Fault Prediction of Electromagnetic Launch System Based on Knowledge Prediction Time Series

The fault prediction of the electromagnetic launch (EML) system is an important guarantee to improve the reliability of the system, but there is no mature method that can be directly applied. Combined with the engineering practice of large-scale EML system, a fault prediction method based on knowledge prediction time series is proposed. First, the high-frequency waveform collected in each launch is extended into a time series along the number of launches; second, an intelligent waveform features extraction expert system is constructed to realize feature extraction; third, multidimensional feature sequence prediction and waveform prediction are realized by using two neural networks, respectively; finally, fault prediction is realized by associating the fault diagnosis knowledge. The temperature rising test data of a railgun system for 15 consecutive launches and the recoil stroke test data of noncontinuous 78 launches are used as the input source of the proposed algorithm. The results show that the proposed algorithm can automatically extract the features with fault trend. The single step prediction error of features is less than 1.47%, and the mean square error of curve prediction is half of the results of SARIMA prediction algorithm. Through the temperature rise curves, the proposed algorithm can predict the fault free of the 14th and 15th launch. According to the recoil stroke curves, the fault of the launcher of the 75th launch is predicted. The actual analysis shows that the fault prediction accuracy is high, and the algorithm can significantly improve the system reliability after being applied to engineering.