Professional Documents
Culture Documents
Abstract
A numerical algorithm to analyze an elasto-plastic contact problem using the smoothed particle hydrodynamics (SPH) method is
proposed. In order to solve the contact problem with SPH when both bodies are deformable, a variational equation based on the virtual
work principle is derived and its solution is obtained by the penalty method. Methods to find the boundary particles and to determine the
boundary normal vector are reviewed to apply to the contact algorithm. In calculating the actual penetration and penetration rate, a
method that can determine the actual normal boundary vector through the contact relationship with a contact surface is proposed. A
numerical simulation is conducted to validate the proposed method in cylindrical coordinates. A steel ball 10 mm in diameter impacting a
thin steel plate of 1 mm thickness at a high velocity such as 200 m/s is chosen to verify that the contact algorithm can be applied to the
penetration problem. The final shape results obtained by the proposed contact algorithm are more similar to the experimental results
than the conventional SPH analysis results.
r 2007 Elsevier Ltd. All rights reserved.
0734-743X/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijimpeng.2007.04.009
ARTICLE IN PRESS
S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588 579
rr
value for velocity and stress between each particle, assum- dU ri 2X si I1
¼ 2 m j W c 2 ri rj
ing that contacts occur independently as with fluid particles. dt h j ri I0
Contact problems are inherently non-linear and have a
srr
j I0 þ I2 I1
variational inequality because the actual surface on which þ 2 ri rj
these bodies meet is generally unknown a priori and has to be rj 2I 0 I0
! )
determined as part of the solution [8,9]. The penalty method syy I0 I2 srz srr
I
j i j 1
used as the solution technique in this study is used to convert þ 2 ri þ þ 2 zij þ Pij ,
rj 2I 0 r2i rj I 0
the inequality constraint into a variational equality on the
boundary. In this method, both a penetration and a ð4Þ
penetration rate are introduced and a method to determine
rz
their parameters is proposed. It is also described how to dU zi 2X si I1
determine normal vectors on the boundary, especially in ¼ 2 m j W c 2 ri rj
dt h j ri I0
cylindrical coordinates, how to judge whether the actual ! )
contact occurs or not, and how to reconstruct the actual srz
j I1 szz
i
szz
j
þ 2 ri rj þ þ 2 zij þ Pij ,
boundary normal vector that is required to compute a real rj I0 r2i rj
penetration. A penetration problem is studied to validate
ð5Þ
whether the algorithm can be applied to a problem with
fracture in cylindrical coordinates. Body forces and the rr
traction enforced on bodies are ignored. dE i 2X si r I1
¼ 2 m j W c 2 U i ri rj
dt h j ri I0
2. The method I1 1 I2
þ U rj ri rj 1 þ
I0 2 I0
2.1. Review of SPH theory rz
s I1 I1
þ i2 zij U ri U rj þ U zij ri rj
ri I0 I0
SPH is a Lagrangian method which uses particles instead zz yy
of a background mesh to compute the spatial derivative. s s I 0 I 2 Pij
þ i2 U zij zij i2 U rj rj þ , ð6Þ
Physical quantities of each particle are calculated from the ri ri 2I 0 2
relation of neighbor particles using the kernel function for
where Wc is the 2D cylindrical kernel function, Ik
interpolation. Both the kernel approximation to estimate the
(k ¼ 0,1,2,y) is the modified Bessel function, zij ¼ zjzj,
physical quantities of a particle and the particle approxima-
Uijz ¼ UizUjz, and Pij is an artificial viscosity, which is
tion to convert an integral of a continuous system into the
introduced to reduce numerical oscillation occurring when
discrete form are used for numerical calculation [4].
two particles approach each other, suggested by Monaghan
Using the concept of unitary volume and kernel
and Gingold [11]. Libersky and Petschek [3] have for-
approximation to represent the form of particle approx-
mulated the elasto-plastic model in SPH in order to
imation, expressions for a function at any point X0, u(X0),
consider the strength of materials impacting under a
and its derivative are
velocity of about 2 km/s. Total stress sab, in which
X
n compressive stress is taken to be positive, is expressed in
uðxi Þ ¼ uðxj ÞW ðxi xj ; hÞV j , (1) terms of a deviatoric stress tensor Sab and hydrostatic
j¼1
pressure P as follows:
X
n
sab ¼ Pdab S ab , (7)
ruðxi Þ ¼ uðxj ÞrW ðxi xj ; hÞV j , (2)
j¼1 where dab is Kronecker’s delta.
