Implementation of a piecewise Drucker-Prager

model in Abaqus
Kwangmin Lee *, Karuppasamy Pandian Marimuthu *, Hyungyil Lee *
* Department of Mechanical Engineering, Sogang University, Seoul 04107, Rep. of Korea

Abstract: Due to size effect in nanotechnology, an appropriate pressure-dependent model is

required to define the material yielding in ceramics. Current pressure-dependent models, such as
the Mohr-Coulomb and Drucker-Prager (DP) models, have specific shapes and thus cannot be
flexibly applied to ceramic materials. This study develops a constitutive model, an integration
algorithm, and consistent tangent operators for a piecewise Drucker-Prager (PDP) model. The
constitutive model with considering strain hardening with yield flows is derived. The integration
algorithm is developed based on return mapping (to smooth portion, apex and corner). The
consistent tangent operators are established for each return mapping case. The PDP model is then
implemented in commercial finite element code (using Abaqus) by developing a user material
subroutine (UMAT). This UMAT is verified for bilinear and extended DP models through finite
element analysis (FEA) of a hydrostatic test with a single element. We expect that the UMAT for
the PDP model can be used to describe the plastic behavior of pressure-dependent materials
accurately.

Keywords: Piecewise Drucker-Prager yield model, Numerical implementation, Constitutive

modeling, UMAT, FEM.

1. Introduction
The classical Drucker-Prager (DP) model is a pressure-dependent yield model in which yield
strength and hydrostatic pressure are linearly related (Drucker and Prager, 1952). In addition to the
linear DP model, extended DP models also have hyperbolic and general exponent forms (Abaqus,
2013). However, since existing models have a limitation that cannot be modified to fit the yielding
model of any material, a piecewise Drucker-Prager (PDP) criterion is required. Most commercial
finite element analysis (FEA) programs do not have a built-in PDP model, although a simple PDP
model where strain hardening is not implemented is available in the commercial finite element
software, Autodyn (Ansys, 2015). To establish and popularize a user subroutine for an enhanced
PDP model, more careful and detailed research about constitutive equations, integration algorithm
and consistent tangent operators of the PDP model is required.
The present study develops an elasto-plastic constitutive model, an integration algorithm based
on return mapping, and consistent tangent operators for PDP model in which strain hardening can
be considered. The return mapping method and tangent operators are developed at the intersection
where piecewise linear DP models meet. Based on the developed constitutive model and

2018 Science in the Age of Experience 1

http://www.3ds.com/events/science-in-the-age-of-experience
numerical implementation techniques, a user material subroutine (UMAT) for the PDP model is
developed and then verified by using single element triaxial finite element (FE) simulations.

2. Piecewise Drucker-Prager model

In some materials, an extended pressure-dependent model is needed to describe material’s
plastic behavior (Ma et al., 1998; Milani and Lourenco, 2009). However, since it is difficult to
directly make a UMAT for the extended pressure-dependent model by using its constitutive
equations due to the complexity of update formula, we develop PDP model by combining several
linear DP models to describe extended DP model.

2.1 Piecewise Drucker-Prager (PDP) constitutive model

In a PDP model (Fig. 1), multiple linear DP models are combined.

J 2 s
linear Drucker-Prager model

corner
smooth portion
piecewise Drucker-Prager model

apex

p
Fig. 1 Schematic comparison of linear and piecewise Drucker-Prager models

The yield function of the PDP model is defined by

Φ i  , ci   J 2  s ( )   i p     i ci  i  1, 2, , n (1)

Here

1
J2  s    p I I  1 1 1 0 0 0 
T
s:s ; ; (2)
2

where s is the deviatoric stress,  is the stress, p is the hydrostatic pressure and ci is the cohesion
of the material. The ηi and  i are constants derived from approximation to the piecewise Mohr-
Coulomb (PMC) yield model (Fig. 2). The formulas for the outer edges are
2018 Science in the Age of Experience 2
http://www.3ds.com/events/science-in-the-age-of-experience
 6sin i 6cos i
i  , i   i  1, 2, , n (3)
3  3  sin i  3  3  sin i 

whereas the formulas for the inner edges are

6sin i 6cos i
i  , i   i  1, 2, , n (4)
3  3  sin i  3  3  sin i 

where  i is friction angle in the PMC model.

