Neural network-driven optimization of electromagnetic and thermal performance in traction induction machines through rotor design modifications

Squirrel-cage induction machines (SCIMs) are widely used in traction and industrial applications owing to their robustness, simple construction and cost-effectiveness. However, temperature rise within the machine can negatively impact performance, reduce reliability and shorten operational lifespan, making thermal considerations essential during the design process. Traditional methods for optimizing rotor bar number and shape focus on electromagnetic performance, often overlooking thermal effects, limiting practical effectiveness. Considering both electromagnetic and thermal behaviours substantially increases computational demands, making iterative finite element analysis (FEA) impractical. This article introduces a neural network-based modelling and optimization framework for SCIMs in traction applications. By evaluating multiple rotor bar configurations under fixed design parameters, the framework efficiently refines rotor bar dimensions, enhancing performance while controlling losses and temperature. Generalizability is demonstrated through a case study with distinct specifications. Benchmarking against direct FEA optimization shows substantial computational savings with comparable accuracy, offering an effective approach for thermally resilient machine design.