Institute of Mechanics and Computational Mechanics

Shape Optimization of Elastoplastic Structures

under Shakedown Conditions
Shakedown theory py
With δx, δuj and δzi all variations of ob-
A system under varying loads is said to L jective and constraints can be computed.
shakedown (elastically) if after initial yield-
ing the structure behaves elastically again.
Examples
In 1938 Melan formulated the following px L
D
px

Initial mesh, 88 elements Considering again the square

static shakedown theorem for structures plate from Fig. 4 the follow-
with linear unlimited kinematic hardening ing optimization problem can
behavior dependent on the yield function Φ: y py be formulated:
6 D/L=0.2
-
A system will shakedown elastically, if x

Objective:
Geometry and loading Refined mesh, 1408 elements
there exists a time-independent residual weight G → min
stress field ρ̄(x), a time-independent mi- Fig. 4: Perforated plate with central Constraint:
cro stress field χ̄(x) and a factor m > 1 circular hole load factor β ≥ 0.25
such that the condition For linear unlimited kinematic hardening Design variables:
control points S = (xm , yn ),
Φ[mσ e (x, t), ρ̄(x), χ̄(x), σo ] ≤ 0, material the balance equation for the resid- m=B,C,D,E
ual stress ρ can be eliminated by intro- n=A,B,C,D
is fulfilled ∀P (t) ∈ M and ∀x ∈ Ωo .
ducing the free internal stress y = ρ − χ. bounds 0.0 L ≤ Sk ≤ 0.35 L
A typical 1D stress-strain diagram for linear The global shakedown factor β can then be k=1,...,8
hardening material is shown in Fig. 1. calculated from a sequence of shakedown Results for the improved geometry and the
s computations defined at the Gauss points i corresponding distribution of the shakedown
e
s of the discretized structure. The dimension factor β for two different loadings are shown
σe : elastic stress
r
ρ : residual stress of these local problems is very small when in Fig. 5 and Fig. 6.
s χ : micro stress
so c
Φ : yield function compared to the global shakedown problem.
σo : initial yield stress
Tab. 1: Computed shakedown factors β
e e

Fig. 1: Stress-strain diagram of a

linear-elastic, linear hardening material β
11.52
10.78
β
34.57
32.29
10.04 30.00
9.30 27.72
8.56 25.44
7.82 23.15
7.08 20.87
6.34 18.58
5.60 16.30

Load domains M
4.85 14.01
4.11 11.73

P 2 px = 0, py = 1 px = 1, py = 1
3.37
2.63
1.89
9.44
7.16
4.88

P 3
1.15 2.59

considered here P 4
Zouain et al. (2002) 0.594 0.429
P(t) Zhang (1995) 0.596 0.431
are supposed to be Garcea et al. (2005) 0.604 0.438 y
Initial design Improved design
Gross-Weege (1997) 0.614 0.446 6
convex polyhedra P
-
Present∗
x
M 1
0.610 0.440 px = 0, −0.5 ≤ py ≤ 1.0
P 2 ∗ overestimation ≈ 2.5%
spanned by the vec- P 1

tors corresponding M : load domain Numerical results of shakedown factors for Fig. 5: Distribution of shakedown factors
Pj : load vertices
to the load vertices a perforated square plate, cf. Fig. 4, are for proportional loading py
Pj , cf. Fig. 2. Fig. 2: Load domain shown in Tab. 1. Comparison of the results in Fig. 5 shows
that large savings can be achieved, while the
Shakedown analysis Sensitivity analysis improved structure is still safe for any load
P2 The maximal en- For sensitivity analysis two quantities, path within the load domain.
largement of the namely the variation of the displacements
Mβ
load domain M, uj and the variation of the primary variables
characterized by a of the shakedown analysis zi w.r.t. the de-
Mo
parameter β, such sign variables must be computed.
P that the system still
1 The balance equation for the elastic stress β β
0.91 1.05
0.87 0.99

σ ej of load vertex j is denoted with Gje . Its

0.83 0.94

Fig. 3: Initial (Mo ) shakes down, is de-

0.79 0.88
0.75 0.82
0.71 0.77
0.67 0.71
0.63 0.65
0.59 0.60
0.54 0.54
0.50 0.49

and maximal (Mβ ) noted as the shake- total variation implicitly defines the required 0.46
0.42
0.38
0.34
0.43
0.37
0.32
0.26

load domain down analysis. variation of the displacements δuj

y
Initial design Improved design
Gje = Ĝje (x, uj ) = 0, 6
The maximization problem for β is con- ∂Gje ∂Gje
-
x
−0.5 ≤ px ≤ 1.0, −0.5 ≤ py ≤ 1.0
strained by balance equations for residual δGje = ∂x
δx + ∂uj
δuj = 0.
and elastic stresses and by yield conditions Fig. 6: Distribution of shakedown factors
The Lagrangian for the shakedown analy-
sis of the ith Gauss point is denoted by for non-proportional loading px and py
max β
Li . Its total variation Li w.r.t. the pri- For the non-proportional loading savings are
R
subj. to Grad η : ρ̄(x)dVo = 0,
ΩRo mary variables zi = (βi , yi ) equals zero at smaller. This is due to the fact that the load
Grad η : σ e (x, t)dVo − G ext = 0, a KKT-point. The total variation of Li
Ωo
factor for the initial geometry is very close
Φ(βσ e (x, t), ρ̄(x), χ̄(x), σo ] ≤ 0. implicitly defines the total variation of the to the aimed load factor β = 0.25.
primary variables zi By comparison of the results for propor-
For convex polyhedric load domains, cf. tional and non-proportional loading it be-
Fig. 2, the time variable t can be elim- Li = −βi + λj Φj (x, uj , βi , yi ),
comes obvious that shakedown constraints
inated and only the elastic stresses σ ej Li = δzi Li (x, uj , zi , λj ) = 0, are essential for deriving robust and reli-
corresponding to the vertices j of the ∂Li ∂Li ∂Li able geometries from shape optimization of
load domain Mo must be computed. δLi = δx + δuj + δzi = 0.
∂x ∂uj ∂zi elastoplastic structures.

Dr.-Ing. K. Wiechmann

c Copyright September 2005, IBNM