Bearing Fault Diagnosis Based on Improved Denoising Auto-encoders

Abstract

Most of the fault characteristics of the wind power were manually marked, and the characteristics of manual labeling were based on expert experience, and in some cases, the operation law of the equipment cannot be objectively reflected. Therefore, an improved Denoising Auto-Encoders for multi-sensor data fusion diagnosis (IDAE) method was proposed. A multi-sensors data was constructed by one-dimensional layer-by-layer stacking to construct a two-dimensional matrix to realize data fusion and ensure the robustness of fault diagnosis. Then using the unsupervised learning ability of the convolutional Auto-Encoding neural network enables the network to automatically extract fault features from the unlabeled data, ensuring the comprehensiveness, objectivity and adaptability of the fault features. Experiments on the actual historical data of Huarui FL1500 wind turbine in a wind farm in Shandong show that the proposed method has better robustness and automation in fault diagnosis of bearing fault diagnosis.