rigid wall
three coordinates.
The maximum principal stresses (SP3) by the static The displacement for the bottom surface of the elastic
algorithm are shown in Fig. 4. The original gap between cylindrical disk is set to be 0.02 m. The total time is 1 s,
the model and the rigid wall is 0.01 m. A static displacement which is much larger than in the fast case (0.002 s).
of 0.02 m is applied on the bottom surface of the large Fig. 8a shows the displacement U2 of node 311, whilst
cylinder. Fig. 8b shows the stress component S22 of element 173:
An initial velocity of 10 m sÿ1 is prescribed. Fig. 5a shows differences are apparent.
the agreement of the contact reaction force RF2 for the Fig. 8b shows the stress component S22 of element 409.
results by the two methods. The results show good agreement. Figs. 8c and d compare
The displacements U1, U2 and U3 of node 311 are shown the maximum principal stress SP3. The regularities are the
in Figs. 5a±c. U3 shows a large difference, but the absolute same as mentioned in the last section. The perfect agreement
value of U3 for node 311 is relatively small because node between Fig. 8c and Figs. 6a and b show that the results by
311 is on the axisymmetric axis. Due to the fact that the implicit methods are very accurate, where under these
displacement of the rigid body in direction 2 is compara- conditions inertia effects can be neglected.
tively large relative to the deformation, U2 shows an insig- The stable time step for explicit is 7 10ÿ6 s for the slow
ni®cant change, but Figs. 5d±f show apparent differences of case and 8:5 10ÿ6 s for the fast case. For the slow case, the
stresses of element 173 by these two methods. Although at time increment 139935 is much more than that of the fast
some region such as node 331, the values of U3 are quite case. Thus the CPU cost increases dramatically from 7 min
different for the two methods, the whole displacement to 1 h 17 min. The CPU cost for the implicit method is
pattern of U3 is similar, as shown in Figs. 6a and b. 10 min.
On the other hand, the stress components S11, S22 and
S33 of another element 409 which connects to the contact
surface, show agreement by the two methods in Figs. 7a±c. 4. Closure
Displacements have higher accuracy than stresses, thus the
three displacements U1, U2 and U3 of node 709 obviously Fast and slow linear contact problems have been analyzed
agree. The results are testi®ed in Figs. 6a and b. For stress by different methods in ABAQUS. For the fast case, the
distribution, the regularity is the same. Figs. 7d and e show advantages of the explicit method are apparent within the
that the pattern of stress distribution is similar for most desirable tolerance. The cost of the explicit method is much
regions, especially at regions where the stresses are com- less than that of the implicit method. Due to numerical
pressive and the absolute values are large. This is a com- damping, amplitude decay is observed for the implicit
pressive model, the absolute values of tensile stresses being method.
much smaller than the compressive stresses. The differences For the slow case, the solutions are more unstable because
shown in Figs. 7d and e are all in the region of tensile high frequency numerical noise becomes more important.
stresses. The numerical damping induced in the implicit method
For this fast impact problem, the total time is usually very shows its function to remove the noise and keep the results
short, the computational cost of the explicit begin about one more accurate. Because the explicit method is conditionally
tenth that of the implicit. stable, the stable time period is much smaller than that of the
J.S. Sun et al. / Journal of Materials Processing Technology 105 (2000) 110±118 117
Fig. 7. (a) The stress component S11 of element 409; (b) the stress component S22 of element 409; (c) the stress component S33 of element 409; (d) the
maximum principal stress SP3 of the impacting disk by explicit; (e) the maximum principal stress SP3 of the impacting disk by implicit.
118 J.S. Sun et al. / Journal of Materials Processing Technology 105 (2000) 110±118
Fig. 8. (Continued ).
References
[1] ABAQUS User's Examples and Theory Manual, Version 5.7. Hibbitt,
Karlsson & Sorensen Inc., 1998.
[2] N. Rebelo, J.C. Nagtegaal, L.M. Taylor, Comparison of implicit and
explicit ®nite element methods in the simulation of metal forming
processes, in: Chenot, Wood, Zienkiewicz (Eds.), Numerical
Methods in Industrial Forming Processes, 1992, pp. 99±108.
[3] Hibbitt, Karlsson & Sorensen Inc., Application of implicit and
explicit ®nite element techniques to metal forming, J. Mater. Process.
Technol. 45 (1994) 649±656.
[4] H.M. Hilber, T.J.R. Hughes, Collocation, dissipation and `overshoot'
for time integration schemes in structural dynamics, Earthquake Eng.
Struct. Dyn. 6 (1978) 99±117.
[5] T.J.R. Hughes, The Finite Element Method Ð Linear Static and
Fig. 8. (a) The displacement U2 of node 311; (b) the stress component S22 Dynamic Finite Element Analysis, Prentice-Hall, Englewood Cliffs,
of element 173; (c) the stress component S22 of element 409; (d) the NJ, 07632, 1987.
maximum principal stress (SP3) of the slow contact disk by explicit; (e) [6] Z.H. Zhong, Finite Element Procedures for Contact Ð Impact
the maximum principal stress (SP3) of the slow contact disk by implicit. Problems, Oxford University Press, New York, 1993.