Setting up a Gradient-Based Optimization Study
A gradient-based optimization study is an optimum search procedure based on the gradients (sensitivities) of geometric parameters. Simcenter STAR-CCM+ computes the parametric gradients for each geometric parameter, based on the geometric sensitivity of the parameter and the adjoint surface sensitivity of the entire geometry towards the optimization objective.
- A parametrized CAD-Model, either prepared in Simcenter STAR-CCM+ using 3D-CAD or imported using CAD Clients.
- The surface mesher Geometric Sensitivity
- The physics model Adjoint Surface Sensitivity
- To define a parametrized CAD model:
- Create the geometry in 3D-CAD and define design parameters for the geometric features that you want to optimize in the study. See Using 3D-CAD and Working with Design Parameters.
- Create the geometry in an external CAD package that has a Simcenter STAR-CCM+ Client. In the CAD model,
include the relevant design parameters. Transfer the geometry in Simcenter STAR-CCM+ using the Simcenter STAR-CCM+ Client for the CAD
package.
This approach transfers the geometry as a CAD Client model in Simcenter STAR-CCM+. The design parameters are listed under CAD Parameters node. See CAD Client Design Parameters.
- As this type of optimization uses global parameters, replace the parameters from either the 3D-CAD model (Design Parameters) or the imported CAD Client model (CAD Parameters) with global parameters. For instructions, see Replace Parameters.
- To compute the geometric sensitivity, include the Geometric Sensitivity mesher in the relevant mesh operation. For detailed instructions, see Geometric Sensitivity.
- To obtain the surface sensitivity information, activate the adjoint solver and create cost function reports. Only the reports used as adjoint cost function can be selected later as objectives or constraints for the optimization.
- Create reports for all objectives and constraints.
- In the relevant physics continuum, include the adjoint Surface Sensitivity model. For details, see Computing the Surface Sensitivity.
- Add reports as adjoint cost functions. See Creating the Adjoint Cost Functions.
- Compute surface sensitivities of the geometry with respect to specified cost functions. See Computing the Surface Sensitivity.
- Create the relevant scenes and plots that you want Design Manager to use for post-processing.
- Make sure that the stopping criteria are appropriate for running various designs.
- To start a gradient-based optimization:
- Run the design study. See Running a Design Study.When using the SQP method, the optimization search is sequential. The study runs serial—only one Simultaneous Jobs at a time.
- To analyze the optimization results:
- To visualize the optimization path, create a history plot for the objective. When searching for minimum, Design Manager reduces the objective within each outer iteration.
- Create a sensitivity plot for all the geometric parameters the optimizer accesses. When the local minimum is reached, all the sensitivities approach to zero (in absence of constraints).
For more details, see Plot Setup Dialog Reference.