GEAR FAULT SEVERITY CLASSIFICATION USING ACOUSTIC EMISSION PEAKS AND POINCARÉ PLOTS

Acoustic Emission (AE) signals are broadband signals used for fault detection in rotating machinery. Acquiring AE signals requires high sampling frequencies, often exceeding 1 MHz, which necessitates significant memory resources for storage. This limitation renders such signals impractical for fault condition monitoring in applications where remaining useful life prediction is crucial. This research proposes a novel peak-based signal representation for AE signals. This representation significantly reduces the memory space required for storage. Furthermore, we propose a feature set derived from this representation, combining Poincaré plot-based and statistical features. We evaluate its effectiveness for classifying broken tooth fault severity in a spur gearbox. Using Random Forest models, we achieve the highest classification accuracy of 94.46%.