MODELING OF CONCRETE MATERIALS AND STRUCTURES

Kaspar Willam
University of Colorado at Boulder

Class Meeting #5: Integration of Constitutive Equations

Structural Equilibrium:
Incremental Tangent Stiffness and Residual Force Iteration
Radial Return Method:
Elastic Predictor-Plastic Corrector Strategy
Algorithmic Tangent Operator:
Consistent Tangent with Backward Constitutive Integration

Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

STRUCTURAL EQUILIBRIUM
Out-of-Balance Force Calculation:
u) = F int F ext 0
Out-of Balance Residual Forces:
R (u
P R
Internal Forces: F int = e V B t dV
P R
P R
t
External Forces: F ext = e V N b dV + e S N tt dS
Rate of Equilibrium:
XZ
e

B t dV =

XZ
e

N tb dV +

XZ
e

N tt dS

1. Incremental Methods of Numerical Integration: Path-following continuation

strategies to advance solution within t = tn+1 tn
(a) Explicit Euler Forward Approach: Forward tangent stiffness strategy (should
include out-of-balance equilibrium corrections to control drift).
(b) Implicit Euler Backward Approach: Backward tangent stiffness
(requires iteration for calculating tangent stiffness at end of increment; Heuns
method at midstep, and Runge-Kutta h/o methods at intermediate stages).

INCREMENTAL SOLVERS
Euler Forward Integration: Classical Tangent stiffness approach
P R

Internal forces: F ext = e V B t dV

Tangential Material Law: = E tan
Tangential Stiffness Relationship:
K tanu = F

where K tan =

XZ
e

Incremental Format:

R un+1
un

Euler Forward Integration:

B tE tanB dV

u = F n+1 F n
K tandu
K ntan

P R
e

t n
B
E tanB dV
V

u = F
F where E ntan = E tan(tn)
K ntanu
Note: Uncontrolled drift of response path if no equilibrium corrections are
included at each load step.
Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

2. ITERATIVE SOLVERS
Picard direct substitution iteration vs Newton-Raphson iteration within
t = tn+1 tn
Robustness Issues: Range and Rate of Convergence?
Newton-Raphson Residual Force Iteration:

u) = 0
R (u

Truncated Taylor Series Expansion of the residual R around u i1 yields

R (u
u)i = R (u
u)i1 + (
ui) = 0 solve
Letting R (u

R i1 i
R
u ui1]
) [u
u
u

R i1 i
R
u u i1] = R
R (u
u)i1
) [u
u
u

Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

NEWTON-RAPHSON EQUILIBRIUM ITERATION

Assuming conservative external forces:

R
R
u
u

d
d
u
d du

P R
e

t d

B
u dV
V
du

Chain rule of differentiation leads to

= E tand = E tan B du
u such that
d

d
u
du

= E tan B .

Tangent stiffness matrix provides Jacobian of N-R residual iteration,

R
t
R
R
=
K
where
K
=
B
E tanB dV
tan
tan
u
V
u
Newton-Raphson Equilibrium Iteration:
ui u i1] = R
R (u
u)i1
K i1
tan [u
For i=1, the starting conditions for the first iteration cycle are,
Z
u1 u n] = F n+1 B t n
K ntan[u
V

First iteration cycle coincides with Euler forward step, whereby each equilibrium
iteration requires updating the tangential stiffness matrix.
Note: Difficulties near limit point when det K tan 0

RADIAL RETURN METHOD OF J2-PLASTICITY I

Mises Yield Function:

1
1 2
s
s
s
F (s ) =
: Y = 0
2
3
Associated Plastic Flow Rule:
F
 p = s where m =
=s
ss
Plastic Consistency Condition:
F

F =
: s = s : s = 0
ss
Deviatoric Stress Rate:
s]
s = 2G [e e p] = 2G [e s
Plastic Multiplier:
s : e
=
s :s