where W(x0x,h) is a kernel or smoothing function, h is a The equation of state (EOS) describes the relation-
smoothing length, and Vj represents the particle volume j. ship between pressure, density, and internal energy of a
The smoothing length controls the size of the compact material in a physical system. When the pressure is not too
support domain O. In order to analyze axisymmetric high, the Mie–Gruneisen equation [4] can be successfully
problems, the method proposed by Petschk et al. [10] is used to represent the deformation property of metallic
adopted. We can obtain the governing equations for the r materials.
and z directions using tensor transformation and the
symmetry of cylindrical coordinates:
2.2. Elasto-plastic model
dri 2 X mj I1
¼ ri 2 W c U ri ri rj
dt h j rj I0 The stress in a material during elastic deformation has
the relationship of linear proportion to the strain, which is
I1 called Hooke’s law. A unique strain can thus be determined
þ U zij zij U rj ri rj , ð3Þ
I0 when a stress is given. However, once a material reaches the
ARTICLE IN PRESS
580 S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588
yield state exceeding the elastic state, the plastic deformation is the variation of admissible velocity satisfying both the
process represents history or a path-dependent phenomenon. velocity condition and the strain rate–velocity relation.
A material being deformed plastically under a multi-axial Integrating over the domain O and the boundary G, it is
stress condition transitions from elasticity to plasticity at the given as
limitation of combined stress. The von Mises yield criterion is Z Z
used to check the limitation and the exponential work sab;b dva dV þ rba dva dV
hardening model as the elasto-plastic constitutive model is O
Z O
Z
applied. Because a stress state computed during any load þ rv dv dV þ ðta t̄a Þdva dA ¼ 0.
a a
ð9Þ
instant by an explicit method, for instance SPH, may not O G
strictly satisfy the yield condition, it is necessary to conduct
iterations to be able to satisfy the yield condition at a given Applying Gauss’ divergence theorem for the first term on
instant. The radial return method [12] regarded as a simple the left-hand side, we can obtain the equation of virtual
and efficient algorithm is adopted as the time integration of work. The penalty method is used to convert the inequality
the constitutive equation. constraint, which is caused by the kinematic constraint
The Prandtl–Reuss equation, called plastic flow theory and condition imposed on the contact problem, to an equality
known as the representative constitutive equation, is also constraint on the boundary. Therefore, the general varia-
used and is written as the following Eq. (8). This equation tion form can be written as the following equation, ignoring
assumes that the plastic strain increment is in proportion to the body force ba:
the deviatoric stress for an elasto-plastic material. Z Z
ab
dW ¼ sab dva;b dV þ rva dva dV
dp ¼ S ab dl, (8) O
Z O
p
where de denotes the plastic strain increment and dl þ ðl1 pn þ l2 pn Þna dv dG; ð10Þ
Gc
is a parameter that can be determined using the yield
criterion. where p is the penetration, p is the penetration rate, and l1
and l2 are the parameters for penetration and penetration
rate, respectively. Transforming the weak form of Eq. (10),
2.3. Contact theory
which is derived from the virtual work principle, into the
SPH formulation using the approximations in Eqs. (1) and
An interaction between contact bodies can be handled by
(2), we can represent it as
a particular algorithm called the slideline algorithm [13] in
numerical analysis. The main concept of this algorithm is to "
XN X X ab
impose the restriction that two bodies cannot occupy the vai rj W ij V j V i þ sj rW ij V j V i
same space at the same time. FEM or FDM imposes the i j2M i j2M i
3
condition of whether a penetration between finite element X
lattices whose consecutiveness is given on the contact þ ðl1 pn þ l1 pn Þni W ij Ai V j 5dvai ¼ 0, ð11Þ
surface occurs or not. However, SPH requires new contact j2M Bi c
conditions and consideration because the contact surface is
modeled by discontinuous particles. These characteristics where Mi and M Bi c denote the neighbor particles and
are illustrated schematically in Fig. 1. neighbor boundary particles within a 2-h distance of
The variational equation based on the virtual work particle i, respectively, and na is the boundary normal
principle can be derived to obtain the numerical formu- vector. Therefore, the momentum equation acting between
lation for SPH [14]. In order to obtain the weak form for two bodies in contact can be written as the following:
the contact problem, we can multiply the equation of
motion and the boundary constraint equation by dva, which mi vi ¼ Fint con
i Fi , (12)
Fig. 1. Comparison of contact concept between FEM and SPH: (a) slideline contact algorithm in FEM and (b) contact in SPH.
