Three-Stage Framework for Predicting Fault Modes and Remaining Useful Life in Aircraft Engines

The prediction of Remaining Useful Life (RUL) for aircraft engines has been a focal point in predictive maintenance research, given its critical role in reducing maintenance costs and ensuring aviation safety. Accurately forecasting RUL is challenging as aircraft engines experience multiple fault modes during continuous operation, each impacting prediction accuracy. To address this, we propose a three-stage RUL prediction framework. In the first stage, an autoencoder reduces the dimensionality of the sequence data. In the second stage, time series clustering are applied to the reduced sequences to label the fault modes. Finally, in the third stage, a joint model is trained to simultaneously predict both the fault mode and the RUL. Experiments conducted on the public C-MAPSS dataset demonstrate that our method achieves performance gains of 4.1% and 13.1% in Root Mean Square Error and Scoring Function, respectively, with a fault mode prediction accuracy of 98.8%. Compared to other studies, our approach not only improves predictive performance but also effectively identifies the fault modes and RUL of the samples.