A multi-rate sampling data fusion method for fault diagnosis and its industrial applications

Abstract

The multi-sensor data fusion based data-driven fault diagnosis method is a promising approach to detect faults of complex systems. However, in the actual industrial environment, the sampling rate of different sensors is often inconsistent. In order to apply this kind of data to fault diagnosis, the traditional methods are to preprocess it and convert it into single sampling rate data. However, these methods are all machine learning methods, which rely on manual feature extraction. To the best of our knowledge, few works have used deep learning (DL) methods to solve this problem. To fill this gap, a novel multi-rate sampling data fusion method for fault diagnosis is proposed in this paper. In the proposed method, signals with different sampling rates are fused. First, a convolutional neural network (CNN) is adopted to learn features from raw data automatically. Then, a long short-term memory (LSTM) network is utilized to mine the time correlation in extracted features and encode the temporal information. The methodology is validated on a public experimental dataset and data from a real industrial scenario. The proposed method is compared with some state-of-the-art machine learning (ML) and DL methods, the results show that the proposed method can distinguish different conditions satisfactorily and has the best diagnostic accuracy among all comparison methods.