A novel optimized multi-kernel relevance vector machine with selected sensitive features and its application in early fault diagnosis for rolling bearings

Abstract

Since the vibration signal of mechanical equipment with early faults is highly similar to that of mechanical equipment under the normal state, it is still a great challenge to extract sensitive features from the original vibration signal to execute the intelligent fault diagnosis for mechanical equipment. An early fault diagnosis method for rolling bearings based on multi-kernel relevance vector machine and multi-domain features was proposed in this paper. In order to reflect the time-varying characteristics, the vibration signals and operation status of rolling bearings in time series were incorporated into the process of early fault diagnosis. The three steps of the early fault diagnosis method for rolling bearings were as follows. Firstly, the vibration signals of rolling bearings during operation were measured online. Secondly, the original vibration signals were decomposed by wavelet packet transformation. The fault features were extracted from sensitive frequency band by time domain statistical analysis as well as, frequency domain statistical analysis, and then the multi-domain feature set was constructed to fully characterize the intrinsic properties of vibration signals. The correlation analysis was adopted to eliminate insensitive features from original multi-domain feature set. The low-dimensional feature set that are highly sensitive to early failures was reconstructed to improve the computational efficiency for subsequent fault diagnosis. Finally, the intelligent fault diagnosis was carried out based on the multi-kernel relevance vector machine model. The performance of this proposed method has been validated in practical rolling bearing fault diagnosis. The results show that the proposed method can achieve higher diagnosis accuracy for rolling bearing under different working conditions than traditional single-kernel model and is effective in early fault diagnosis.