Physics-Constrained Deep Operator Network for Remaining Useful Life Prediction of Li-ion Batteries with Early-Cycle Data
Authors:
- Min-Min ZhouMZMin-Min Zhou
- Wangqi WuWWWangqi Wu
- Chun ZhuCZChun Zhu
- Jiaze MaJMJiaze Ma
- Lunbo DuanLDLunbo Duan
- Jiwei YaoJYJiwei Yao
Preprint
- 24 June 2026
- preprint
- Published by Elsevier in SSRN Electronic Journal
Abstract
Accurate remaining useful life (RUL) prediction of lithium-ion batteries (LIBs) is critical for safe, reliable, and cost-effective operation of battery-powered systems. Existing data-driven approaches often require long-term cycling histories and may produce physically inconsistent predictions, while physics-based models can be limited by complexity and parameter-identification challenges. This paper proposes a physics-constrained deep operator network (PC-DON) for early-cycle battery RUL prediction. The framework uses a branch network to learn cell-specific degradation patterns from early-cycle incremental capacity curves and a trunk network to represent cycle-level degradation states. Three physics-based constraints are designed and evaluated to enhance physical consistency and generalization. The method is tested on 124 commercial LIB cells and compared with a benchmark physics-informed neural network. Results demonstrate improved accuracy and robustness. Additional analyses of physics-loss combinations, early-cycle selection, and SHAP-based feature importance confirm the framework’s effectiveness, interpretability, and potential for early-stage battery health prognosis.Keywords
This publication has 17 references indexed in Scilit:
- A Physics-Constrained Bayesian neural network for battery remaining useful life predictionApplied Mathematical Modelling, 2023
- Augmented model-based framework for battery remaining useful life predictionApplied Energy, 2022
- A review of lithium-ion battery safety concerns: The issues, strategies, and testing standardsJournal of Energy Chemistry, 2021
- Learning nonlinear operators via DeepONet based on the universal approximation theorem of operatorsNature Machine Intelligence, 2021
- Data-driven prediction of battery cycle life before capacity degradationNature Energy, 2019
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equationsJournal of Computational Physics, 2019
- A new hybrid method for the prediction of the remaining useful life of a lithium-ion batteryApplied Energy, 2017
- Degradation diagnostics for lithium ion cellsJournal of Power Sources, 2017
- Ageing mechanisms in lithium-ion batteriesJournal of Power Sources, 2005
- Issues and challenges facing rechargeable lithium batteriesNature, 2001