Study On The Fault Feature Extraction And Remaining Life Prediction Of Rolling Bearing Based On Vibration Signal

Rolling element bearing is one of the most widely used and easily damaged parts in rotating machinery, which is of great importance to guarantee the reliability of the whole machine. When the failure degradation indication has been detected, it is essential to implement the diagnosis the bearing faults and estimate the remaining useful life (RUL) based on the condition monitoring. This paper studied the feature extraction and the RUL estimation based on the bearing vibration signal. The article is organized as follows:(1) Based on the Case Western Reserve University seeded fault test data, multiple features selected from the time-frequency domain of rolling bearing vibration signals are analyzed and compared. The Hilbert envelope spectrum is used to demodulate the raw signal and the wavelet packet analysis is applied to extract the frequency band energy of vibration signal. Since the statistical parameters unavoidably exists irrelevant and redundant components, a novel method based on variance score and Principle component analysis (PCA) is proposed to extract the PCs to represent the fault information. The proposed method is verified with the engineering bearing data. The main advantage of this method is using the simple time-frequency statistic parameters to diagnose the bearing faults.(2) The RUL of the rolling element bearing indicates its ability of surviving the operation in the future. This paper implements the proportional hazards model (PH) and logistic regression model to estimate the RUL. First, extract the sensitive frequency band energy and root mean square parameters as covariates. Build the model using the life history data and condition monitoring data of the same components. Then carry out the real-time assessment of the RUL based on the monitoring data. Comparisons are made for the two models regarding their effectiveness and computation effort. A full life data of bearing tests is provided to demonstrate the proposed approach in practical use. The result shows that the PH model are capable of providing more accurate RUL prediction to support timely maintenance decisions.