Comparison of denoising schemes and dimensionality reduction techniques for fault diagnosis of rolling element bearing using wavelet transform

This paper presents the evaluation of five wavelets-based denoising schemes in order to select the best possible scheme for denoising bearing vibration signals and dimensionality reduction techniques using artificial neural network (ANN). Vibration signals from four conditions of rolling element bearing (REB) namely normal (N), defect on inner race (IR), defect on ball (B) and defect on outer race (OR) have been denoised using interval-dependent denoising scheme, which is the best possible scheme. The denoised signal is subjected to discrete wavelet transform (DWT) to extract 17 statistical features. Principal component analysis (PCA)-based dimensionality reduction technique (DRT) namely PCA alone, Kernel-PCA (KPCA) alone, PCA using SVD and KPCA using SVD have been used for reducing the dimension of the features. It is found that KPCA using SVD resulted in highest prediction accuracy using ANN, making it suitable for effective REB fault diagnosis.