Issues with Discrete Adjoint Objective Function in SU2 for Turbomachinery Optimization

[xiapxiaoyu1981]

Hi everyone,

I am trying to perform shape optimization for turbomachinery using the discrete adjoint method. From what I have learned, the discrete adjoint functionality in the latest version of the turbine module is currently broken (the developers are working on a fix, but it will take time). Based on recommendations, I have explored older versions of SU2, including V7.5.1_master and V5_feature_turbomachinery.

In V7.5.1_master, there are three built-in objective functions: ENTROPY_GENERATION, FLOW_ANGLE_OUT, and MASS_FLOW_IN. I tested them using the configuration and mesh files from TestCases/disc_adj_turbomachinery, making only minor modifications to the FFD settings and the objective function type. So, I believe the test cases should be reasonable.

For ENTROPY_GENERATION and FLOW_ANGLE_OUT, the adjoint solver in SU2 produces sensitivity results that match well with finite-difference approximations. However, MASS_FLOW_IN shows a significant discrepancy. Since I don’t frequently use MASS_FLOW_IN, I ignored this issue.

Following the implementation of the built-in objective functions (ENTROPY_GENERATION and FLOW_ANGLE_OUT) and referring to V5_feature_turbomachinery, I added TOTAL_PRESSURE_LOSS and KINETIC_ENERGY_LOSS to the V7.5.1_master source code. Unfortunately, the accuracy of the sensitivity verification is quite poor.

I have a basic understanding of discrete adjoint automatic differentiation principles, but I am unfamiliar with the specific code implementation. At this stage, my understanding is that in void CDiscAdjSinglezoneDriver::SetObjFunction(), ObjFunc retrieves the corresponding objective function using a specific get method, and the variable returned by the get method is implemented in output_physics.cpp. Am I missing any important steps?

Additionally, I understand that SU2 uses CoDiPack for automatic differentiation. Do I need to study CoDiPack in detail to solve this issue? (If possible, I would prefer to avoid this.)

Any help would be greatly appreciated!