Design Optimization of Traction Motors using a Quasi-Monte Carlo-based Two-Step Method

In the past decade, various approximation techniques have been thoroughly studied to minimize the computational cost for design optimization of permanent magnet synchronous machines using finite element analysis (FEA). However, most approximation techniques face the "curse of dimensionality&quo...

Full description

Saved in:
Bibliographic Details
Main Authors: Choi, Mingyu, Choi, Gilsu, Bramerdorfer, Gerd
Format: Conference Proceeding
Language:English
Subjects:
artificial neural network
Artificial neural networks
Computational efficiency
Computational modeling
Interior permanent magnet synchronous machine
multi-objective design optimization
Predictive models
quasi-Monte Carlo method
Simulation
Space exploration
surrogate model
Traction motors
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the past decade, various approximation techniques have been thoroughly studied to minimize the computational cost for design optimization of permanent magnet synchronous machines using finite element analysis (FEA). However, most approximation techniques face the "curse of dimensionality" problem, in which computation time increases exponentially with the number of input parameters considered. In this paper, we present a two-step design optimization approach based on artificial neural networks (ANNs) and quasi-Monte Carlo (QMC) methods for traction motor applications. First, a surrogate model is constructed using ANN to approximate the predictions by FEA. The proposed two-step method improves the estimation accuracy by starting with a small number of initial samples and then iteratively selecting additional samples near the Pareto front obtained based on the initial sampling. This allows for significant expansion and refinement of the Pareto front. Finally, the results of the proposed algorithm were compared with those of non-dominated sorting genetic algorithm II (NSGA-II). The results show that comparable results are obtained at a significantly lower computational cost.
ISSN:2642-5513
DOI:10.1109/ICEMS56177.2022.9983402