Beruflich Dokumente
Kultur Dokumente
1, 2008 31
Yucheng Liu
Department of Mechanical Engineering,
University of Louisville,
Louisville, KY 40292, USA
Email: y0liu002@louisville.edu
1 Introduction
Finite element analysis (FEA) algorithm has been integrated with many engineering
software tools for solving most engineering problems. An efficient computer aided
engineering (CAE) tool for static, transient, and dynamic analysis can be extensively
applied in engineering, including model design, optimising, and simulation. ANSYS and
LS-DYNA are the two general-purpose finite element analysis tools that are widely used
for most engineering areas. ANSYS is developed by ANSYS, Inc., and generally used for
numerically solving a wide variety of engineering problems, which include: static and
dynamic structural analysis, heat transfer and fluid problems, and acoustic and
electro-magnetic problems. LS-DYNA is developed by LSTC and mostly applied for
engineering problems that include non-linear finite element analyses, such as automotive
crashworthiness, occupant safety, and sheet metal forming. Linde et al. (2006) and Rust
and Schweizerhof (2003) discussed general properties of ANSYS and LS-DYNA and
demonstrated the benefit of using LS-DYNA to perform quasi-static limit load analyses
in the literatures, the researchers recommended a proper combination of ANSYS and
LS-DYNA, and showed how to apply it to perform dynamic analysis by static analysis
steps. This paper mainly focuses on the capabilities of both software of solving structural
analysis problem, including static and dynamic analyses and exposes the reason by
hand-solving a simple two degrees of freedom problem.
ANSYS is an implicit FE program which uses a static load step with the default
incremental iterative solution. In ANSYS, the Newmark time integration method is
employed for the solution of the transient dynamic equilibrium equation [equation (1)].
This method uses finite difference expansions in the time interval ∆t, which will be
demonstrated in following text. ANSYS also uses the Newton-Raphson method to solve
nonlinear problems, in which method, applied load is divided into a series of load
increments and the load increments then are applied over several load steps (Advance
Analysis Technique Guide). Contact analysis is performed with penalty and Lagrange
methods, where only two potential contact surfaces must be specified.
LSDYNA is an explicit FE program and is designed for transient dynamic analysis of
highly nonlinear problems. In LS-DYNA, the transient dynamic equilibrium equation is
solved by the central difference method. Similar to ANSYS, LS-DYNA also uses
Newton-Raphson method to solve the nonlinear problems, including the contact and
impact problems. For the contact problems, three distinct methods are used to handle
such problems, includes the kinematic constraint method, the penalty method and the
distributed parameter method.
Among all the theories and approaches implemented in the two FE programs, the time
integration method is the crucial scheme that decides different properties of both
programs in performing different structural analyses, including static and transient
dynamic analysis. Following example shows how ANSYS and LS-DYNA are used to
prepare static and dynamic analyses and the theoretical parts are discussed later by
exposing different time integration processes.
3 Example
This section presents a simple impact problem to compare general properties of implicit
analysis (ANSYS) and explicit analysis (LS-DYNA). In this example, a thin-walled beam
which is a hollow-box structure enclosed with all surfaces surfers quasi-static load and
dynamic load separately. This beam is made of mild steel, whose Young’s modulus is
200GPa, Poisson’s ratio is 0.3, yield stress is 250MPa and hardening modulus is 630MPa
(Figure 1). ANSYS and LS-DYNA are used to perform the whole static and dynamic
analyses, from creating the FE model till generating the final results.
ANSYS and LS-DYNA used for structural analysis 33
Table 1 lists the static analysis results obtained from ANSYS, however, this model
entirely collapsed and yielded huge deformation if running the analysis using LS-DYNA,
which could not be correct. To affirm this judgment, the FE model’s stiffness is evaluated
using popular theories. According to classic mechanics of materials, the axial, bending,
and torsional stiffness of this thin-walled steel beam are:
K axial = EA / L (1)
K bending = 3EI / L3
(2)
K torsion = GJ / L (3)
The material of beams is mild steel, whose Young’s modulus E is 200GPa and shear
modulus G is 79GPa. For thin-walled square sectional beam with its width b and
thickness t, its area moment of inertia and polar moment of inertia are:
I = (b 4 − (b − 2t ) 4 ) / 12 (4)
J = (b 4 − (b − 2t ) 4 ) / 6 (5)
Therefore, with given beam parameters b = 60mm, t = 1.5mm, and L = 300mm, the axial
stiffness should be around 234000 KN/m; the bending stiffness should be around 4452
KN/m; and the torsional stiffness should be about 106KN-m/rad. Compare with Table 1,
it can be found that both axial and bending stiffness results obtained from computer
analyses fully agree to the theoretical evaluation. Thus, the accuracy of the ANSYS
results is verified through the theoretical evaluation.
