(375ab) Enhancing Li-Ion Battery Remaining Useful Life Prediction with a Physics-Constrained Bayesian Recurrent Neural Network

The prognosis of lithium-ion (Li-ion) batteries' Remaining Useful Life (RUL) is pivotal, given the expanding reliance on these energy storage systems across a diverse range of applications. Accurate RUL prediction not only elevates the operational efficiency and safety of devices powered by Li-ion batteries but also contributes to sustainable energy management practices. Despite significant advancements, existing RUL estimation methodologies, encompassing physical-based, data-driven, and hybrid approaches, exhibit limitations in adaptability, interpretability, and precision under varying operational conditions.

Physical-based methods, grounded in electrochemical models and empirical formulations derived from physics laws, offer insights into the fundamental degradation reactions within batteries. These methods, however, often lack flexibility and struggle to generalize across different battery types and usage conditions [1–7](References 2-8). Conversely, the advent of machine learning (ML) and the availability of open-source battery operation data have propelled data-driven approaches. These methods leverage operational data to forecast RUL, benefiting from ML's pattern recognition capabilities. Despite their promise, such approaches are critiqued for their opaque decision-making processes and diminished efficacy on previously unseen data [8][8–11](References 9-13).

Hybrid approaches attempt to bridge the gap by integrating physical models with data-driven techniques, aiming to combine the best of both worlds. By infusing physical insights into ML models, these methods seek to improve prediction accuracy and generalizability while enhancing interpretability through physics-based constraints [12–16]. However, challenges persist, including the intricate calibration of hybrid models and the quantification of predictive uncertainty, which are crucial for reliable RUL estimation.

Addressing these challenges, our study introduces a novel Physics-Constrained Bayesian Recurrent Neural Network (RNN) approach for Li-ion battery RUL prediction. This approach innovatively combines the robust predictive capabilities of Bayesian RNNs with physics-based constraints to enhance prediction accuracy, interpretability, and reliability. The proposed method focuses on several key advancements:

• Uncertainty Quantification: By incorporating Bayesian inference, our model quantifies uncertainty in RUL predictions, offering a probabilistic assessment that is invaluable for risk management and decision-making.
• Physics-Constrained Modeling: Leveraging empirical models to enforce physical constraints within the neural network ensures that predictions adhere to fundamental physical principles, enhancing model reliability and interpretability.
• Adaptability and Continuous Learning: The model is designed to update its parameters continuously with new operational data, facilitating adaptability to evolving battery usage patterns and conditions.

The proposed Physics-Constrained Bayesian RNN approach represents a significant leap forward in the field of battery health management. By addressing the limitations of existing RUL prediction methodologies, this approach promises to improve the safety, efficiency, and longevity of Li-ion batteries, which are at the heart of modern electronic devices, electric vehicles, and renewable energy storage solutions. Our findings have broad implications for the design and operation of battery-powered systems, offering a pathway to more sustainable and reliable energy storage.

Reference:

[1] Micea MV, Ungurean L, Cârstoiu GN, Groza V. Online state-of-health assessment for battery management systems. IEEE Trans Instrum Meas 2011;60:1997–2006. https://doi.org/10.1109/TIM.2011.2115630.

[2] Saha B, Poll S, Goebel K, Christophersen J. An integrated approach to battery health monitoring using Bayesian regression and state estimation. AUTOTESTCON (Proceedings) 2007:646–53. https://doi.org/10.1109/AUTEST.2007.4374280.

[3] Schmalstieg J, Käbitz S, Ecker M, Sauer DU. A holistic aging model for Li(NiMnCo)O2 based 18650 lithium-ion batteries. J Power Sources 2014;257:325–34. https://doi.org/10.1016/J.JPOWSOUR.2014.02.012.

[4] Karger A, Schmitt J, Kirst C, Singer JP, Wildfeuer L, Jossen A. Mechanistic calendar aging model for lithium-ion batteries. J Power Sources 2023;578:233208. https://doi.org/10.1016/j.jpowsour.2023.233208.

[5] Ning G, White RE, Popov BN. A generalized cycle life model of rechargeable Li-ion batteries. Electrochim Acta 2006;51:2012–22. https://doi.org/10.1016/j.electacta.2005.06.033.

[6] Ng SSY, Xing Y, Tsui KL. A naive bayes model for robust remaining useful life prediction of lithium-ion battery. Appl Energy 2014;118:114–23. https://doi.org/10.1016/j.apenergy.2013.12.020.

[7] Lui YH, Li M, Downey A, Shen S, Nemani VP, Ye H, et al. Physics-based prognostics of implantable-grade lithium-ion battery for remaining useful life prediction. J Power Sources 2021;485:229327. https://doi.org/10.1016/j.jpowsour.2020.229327.

[8] Lin CP, Cabrera J, Yang F, Ling MH, Tsui KL, Bae SJ. Battery state of health modeling and remaining useful life prediction through time series model. Appl Energy 2020;275:115338. https://doi.org/10.1016/j.apenergy.2020.115338.

[9] Zhou J, Wang S, Cao W, Xie Y, Fernandez C. State of health prediction of lithium-ion batteries based on SSA optimized hybrid neural network model. Electrochim Acta 2024;487:144146. https://doi.org/10.1016/j.electacta.2024.144146.

[10] Xie P, Pang X, Wang C, Yang W, Zou H, Zhao W, et al. A sequence to sequence prediction model for remaining useful life of lithium-ion batteries with Bayesian optimisation process visualization. J Energy Storage 2024;87:111346. https://doi.org/10.1016/j.est.2024.111346.

[11] Zhang Y, Xiong R, He H, Pecht MG. Long short-term memory recurrent neural network for remaining useful life prediction of lithium-ion batteries. IEEE Trans Veh Technol 2018;67:5695–705.

[12] Najera-Flores DA, Hu Z, Chadha M, Todd MD. A Physics-Constrained Bayesian neural network for battery remaining useful life prediction. Appl Math Model 2023;122:42–59. https://doi.org/10.1016/j.apm.2023.05.038.

[13] Arias Chao M, Kulkarni C, Goebel K, Fink O. Fusing physics-based and deep learning models for prognostics. Reliab Eng Syst Saf 2022;217:107961. https://doi.org/10.1016/j.ress.2021.107961.

[14] Thelen A, Lui YH, Shen S, Laflamme S, Hu S, Ye H, et al. Integrating physics-based modeling and machine learning for degradation diagnostics of lithium-ion batteries. Energy Storage Mater 2022;50:668–95. https://doi.org/10.1016/j.ensm.2022.05.047.

[15] Navidi S, Thelen A, Li T, Hu C. Physics-informed machine learning for battery degradation diagnostics: A comparison of state-of-the-art methods. Energy Storage Mater 2024;68:103343. https://doi.org/10.1016/j.ensm.2024.103343.

[16] Shi J, Rivera A, Wu D. Battery health management using physics-informed machine learning: Online degradation modeling and remaining useful life prediction. Mech Syst Signal Process 2022;179:109347. https://doi.org/10.1016/j.ymssp.2022.109347.