ECLSTM: An Efficient Channel Attention-based Spatio-temporal Fusion Method for Fault Detection of Instruments

Abstract

The instrument fault detection has the problems of incomplete temporal and spatial feature extraction and poor feature expression ability of instrument data. To solve the above-mentioned, this research proposed a new intelligent fault detection method for instruments with an efficient channel attention-based convolutional neural network and long short- term memory neural network (ECLSTM). Firstly, the convolutional neural network (CNN) was used to extract the spatial features of different instruments, and the long short- term memory neural network (LSTM) was applied to extract the temporal features of the sequence. Secondly, to improve the expression ability of the features extracted by the CNN, the efficient channel attention was used to selectively weight the spatial features. At last, the healthy indicator was constructed by predicting the value of the instrument through the network model, calculating the threshold value by the exponentially weighted moving average (EWMA), and comparing with the threshold value to achieve instrument fault detection. The effectiveness and progressiveness of the method were verified by experiments.