CA2b. NON-LINEAR FINITE ELEMENT ANALYSIS OF PLATE SUBJECTED TO SHEAR LOADING
Problem definition
In this assignment, you are going to implement the multiaxial elastic-plastic material response with linear isotropic hardeninginto a constitutive driver. In the spirit of the developments in computer assignment CA2a, we shall analyze the shear loadedcolumn in Fig. 1. In order to simplify the analysis, the columns may be considered as a "plate" in a state of plain strain. Asmay be seen in the Figure below, the plate is loaded by a shear load P[t], and the behavior of the plate during one loadingcycle will be analyzed whereby
P@tD = Pmax sinB 2 Π
TtF with T = 1
where Pmax is the maximum value of the load amplitude. It is determined by the bending moment at onset of plastic deforma-
tion according to beam theory My, cf. computer assignment 1, whereby
Pmax = 3My
awith My = MinBh
2 Ibl2 + 4 bl bu + bu2M Σy
12 H2 bl + buL ,h2 Ibl2 + 4 bl bu + bu
2M Σy
12 Hbl + 2 buL FThe height, width and thickness of the plate are a, h and t, respectively. Hence, the geometry data are chosen as follows:
a = 60 mm, h = 20 mmm, t = 10 mm
In the assignment, special focus will be placed on the assessment of the residual stress state of the column (obtained at theend of the time interval, i.e. t=1) as compared to the result obtained from the corresponding previous 1D analysis based onthe Euler-Bernoulli beam theory assumption.
Figure 1: "Plane strain" column subjected to prescribed shear loading
0.2 0.4 0.6 0.8 1.0t
-1.0
-0.5
0.5
1.0P@tD @´PmaxD
Cyclic prescribed shear load considered
Figure 2: Cyclic prescribed shear load variation versus time
The material behavior is described by the experimentally determined cyclic stress-strain curve shown in Figure 3. Like, inthe previous assignment, let us consider only the behavior in the first loading cycle, cf. Figure 3. In order to model thedescribed behavior, and at the same time simplifying our analysis, we shall consider the linear isotropic hardening rulecombined with the plasticity concept, i.e. r = 1 is used as a model parameter. Otherwise, the same model parameters as in
CA1 are used in the present assignment. The model response was fitted to the experimental curve using "manual" curvefitting and the resulting matching with r = 1 is shown in Fig. 4 below. It may be observed that the model response is far
from "perfect" when r = 1 is used. In particular,, the yield stress after one cycle is significantly overestimated by the linear
isotropic hardening model. The model parameters were obtained as: E = 200 GPa, H = E 10, Σy = 540 MPa and r =1,
corresponding to linear isotropic hardening.
Figure 3: Cyclic behavior of a typical railway steel in strain-controlled uniaxial test. Experimental results are due to Johan
Ahlström, Birger Karlsson, Wear 258, 1187–1193, (2005).
-0.010 -0.005 0.005 0.010Ε@tD
-500
500
Σ@tDIsotropic and kinematic hardening plasticity: cyclic loading
Figure 4: Linear isotropic hardening model response as compared to experimental values for the strain amplitudeDΕf 2 = 1 %, The model response has been obtained using the same parameters as in CA1 except for r=1.
Exercise problems
As alluded to in the problem definition, consider the assignment problems:
è Implement the elastic response of the model behavior. Check the response when Pmax =My
a. Almost no plastic
response should appear in the plate for this load magnitude? Discuss the result.
2 CA2.nb
è Implement the elastic-plastic response of the model using the discussed linear isotropic hardening law. Check the
response after one loading cycle when Pmax = 3My
a. In particular, consider the residual stress state at the base of
the column from the computed reaction forces. Discuss the obtained residual stress state in comparison with the corresponding result obtained from the 1D beam model behavior, already developed in the previous assignment.
Numerical procedures
Equilibrium condition
The equilibrium conditions for the considered column may be formulated as the balance of internal and external virtualwork:
àBΣ@uD : Ε@uD âB = à
Bf × u âB + à
G
t × u âB
where B is the 3D domain of the solid plate and ¶B is the external boundary thereof. In the present analysis, the plane strain
symmetry condition may be assumed, whereby the 3D problem may be reduced to a 2D problem. Hence, we may considerthe 3D domain restricted to the 2D region W so that B ® t ´ W, where t is the plate thickness. Likewise, the external
boundary may be reduced to ¶B ® t ´ G.
