Low-speed rolling bearing fault diagnosis based on EMD denoising and parameter estimate with alpha stable distribution

• Published in: Journal of mechanical science and technology Vol. 31; no. 4; pp. 1587 - 1601
• Main Authors: Xiong, Qing, Xu, Yanhai, Peng, Yiqiang, Zhang, Weihua, Li, Yongjian, Tang, Lan
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society of Mechanical Engineers 01.04.2017; Springer Nature B.V; 대한기계학회
• Subjects: Acceleration; Bearings; Computer simulation; Control; Damage; Damage assessment; Dynamical Systems; Empirical analysis; Engineering; Fault diagnosis; Feature extraction; Industrial and Production Engineering; Kurtosis; Mechanical Engineering; Noise reduction; Parameter estimation; Parameter identification; Parameter sensitivity; Parameters; Particle swarm optimization; Probability density functions; Roller bearings; Rolling bearings; Sensitivity analysis; Stability analysis; Test procedures; Vibration; Wheelsets; 기계공학; Fault diagnosis; Low-speed rolling bearing; Empirical mode decomposition; Alpha stable distribution; Least squares support vector machine; Particle swarm optimization algorithm
• ISSN: 1738-494X; 1976-3824
• DOI: 10.1007/s12206-017-0306-y

When low-speed rolling bearings fail, it is hard to diagnose the extent of their damage. We developed a test rig to simulate the lowspeed rolling bearing operating condition, where bearings with various fault states are installed on the test wheelset and subjected to the same external loading condition. The collected bearing box acceleration time histories are processed with the Empirical mode decomposition (EMD) method combined with kurtosis criterion to filter the trend and noise components. Five characteristic parameters of Alpha stable distribution (ASD) are identified by fitting the ASD distribution to the vibration acceleration signals and computing the Probability density function (PDF). To highlight the advantage of ASD method in feature extraction, kurtosis also has be calculated. Through sensitivity and stability analysis of the six parameters and utilization of Least squares support vectors machine (LSSVM) with Particle swarm optimization (PSO), three most sensitive and stable feature parameters including the characteristic exponent α , the scale factor γ and the peak value of the PDF h are located and applied to evaluate the low-speed rolling bearings’ damage position and damage extent. The proposed method was validated by test data, and the results demonstrated that the ASD characteristics combined with PSO-LSSVM can not only achieve fault diagnosis of low-speed rolling bearings' damage position and damage extent, but also have better diagnosis accuracy and operational efficiency than other methods.