Blind Deconvolution Based on Modified Smoothness Index for Railway Axle Bearing Fault Diagnosis

Blind deconvolution is a widely used technique for fault diagnosis of rolling bearings. Traditional blind deconvolution methods, such as minimum entropy deconvolution, are susceptible to random transients, making it difficult to extract fault features of railway train axle bearings under strong external shock conditions. Deconvolution methods that take the fault characteristic frequency of interest as an input parameter, such as maximum second-order cyclostationarity blind deconvolution, can alleviate this deficiency, however, the bearing fault features are difficult to be extracted when the specified characteristic frequency deviates from the actual value greatly. To overcome these problems, the modified smoothness index of the squared envelope and the modified smoothness index of the squared envelope spectrum are proposed as objective functions of the deconvolution algorithms, allowing two new blind deconvolutionmethods to be developed for railway axle bearing faults diagnosis. The two proposed blind deconvolution methods are robust to random transients and do not require the characteristic frequency of interest as an input parameter. The fault diagnosis performance of the two proposed methods is verified using the experimental data of actual railway axle bearings and compared with the state-of-the-art deconvolution methods. The results show that the two proposed blind deconvolution methods can adaptively extract repetitive transient features from noisy vibration signals and effectively diagnose different faults of railway axle bearings.

Automated summary

This article proposes two new blind deconvolution methods for diagnosing faults in railway axle bearings. These methods can adaptively extract repetitive transient features from noisy vibration signals and effectively diagnose different faults of railway axle bearings.