Impact of Phase Current Sequencing in Machine Learning Models for Induction Motor Fault Detection

Induction Motors (IMs) are widely preferred in industrial, commercial, and residential settings due to their simplicity, reliability, cost-effectiveness, and low maintenance needs. Their durable design, versatility in different environments, and efficiency make them well-suited for a wide range of applications, contributing to their global popularity. Consequently, fault diagnosis and Condition-Based Monitoring (CBM) have become crucial research areas for developing non-invasive, online detection methods, with numerous studies and advancements reported in literature. The primary purpose of these efforts is to prevent the consequences of unexpected failures in critical equipment, which can lead to extended process shutdowns and costly repairs. However, existing techniques that use multiple phase current signals as inputs to Machine Learning (ML) models maintain the phase sequencing of the motor. This practical limitation requires operators to install current sensors on specific motor windings and restrict the motor to operate in a single direction. This article aims to highlight the negative effects of training models without augmenting phase current sequences and explores the potential benefits of models that incorporate such phase current sequence augmentation for better generalization.