Combining generalized fault trees and k-LSTM ensembles for enhancing prognostics and health management

Accurately predicting the remaining useful life (RUL) is essential for effective prognostics and health management, enabling optimal maintenance scheduling. However, the black-box nature of many deep learning models hinders their adoption by maintenance practitioners. This study presents k-LSTM-GFT, a novel hybrid framework that integrates automated data-driven generalized fault tree (GFT) construction with a k-fold ensemble of LSTM models to enhance both interpretability and prediction accuracy. Unlike prior work, our method automatically constructs fault trees from sensor data, simultaneously enabling interpretable failure pathway analysis and enriching temporal features for RUL estimation, a unified integration that constitutes a key contribution of this work. Experimental results on benchmark datasets (C-MAPSS and N-CMAPSS) show that k-LSTM-GFT outperforms state-of-the-art models, reducing RMSE by over 18% in multi-fault scenarios. Validation on a real-world injection molding machine dataset further confirms its industrial applicability. By combining model transparency with robust performance, k-LSTM-GFT offers a practical solution tailored to the needs of maintenance practitioners. • A mixture of generalized fault trees (GFTs), and LSTM for RUL prediction. • Automatic failure analysis, enabled by the mixed model. • Ensemble method to enhance the performance of the deep learning model. • Validation on C-MAPSS, N-CMAPSS, and a real industrial dataset. • Results on the more realistic datasets outperform state-of-the-art solutions.