Multiobjective electric machine optimization for highest reliability demands

Abstract

This article is about illustrating a workflow for incorporating reliability measures to typical electric machine design optimization scenarios. Such measures facilitate comparing designs not only for rated conditions, but also allow to analyze their performance in the presence of unevitable tolerances. Consequently, by additionally considering reliability or robustness as objectives compared to conventional optimization scenarios, designs featuring low parameter sensitiveness can be obtained. The analysis of the design's reliability as part of solving optimization problems involves a significant increase in required numerical evaluations. To minimize the associated prolongation of the runtime, an approach featuring a design of experiments based reduction of required computations and a consequent surrogate modeling technique is presented here. After successful training, the metamodel can be applied for fast evaluating lots of different parameter combinations. A test problem is defined and analyzed. Based on the observed findings, the necessity of incorporating robustness evaluations to machine design optimization becomes evident. In addition, the derived models allow for studying the impact of any tolerance-affected parameter on the machine performance in detail. This facilitates further beneficial studies, as for instance the analysis of selected changes of tolerance levels rather than a general minimization of the respective ranges which usually is associated with high production cost.