Design of an Improved Physics-Aware Early-Life Prognostic Framework for Reliable Remaining Useful Life Prediction of Lithium-Ion Batteries in Real-Time Battery Management Systems

Abstract

Unanticipated lithium-ion battery RUL deterioration creates economic loss, reliability issues, and safety hazards in electric mobility, grid-scale storage, and safety-critical cyber-physical systems. Data-driven prognostics has improved, but early-life prediction, physical interpretability of learnt representations, and real-time Battery Management System (BMS) deployability remain difficult challenges. Many models rely on raw voltage, current, and temperature inputs drawn from a large share of battery lifespan data, but deep learning approaches show unstable generalization and opaque decision mechanisms when operating conditions differ from training regimens. To address these issues, this study introduces a physics-aware and learning-driven prognostic system that uses degradation saliency analysis, constrained feature representation, early-life inference, and real-time validation. Cycle-Resolved Thermo-Electrochemical Saliency Decomposition (CR-TESD) decomposes charge–discharge cycles into phase-aware degradation saliency maps, revealing aging-relevant information even in early operating phases. For compact, physically consistent health trajectories, the Physics-Anchored Feature Manifold Compression Network (PA-FMCN) embeds saliency-weighted features and applies electrochemical admissibility and monotonic aging constraints during latent-space construction. The Progressive Early-Life RUL Bootstrap Learner (PEL-RBL) estimates RUL with limited lifespan data by gradually aligning early degradation traits with end-of-life patterns. For robustness and interpretability, the Physics-Guided Temporal Hybrid Prognostic Network (PG-THPN) refines long-term temporal dependencies and nonlinear deterioration dynamics using GRU–CNN temporal learning combined with physics-derived degradation states. Finally, the BMS-Aware Scalability and Trust Validation Framework (BMS-STVF) evaluates computational efficiency, latency, and consistency under streaming BMS constraints for real-world application. Early-life RUL accuracy, generalization stability, and inference efficiency outperform machine learning and deep learning baselines on benchmark battery aging datasets. The methodology improves battery prognostics beyond predictive performance by providing physically grounded, uncertainty-aware, deployment-ready RUL estimates for next-generation intelligent BMS and proactive energy system management.

Introduction

The paper presents a physics-aware, machine learning-based framework for predicting the Remaining Useful Life (RUL) of lithium-ion batteries. Lithium-ion batteries are widely used in electric vehicles, renewable energy systems, aerospace applications, and portable electronics because of their high energy density and long lifespan. Accurate RUL prediction is essential for improving battery safety, reliability, maintenance planning, warranty assessment, and second-life utilization.

Traditional battery prognostic methods rely on physics-based electrochemical models, equivalent circuit models, and empirical degradation equations. Although these methods are interpretable, they require extensive calibration, expert knowledge, and assumptions that often fail under varying battery chemistries and operating conditions. Moreover, they struggle to capture nonlinear degradation caused by fluctuating loads, environmental factors, and manufacturing variations.

To overcome these limitations, data-driven machine learning and deep learning techniques such as Support Vector Machines, Random Forests, Gaussian Process Regression, Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), Transformers, and hybrid models have been developed. While these approaches generally achieve high prediction accuracy, they often depend on large amounts of late-life degradation data, lack physical interpretability, require substantial computational resources, and are difficult to deploy in real-time Battery Management Systems (BMS).

