Reduced stiffness matrix for nonlinear analysis

[ANick]

Dear all,

I generate reduced stiffness matrix and a load vector using *SUBSTRUCTURE GENERATE (the while code is presented below) card and afterwords I do some operations in matlab.

In order to validate the stiffness matrix and the load vector I made a simple test: idea is to compare nodal displacements computed in ABAQUS with displacements computed by solving
Ku = f manually (in matlab). I found a very interesting effect:
for linear analysis, error is about 10^-5, but for non-linear analysis it can reach even 300%. Where ABAQUS results are used as a reference data. Nonlinear analysis is convergent. During matrix generation unsymmetrical solver is used and of course the report file with displacements is generated after the last increment. Substructure generation step is also called after all loading steps, so the whole model is pre-loaded correctly. Load cases for the substructure and main modelling steps are the same. Theoretically these displacements should be equal, aren't they?

Nodeset (Model-1_retained) for the substructure looks like for example 1/4 of the whole mesh. But it does really not make to much difference - error is still very high.

What can cause this problem?


Part of input file for substructure generation:
[tt]*Step, UNSYMM=YES
*SUBSTRUCTURE GENERATE, RECOVERY MATRIX=NO, TYPE=Z0001, OVERWRITE
*SUBSTRUCTURE LOAD CASE, NAME=force
*Dsload
////////
*RETAINED NODAL DOFS, SORTED=NO
Model-1_retained
*SUBSTRUCTURE MATRIX OUTPUT, SLOAD=YES, STIFFNESS=YES,FILE NAME=outputfilename
*End Step[/tt]

 

[portliner]

ABAQUS treats the substructure as linear, plastic material properties are ignored