Bearing defects classification using wavelet time scattering features and machine learning techniques

Rolling element bearings (REBs) are critical components used in almost all rotating machinery. Small defects, such as pits and spalls, are formed on bearing surfaces due to cyclic loading, results in failure in these rotating machinery and sometime production downtime. Early detection of these small defects is essential to avoid such failure in machinery. In this work, a technique for bearing fault classification using the Wavelet Scattering Transform (WST) and machine learning is proposed. The proposed technique is based on principles of signal processing i.e. wavelet transforms, extracts features that are stable under small deformations and invariant to time shifts. These features are captured automatically from the WST images and are used to train conventional machine learning techniques such as Support Vector Machine, Artificial Neural Network and Decision Tree. The proposed methodology is validated using experimental data from the Case Western Reserve University Bearing Data Centre. Results indicate that WST-based feature extraction, with various classification algorithms, significantly improves the accuracy of bearing fault diagnosis, offering a robust solution for early defect detection.