Next, finite element discretization is made with respect to the computational domain W into finite elements
We, e = 1, ... NEL. It is assumed that each element has the interpolation
u = âI=1
NODE
NI@xD uI Þ u = âI=1
NODE
NI@xD uI Þ l =¶u
¶x= â
I=1
NODE
uI Ä gI with gI =¶NI
¶x
where uI are the nodal displacement variables, and 8NI@xD<I=1,2,NODE are the element interpolation functions. In the
assignment, we restrict to CST elements corresponding linear displacement approximation within each triangle, i.e.NODE = 3. To this end, a local Ξ1, Ξ2-coordinate system is introduced, where the shape functions NI@Ξ1, Ξ2D are defined
(within a base triangle element shown in Fig. 5 below) as
N1 = 1 - HΞ1 + Ξ2L, N2 = Ξ1, N3 = Ξ2 Þ gI =¶NI
¶x= â
i=1
2 ¶NI
¶Ξi
Gi with Gi =¶Ξi
¶x
0.00.5
1.0Ξ1 0.0
0.5
1.0
Ξ20.00.51.0
,
0.00.5
1.0Ξ1 0.0
0.5
1.0
Ξ20.00.51.0
,0.0
0.51.0
Ξ1 0.0
0.51.0
Ξ20.00.51.0
Figure 5 Linear basis functions pertinent to the Constant Strain Triangle element
The corresponding discretized version of the equilibrium condition state above, cf. the lectures x-x, may be stated as balancebetween internal and external finite element nodal forces.
g@uD = b@uD - f@tD = 0 with b@u, tD = A
e=1
NEL àWe
Bet
Σ âW , f@tD = A
e=1
NELfeext
The problem is to find the total "displacement" u = 8uI<I=1,.. NNO so that the vector of out-of-balance forces g@uD = 0.
To this end, the non-linear solution strategy is used as discussed below.
Non-linear solution strategy - Newton's method
In order to solve the non-linear problem, we consider the balance of internal and external forces for each given time step as
CA2.nb 3
g@uD = b@uD - f@tD = 0
with
u = nu + v Dt and t = n+1t
For each time t = n+1t, consider the algorithm defined as: for each iterate i, evaluate the improved displacement incrementi+1
Ξ from the Newton step
iK i+1Ξ = -gAnu + iv DtE, i+1v = iv + i+1
Ξ
where K is the Jacobian (or tangent stiffness) matrix of the considered component. It is defined from the linearization
dg@uD = db@uD = K du
The tangent stiffness K is defined by
K =¶b
¶u= Dt
¶b
¶v=. ..
Preliminary results
Preliminary results in the form of a comparison between the residual stress states obtained from 1D-beam analysis modelingand the multiaxial finite element modeling are given in the Figure below.
-10 -5 5 10z
-400
-200
200
400Σ@zD
Fiber-stress response over cross section
From 1D beam-analysis
From FE analysis
Implementation: Pseudo code
Main routine: Input data and solution independent arrays
lh = 60; % horisontal lengthlv = 20; % vertical lengtht= 10;h=lh; bl=t; bu=t;nelv = 20 % # elements verticallynelh = 10 % # elememts horisontally
[nel, nno, edof, ex, ey, nol1, nol2, nol3, nol4] = plate_mesh_CST (nelv, nelh, lh, lv);
nva = 2*nno;bc = [2*nol1-1 2*nol1; zeros (1, length (nol3)) zeros (1, length (nol3))]';bcload = 2*nol3 - 1;
% loads and other constantsmtim = 400; % total number of timesteps
∆ = 10-6; % tolerance for equilibrium iterationsmite = 50; % max number of iterations
My=min(...
((bl^2 + 4*bl*bu + bu^2)*h^2*sigy)/(12.*(2*bl + bu)),... ((bl^2 + 4*bl*bu + bu^2)*h^2*sigy)/(12.*(bl + 2*bu)));
Pmax = 3 * Mel/lv;
% Material parameters for elastic plastic material
E = 200 103; Ν = 0.3; H = E/10; r = 1; Σy= 540; mp = [E Ν H r ΣyE;% Initiation of variablesu = zeros (nva, 1); %displacementv = zeros (nva, 1); %velocityf = zeros (nva, 1); %applied nodal loads
nΣ = zeros (4, nel); Σ = n
Σ;nΚ = zeros (1, nel); Κ = nΚ;
pdof = bc (: , 1); % prescribed degrees of freedomfdof = [1 : nva]'; fdof (pdof) = []; % "free" degrees of freedom
4 CA2.nb
lh = 60; % horisontal lengthlv = 20; % vertical lengtht= 10;h=lh; bl=t; bu=t;nelv = 20 % # elements verticallynelh = 10 % # elememts horisontally
[nel, nno, edof, ex, ey, nol1, nol2, nol3, nol4] = plate_mesh_CST (nelv, nelh, lh, lv);
nva = 2*nno;bc = [2*nol1-1 2*nol1; zeros (1, length (nol3)) zeros (1, length (nol3))]';bcload = 2*nol3 - 1;
% loads and other constantsmtim = 400; % total number of timesteps
∆ = 10-6; % tolerance for equilibrium iterationsmite = 50; % max number of iterations
My=min(...
