LSTM and WaveNet Implementation for Predictive Maintenance of Turbofan Engines

Abstract

With the development of technology, the condition analysis of industrial machines over sensor data has become more commonly used. These developments have made the processing and interpretation of sensor data a new problem to be solved. In this study, Long Short Term Memory (LSTM) and WaveNet are used to produce solutions for predicting remaining useful life. Experimental studies were performed on jet engine sensor data containing degradation and failures using subsets FD001 and FD003 in C-MAPSS Turbofan engine dataset. The results obtained were analyzed using various evaluation metrics and graphs. Mean square error (MSE) values obtained from LSTM tests are 11.02 for FD001 and 26.89 for FD003. MSE values obtained from WaveNet tests are 10.85, 20.71 for FD001 and FD003 respectively. Finally, tests are also evaluated by using decision level fusion and applying weighted sum rule to LSTM and WaveNet models. The results obtained clearly showed that the decision level fusion system have promising results for predictive maintenance.