Coupled electromagnetic/thermal machine design optimization based on finite element analysis using high-throughput computing

Comprehensive optimization of an electrical machine design requires that its electromagnetic (EM) and thermal performance must be analyzed and optimized simultaneously since electric machines are heavily constrained by thermal limits. This research presents a coupled EM/thermal model that can efficiently identify the maximum current density for a given machine during static operation, together with the integration of this coupled model into an iterative machine design optimization program. An artificial neural network (ANN) that is capable of effectively characterizing input/output relationships for nonlinear multivariable functions is incorporated into the optimization program, resulting in a significant reduction of the total computation time.;For demanding applications such as traction motors, the electric machine is frequently required to run at peak power conditions for short periods of time, causing large thermal swings. In addition to steady-state operating conditions, a transient version of the coupled EM/thermal model has been developed. This makes it significantly easier for machine designers to maximize the winding current density to achieve the highest possible torque/power ratings within thermal limits set by the winding insulation or demagnetization threshold requirements. Furthermore, this transient model has been integrated into the optimization program to give it the capability of optimizing the machine designs for both steady-state and short-duration transient operating conditions.;Although finite element (FE) analysis is a powerful analytical tool for electric machines, it is rarely used in iterative machine design optimization programs since it is computationally intensive, requiring excessive calculation times. This research introduces an approach for overcoming this obstacle using a high-throughput computing (HTC) environment that harnesses the parallel processing capabilities of large numbers of computers to evaluate many candidate designs simultaneously. Differential evolution has been selected as the optimization algorithm that applies FE analysis to maximize the electromagnetic performance according to an objective function in a computationally-efficient manner. Tests comparing the computational speeds achieved using the same optimization software with the HTC resources and a single computer have demonstrated a major reduction (approx. 30:1) of the computation time using the HTC approach.