 3

piecewise Drucker-Prager
piecewise Mohr-Coulomb
(outer edge)

piecewise Drucker-Prager
(inner edge)

 1  2

Fig. 2 -plane section of piecewise Mohr-Coulomb surface and

piecewise Drucker-Prager approximations

In the associative DP model, the yield function of Eq. (1) is employed as flow potential;
correspondingly, three distinct plastic flows are described as follows.

(i) The plastic flow on the piecewise smooth portions of the yield surface is

 p   Ni  i  1, 2, , n (5)

where  p is plastic strain rate,  is the plastic multiplier and N is the flow vector. The flow vector
is
2018 Science in the Age of Experience 3
http://www.3ds.com/events/science-in-the-age-of-experience
  i 1 
Ni   s i I  i  1, 2, , n (6)
  2 J 2 s  3

(ii) At the corner where two linear DP models meet, the plastic flow is

 p   i Ni   i 1 Ni 1  i  1, 2, , n  1 (7)

where N i and N i 1 are the flow vector of the ith and i+1th linear DP model.

(iii) The plastic flow vector at the apex singularity is a sub-gradient of the first linear DP model
flow potential Ψ1. The effective plastic strain rate, in this case, is (de Souza Neto et al., 2008)

 p  1 (8)

Since volumetric plastic strain rate  vp is given as

 vp   1 (9)

the effective plastic strain rate can thus be expressed as

1 p
p v (10)
1

To avoid excessive dilatancy, the non-associative flow rule is also used in the present PDP
model as in the linear DP model. Since the PDP model is an approximation of the PMC model, the
PDP yield function, as flow potential, is used with the dilatancy angle  i instead of the friction
angle  i (i <  i); that is

 i   , ci   J 2  s ( )   i p  i  1, 2, , n (11)

where  i is obtained by replacing  i with  i in the definition of ηi given by Eq. (3) or (4). In other
words, if the outer cone approximation to the PMC criterion is applied, then

6sin i
i   i  1, 2, , n (12)
3  3  sin i 

while if the inner cone approximation is employed, then

2018 Science in the Age of Experience 4

http://www.3ds.com/events/science-in-the-age-of-experience
 6sin i
i   i  1, 2, , n (13)
3  3  sin i 

In the non-associative DP model, by using i and  i instead of i and i in Eqs. (5)-(10),

corresponding flow vectors can be obtained.

2.2 Integration algorithm for the PDP model

The general return-mapping update formula for the stress tensor of materials is

1  D :  
 n 1   ntrial e p
(14)

where D e is an elasticity matrix. Since the flow vectors are different according to the location of
the PDP model, three explicit forms exist for the return-mapping algorithm.

2.2.1 Return to the piecewise smooth portions

On the piecewise smooth portions, the flow vector is defined, as described in section 2.1. Then,
plastic strain increment is

 1  
  p   Nin 1    s n 1  i I   i  1, 2, , n (15)
 2 J 2 s 3 
 

The corresponding updated stress is

n 1    2G  N d  n 1  K  N v  n 1 
 n 1   trial  i i 

 G K   i  1, 2, , n (16)
n 1  
  trial  s n 1  i I 
 2 J 2  s  3 


where G is the shear modulus, K is the bulk modulus, N d is the deviatoric component of flow
vector, and N v is the volumetric flow vector. Eq. (16) can be simplified by noting that the
following equality holds due to the definition of J2
trial
s n 1 s n 1

J 2  s n 1  J 2  s ntrial
1 
(17)

Then, substituting Eq. (17) into Eq. (16) provides the updated stress, expressed as

2018 Science in the Age of Experience 5

http://www.3ds.com/events/science-in-the-age-of-experience
  
 G K i 
1  
 n 1   ntrial 1 
s ntrial I  i  1, 2, , n (18)
2 
 2 J s trial 3 
 

The components of updated deviatoric stress sn+1 and hydrostatic stress pn+1 are then

 
G  trial
s n 1  1  s ; 1  K i 
pn 1  p ntrial  i  1, 2, , n (19)
  n 1
J 2  s ntrial
1 
 

The consistency condition is

Φ in +1 J 2  s n 1   i pn 1   i ci  np1   0  i  1, 2, , n (20)

Here the update effective plastic strain is

 np1   np   P (21)

with

 P    (22)

Applying Eq. (19) to the consistency condition yields the following equation for Δγ:

Φi     J 2  s ntrial
1   G  i  p n 1  Ki     i ci   n   i    0
trial p
 i  1, 2, , n (23)

By solving Eq. (23), the stress is updated by Eq. (19).

