Basic Workflow of a CAD-based MDO framework

Last year I had the opportunity to present an application using the CAD-based adjoint multidisciplinary optimization framework at the GPPS 2018 conference in Montreal. This year, the optimization has been further extended by Lasse Mueller to perform a multipoint optimization, which will be presented at the ASME TurboExpo 2019 this summer. I thought this would be a good time to post a short description of the CAD-based MDO framework.

The buzzwords CAD-based adjoint multidisciplinary framework reveal the basic steps:

1. Start with CAD design parameters. These are typically paremeters that engineers can use to define the shape of the geometry. For instance, for a radial turbine these could include the blade angle and thickness distributions. These are essentially the design parameters, i.e., the parameters that will be updated by the optimizer to minimize the objective.
2. Based on these CAD design parameters, the CAD kernel generates a CAD geometry and the shape can be visualized.
3. Based on this geometry, computational meshes for the different disciplines are generated. In the two applications mentioned above, these would be fluid and solid disciplines. In this case, the meshes are different (one structured, the other unstructured), but both mesh generators are based on the same geometry, which essentially acts as an interface between the two disciplines.
4. The meshes are then used to run the CFD and CSM solvers to compute the required quantities of interest. These could include aerodynamic efficiency from the CFD side as the objective and maximum von Mises stress from the CSM side as a constraint.
5. An adjoint evaluation or reverse run of steps 4.-1. to efficiently compute gradients of the performance parameters with respect to the CAD design parameters.
6. The optimizer uses the results from steps 4 and 5 to perform an optimization step and outputs an improved iteration of the CAD design parameters
7. Repeat until convergence.