Localized fault detection and diagnosis in

This paper puts together a collection of a number of previously proposed signal processing techniques to detect and diagnose faults in rolling element bearings. The collection of previously proposed signal processing algorithms contains two main phases. Phase one includes a surveillance and diagnosis stage using time, frequency and an envelope analysis over the full frequency bandwidth. The second phase includes a more specific and impulsiveness-targeted analysis which starts by separating the signal into deterministic and random parts using time synchronous averaging (TSA) and/or discrete random separation (DRS). It then utilizes Spectral Kurtosis (SK) analysis (Kurtogram) for a more concise and automated envelope analysis. The SK-algorithm, encoded as AR-MED-SK analysis, includes prewhitening using an autoregressive model (AR), Minimum entropy deconvolution (MED) and SK analysis. The two-phase setup hasn’t been combined in a single algorithm previously and this paper points out a number of hints to produce a reliable diagnosis. The algorithm is explained using a signal from a helicopter gearbox that has a defective planetary bearing (both inner race and roller faults), with clear identification of the source of fault.