2.2.2 Return to the apex

At the apex, the consistency condition of Eq. (20) in the case of i =1 is reduced to

c1  np   p
 
1
1  K  v  0
 p ntrial p
(24)
1

Further, with the introduction of the discretized form of Eq. (10) for the non-associative DP model
to Eq. (24), the final return-mapping equation for the DP apex can be obtained as

r   np   c1  np  1 p    p ntrial
1  K  v  0
p
(25)

2018 Science in the Age of Experience 6

http://www.3ds.com/events/science-in-the-age-of-experience
where

1  1 /1 ,   1 /1 (26)

This is geometrically shown in Fig. 3. After the solution of Eq. (25) is obtained for  vp , the
updated stress and effective plastic strain are obtained as

 np1   np  1  vp
 n 1   p ntrial
1  K  v  I
p (27)

J 2 s

 K  vp I
pntrial
+1 n+1= pn+1 I p

Fig. 3 Piecewise Drucker-Prager model; return mapping to apex

2.2.3 Return to the corner

A pressure at a corner, where the ith and i+1th linear DP models are merged, is defined as critical
pressure pc. If p > pc, the ith linear DP model is applied, and if p ≤ pc, the i+1th linear DP model is
applied to the material.
On return to the piecewise smooth portions, pn+1 is always smaller than pntrial +1 by Eq. (19). Thus,

in the PDP model, pn+1 can be smaller than pc even when pntrial +1 > p c , which is inappropriate return
mapping (Fig. 4a). To solve this problem, two plastic multipliers ( and  ) for the ith and i+1th
i i+1

linear DP models (which may be nonzero) are used. Then, the incremental plastic strain is

 p    i N i   i N i   i 1 N i 1 (28)

where N i and N i 1 are the normal vectors to the ith and i+1th linear DP models, respectively. Recall
that the general updated return-mapping formula for the stress tensor is

1  D :  
 n 1   ntrial e p
(29)

2018 Science in the Age of Experience 7

http://www.3ds.com/events/science-in-the-age-of-experience
With Eq. (28), Eq. (29) is expressed as

1  D :   N n 1
 n 1   ntrial e i i
(30)

Then, the corresponding updated formula is

1     2G  N d  n 1  K  N v  n 1 
i 
 n 1   ntrial i i
(31)

The components of updated deviatoric stress sn+1 and hydrostatic stress pn+1 are then


 G  i   i 1 
s n 1  1 
  strial
 J  s ntrial  n 1
1  (32)
 
1  K i   i 1  
i 1
pn 1  pntrial i

At the corner, the updated stresses are such that the equation of the ith linear DP model, i
and the equation of the i+1th linear DP model,  i+1are simultaneously fulfilled (Fig. 4b).
These two equations have to be solved for i and i+1

Φi   i ,  i 1   J 2  s ntrial
1   G    
i i 1
  i  pntrial1  K (i  i  i 1 i 1 ) 
 i ci  np1   p

(33)
Φi 1   i ,  i 1   J 2  s ntrial
1   G    
i i 1
  i 1  pntrial1  K (i  i  i 1 i 1 ) 
 i 1 ci 1  np1   p 

Note that principle of this return mapping to the corner is similar to the modified DP/cap model,
although when checking plastic admissibility in the PDP model, the pressure should also be
considered with the yield function. This is because, in the modified DP/cap model, yield stress
monotonically decreases after the corner (de Souza Neto et al., 2008), but in the PDP model, yield
stress can increase.