The proposed research introduces a unified, end-to-end battery prognostic framework that combines electrochemical knowledge with machine learning to improve early-life prediction, interpretability, and deployment readiness. Rather than treating RUL estimation as a single prediction task, the framework consists of five tightly integrated stages:

1. Cycle-Resolved Thermo-Electrochemical Saliency Decomposition (CR-TESD): Identifies important degradation regions by analyzing voltage, current, and temperature variations during battery charge–discharge cycles.
2. Physics-Anchored Feature Manifold Compression Network (PA-FMCN): Compresses battery degradation data into a compact latent representation while enforcing physically meaningful aging behavior through electrochemical constraints.
3. Progressive Early-Life RUL Bootstrap Learner (PEL-RBL): Predicts battery RUL using only early-life cycling data by aligning early degradation patterns with complete battery life trajectories, enabling accurate early-stage prognosis.
4. Physics-Guided Temporal Hybrid Prognostic Network (PG-THPN): Integrates convolutional neural networks (CNNs), gated recurrent units (GRUs), and physics-informed degradation information to model temporal battery aging while maintaining interpretability and robustness.
5. BMS-Aware Scalability and Trust Validation Framework: Evaluates computational efficiency, inference speed, consistency, and deployment feasibility to ensure compatibility with real-time Battery Management Systems.

The paper also reviews 30 recent battery RUL prediction models, including hybrid neural networks, Transformers, CNNs, LSTMs, attention mechanisms, particle filters, federated learning, probabilistic models, and ensemble approaches. Although these models have improved prediction accuracy, they commonly suffer from limitations such as high computational complexity, dependence on large datasets, poor early-life prediction capability, limited interpretability, sensitivity to noise, and challenges in embedded system deployment.

The proposed architecture addresses these shortcomings by integrating degradation saliency analysis, physics-guided feature learning, early-life prediction, hybrid temporal modeling, and deployment-oriented validation into a single analytical pipeline. Mathematical formulations are provided to describe degradation saliency, physics-constrained latent representation learning, early-life RUL estimation, and temporal state evolution, ensuring that learned battery representations remain physically consistent while accurately modeling degradation.

Conclusion

This paper presented a unified, physics-aware, deployment-oriented lithium-ion battery Remaining Useful Life prediction method addressing early-life prognostics, interpretability, robustness, and real-time applicability. Cycle-resolved thermo-electrochemical saliency analysis, physics-anchored manifold learning, progressive early-life inference, and hybrid temporal modeling together form a structured analytical pipeline for battery prognostics. Experimental results on benchmark battery aging datasets indicated a consistent and significant performance improvement: the proposed model outperformed comparable approaches by 20–40% across full-lifecycle prediction, achieving an RMSE of 9.3 cycles and an MAE of 6.8 cycles. Early-life RUL prediction using only 20% of operational cycles showed strong accuracy (RMSE = 11.8 cycles, R² = 0.88), confirming the framework\'s ability to extract long-term degradation information from small samples. Beyond accuracy, the proposed model increased resilience and stability, maintaining low prediction error across thermal conditions and reducing RUL prediction variance to roughly half that of deep learning baselines. Narrow confidence intervals (±9 cycles) strengthen trustworthiness for decision-critical applications. Real-time Battery Management Systems were also demonstrated to be feasible, with inference latency under 12–15 ms per cycle and a model footprint of 4.6 MB, showing that the proposed architecture balances predictive performance, physical consistency, and computational efficiency. The success of this integrated architecture lays a foundation for more sophisticated, dependable, and physically interpretable battery health monitoring.

