SPLSTM: A Prediction Model for Operational Faults in Rotating Machinery

The study of rotating machinery fault diagnosis and prediction technology is of great significance to ensure equipment stability and engineering safety. However, the operating signals of rotating machinery obtained through the sensors are highly nonlinear. How to improve the accuracy of signal nonlinear prediction has become an essential topic in fault prediction research nowadays. For this reason, this article researches rotating machinery fault prediction based on the high nonlinear prediction accuracy of long short-term memory (LSTM) networks. First, in the idea that rotating machinery reaches the critical degradation stage in advance, its operational faults are defined as sudden faults, degradation faults, and early weak faults. Second, to address the diversity of signals over the whole life of rotating machinery, the performance degradation index mean grey relation index (MGR) is proposed, and the operational phases of rotating machinery are redefined by dividing the initial and aggravated degradation phases. Then, to address the problem that early weak faults are difficult to predict accurately, superposition (ST1) and progressive (PT2) dual prediction channels are constructed, the comparison and selection mechanism between the channels is designed, and the superposition and progressive prediction LSTM network prediction model (SPLSTM) is proposed. Finally, the feasibility and effectiveness of SPLSTM in predicting operational faults of rotating machinery are verified by test-bed and wind power engineering tests. In the test-bed test, the prediction accuracy of the early weak fault interval is 95.56%. In the wind turbine test, the prediction accuracy of early weak and aggravated degradation faults is 87.81% and 93.76%, respectively.