Interactive multiobjective optimization of an extremely computationally expensive pump design problem

Abstract

The hydraulic design of a pump is a challenging optimization problem. It has multiple conflicting objective functions based on computationally very expensive (16–20 hours) numerical simulations, and simulation failures, meaning that simulation calls can be unsuccessful. In this article, a surrogate-assisted evolutionary interactive multiobjective optimization method is applied to designing a pump stator. A decision maker's preferences are iteratively incorporated into the solution process and the advantages of an interactive method are demonstrated in two areas: (1) reducing the computation time; and (2) finding a preferred solution that reflects the decision maker's preferences with a low cognitive load. The decision maker was satisfied with the interactive solution process and the final solution reflected his preferences well. Additionally, because he was familiar with the domain of the problem, the preferences he provided guided the search in directions where no failed simulations were encountered. Importantly, the applied method could save days of computation time.