Stress Constrained Topology Optimization Using P-NORM

Mohan Parthasarathy
Mohan Parthasarathy
Altair Employee
edited October 2023 in Altair HyperWorks

Stress constraint for topology optimization (referred to as DTPL stress constraint) has been available in Optistruct since version 8.0. This stress constraint is defined through the STRESS field in the DTPL bulk data card of the topology design variable. It is limited to a single value of von Mises stress for the entire model, including the design and non-design spaces. The DTPL stress constraint has some built-in intelligence to filter out stress concentrations.  So even if DTPL stress constraint is included during topology optimization, the optimal load path may contain stress concentrations and the user should consider performing size/shape/free-shape optimization, whichever is applicable, before releasing the design for production.

In Optistruct v14.0, DRESP1 stress constraint was made available for topology optimization. Instead of applying the stress constraint on individual elemental stresses, a single stress-NORM aggregate is internally computed for the elements included in the DRESP1 stress response, and a constraint is applied on the stress-NORM aggregate. This does not have the limitations of the DTPL stress constraint. Stress concentrations are not filtered out in this case. As the user-adjustable stress-NORM parameter is gradually increased from the default value, stress concentrations or singularities can be eliminated, and a more uniform stress distribution can be obtained. The user should still consider performing size/shape/free-shape optimization, whichever is applicable, to address any areas of stress concentration, before releasing the design for production.

In this document, the difference between the DTPL and DRESP1 stress constraints and the elimination of stress concentration or singularity using the DRESP1 stress constraint with stress-NORM aggregate are illustrated with the help of numerical examples. The examples do not use production parts and hence size/shape/free-shape optimization was not performed for the examples.