Stator Inter-Turn Fault Severity Classification Using Neural Based Approaches

The frequent failure of Induction Machines (IMs) can result in serious accidents that cause monetary losses and environmental harm. Despite their excellent reliability, these electromechanical devices are prone to various errors. This is because IMs are commonly used as the “utility player” in multiple applications and are vulnerable to stress and severe working conditions. An efficient and trustworthy diagnosis is required to overcome such problems. This paper therefore, presents a Neural-based classification approaches to extract characteristics from Stator Inter-Turn Faults (SITF) frequency distribution. The conditions, such as the healthy and SITF working states, are determined by understanding the attributes from the 20 kHz sample frequencies categorized from three severity levels (severity levels 1, 2, and 3). Principal Component Analysis (PCA) was additionally used to study the geometry of the sample data. The testing results from this study validated the robustness and durability of the proposed approach, which could detect IM faults with an accuracy rate of 95.2 %, concluded by the Multilayer Perceptron (MLP) method.