Reduction of Memory Consumption in Time Domain Adjoint Variable Method for Topology Optimization of Squirrel-cage Induction Motor Driven by Pulse Width Modulation Inverter

抄録

Among the various types of motors, the induction motors (IMs) are the most commonly used. One approach to improve the performance of IMs is involved using a design optimization method that combines electromagnetic field analysis with topology optimization (TO). The gradient-based method is considered superior to the other methods from the viewpoint of the convergence speed and elapsed time when dealing with IMs with long transient phenomena. The time domain adjoint variable method (TDAVM) is generally used to calculate the gradients of the objective function based on time integration. TDAVM is highly efficient in calculating objective function gradients; however, it requires the storage of state variables at all time steps for the temporal discretization. In this study, a memory reduction technique in which state variables are approximated by a discrete cosine transform (DCT-II) is proposed to reduce the memory consumption of the TDAVM in TO for an IM driven by a three-phase pulse width modulation inverter, which is applied to the strongly coupled analysis of magnetic fields and electric circuits. By applying the DCT-II to the TDAVM, an optimized structures similar to those derived from the conventional TDAVM are obtained. Moreover, 85.6% memory reduction was achieved.