Fault Diagnosis Based on Sensitive SVD and Gaussian Process Latent Variable Model

To solve the problems that the system running state feature in fault diagnosis is sometimes masked by noise and its high dimensionality decreases the fault recognition degree. A fault diagnosis method based on Sensitive Singular Value Decomposition (Sensitive SVD) and Gaussian Process Latent Variable Model (GPLVM) is proposed. The method firstly performs Sensitive SVD analysis on the vibration signal, extracts various time domain features from the reconstructed signal, constructs a high-dimensional feature set, and uses GPLVM to reduce the dimensionality, and then use the reduced feature to establish the Extreme Learning Machine (ELM) fault diagnosis model. The rolling bearing fault detection test shows that the proposed method can effectively reduce the redundancy of features, and the established fault diagnosis model has higher identification accuracy.