Table 1 Static analysis results from ANSYS and LS-DYNA
ANSYS LS-DYNA
Axial stiffness (KN/m) 2.4×105 Fail to obtain meaning stiffness
Bending stiffness (KN/m) 4350 and frequencies using LS-DYNA
Note: ANSYS, LS-DYNA, from Zhong’s work (Zhong, 1993) (from top to bottom).
36 Y. Liu
Figure 3 Analysis results from LS-DYNA (a) displacements, (b) crushing forces, (c) absorbed
energies (see online version for colours)
160
140
120
Displacement (mm)
100
80
60
40
20
0
0 0.002 0.004 0.006 0.008 0.01 0.012
Tim e (sec)
(a)
90
80
70
60
Force (kN)
50
40
30
20
10
0
0 0.002 0.004 0.006 0.008 0.01 0.012
Tim e (sec)
(b)
7
5
Energy (kJ)
0
0 0.002 0.004 0.006 0.008 0.01 0.012
Tim e (sec)
(c)
ANSYS and LS-DYNA used for structural analysis 37
As mentioned in the text, the primary differences between the ANSYS and the
LS-DYNA are different solution procedures and time integration methods they are using.
The ANSYS is an implicit solver which applies Newmark method whereas the
LS-DYNA is an explicit solver and employs central difference method. This section
thoroughly demonstrates the different integration methods used by ANSYS and
LS-DYNA, and then continues to discuss and compare the attributes of both implicit and
explicit solution procedures.
the Newmark integration scheme assumes that the state of displacement, velocity, and
acceleration at time t and t + Δt is:
• • •• ••
U i +1 = U i + [(1 − δ ) U i + δ U i +1 ]Δt (2)
• 1 •• ••
U i +1 = U i + U i Δt + [( − α ) U i + α U i +1 ]Δt 2 (3)
2
where α and δ are parameters that can be determined to obtain integration accuracy and
stability, in the Newmark method α = 1/4 and δ = 1/2. In solving the displacement,
velocity, and acceleration at time t + Δt, the equilibrium equation (1) can be written as:
•• •
M U i +1 + C U i +1 + KU i +1 = Ri +1 (4)
••
To solve (4), the acceleration U i +1 is first solved from (3) in terms of the displacement
•• •• •
U i +1 and then substituted into (2). Thus, the acceleration U i +1 and velocity U i +1 are
••
solved, both in terms of the unknown displacement U i +1 only. The solved U i +1 and
•
U i +1 then are substituted into (4) to entirely solve the displacement U i +1 . Afterwards,
•• •
U i +1 and U i +1 then can be calculated based on the solved U i +1 using (2) and (3).
• 1
Ui = (U i +1 − U i −1 ) (6)
2Δt
which include the states at time t, t + Δt, and t – Δt. For the governing equilibrium
equation at time t
•• •
M U i + CU i + KU i = Ri (7)
• • •• ••
xi +1 = xi + [(1 − δ ) x i + δ x i +1 ]Δt (11)
• 1 •• ••
xi +1 = xi + x i Δt + [( − α ) x i + α x i +1 ]Δt 2 (12)
2
Substitutes equations (11) and (12) into (10) and using α = 1/4 and δ = 1/2, following
relation can be established as:
ANSYS and LS-DYNA used for structural analysis 39
⎡ •• ⎤ ⎡ •• ⎤
⎢ x i +1
⎥ ⎢ xi ⎥
⎢• ⎥ ⎢• ⎥
⎢ xi +1 ⎥ = A ⎢ x i ⎥ + Lri +1 (13)
⎢x ⎥ ⎢x ⎥
⎢ i +1 ⎥ ⎢ i⎥
⎣ ⎦ ⎣ ⎦
where
⎡ β β ⎤
⎢ −0.25 β − − 2 ⎥
Δt Δt
⎢ ⎥
⎢ β ⎥
A= Δt [ 0.5 − 0.125β ] 1 − 0.5β − (14)
⎢ 2 Δt ⎥
⎢ 2 ⎥
⎢ Δt [ 0.25 − 0.0625 β ] Δt [1 − 0.25 β ] 1 − 0.25 β ⎥
⎢⎣ ⎥⎦
and
1 1
β =( + ) −1 (15)
ω Δt 4
22
xi +1 = ( 2 − ω 2 Δt 2 ) xi − xi −1 + Δt 2 ri (19)
Substitutes this A into the equation Au = λu, and solve for the inequality λ ≤ 1 . From
the mathematical results it can be found that the inequality can be satisfied only when Δt
≤2/ωn (where ωn is the largest natural circular frequency), which shows the explicit
analysis is conditionally stable. The severe time step restriction is one of primary
shortcomings in the use of the explicit integration method.
