CAD-Based Shape Optimization Of An Exhaust Manifold Using The Adjoint Solver
Résumé
The simulation process of CFD is usually starting from a CAD model. Due to the computational cost of large-scale parametric optimizations that use design of experiments and surrogate modeling, roll-out methods in CFD optimizations face difficulties. The Adjoint solver provides a more efficient approach whose cost is independent of the number of design parameters. However, the sensitivity of the cost function (with respect to mesh coordinates) is used by a mesh morpher that cannot respect manufacturing constraints. Thus, we propose a cutting edge method to extend these gradients to CAD parameters. Indeed, we compute the sensitivity of the mesh coordinates with respect to CAD parameters thanks to a harmonic projection. This sensitivity is coupled to the results of STAR-CCM+® Adjoint solver. Finally, one computation provides both output data (pressure drop, forces…) and the sensitivity of the cost function with
respect to CAD parameters. This method has been applied on shape optimization of an exhaust manifold (with 50 parameters). A 34% improvement for the pressure drop has been obtained in 8 iterations all while taking into account manufacturing constraints included in the CAD model.