Rolling Bearing Fault Diagnosis Based on DTCWPT and DBN

This paper constructs a fault analysis and identification model by employing the dual-tree complex wavelet packet transform (DTCWPT) together with the deep belief network (DBN), aiming to achieve precise fault diagnosis of rolling bearings. The DTCWPT decomposition of the vibration signals was first performed, and an initial feature set of the fault pattern was developed by extracting the features of initial vibration signals under different frequency bands. Subsequently, the Laplacian Score (LS) approach was utilized to detect the fault-sensitive characteristics within the original high-dimensional feature collection. Furthermore, leveraging depth learning techniques which are proficient in high-dimensional data manipulation and nonlinear data analysis, an adaptive exploration of fault characteristics and an intelligent discrimination of faults were carried out with the Dunn Validity Index and standard deviation ratio in the context of DBN. Four tests for different cases were performed using the previously reported bearing data. The experimental outcomes manifested that the LS approach is efficacious in extracting fault-sensitive features, and the DBN model is capable of enhancing the precision of fault recognition.