Probabilistic machine learning framework for peripheral dimension optimization of five-phase induction motors

This article proposes a probability-based Machine-learning (ML) framework for peripheral-dimension optimization of a Five-Phase Induction Motor (FPIM). The proposed approach investigates the influence of Inner Stator Diameter ( D ), Core Length ( L ), and L/D ratio on motor-performance parameters, including efficiency ( η ), power factor ( p f ), and magnetic flux density ( B a v ). Machine volume ( D 2 L ) is maintained constant. A dataset consisting of 1000 electromagnetic simulation cases was generated using detailed Finite Element Modelling (FEM) in ANSYS Maxwell. The generated dataset was subsequently used for training and evaluation of multiple regression-based ML models, including Linear Regression (LR), Decision Tree Regressor (DTR), Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Regression (SVR), and Gaussian Process Regression (GPR). Unlike conventional deterministic optimization approaches, proposed framework treats inverse mapping between performance parameters and peripheral dimensions as nonlinear and partially non-unique. Accordingly, probabilistic and uncertainty-aware learning strategies are incorporated to identify statistically favourable and practically feasible L/D regions rather than a single deterministic optimum. Repeated k -fold cross-validation, sensitivity analysis, and comparative model evaluation were performed to assess prediction robustness, generalization capability, and stability. The probabilistic GPR-based framework demonstrated superior overall prediction consistency and robustness for the considered inverse-design problem. The developed Artificial Intelligence (AI) framework is subsequently applied to the peripheral-dimension optimization and performance-oriented design selection of FPIM. Comparative analysis demonstrates improved flexibility in performance-oriented design selection and enhanced capability for handling nonlinear electromagnetic relationships in FPIMs. The obtained optimized motor configuration exhibits stable operating characteristics with acceptable η , p f , current, and torque-ripple performance under investigated loading conditions.