2018 Science in the Age of Experience 8

http://www.3ds.com/events/science-in-the-age-of-experience
 J 2 s

 ntrial
+1
ith
linear
Drucker-Prager model

i+1th linear
Drucker-Prager model

pntrial
+1 pc pn+1 p
(a)

J 2 s

 ntrial
+1
ith
linear
Drucker-Prager model

i+1th linear
Drucker-Prager model

pntrial
+1 pn+1 = pc p

(b)

Fig. 4 Piecewise Drucker-Prager model; (a) inappropriate return mapping,

(b) appropriate return mapping

2.2.4 Selection of the appropriate return mapping

Three return mappings are required in the PDP model. The selection procedure is summarized
as follows. First, the values of pn+1 and pc should be compared. If pn+1 ≥ pc and

J 2  s n 1   J 2  s ntrial
1   G   0 (34)

2018 Science in the Age of Experience 9

http://www.3ds.com/events/science-in-the-age-of-experience
then return mapping to the piecewise smooth portion is applied. In the above requirements, if
condition pn+1 ≥ pc changes as pn+1  pc, then return mapping to the corner should be applied.
Otherwise, the return mapping to the apex has to be applied.

2.3 Consistent tangent operator

The elastoplastic tangent operator of the linear DP model is used in the PDP model. The
elastoplastic tangents associated with the linear DP model have three possible forms. On the
piecewise smooth portion return, associated elastoplastic tangent consistent is (de Souza Neto et
al., 2008)

     
Dep  2G 1   I d  2G   GA  DD
 2  ed trial   2  ed trial   i  1, 2, , n
 n 1   n 1  (35)
 2GAK i D  I  i I  D   K 1  K i i  I  I

where  ed is the deviatoric component of elastic strain, D and A is defined by

 ed trial
n 1 1
D ; A (36)
 ed trial
n 1
G  K
i i  i H i
2

At the apex, the associated elastoplastic tangent consistent is

 K 
D ep  K 1   II  i  1, 2, , n (37)
 K   i i H i 

At the corner, due to the complexity of the formulas in the subsequent derivations, we define
several parameters as follows:

const1  G  K  i i  ( i ) 2 H
const2  G  K  i i 1   i i 1H
(38)
const3  G  K  i 1 i   i i 1H
const4  G  K  i 1 i 1  ( i 1 ) 2 H

2018 Science in the Age of Experience 10

http://www.3ds.com/events/science-in-the-age-of-experience
 1
Q
 const1  const4  const2  const3
V  2 K i 2i 1   1   2i i H   1   
W  G i i   11     K 2i 2i 1  2  1   i i 2i H    2      (39)
X  K i i 1     1  H  2i   1
2

Y  2G i   1  K 2i i 1  2  1  H i 2i  1   2   2 

where H is the hardening modulus. Then, the associated elastoplastic tangent consistent is

   i   i 1      i   i 1  
D ep  2G 1   I  2G   G Q X D  D
2  d n 1   2  d n 1
d
 e trial e trial
  i  1, 2, , n  1 (40)
 2G Q K  Y D  I  V I  D   K 1  K W Q  I  I

3. Constitutive programming
Abaqus (2013) provides a useful user subroutine interface called UMAT that allows one to
define complex or novel constitutive models that are not available with the built-in Abaqus
material models. UMATs are written as FORTRAN code and then linked and compiled by Abaqus
during numerical simulations.
We develop the UMATs for linear, bilinear and piecewise DP model for implementing the
plastic behavior of pressure dependent materials, and those are available in Mendeley Data’s
‘UMATs for linear, bilinear and piecewise Drucker-Prager models, Lee et al. (2018)’.

4. UMAT program verification

The UMAT for both linear and piecewise DP models are verified. The UMAT for the linear DP
model is compared with the constitutive equations of several materials. The UMAT for PDP
model is verified by comparing the FEA results with the extended DP criterion

4.1 FE model
A three-dimensional (3D) FE model with single 8-node elements is created for single element
triaxial test simulation by using commercial software Abaqus 6.13 program (Abaqus, 2013). The
boundary conditions about the xy-plane, xz-planes and yz-planes are applied on five surfaces to fix
the element. Pressure is applied to the single extra surface of the xy-plane.

2018 Science in the Age of Experience 11

http://www.3ds.com/events/science-in-the-age-of-experience
4.2 Linear Drucker-Prager model
A linear DP model subroutine of the HYPLAS program (de Souza Neto et al., 2008) is used to
develop the UMAT for the linear DP model.
Various pressure-dependent material properties (Table 1) obtained from the literature and are
used in the single element FE model. The zero-pressure yield (equals the cohesion of the material)
o is related to the uniaxial compression yield strengths yc, if hardening is defined by the yc, as

 1 
 o  1  tan   yc (41)
 3 

where  is friction angle of the material. The numerical results are then compared with the
constitutive equations of the linear DP model for various materials, as shown in Fig. 5. The result
shows that after yielding begins, the von Mises stress  from the UMAT models is identical to
that of the linear DP models.

