Aircraft Engine Remaining Useful Life (RUL) Prediction using Deep Learning

To effectively conduct predictive maintenance on aircraft engines, it is important to accurately predict Remaining Useful Life (RUL). This paper presents a framework that combines different models into a hybrid stacking model using an XGBoost meta-learner; specifically, GRUs (Gated Recurrent Units) and Transformers. The purpose of using these models is to utilize both temporal dependencies and the nonlinear interaction between the features of the multivariate time-series data. Using empirical validation with experiments utilizing the NASA C-MAPSS Jet Engine Stimulated dataset, it has been demonstrated that the proposed model produces significantly superior results with RMSE = 19.53 and R² = 0.7986, compared to any of the individual or conventional ensemble predictions made with individual predictive models. In addition to providing a method for predicting RUL accurately, an uncertainty estimation method based on model disagreement is also presented to help improve prediction reliability. The empirical results demonstrate that the proposed approach is a viable and robust option for use in real-world prognostic applications.