Reducing Development Time of Electric Machines with SyMSpace

摘要

This paper presents methods to accelerate the optimization of electrical machines using the software tool SyMSpace. Due to the nonlinear properties of soft magnetic materials, finite element analysis (FEA) is typically used for the simulation of electrical machines. For a complete optimization run hundreds to several thousand FEA calculations are required, which are computationally very expensive. Simple measures such as consideration of symmetries in the geometry to more sophisticated techniques like generation of a surrogate motor model can easily achieve a significant reduction in the calculation effort. By means of novel optimization algorithms specially designed for electrical machines, it is possible to achieve faster convergence of the Pareto front. To further speed-up the optimization a nonlinear mapping between the optimization variables and objectives based on artificial neural networks (ANNs) is derived during the optimization run to cut down the simulation time significantly. Once the optimization has converged, the most suitable machine for the particular application can be selected from the Paret front for further detailed analysis. For example, it is possible to generate an accurate motor model for further dynamic simulations in the form of a functional mock-up unit (FMU). Additionally, it is also possible to create data files for rapid prototyping fully automatically. This comprises, for example, data files for laser cutting, STL files for 3D printing of insulation parts and generation of program code for a needle winding machine.