| Title |
Predicting Remaining Life of Aircraft Turbine Engines using EBM-LSTM Hybrid Models |
| Authors |
김기현(Kihyun Kim) ; 김주현(Juhyun Kim) ; 임대영(Daeyeong Lim) ; 권종원(Jongwon Kwon) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.10.76 |
| Keywords |
Remaining useful life; Explainable artificial intelligence; Feature selection; Ensemble techniques |
| Abstract |
This paper proposes an EBM-LSTM hybrid methodology that combines the Explainable Boosting Machine (EBM), an explainable artificial intelligence (XAI) model, with the Long Short Term Memory (LSTM), a time series prediction model, to simultaneously improve interpretability and prediction accuracy in aircraft turbine engine Remaining Useful Life (RUL) prediction. The proposed methodology utilizes the variable importance provided by EBM to select variables that significantly influence RUL prediction from high-dimensional sensor data, which are then used as inputs for the LSTM model. This process ensures transparency in variable selection while reducing model complexity and training costs by eliminating unnecessary variables. To validate the effectiveness of the proposed methodology, NASA's C-MAPSS dataset was utilized, and the performance of the EBM-LSTM hybrid model was compared and analyzed against individual EBM and LSTM models as well as previous studies using deep learning approaches. Experimental results showed that through the variable selection process, dimensionality reduction from 14 to 8 input variables resulted in only a 0.46% decrease in coefficient of determination (R2) performance, maintaining the model's prediction performance while demonstrating efficiency and robustness. The proposed EBM-LSTM hybrid methodology is expected to contribute to enhancing the field applicability of AI-based predictive maintenance systems in aerospace and various industrial sectors requiring high reliability by improving the reliability of RUL prediction models. |