Issue with Stress Constraints in Topology Optimization

Does anyone have an idea why it is not possible to get a meaningful result when considering stresses as constraints in a topology optimization whereas mass the objective for optimization is? i always get error message optimization infeasible although there is no high stress in design variable before the topology optimization.

Find more posts tagged with

English

OptiStruct

v2021.2

Accepted answers

Adriano_Koga

what stress constraint are you using?

There are 2 types:

global stress constraint (Defined inside the design variable) >> acts globally, for all elements, and for all loadsteps. Often is affected by stress singularities, even outside the design domain.

local stress constraints (defined by a response Static Stress and Design Constraint) >> acts specifically at the regions selected and it is possible to exclude singular regions from the calculation to avoid misleading the optimizer.

The first approach is simpler, and computationally less expensive. The second one is more detailed and you have more control, but it is computationally more expensive, specially if your topology domain (number of elements) is too high, as sensitivities computation is really intensive.

All comments

Adriano_Koga

what stress constraint are you using?

There are 2 types:

global stress constraint (Defined inside the design variable) >> acts globally, for all elements, and for all loadsteps. Often is affected by stress singularities, even outside the design domain.

Karshenas Reza

what stress constraint are you using?

There are 2 types:

global stress constraint (Defined inside the design variable) >> acts globally, for all elements, and for all loadsteps. Often is affected by stress singularities, even outside the design domain.

local stress constraints (defined by a response Static Stress and Design Constraint) >> acts specifically at the regions selected and it is possible to exclude singular regions from the calculation to avoid misleading the optimizer.

The first approach is simpler, and computationally less expensive. The second one is more detailed and you have more control, but it is computationally more expensive, specially if your topology domain (number of elements) is too high, as sensitivities computation is really intensive.

great, thaks