ARTICLE IN PRESS
S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588 581
i,vi, using Eq. (12) and determine the velocity and position where MiR denotes real particles, MiI imaginary particles,
by time integration. volume Vj is dr dz, and Wij does not use the cylindrical
Gaussian type but 2D cubic B-spline type. Since the
2.4. Contact algorithm imaginary particles are symmetrical to the z-axis, for the
imaginary particle j, the distance between particles is ri+rj
Contact between particles is between two circles of radius instead of rirj. If the boundary particle is detected through
d/2, which is easily determined by checking the inter- the same method in 2D, we can determine the boundary
particle distance, and only the interaction of two particles is unit normal vector, similarly to Eq. (18).
considered at any one time. This approach adopted in this
paper has similarities with the pinball contact algorithm [7] ni ¼ ðrc=jrcjÞi ,
or Reimann solution [2]. This can be extended to hyper- where
velocity large deformation analysis as well as to elasto-
X X
plastic deformation, and it can utilize the properties of SPH ðrcÞi ¼ cj rW ij V j cj rW ij V j . (20)
particles. j2M R j2M Ii
i
Whether boundary particles of two bodies are really in
contact or not must be determined to construct the The boundary normal vector determined by this method
relationship between elasto-plastic bodies in SPH. The in a cylindrical model is illustrated in Fig. 2(b). Once the
boundary normal vectors for boundary particles of each normal vector of a boundary is determined, it should be
body should be calculated, and whether or not the particles checked whether boundary particles actually contact or not.
ARTICLE IN PRESS
582 S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588
Fig. 2. (a) Calculation of boundary normal vector in cylindrical coordinate. (b) Normal vectors on boundary of material (circle) in cylindrical coordinate.
Fig. 3. Detection of real contact: a case of (a) contact and (b) no contact.
For this purpose, the following criterion is first performed: where y denotes the angle formed with the normal
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
vectors of particles i and j. If the angle obtained by
di dj 2 di þ dj 2 Eq. (22) is less than yc, the average normal vector of
jrij jp Max ; þ . (21)
2 2 2 the particle, nav, is computed as in Eq. (23a). If the
maximum among three values is rij ni, then nav ¼ ni.
This criterion allows for unequal representative diameters
On the other hand, if the angle is larger than yc, it is
and for lateral offsets between approaching particles, like the
calculated as in Eq. (23b). If the contact surface is not
examples shown in Fig. 3. In this figure, (a) represents actual
very irregular, ycE701 is considered a reasonable criterion
contact, while (b) shows a case of no contact.
angle.
However, the above criterion is only to judge actual
The average normal vector determined by the above
contact and not to get the average normal vector nav, which
method is used to compute the contact force in Eq. (16). In
is the normal vector for an actual contact boundary. In this
this equation, the penetration is a quantity produced by
paper, the following method is suggested to determine this
moving particles from step n, at which they satisfy the
vector:
contact criterion, to step n+1. The penalty value will be
jyj ¼ ffðni ; nj Þ, (22) imposed only if it has a positive value. On the other hand, if
(i)
ni nj ni nj
jyjoyc if max rij ni ; rij nj ; rij then nav ¼ ni ; nj ; , (23a)
jni nj j jni nj j
Table 1
Analysis conditions for cylinder impact test by SPH
Table 4
Material properties and constants for Mie–Gruneisen EOS of steel
7850.0 3600.0 1.90 1.70 0.30 200.0 77.0 600.0 275.0 0.36
Fig. 12. Simulated penetration with time using conventional SPH (RSPH).
ARTICLE IN PRESS
S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588 587
Fig. 13. Simulated penetration with time using the contact algorithm (CSPH).
ARTICLE IN PRESS
588 S. Seo et al. / International Journal of Impact Engineering 35 (2008) 578–588
Fig. 14. Comparison of simulated final deformed shapes and experimental (not to scale): (a) conventional SPH (RSPH), (b) SPH added to the contact
algorithm (CSPH), and (c) experimental results.