Table 1 Material properties of pressure dependent materials obtained from the literature
material Young’s modulus E (GPa) Poisson’s ratio  yc (GPa)a  (°)a
Starphire 72.1b 0.222b 2.00 49
b
Borofloat 62.2 0.195b 1.75 52
c c
Zr65Cu15Al10Ni10 83.0 0.369 1.80 14
Mg58.5Cu30.5Y11 53.9d 0.318d 0.94 29
a
Rodríguez et al, 2012; b Dannemann et al., 2012; c Plummer et al., 2011; d Zheng et al., 2006

16 8
14 7
starphire Mg58.5Cu30.5Y11
12 model 6 model
10 UMAT 5 UMAT
v GPa)
v GPa)

8 4
6 borofloat 3
4 2
Zu65Cu15Al10Ni10
2 1
0 0
0 2 4 6 8 10 0 2 4 6 8 10
p (GPa) p (GPa)
(a) (b)
Fig. 5 Comparison results of UMAT code with linear DP model for several
materials; (a) ceramic glasses, (b) bulk metallic glasses

2018 Science in the Age of Experience 12

http://www.3ds.com/events/science-in-the-age-of-experience
4.3 Piecewise Drucker-Prager model
The PDP model UMAT is developed to implement a pressure-dependent yield model that
cannot be described by using existing DP models such as linear, hyperbolic, or exponent forms.
The first example is for bilinear DP constitutive equation for failed borosilicate glass (Chocron et
al., 2010). The constitutive equation is

 0.038  1.2 p p  1.72 GPa

 (42)
 2.1 GPa p  1.72 GPa

The Young’s modulus E is 62.3 GPa, and the Poisson’s ratio  is 0.2. The comparison of results
from the UMAT code with the DP constitutive equation for failed borosilicate glass is shown in
Fig. 6a. The results show that after yielding begins,  from UMAT model are identical to that of
the bilinear DP model for failed boro-silicate glass.
The second example is for an extended Mohr-Coulomb model (Shafiq and Subhash, 2016),
which is a generalized constitutive model for brittle ceramics that is expressed as

 k
 P 
 P 

   HEL  a  be  HEL   (43)
 
 

where  is shear stress, HEL is equivalent shear stress at the Hugoniot elastic limit (HEL) given by
HEL = HEL / 2, a = 1.15, b = -1.06 and k = 1.78. Then multiplying both sides of Eq. (43) by 2, we
obtain the extended Drucker-Prager (EDP) model, expressed as

 k
 P 
 P 

 v   HEL  a  be  HEL   (44)
 
 

The two pressure-dependent material properties obtained from the literature are listed in Table 2.
The result of the UMAT code and the constitutive equation for EDP model is compared in Fig. 6b.
The result shows that after yielding starts, the  from the UMAT model are identical to that in the
EDP model for SiC and soda-lime glass.

Table 2 Material properties of pressure-dependent materials obtained from the literature

material E (GPa)   HEL (GPa) PHEL (GPa)
soda-lime glass 69.0a 0.23a 4.54b 2.92b
SiC 401.2c 0.186c 13d 5.9d
a
Chen et al., 1995; b Holmquist et al., 1995; c Shackelford and Alexander; 2000, d Cronin et al., 2003

2018 Science in the Age of Experience 13

http://www.3ds.com/events/science-in-the-age-of-experience
 4 16
14
3 12 SiC
model
UMAT 10
v GPa)

v GPa)
2 8
6 soda-lime glass
1 4 model
2 UMAT

0 0
0 2 4 6 8 10 0 2 4 6 8 10 12 14 16 18 20
p (GPa) p (GPa)
(a) (b)
Fig. 6 Comparison results of UMAT code; (a) bilinear DP model for failed
borosilicate glass, (b) EDP model for SiC and soda-lime glass

5. Summary
This study has derived constitutive equations for a piecewise Drucker-Prager (PDP) model and
has developed integration algorithm based on the notion of return mapping to the smooth portion,
apex, and corners. A consistent tangent operator with the developed integration algorithm was
formulated for each return mapping case. The PDP model is then applied in commercial finite
element code (using Abaqus) by developing a user material subroutine (UMAT). This UMAT for
PDP model is verified through finite element analysis (FEA) of a hydrostatic test with a single
element.

