Sei sulla pagina 1di 9

5

Application of
Newton Raphson Techniques
to Elasto-Plastic Models with
Constant Yield Surfaces

CE 6102
Geotechnical Analysis

5-1
Table of Contents

1 Typical Elasto-Plastic Yield Surfaces 5-1

2 Non-Linear Correction with a Fixed Yield Surface 5-3

5-2
1 Typical Elasto-Plastic Yield Surfaces

Elasto-plastic yield surfaces can either be fixed (i.e. constant) or changing. Fixed yield
surfaces are generated by elastic-perfectly plastic constitutive models such as
(a) Mohr-Coulomb model (we have done this before).
(b) Drucker-Prager model (without cap). This is very similar to the Mohr-Coulomb
model except that it has a circular generalization, i.e. on the -plane, the cross-
section of the yield surface looks like a circle, rather than a hexagon (see Figure
5.1 below). Some variants of the Drucker-Prager model has a non-circular
cross-section in the -plane (see Figure 5.2 below). The use of a curvilinear
cross-section rather than hexagonal cross-section was motivated primarily by
the ease of programming a smooth curvilinear function rather than a series of
straight lines.
(c) Tresca model. This is actually Mohr-Coulomb with = 0 (see Figure 5.3). If you
use Plaxis method B with c = cu and u = 0, then this is basically the model you
are using.
(d) Von-Mises model. This is actually the Drucker-Prager circularly generalized
model with = 0 (see Figure 5.3).

Figure 5.1 Mohr-Coulomb and Drucker-Prager Yield Surface in plane

5-1
Figure 5.2 Circular and Non-circular Drucker-Prager Yield Surfaces in
plane

Figure 5.3 Different Yield Surfaces in the plane

There is actually another Drucker-Prager Cap model, which is essentially the Drucker-
Prager with an added compression cap. In this model, the compression cap can expand
while the shear yield surface (i.e. the side of the cone, defined by the friction angle) may
or may not change. We will not deal with it here since it is not commonly implemented in
softwares (apart from possibly ABAQUS).

5-2
There are many, many constitutive models available which involve changing yield
surfaces. Indeed, most advanced constitutive models involve changing yield surfaces.
One of the simplest constitutive models with changing yield surface is the Cam Clay
model (or modified Cam Clay model). The yield surface can enlarge or contract
depending upon what state the soil is in and how it is loaded (e.g. see Figure 5.4 below).
In the case of the Cam Clay and modified Cam Clay, loading on the wet side of critical
causes expansion of the yield surface whilst loading on the dry side of critical causes
contraction of the yield surface.

Figure 5.4 Yield Surfaces of the Cam-clay Model

2 Non-Linear Correction with a Fixed Yield Surface

The problem with the tangent stiffness projection on a fixed yield surface is illustrated in
Figure 5.5 below. As can be seen, there will always be an overshoot regardless of
whether one starts with the increment below the yield surface (and therefore tangent
stiffness would have used the elastic stress-strain matrix De) or on the yield surface (i.e.
tangent stiffness would have used the tangential elasto-plastic stress-strain matrix Dep).

5-3
Figure 5.5 Tangent stiffness projected stress increment vector in principal stress
space and deviatoric plane

If these errors are not corrected, they will gradually accumulate so that the final stress
state will end up well outside the yield space. There are several approaches to
correcting these errors. One way is to use an improved method of projection such as the
Consistent Tangent Matrix method which involves using the parameters for the end
point (say B) to correct for the drift. However, a more common approach is to correct the
stress state directly back to the yield surface. There are several methods of back-
projection and these are discussed by Potts and Gens (1985) [A Critical Assessment of
Methods of Correcting for Drift from the Yield Surface in Elasto-Plastic Finite Element
Analysis, published in International Journal for Numerical and Analytical Methods in
Geomechanics, v. 9, p. 149 159]. Potts and Gens (1985) examined five different
methods of back-projection, namely
(a) Projecting back reducing the deviator stress q at constant mean effective
stress p (e.g. from stress state B to B1 in Figure 5.6).

5-4
(b) Projecting back by back-scaling along the accumulated effective stress vector
(e.g. from stress state B to B2 in Figure 5.6).

Figure 5.6 Reducing q at constant p'

(c) Projecting back along the plastic flow direction. This is equivalent to scaling
G
back the stress along the plastic strain increment p ( ), in which G is
'

the plastic potential. This leads to

G
C ' = B ' C (5.1)
'
in which C is an unknown scalar quantity and the subscript C denotes the
corrected state. Since the corrected stress state must satisfy the yield
condition

G
F( C ') = F B ' C =0 (5.2)
'
Expanding Eq. 5.2 as a Taylor Series and neglecting the higher order terms
leads to
F( B ' )
C =
F G

' '

5-5
F( B ' )
= (5.3)
bTa
The b and a vectors can be evaluated at the starting point of the step to give
an initial estimate of C.

(d) Projecting back along the total strain increment direction. This method is
similar to that in (c) except that the total strain increment = e + p is
used. This gives
C ' = B ' C (5.4)
and
F( B ' )
C = (5.5)
F

'
F( B ')
= (5.6)
b T
(e) Correct projecting back. In this method, the assumption is made that there is
no change in the total strain increment vector during the correction process,
so that any changes in the elastic strain increment must be balanced by an
equal and opposite plastic strain increment. Let e be the change in elastic
strain increment during the correction, then

e = D e ( C ' B ' ) (5.7)

p = e = D e ( C ' B ' ) (5.8)

G p
p = C C B
p
(5.9)
'
Combining Eqs. 5.8 and 5.9 leads to

G
C ' = B ' CD e (5.10)
'
The corrected stress state must satisfy the yield condition, i.e.

G
F B ' CD e = 0 (5.11)
'

5-6
Expanding as a Taylor Series and neglecting the higher order terms leads to
F( B ')
C = (5.12)
b TD ea
so that Eq. 5.10 becomes
F( B ')a
C ' = B ' De (5.13)
b TD e a

All these equations are for elastic-perfectly systems only, but they can be extended to
hardening plasticity as well. Potts and Gens recommended methods (c) and (e).

In addition to the above methods, there is actually another simpler method of sub-
stepping which is heuristically sound and readily applicable, albeit with a bit more
computation efforts. This method of sub-stepping involves breaking up the strain
increment vector into smaller sub-steps and evaluating the stress-strain matrix and
stress increment at each sub-step. The steps involved after evaluation of the
displacement increment u (i.e. the primary unknown) are as follows:

(a) Evaluate the strain increment vector = [B] u.



(b) Divide into smaller sub-steps i.e. = in which n is the number of sub-
n
steps.
(c) For each sub-step:
i. Evaluate ' = Dep
ii. Accumulate the stress 'new = 'old + '
iii. Evaluate Dep at the new stress state new.
iv. Repeat (c) for the next sub-step.

This method is the reference method for comparison used by Potts and Gens and is also
the method implemented in GeoFEA for the Cam Clay model. It does really make much
difference to the overall time requirement since this part of the analysis cycle usually
takes a relatively short time, compared to that required by the solution process. It is
applicable to both elastic-perfectly plastic as well as strain-hardening plastic materials.

5-7

Potrebbero piacerti anche