A Deep Residual Convolutional Neural Network based Bearing Fault Diagnosis with Multi-Sensor Data

Abstract

Accurate bearing fault diagnosis is essential and significant for the safety and reliability of industrial rotating systems. To monitor the health condition of bearings in full scales, multi-sensors are always assembled in different locations of bearings. The data from different types or different locations of sensors have different sensitivity for fault diagnosis and localization, which makes feature extraction from multi-source data is critical for diagnosis. Deep learning based approaches have been widely used in bearing fault diagnosis, and have achieved significant successes. Deep residual convolutional neural network, as an improved structure of convolutional neural networks, has powerful learning ability when dealing with a deeper network. To improve the training efficiency and accuracy of diagnosis and make better use of bearing vibration data, this paper proposes a deep residual convolutional neural network based bearing fault diagnosis in which multisensor data are converted and combined into grayscale images for feature extraction and diagnosis. The proposed approach is verified using the data of tapered roller bearings, which are tested under different rotating speeds and loads. Experimental results and comparisons show that the proposed approach can achieve promising diagnosis accuracy with high efficiency.