RADIAL RETURN METHOD OF J2-PLASTICITY II

Incremental Format:
ss = 2G [ee ss]
Elastic Predictor-Plastic Corrector Split:
(a) Elastic Predictor: s trial = s n + 2Gee
(b) Plastic Corrector: s n+1 = s trial 2Gsstrial = [1 2G]sstrial
Full Consistency: Fn+1 = 21s n+1 : s n+1 13 Y2 = 0
Quadratic equation for computing plastic multiplier 1 =? and 2 =?
1
1
[sstrial 2Gsstrial ] : [sstrial 2Gsstrial ] = Y2
2
3
q
1
[1 23 Y: ]
Plastic Multiplier: min = 2G
trial

trial

Radial Return: represents closest point projection of the trial stress state
onto the yield surface. Final stress state is the scaled-back trial stress,
r
2
Y
s n+1 =
s trial

3 s trial : s trial

GENERAL FORMAT OF PLASTIC RETURN METHOD

Incremental Format:
= E : [ m
m]

Elastic Predictor-Plastic Corrector Split:

(a) Elastic Predictor: trial = n + E : 
E :m
(b) Plastic Corrector: n+1 = trial E
n + E :  E
E : m) = 0
Full Consistency: Fn+1 = F (
(i) Explicit Format: m = m n (or evaluate m at m = m c or m = m trial )
Use N-R for solving =? for a given direction of plastic return e.g. m = m n.
m = m n+1 for BEM).
(i) Implicit Format: m = m n+ where 0 < 1 (m
n + E :  E
E : m n+) = 0
Fn+1 = F (
Use N-R for solving =? in addition to unknown m = m n+
Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

ALGORITHMIC TANGENT STIFFNESS

Consistent Tangent vs Continuum Tangent:
Uniaxial Example:
= Etan where Etan

0 
= E0[1 ] hence = 0[1 e 0 ]
0

n+1
Fully Implicit Euler Backward Integration: Etan = Etan
in t = tn+1 tn,

n+1

n+1

][n+1 n]
= n + E0[1
0

Algorithmic Tangent Stiffness: Relates dn+1 to d n+1 at tn+1

d

n+1

alg
dn+1
Etan

where

alg
Etan

alg

Ratio of Tangent Stiffness Properties:

Etan
Etan

1
E
1+ 0 
0

E0[1

n+1
0 ]

1 + E00 
0.7

Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

ALGORITHMIC TANGENT STIFFNESS OF J2-PLASTICITY

Incremental Form of Elastic-Plastic Split:
ss = 2G [ee ss]
Fully Implicit Euler Backward Integration: for t = tn+1 tn,
s n+1 = s n + 2G[een+1 e n] 2Gssn+1
Relating dssn+1 to deen+1 at tn+1, differentiation yields,
dssn+1 = 2G deen+1 d2Gssn+1 2Gdssn+1
Algorithmic Tangent Stiffness Relationship:
dssn+1 =

2G
s n+1 s n+1
[II
] : deen+1
1 + 2G
s n+1 : s n+1

Ratio of Tangent Stiffness Properties:

alg
G tan
G cont
tan

1
1+2G

when ||ssn+1|| ||ssn||.

Class #5 Concrete Modeling, UNICAMP, Campinas, Brazil, August 20-28, 2007

CONCLUDING REMARKS
Main Lessons from Class # 5:
Nonlinear Solvers:
Forward Euler method introduces drift from true response path.
Newton-Raphson Iteration exhibits convergence difficulties when
K tan 0 (ill-conditioning).
CPPM for Computational Plasticity:
Analytical Radial Return solution available for J2-plasticity and
Drucker-Prager. Generalization leads to explicit and implicit plastic
return strategies which are nowadays combined with the incremental
hardening and incremental stress residuals in a monolithic Newton
strategy.
Algorithmic vs Continuum Tangent:
For quadratic convergence tangent operator must be consistent with
integration of constitutive equations. The algorithmic tangent
compares to secant stiffness in increments which are truly finite.