((bl^2 + 4*bl*bu + bu^2)*h^2*sigy)/(12.*(2*bl + bu)),... ((bl^2 + 4*bl*bu + bu^2)*h^2*sigy)/(12.*(bl + 2*bu)));
Pmax = 3 * Mel/lv;
% Material parameters for elastic plastic material
E = 200 103; Ν = 0.3; H = E/10; r = 1; Σy= 540; mp = [E Ν H r ΣyE;% Initiation of variablesu = zeros (nva, 1); %displacementv = zeros (nva, 1); %velocityf = zeros (nva, 1); %applied nodal loads
nΣ = zeros (4, nel); Σ = n
Σ;nΚ = zeros (1, nel); Κ = nΚ;
pdof = bc (: , 1); % prescribed degrees of freedomfdof = [1 : nva]'; fdof (pdof) = []; % "free" degrees of freedom
Time stepping and non-linear Newton solution procedure
T = 1; Dt = T/mtim; time = 0;
for itim = 1 : mtim % for all time steps
time = time + Dt;
Ω = 2*Π/T; %one loading cyclev (bc (: , 1)) = bc (: , 2);f(nol3*2-1)=Pmax/length(nol3)*sin(omega*time); %distribute load in x-dir. on
line 3
err = 1; stop_eqiter = 0; iite = 0;while ((stop_eqiter == 0) & (iite < mite)) % equilibrium iterations
iite = iite + 1;K = sparse (nva, nva);g = zeros (nva, 1);for iel = 1 : nel % for all elements
leva = edof (iel, 2 : 7);[B, Area] = B_matrix (ex (iel, :), ey (iel, :));
due = v (leva)*Dt; Σtr = nΣ(: , iel) + Ee*B*due;% trial stress
[Σ (: , iel), Κ (: , iel), E_t] = IO_V _Mises(Σtr, nΚ (: , iel), mpL;
fext = zeros (6, 1);g (leva) = g(leva) + HBt Σ (: , iel)*Area*t - fext);K (leva, leva) = K (leva, leva) + Dt BtE_t B*Area*t;
end
g=g-f;
err = norm (g (fdof), 2); % Check convergenceif err < ∆
stop_eqiter = 1;else
Ξ = -K (fdof, fdof)\g (fdof);v (fdof) = v (fdof) + Ξ; % Update displacement increment (free nodes)
end
end
u = u + v*Dt; % update total displacementnΣ=Σ; n
Κ=Κ; % Save data from previous time stepend% post-processing...stop
CA2.nb 5
T = 1; Dt = T/mtim; time = 0;
for itim = 1 : mtim % for all time steps
time = time + Dt;
Ω = 2*Π/T; %one loading cyclev (bc (: , 1)) = bc (: , 2);f(nol3*2-1)=Pmax/length(nol3)*sin(omega*time); %distribute load in x-dir. on
line 3
err = 1; stop_eqiter = 0; iite = 0;while ((stop_eqiter == 0) & (iite < mite)) % equilibrium iterations
iite = iite + 1;K = sparse (nva, nva);g = zeros (nva, 1);for iel = 1 : nel % for all elements
leva = edof (iel, 2 : 7);[B, Area] = B_matrix (ex (iel, :), ey (iel, :));
due = v (leva)*Dt; Σtr = nΣ(: , iel) + Ee*B*due;% trial stress
[Σ (: , iel), Κ (: , iel), E_t] = IO_V _Mises(Σtr, nΚ (: , iel), mpL;
fext = zeros (6, 1);g (leva) = g(leva) + HBt Σ (: , iel)*Area*t - fext);K (leva, leva) = K (leva, leva) + Dt BtE_t B*Area*t;
end
g=g-f;
err = norm (g (fdof), 2); % Check convergenceif err < ∆
stop_eqiter = 1;else
Ξ = -K (fdof, fdof)\g (fdof);v (fdof) = v (fdof) + Ξ; % Update displacement increment (free nodes)
end
end
u = u + v*Dt; % update total displacementnΣ=Σ; n
Κ=Κ; % Save data from previous time stepend% post-processing...stop
Stress integration
function@Σ, Κ, EtD = IO_V _Mises HΣtr, nΚ, mpL
% Constitutive driver for von Mises model with linear isotropic hardening
% Note : Stress and strain by Voigt - representation Hfor input outputL!
% Σ = @Σ11, Σ22, Σ33, Σ12Dt
E = mp H1L; Ν = mp H2L; H = mp H3L; Σy = mp H5L;K = E 3 H1 - 2 * ΝL; G = E 2 H1 + ΝL;v1 = @1, 1 , 1, 0Dt; v2 = @0, 0, 0, 1Dt;
I = diag Hv1 + .5 * v2L;Idev = I -
1
3v1 * v1t;
Ee = 2 G Idev + K v1 * v1t;
Σdevtr =. ..;
Σetr =. ..;
Φtr = Σetr - Σy - nΚ;
if Φtr £ 0
% elastic loading
Σ = Σtr;
Κ = nΚ;
Et = Ee;
else
% plastic loading
Μ =. ..;
Σ =. ..;
Κ =. ..;
Et = Ee - ...;
end
Close up of FE element analysis
6 CA2.nb