Accurate Estimation of Peak Performance in Automotive Electric Machines Using Advanced AI-Based Deep Saturation Extrapolation Technique

Amit Roy, R. Kumar, G. Vakil, Gaurav Kumar, M. Khowja, Kuldeep Singh, T. Zou, C. Gerada, A. Choubey

High-performance electric steels are used in automotive electric machines to realize superior efficiency and power density. These laminations are pushed to deep saturation during peak operation of the electric machine. The lack of material data on deep saturation results in inaccurate modeling of peak operating conditions of automotive electric machines. This has led to the development of several empirical models for extrapolating the available B–H curves. These empirical models require additional test data and are found to have limited accuracy in deep-saturation prediction. Additionally, experimental deep saturation characterization requires dedicated high-power amplifier-based test equipment, which is not always available. To overcome these challenges, this article introduces a practically impactful AI-based application using a multimaterial artificial neural network (ANN) model to predict the deep saturation behavior of high-performance electric steels accurately. Several machine learning (ML) algorithms are studied with single- and multimaterial data input strategies. Among them, the ANN model with multimaterial input is found to consistently outperform the others in predicting the saturation region, indicating its reliability and generalizability. Finally, the performance of the proposed multimaterial input-based ANN model is demonstrated by predicting the peak torque of a prototyped and experimentally validated high-speed permanent magnet-assisted synchronous reluctance (PM-SynRel) machine.