Optimized Fault Classification in Electric Vehicle Drive Motors Using Advanced Machine Learning and Data Transformation Techniques

S. Thirunavukkarasu, K. Karthick, S. K. Aruna, R. Manikandan, Mejdl Safran

The increasing use of electric vehicles has made fault diagnosis in electric drive motors, particularly in variable speed drives (VSDs) using three-phase induction motors, a critical area of research. This article presents a fault classification model based on machine learning (ML) algorithms to identify various faults under six operating conditions: normal operating mode (NOM), phase-to-phase fault (PTPF), phase-to-ground fault (PTGF), overloading fault (OLF), over-voltage fault (OVF), and under-voltage fault (UVF). A dataset simulating real-world operating conditions, consisting of 39,034 instances and nine key motor features, was analyzed. Comprehensive data preprocessing steps, including missing value removal, duplicate detection, and data transformation, were applied to enhance the dataset’s suitability for ML models. Yeo–Johnson and Hyperbolic Sine transformations were used to reduce skewness and improve the normality of the features. Multiple ML algorithms, including CatBoost, Random Forest (RF) Classifier, AdaBoost, and quadratic discriminant analysis (QDA), were trained and evaluated using Bayesian optimization with cross-validation. The CatBoost model achieved the best performance, with an accuracy of 94.1%, making it the most suitable model for fault classification in electric vehicle drive motors.