6. Acknowledgment
This research was supported by the Basic Science Research Program through the National
Research Foundation of Korea (NRF-2017R1A2B3009706).

7. References
1. Abaqus User’s Manual, Version 6.13-2, Dassault Systémes Simulia Corp., Providence, RI.
2. Ansys 16.0 release documentation, “Theory and Modelling Guide,” ANSYS, Inc.,
Canonsburg, PA, 2015.
3. Chen, S.Y., Farris, T.N., and Chandrasekar, S., “Contact mechanics of Hertzian cone
cracking,” International Journal of Solids and Structures, Vol. 32, pp. 329−340, 1995.

2018 Science in the Age of Experience 14

http://www.3ds.com/events/science-in-the-age-of-experience
 4. Chocron, S., Anderson JR, C.E., Nicholls, A.E., and Dannemann, K.A., “Characterization of
confined intact and damaged borosilicate glass, Journal of the American Ceramic Society,”
Vol. 93, pp. 3390−3398, 2010.
5. Cronin, D.S., Bui, K., Kaufmann, C., McIntosh, G., Berstad, T., and Cronin, D.,
“Implementation and validation of the Johnson-Holmquist ceramic material model in LS-
DYNA.,” 4th European LS-DYNA Users Conference, pp. 47−60, 2003.
6. Dannemann, K.A., Anderson Jr, C.E., Chocron, S., and Spencer, J.F., “Damage development
in confined borosilicate and soda-lime glass,” Journal of the American Ceramic Society, Vol.
95, pp. 721−729, 2012.
7. de Souza Neto, E.A., Peric, D., and Owen, D.R.J., “Computational methods for plasticity:
theory and applications,” Wiley, 2008.
8. Drucker, D.C., and Prager, W., “Soil mechanics and plastic analysis or limit design, Quarterly
of Applied Mathematics,” Vol. 10, pp. 157−165, 1952.
9. Holmquist, T.J., Johnson, G.R., Lopatin, C., Grady, D., and Hertel E.S. Jr, “High strain rate
properties and constitutive modeling of glass,” 15th International Symposium on Ballistics,
Jerusalem, 1995.
10. Lee, K., Marimuthu, K.P., and Lee, H., “UMATs for linear, bilinear and piecewise Drucker-
Prager models,” Mendeley Data, v5, http://dx.doi.org/10.17632/5ccwt5fv7y.5, 2018.
11. Ma, G.W., Hao, H., and Zhou, Y.X., “Modeling of wave propagation induced by underground
explosion, Computers and Geotechnics,” Vol. 22, pp. 283−303, 1998.
12. Milani, G., and Lourenco, P.B., “Blast analysis of enclosure masonry walls using
homogenization approaches,” International Journal for Multiscale Computational Engineering,
Vol. 7. pp. 91−113, 2009.
13. Plummer, J.D., Goodall, R., Figueroa, I.A., and Todd, I., “A study of mechanical
homogeneity in as-cast bulk metallic glass by nanoindentation,” Journal of Non-Crystalline
Solids, Vol. 357, pp. 814−819, 2011.
14. Rodríguez, M., Molina-Aldareguía, J.M., González, C., and LLorca, J., “Determination of the
mechanical properties of amorphous materials through instrumented nanoindentation,” Acta
Materialia, Vol. 60, pp. 3953−3964, 2012.
15. Shackelford, J.F., and Alexander, W., “CRC Materials Science and Engineering Handbook,”
CRC Press, 2000.
16. Shafiq, M., and Subhash, G., “An extended Mohr-Coulomb model for fracture strength of
intact brittle materials under ultrahigh pressures,” Journal of the American Ceramic Society,
Vol. 99, pp. 627−630, 2016.
17. Zheng, Q., Ma, H., Ma, E., and Xu, J., “Mg-Cu-(Y, Nd) pseudo-ternary bulk metallic glasses:
The effect of Nd on glass-forming ability and plasticity,” Scripta Materialia, Vol. 55, pp.
541−544, 2006.

2018 Science in the Age of Experience 15

http://www.3ds.com/events/science-in-the-age-of-experience

View publication stats