Cyclo-non-stationary analysis for bearing fault identification based on instantaneous angular speed estimation

International audienceRolling Element Bearings (REB) are present in most of the rotating machines being the ones in charge of supporting the charge of the shaft, being the cosntant charge of the shaft a reason to make the REB prone to fail. Failures under real conditions such as variable speed/load are a subject of interest in the state of the art in digital signal processing of vibration signals, lastly, defined as a cyclo-nonstationary process due to the intrinsic cyclic behaviour of the REB and the nonstatinarity introduced by the variations of the Instantaneous Angular Speed (IAS). The most direct approach to deal with a REB failure under time-varying IAS, is to compensate the IAS transforming the vibration signal to the angular domain, then highlight the cyclo-stationary part of the signal in the angular domain. To obtain the IAS, the direct approach is to measure the IAS via an encoder to obtain the so-called tachometer signal, to place an encoder usually requires a modification of the machine. In cases where it is not possible, the IAS could be extracted directly from the vibration signal. To extract the IAS from a vibration signal is a hard task due to to the low Signal to Noise Ratio (SNR); consequently, a short- time approach methodology robust to noise for IAS estiamtion termed, Short Time Non-Linear Least Squares (STNLS) estimation is proposed. However, even with the IAS to identify the failure requires an additional step that is to highlight the impulsive behaviour, several techniques in the literature makes use directly or indirectly of the Spectral Kurtosis (SK) to highlight impulsive behaviour. The SK is designed to work under small variations of the IAS, even when the variation on the IAS could be compensated through the transformation to the angular domain; there are components angle-variant like the transfer function, that could mask the impulsive components. Thus, a short-angle method based on the SK named Short Time/Angle Spectral Kurtosis (STSK) is introduced. The STSK method is compared with the traditional approach outperforming in both numerical an a challenging case study of an aircraft engine. Similarly, the STNLS is tested on a numerical database for robustness to noise and in the real signal showing a Mean Square Error bellow 5 × 10−3