Mechanical Fault Detection in Induction Motors Using a Feature-Based Kalman Filter

In this paper, a data-driven algorithm to identify the eccentricity fault in induction motors is proposed. The algorithm is based on the Kalman Filter (KF) and utilizes experimental data collected from healthy and faulty three-phase stator currents at different speeds and load conditions. Additional data processing techniques including Discrete Wavelet Transform (DWT), Power Spectral Density (PSD), and cepstrum are used to extend the dataset. A feature extraction process involving a few statistical measures is applied to this dataset. For each feature, a State-Space Model (SSM) and a KF are formulated. By comparing the resulting output of the SSMs with the estimated output from KFs, a measure to identify an eccentricity fault is obtained. This method was tested on various operating modes of an induction motor, demonstrating its effectiveness in distinguishing healthy data from those indicating an eccentricity fault.