Point processes for bearing fault detection under non-stationary operating conditions

Bearing faults represent the most frequent mechanical faults in rotational machines. They are characterized by repetitive impacts between the rolling elements and the damaged surface. The time intervals between two impacts are directly related with the type and location of the surface fault. These time intervals can be elegantly analyzed within the framework of renewal point processes. With such an approach the fault detection and identification can be performed irrespective of the variability of rotational speed. Furthermore, we show that by analyzing the entropy of the underlying counting process by means of wavelet transform, one can perform fault detection and identification without any information about the operating conditions. The effectiveness of the approach is shown on a data-set acquired from a two–stage gearbox with various bearing faults operating under different rotational speeds and loads.