Large-Scale PFN Fault Diagnosis Method Based on Multidimensional Time Series Anomaly Detection Using Convolutional Neural Network

Abstract

The fast fault diagnosis of large-scale pulse-forming network (PFN) of electromagnetic rail launch system is of great significance. It is an important research direction to diagnose the PFN fault by detecting the anomaly of multidimensional time series in each pulse-formatting unit (PFU). Due to the high sampling rate of time series and the characteristics of aperiodic instantaneous pulse, the current time series diagnosis method is not suitable. Multidimensional coupling aggravates the complexity of the problem. In order to solve the problem, this article proposes a method of multidimensional time series anomaly detection based on convolutional neural network. First, the multidimensional time series is transformed into a gray-scale image, which can reflect the health status of a PFU. Second, a deep convolution neural network is constructed to identify the abnormal images based on the existing data trainers. Finally, a convolution neural network is constructed to locate the fault. The results show that the model can identify 100% of the abnormal PFU waveform, and the fault classification model can also accurately locate the fault, which proves the effectiveness of the proposed algorithm.