GASF–MSNN: A New Fault Diagnosis Model for Spatiotemporal Information Extraction
Siyu Sun, Jia Ren
Abstract
Deep learning has achieved outstanding performance in many fields due to its powerful feature abstraction and extraction capabilities. Various attempts at fault diagnosis in industrial processes based on deep learning have attracted much attention. However, industrial process data are mostly high-dimensional time series, and the types of faults are closely related to the temporal and spatial characteristics of the data. Therefore, how to maximize the utilization of the effective information in the data is the key to fault diagnosis. Based on this thinking, this article proposes a fault diagnosis algorithm based on a Gramian angular field and multi-scale feature extraction. The temporal information in the data is fully extracted by introducing a Gramian angular field by which the multi-dimensional temporal data are converted to multi-channel two-dimensional image data. Two multi-scale convolution modules are suggested to fuse and extract the spatiotemporal information at different scales. Single-channel convolution and global pooling are selected to complete the final classification task. The proposed algorithm is compared to four typical deep learning algorithms on two benchmarks: the Tennessee Eastman process and the three-phase flow process. The test results show that the proposed algorithm has obvious advantages in terms of average recall and average F1 score.