References

[1] Pratheeba, C., & Sukumar, P. (2025). Hybrid HGRN-SCSO technique for enhanced prediction of remaining useful life in EV batteries. Electrical Engineering, 107(4), 5311-5323. https://doi.org/10.1007/s00202-024-02913-8 [2] Iftikhar, M., Shoaib, M., Altaf, A., Iqbal, F., Villar, S. G., Dzul Lopez, L. A., & Ashraf, I. (2024). A deep learning approach to optimize remaining useful life prediction for Li-Ion batteries. Scientific Reports, 14(1). https://doi.org/10.1038/s41598-024-77427-1 [3] Wang, H., Dai, X., Cui, L., & Gao, M. (2024). A progressive learning residuals based on multivariate Mamba and adaptive singular value decomposition method for remaining useful life prediction of lithium-Ion batteries. Ionics, 31(2), 1731-1746. https://doi.org/10.1007/s11581-024-05981-9 [4] Ibrahim, M., Ansari, S., Ayob, A., Lipu, M. S. H., Abdolrasol, M. G. M., Khawaja, A. W., Khalil, M. A., & Stroe, D. I. (2025). Hybrid optimized remaining useful life prediction framework for lithium-Ion batteries with limited data samples. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-26743-1 [5] Sun, L., Huang, X., Liu, J., Song, J., & Wu, S. (2025). Remaining useful life prediction of lithium batteries based on jump connection multi-scale CNN. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-08619-6 [6] Zhao, F., Gao, D., Cheng, Y., & Yang, Q. (2024). Application of state of health estimation and remaining useful life prediction for lithium-Ion batteries based on AT-CNN-BiLSTM. Scientific Reports, 14(1). https://doi.org/10.1038/s41598-024-80421-2 [7] Zeng, L., Jiang, Z., & Wang, S. (2025). Remaining useful life prediction of lithium-Ion batteries based on MWASFormer network. Ionics. https://doi.org/10.1007/s11581-025-06794-0 [8] Duan, C., Cao, H., Liu, F., Li, X., Duan, X., & Sheng, B. (2025). Lithium-Ion batteries remaining useful life prediction using a parallel BILSTM-MHA neural network based on a CEEMDAN module. The International Journal of Advanced Manufacturing Technology, 137(7-8), 3359-3386. https://doi.org/10.1007/s00170-025-15169-8 [9] Tang, Y., Zhong, S., Wang, P., Zhang, Y., & Wang, Y. (2024). Remaining useful life prediction of high-capacity lithium-Ion batteries based on incremental capacity analysis and Gaussian kernel function optimization. Scientific Reports, 14(1). https://doi.org/10.1038/s41598-024-74755-0 [10] Sun, L., Huang, X., Liu, J., & Song, J. (2024). Remaining useful life prediction of lithium-Ion batteries combined with SVD-SDAE and support vector quantile regression. Discover Energy, 4(1). https://doi.org/10.1007/s43937-024-00027-7 [11] Chen, L., Bai, L., Wei, X., Li, Y., & Fernandez, C. (2025). A multi-layer kernel extreme learning machine model based on the fusion algorithm for the remaining useful life prediction of lithium-Ion batteries. Ionics, 31(10), 10703-10718. https://doi.org/10.1007/s11581-025-06597-3 [12] Shuai, W., Yiting, L., Shoubin, Z., Lifei, C., & Pecht, M. (2025). Remaining useful life prediction of lithium-Ion batteries using a novel particle flow filter framework with grey model. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-86511-z [13] Yu, S., Zhang, L., Ni, L., & Feng, S. (2025). Bimodal decomposition and LSTM network-based joint prediction of reversible capacity and remaining useful life for Li-Ion batteries. Journal of Power Electronics, 25(10), 1892-1907. https://doi.org/10.1007/s43236-025-01013-5 [14] Wu, L., Zhao, J., Du, X., & Li, B. (2024). Frequency-domain attention mixture of expert models for remaining useful life prediction of lithium-Ion batteries. Signal, Image and Video Processing, 18(11), 8477-8487. https://doi.org/10.1007/s11760-024-03488-4 [15] Wang, Y., Zhang, Y., Gu, B., & Xu, M. (2024). Remaining useful life prediction of lithium-Ion batteries using particle filter algorithm in communication power supply. Russian Electrical Engineering, 95(10), 870-879. https://doi.org/10.3103/s1068371224701128 [16] Li, Z., Shen, S., Ye, Y., Cai, Z., & Zhen, A. (2024). An interpretable online prediction method for remaining useful life of lithium-Ion batteries. Scientific Reports, 14(1). https://doi.org/10.1038/s41598-024-63160-2 [17] Jeha, J., & Priyan, M. S. (2025). Ensemble-optimized learning for accurate remaining useful life estimation of lithium-Ion batteries in electric vehicles. Iranian Journal of Science and Technology, Transactions of Electrical Engineering. https://doi.org/10.1007/s40998-025-00872-9 [18] Zheng, D., Zhang, Y., Guo, X., Ning, Y., & Wei, R. (2025). Research on the remaining useful life prediction method for lithium-Ion batteries based on feature engineering and CNN-BiGRU-AM model. Ionics, 31(6), 5717-5736. https://doi.org/10.1007/s11581-025-06293-2 [19] Zheng, D., Zhang, Y., Deng, W., Guo, X., Ning, Y., & Wei, R. (2025). Remaining useful life prediction approach for lithium-Ion batteries based on feature optimization and an ensemble deep learning model. Ionics, 31(11), 11669-11691. https://doi.org/10.1007/s11581-025-06700-8 [20] Chen, S., & Fang, X. (2025). Remaining useful life prediction of lithium-ion batteries via spatial attention TLSTM and dilated CNN with evolutionary optimization. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-17610-0 [21] Fan, Y., Lin, Z., Wang, F., & Zhang, J. (2025). A hybrid approach for lithium-Ion battery remaining useful life prediction using signal decomposition and machine learning. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-92262-8 [22] Chen, G. (2025). Residual useful life prediction of lithium-Ion battery based on accuracy SoH estimation. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-89727-1 [23] Zhong, R., Hu, B., Feng, Y., Lou, S., Hong, Z., Wang, F., Li, G., & Tan, J. (2024). Lithium-Ion battery remaining useful life prediction: a federated learning-based approach. Energy, Ecology and Environment, 9(5), 549-562. https://doi.org/10.1007/s40974-024-00323-x [24] Gillani, S. W., Shahid, K., Gulzar, M. M., & Arif, D. (2024). Remaining useful life prediction of super-capacitors in electric vehicles using neural networks. Arabian Journal for Science and Engineering, 49(5), 7327-7340. https://doi.org/10.1007/s13369-024-08766-4 [25] Feng, R., Wang, S., Yu, C., & Fernandez, C. (2024). High precision estimation of remaining useful life of lithium-Ion batteries based on strongly correlated aging feature factors and AdaBoost framework. Ionics, 30(10), 6215-6237. https://doi.org/10.1007/s11581-024-05740-w [26] Zhang, H., Guo, K., Chen, Y., & Sun, J. (2024). Remaining useful-life prediction of lithium battery based on neural-network ensemble via conditional variational autoencoder. Applied Intelligence, 55(1). https://doi.org/10.1007/s10489-024-05885-1 [27] Zhang, X., Chen, Z., Li, Y., Zhang, C., Chai, X., Ye, M., & Zhu, M. (2025). AutoDS: a de-stationary transformer method for lithium-Ion battery remaining useful life prediction. Signal, Image and Video Processing, 19(13). https://doi.org/10.1007/s11760-025-04720-5 [28] Pang, X., Li, X., Kim, J., Lee, M., Zhao, Z., Wen, J., & Zeng, J. (2025). Interval prediction strategy for the remaining useful life of lithium-Ion battery pack considering cell inconsistency. Journal of Power Electronics. https://doi.org/10.1007/s43236-025-01072-8 [29] Liu, F., Liu, S., Chai, Y., & Zhu, Y. (2025). Enhanced Mamba model with multi-head attention mechanism and learnable scaling parameters for remaining useful life prediction. Scientific Reports, 15(1). https://doi.org/10.1038/s41598-025-91815-1 [30] He, K., Dai, X., Li, X., Zhou, R., Lin, F., Zhang, W., & Ma, H. (2025). State of health and remaining useful life full lifecycle prediction for lithium-Ion battery based on frequency domain interpolation and phased approach. Ionics, 31(11), 11783-11796. https://doi.org/10.1007/s11581-025-06711-5

Copyright

Copyright © 2026 Vinaya Sonparote. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.