Besides different time integration methods they used, the explicit algorithm also employs
return mapping method to integrate the differential equations between the times t and t +
∆t. This is another significant reason why the explicit algorithm is so economic compares
to the implicit one. As concluded in previous literatures (Lee et al., 1999; Plesek and
Korous, 2002; Leclere et al., 2004; Zhang, 1995), the return mapping method is most
effective, robust, and unconditionally stable compared to other integrate methods. In the
FE analyses, stress updates take place at the Gauss integration points of used finite
elements, where total and incremental strains are calculated. The return mapping
algorithm is to calculate the stress and strain state at time t + ∆t based on the known
stress and strain state at time t. By applying this method, the stress is first evaluated by
solving elastic equations. Afterwards, the obtained stress is taken as an initial condition
for plastic relaxation equations and several plastic corrections are performed in order to
optimise the stress and strain state until the end of the time increment t + ∆t.
From above introduction, it can be seen that the return mapping algorithm is defined
implicitly therefore is unconditionally stable. Meanwhile, the numerical iteration that is
essence of the implicit analysis program is effectively avoided in the return mapping
method. This endows the explicit program solver lots of advantages in solving transient
dynamic analysis because the complicated numerical iteration can be avoided.
6 Conclusions
This paper compares two popular FE programs: ANSYS and LS-DYNA, which are
representatives of implicit and explicit analysis solvers, respectively. From the examples
discussed in this article, it is concluded that in running static analysis, ANSYS should be
used because the implicit analysis is unconditionally stable for any time increment Δt.
The static analysis allows a high Δt during its time integration, whereas the high Δt may
cause unstable solutions if using explicit analysis. On the other side, those dynamic
transient analyses such as high-speed impact problems require small Δt in order to
correctly calculate the varying material properties and geometrics at any time t.
LS-DYNA is best for such analyses, and the explicit analysis can solve such problems
efficiently because it uses central difference method and return mapping algorithm,
therefore the expensive matrix inversions and numerical iteration can be avoided.
Moreover, the small Δt may even cause failure to convergence during the matrix
inversions if using implicit analysis. Summarily, ANSYS and LS-DYNA should be chose
for appropriate FE problems carefully to best utilise their merits.
ANSYS and LS-DYNA used for structural analysis 41
References
Advanced Analysis Techniques Guide, ANSYS, Inc.
Lee, S.W., Yoon, J.W. and Yang, D.Y. (1999) ‘Comparative investigation into the dynamic explicit
and the static implicit method for springback of sheet metal stamping’, Engineering
Computations, Vol. 16, No. 3, pp.347–373.
Leclere, G., Neme, A., Cognard, J.Y. and Berger, F. (2004) ‘Rupture simulation of 3D elastoplastic
structures under dynamic loading’, Computers & Structures, Vol. 82, pp.2049–2059.
Linde, P., Schulz, A. and Rust, W. (2006) ‘Influence of modelling and solution methods on the
post-buckling behaviour of stiffened aircraft fuselage panels’, Composite Structures, Vol. 73,
pp.229–236.
LS-DYNA Theoretical Manual, Livermore Software Technology Corporation
Plesek, J. and Korous, J (2002) ‘Explicit integration method with time step control for
viscoplasticity and creep’, Advances in Engineering Software, Vol. 33, pp.621–630.
Rust, W. and Schweizerhof, K. (2003) ‘Finite element limit load analysis of thin-walled structures
by ANSYS (implicit), LS-DYNA (explicit) and in combination’, Thin-Walled Structures,
Vol. 41, pp.227–244.
Zhang, Z.L. (1995) ‘Explicit consistent tangent moduli with a return mapping algorithm for
pressure-dependent elastoplasticity models’, Computer Methods in Applied Mechanics and
Engineering, Vol. 121, pp.29–44.
