Application of Multi-Scale PCA and Energy Spectrum to Bearing Fault Analysis and Detection in Rotating Machinery

Abstract – This paper introduces various works on fault detection of rotating machines caused by bearings damage. The novelty of this work is the application of new scheme based on the multi-scale principal component analysis MSPCA, and the energy spectrum of details coefficients of the discrete wavelet transform (DWT). The DWT coefficient details are calculate in first stage in order to be used as inputs of the MSPCA scheme and in the second stage they are used to evaluate the spectrum energy. Keywords : Fault analysis, rotating machinery, MSPCA, wavelet transform, spectrum energy, bearing diagnosis. I. I NTRODUCTION olling element bearings are essential elements in most rotating machines. Bearing failures are prone to cause both personal damage and economic loss if not be detected well in advance [1-5]. To avoid such catastrophic failures, it is necessary to develop and implement efficient diagnosis monitoring systems that are independent of operating conditions. Therefore, a significant amount of research efforts have focused on the predictive maintenance of machines. Machine Vibration Signature Analysis (MVSA) provides an important way to assess the health of a machine. In traditional MVSA, the Fourier transform is used to determine the vibration spectrum. Typically, frequencies of bearing defects are identified and compared with initial measurements to detect any deterioration in bearing health. In recent years [5], different technologies have been used in order to process signals produced by dynamical systems. Most of the authors classify the analysis of vibration signature in three approaches. The first one is the time domain, based on statistical parameters such as mean, root mean-square, variance, kurtosis, etc. The second approach proceeds in the frequency domain, where the Fourier transform and its numerous variants have been intensively used. The shortcoming of this approach is that Fourier analysis is theoretically limited to stationary signals, whilst bearing vibrations are typically non-stationary by nature [6]. The third approach is based on the time-frequency analysis such as the short-time Fourier transform (STFT) or the (discrete) wavelet transform (DWT). It has been successfully applied as a fault feature extractor due to its