Wavelet kernel and convolution neural network based accurate detection of incipient stator and rotor faults of induction motor

Abstract Early detection of incipient faults in three-phase induction motors is crucial to enhance system reliability and to minimize unplanned operational interruptions in industrial environments. Traditional diagnostic techniques often struggle to detect incipient faults, especially under fluctuating load conditions and may require complex signal processing or multiple sensors. The paper introduces a method for early detection of faults in three-phase induction motors using Wavelet Kernel-enabled convolutional neural networks (CNNs). The proposed system accurately identifies stator interturn faults in single or multiple phases and broken rotor bar faults, even under varying operating conditions such as load variations. By employing 14 mother wavelets as convolution filters, the method effectively extracts critical features from stator current signatures, streamlining the fault detection and classification process. This technique leverages the deep structures of CNNs to autonomously learn features from current signals, achieving a notable accuracy of above 97% in tests with both simulated model and two different hardware motor setup. The experimental result shows that it is capable of detecting as low as 1–2% of stator interturn fault with varying impedance in short circuit path as well as one broken rotor bar fault. Overall, the proposed method proves to be a powerful tool for the early diagnosis of incipient faults in induction motors with high degree of reliability and effectiveness.