Zhong, Z-H. (1993) Finite Element Procedures for Contact-Impact Problems, Oxford Science
Publications.
Appendix
Example
Given a simple mass-spring system as shown in Figure 4, K1 = 3N/m, K2 = 1.5N/m,
K3 = 1.5N/m, M1 = 1.5kg, M2 = 0.75kg, F1 = 0, and F2 = 7.5N. The governing
equilibrium equation for this system:
0 ⎤ ⎢⎡U 1 ⎥⎤ ⎡ K1 + K 2
••
⎡ M1 − K 2 ⎤ ⎡U1 ⎤ ⎡ F1 ⎤
⎢ + =
⎣ 0 M 2 ⎦⎥ ⎢ •• ⎥ ⎢⎣ − K 2 K 2 + K 3 ⎥⎦ ⎣⎢U 2 ⎦⎥ ⎣⎢ F2 ⎦⎥
⎣U 2 ⎦
which is
⎡ •• ⎤
⎡1.5 0 ⎤ ⎢U 1 ⎥ ⎡ 4.5 −1.5⎤ ⎡U1 ⎤ ⎡ 0 ⎤
⎢ 0 0.75⎥ ⎢ •• ⎥ + ⎢ −1.5 =
3 ⎥⎦ ⎢⎣U 2 ⎥⎦ ⎢⎣ 7.5⎥⎦
⎣ ⎦ U2 ⎣
⎣ ⎦
42 Y. Liu
It is assumed to calculate the response of this system in 12 steps, the Newmark and
central difference methods are used separately.
Before solving the problem, Δt will be determined. Since both methods will use the
same Δt to compare their characteristics, an appropriate Δt has to be calculated to promise
the stability of the central difference method. Generalised eigen problem is firstly solved:
⎡ 4.5 −1.5⎤ ⎡1.5 0 ⎤
⎢ −1.5 ⎥ ϕ = ω2 ⎢ ⎥ ϕ ⇒ ω1 = 2, ω 2 = 5 ⇒ T1 = 4.45, T2 = 2.8
2 2
⎣ 3 ⎦ ⎣ 0 0.75⎦
then Δt can be taken as T2/10 = 0.28 which is far below than the critical value
2/ωn = 0.89.
Solution
a Using Newmark method
•• •
The first step is to calculate U t = 0 , we assume that U t = 0 = 0 and U t = 0 = 0, and then
use
⎡1.5 0 ⎤ •• ⎡ 4.5 −1.5⎤ ⎡U1 ⎤ ⎡ 0 ⎤ •• ⎡0 ⎤
⎢ 0 0.75⎥ U t = 0 + ⎢ −1.5 ⎥ ⎢ ⎥ = ⎢ ⎥ ⇒ U t =0 = ⎢ ⎥
3 ⎦ ⎣U 2 ⎦ ⎣7.5⎦
⎣ ⎦ ⎣ ⎣7.5⎦
Substitutes K, M, R and Δt into equations (2), (3), and (4), we can have:
• • •• ••
U i +1 = U i + 0.14 U i + 0.14U i +1
After performing the iterative calculations, the final results can be obtained.
Time Δt 2Δt 3Δt 4Δt 5Δt 6Δt
U1 0.00673 0.0505 0.189 0.485 0.961 1.58
U2 0.364 1.35 2.68 4.00 4.95 5.34
Time 7Δ t 8Δt 9Δt 10Δt 11Δt 12Δt
U1 2.23 2.76 3.00 2.85 2.28 1.40
U2 5.13 4.48 3.64 2.90 2.44 2.31
ANSYS and LS-DYNA used for structural analysis 43
The two sets of results are compared and plotted in Figures 5 and 6. From the figures, it
can be seen that for this 12-step example, both methods yield very close results and
deliver the same convergent solution.
Figure 5 The displacement for the first degree of freedom (see online version for colours)
3.5 implicit
3 explicit
displacement
2.5
2
1.5
1
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12
time step
44 Y. Liu
Figure 6 The displacement for the second degree of freedom (see online version for colours)
6 implicit
explicit
5
displacement
0
1 2 3 4 5 6 7 8 9 10 11 12
time step
Nomenclature
[M ] Mass matrix
[C ] Damping matrix
[K ] Stiffness matrix
U Displacement vector
• Velocity vector
U
•• Acceleration vector
U
R Load vector of system