Screening of Discrete Wavelet Transform Parameters for the Denoising of Rolling Bearing Signals in Presence of Localised Defects

Abstract

Maintenance scheduling is a fundamental element in industry, where excessive downtime can lead to considerable economic losses. Active monitoring systems of various components are ever more used, and rolling bearings can be identified as one of the primary causes of failure on production lines. Vibration signals extracted from bearings are affected by noise, which can make their nature unclear and the extraction and classification of features difficult. In recent years, the use of the discrete wavelet transform for denoising has been increasing, but studies in the literature that optimise all the parameters used in this process are lacking. In the current article, the authors present an algorithm to optimise the parameters required for denoising based on the discrete wavelet transform and thresholding. One-hundred sixty different configurations of the mother wavelet, threshold evaluation method, and threshold function are compared on the Case Western Reserve University database to obtain the best combination for bearing damage identification with an iterative method and are evaluated with tradeoff and kurtosis. The analysis results show that the best combination of parameters for denoising is dmey, rigrSURE, and the hard threshold. The signals were then distributed in a 2D plane for classification through an algorithm based on principal component analysis, which uses a preselection of features extracted in the time domain.

Keywords

• bearing fault diagnosis
• discrete wavelet transform
• thresholding
• denoising
• intelligent fault diagnosis
• Case Western Reserve University