Fault Diagnosis of Bearing Under Varying Speed Using Hilbert Computed Order Spectrum and Ensemble of Auto-Encoders

Bearings, working under tough conditions, are prone to break down, which cause high maintenance costs and safety accidents. Fault diagnosis technique is of critical significance for bearings. However, due to the variable speed conditions, extracting features from nonstationary vibration signals and constructing a accurate fault diagnosis model are still challenging. This study presents a bearing fault diagnosis approach under varying speed using the Hilbert computed order (HCO) spectrum and ensemble of various types of auto-encoders (EAE). The HCO spectrum, as a preprocessing strategy, is generated through order analysis of the signal after Hilbert demodulation. Subsequently, distinct types of AEs are independently trained to automatically extract features from the order spectrum and accomplish the fault diagnosis task. By a soft voting process, the final classification result of the test sample is obtained. The effectiveness of the proposed bearing fault diagnosis scheme is demonstrated through a widely used bearing dataset.