Hybrid-BiLSTM: An Attention-Enhanced Deep Learning Framework for Bearing Remaining Useful Life Prediction

This paper presented a Hybrid-BiLSTM framework for Remaining Useful Life (RUL) prediction of rolling element bearings, integrating a multi-scale convolutional front-end, dilated Temporal Convolutional Network (TCN) blocks, bidirectional LSTM layers, and a multi-head self-attention mechanism within a single end-to-end architecture. The model was trained on hybrid time- and frequency-domain features, including bearing fault characteristic band energies, extracted from the PRONOSTIA dataset (FEMTO-ST Institute), and optimized using a Huber loss with adaptive learning-rate scheduling and early stopping. Comparative evaluation against SVM, Random Forest, XGBoost, LSTM, and BiLSTM demonstrated that the proposed model achieves superior predictive performance, attaining the highest R² (0.9994) and the lowest MAE (29.34) and RMSE (37.64) among all evaluated models. These results indicate that jointly integrating multi-scale spatial feature extraction, causal long-range temporal modelling, bidirectional sequence learning, and attention-based feature weighting yields measurable improvements in prediction accuracy over architectures that incorporate these components individually or in partial combination. The close alignment between training and validation loss curves further confirms that the model generalizes effectively without significant overfitting. Future work will focus on validating the proposed framework across multiple bearings and operating conditions within the PRONOSTIA dataset, and on evaluating its cross-domain generalization through transfer learning

[1]            A. Gabrielli, M. Battarra, E. Mucchi, and G. Dalpiaz, “Physics-based prognostics of rolling-element bearings: The equivalent damaged volume algorithm,” Mech. Syst. Signal Process., vol. 215, Jun. 2024, doi: 10.1016/j.ymssp.2024.111435. [2]            S. G. Kumbhar, R. G. Desavale, and N. V. Dharwadkar, “Fault size diagnosis of rolling element bearing using artificial neural network and dimension theory,” Neural Comput. Appl., vol. 33, no. 23, pp. 16079–16093, Dec. 2021, doi: 10.1007/s00521-021-06228-8. [3]            V. M. Nistane and S. P. Harsha, “Prognosis of degradation progress of ball bearings using supervised machine learning,” Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body Dynamics, vol. 232, no. 2, pp. 183–198, Jun. 2018, doi: 10.1177/1464419317731046. [4]            Y. Lu, Q. Li, and S. Y. Liang, “Physics-based intelligent prognosis for rolling bearing with fault feature extraction,” International Journal of Advanced Manufacturing Technology, vol. 97, no. 1–4, pp. 611–620, Jul. 2018, doi: 10.1007/s00170-018-1959-0. [5]            K. Adamsab, “Machine learning algorithms for rotating machinery bearing fault diagnostics,” in Materials Today: Proceedings, Elsevier Ltd, 2020, pp. 4931–4933. doi: 10.1016/j.matpr.2020.12.050. [6]            M. G. Alfarizi, B. Tajiani, J. Vatn, and S. Yin, “Optimized Random Forest Model for Remaining Useful Life Prediction of Experimental Bearings,” IEEE Trans. Industr. Inform., vol. 19, no. 6, pp. 7771–7779, Jun. 2023, doi: 10.1109/TII.2022.3206339. [7]            M. S. Rathore and S. P. Harsha, “Prognostic Analysis of High-Speed Cylindrical Roller Bearing Using Weibull Distribution and k-Nearest Neighbor,” J. Nondestruct. Eval. Diagn. Progn. Eng. Syst., vol. 5, no. 1, Feb. 2022, doi: 10.1115/1.4051314. [8]            L. Li, J. Xu, and J. Li, “Estimating remaining useful life of rotating machinery using relevance vector machine and deep learning network,” Eng. Fail. Anal., vol. 146, Apr. 2023, doi: 10.1016/j.engfailanal.2023.107125. [9]            A. Soualhi, K. Medjaher, and N. Zerhouni, “Bearing health monitoring based on hilbert-huang transform, support vector machine, and regression,” IEEE Trans. Instrum. Meas., vol. 64, no. 1, pp. 52–62, Jan. 2015, doi: 10.1109/TIM.2014.2330494. [10]         X. Yan and M. Jia, “A novel optimized SVM classification algorithm with multi-domain feature and its application to fault diagnosis of rolling bearing,” Neurocomputing, vol. 313, pp. 47–64, Nov. 2018, doi: 10.1016/j.neucom.2018.05.002.