Multidisciplinary optimization of the SRV2-O radial compressor using an adjoint-based approach

Abstract

Designing turbomachines requires in many applications to find a suitable compromise between different disciplines, such as aerodynamics, vibration analysis, and structural mechanics. The optimization of such components requires, thus, the concurrent inclusion of all relevant disciplines. This paper presents an efficient gradient-based approach that allows to reach efficiently a compromise when different disciplines are involved. The use of a gradient-based technique guarantees a low computational effort to reach an optimal solution, but faces the challenge of computing the gradients of different disciplines and projecting them on a unified parameter space. This is especially challenging when the solid and fluid domains are governed by different numerical solvers and use different numerical grids. In the present framework, a CAD-based parametrization is used as an intermediate layer that is indifferent to the numerical grid used. This allows to combine sensitivities from different disciplines and grids onto a single set of parameters, and ensures at the same time that at the fluid–solid interface, no overlaps or voids are created between the different numerical grids. The approach is illustrated on the optimization of the SRV2-O radial compressor where concurrently the aerodynamic performance at two different operating points, the impeller moment of inertia, the maximum stress inside the material, and the vibration response of the compressor wheel are considered. Although 112 design variables are used, a 3.5% improvement in efficiency was achieved within less than 25 major iterations, while satisfying all mechanical and aerodynamic constraints. This demonstrates the cost effectiveness of the proposed approach.