GIFT Professor Mian Li's Research Team Publishes Cover Paper on Explainable AI Framework for Battery Storage Maintenance
A research team led by Professor Mian Li at the GIFT-SMART Center, Global Institute of Future Technology, Shanghai Jiao Tong University, together with international collaborators, published a cover paper in Cell Reports Physical Science on June 17, 2026. The paper, titled “A multi-agent AI framework for explainable battery system maintenance,” proposes a novel framework that integrates on‑site data monitoring, professional expertise, and large language model-based (LLM) semantic reasoning to transform inconsistency indicators in large‑scale battery energy storage systems into interpretable, traceable, and executable maintenance guidance. These results establish a scalable pathway for intelligent operation and management of large-scale battery energy-storage systems.
Background and Challenge
Lithium-ion battery energy storage systems have emerged as critical pillars of modern energy infrastructure for grid peak shaving, frequency regulation, and clean energy integration. Large‑scale installations usually contain thousands of cells, where manufacturing variances, uneven operating conditions, and environmental differences gradually amplify cell‑to‑cell inconsistencies over time. Unlike catastrophic faults, battery inconsistency represents a latent “sub‑health” state that indicates performance degradation, hidden risks, or a change of maintenance priorities. Traditional expert-driven operation and maintenance approaches, relying on threshold alarms and statistical features, can identify what is wrong but rarely explain why it occurs, how severe the risk is, or what to do next.
Architecture of the proposed explainable Operation and Maintenance (O&M) framework for BESS applications
Framework Innovation
The proposed multi‑agent AI framework does not replace existing inconsistency evaluation algorithms; rather, it builds on them as front‑end diagnostic modules and further integrates LLM inference with retrieval‑augmented generation (RAG) to form a closed loop from “field signals” to “maintenance guidance.” Users can pose natural‑language queries—for example, identifying which battery packs exhibit the most severe voltage inconsistencies over a given period, or asking about possible mechanisms behind a certain voltage deviation. The system automatically identifies the query type and mobilizes designated agents for data analysis, knowledge retrieval, and information synthesis to generate responses that include the location of anomalies, causal explanations, and maintenance recommendations.
Validation Results
The team validated the framework on an in‑service containerized battery cluster comprising 3,564 lithium‑iron‑phosphate/graphite cells (300 Ah nominal capacity, configured as 9 parallel packs, each with 9 series modules of 44 cells each). The researchers spent eight months on data operations, covering 444 actual cycles. They designed 30 representative queries spanning data‑based, knowledge‑based, and hybrid maintenance scenarios. Results showed that the framework consistently generated high‑quality answers across different LLM backbones and clearly linked inconsistency patterns with potential mechanisms and maintenance measures. Compared with traditional expert-driven workflows that rely on manual data inspection and indicator interpretation, the framework reduced average response time by 83.0% and average maintenance cost by 98.2% for typical query tasks.
Collaboration and Future Research
The study was conducted in collaboration with Global College, SJTU, and the Center for Aging, Reliability and Lifetime Prediction of Electrochemical and Power Electronic Systems (CARL) at RWTH Aachen University. The research focuses on battery lifetime prediction, consistency detection, early safety warning, and collaborative optimization for complex systems. The researchers plan to expand the knowledge base with standards, maintenance manuals, enterprise fault cases and field experience. They also aim to explore cloud‑edge collaboration, localized deployment, and multi-modal interaction to promote the application of explainable AI in renewable energy storage infrastructure.
DOI: https://doi.org/10.1016/j.xcrp.2026.103388
Author Profile
Jingbo Qu
Co‑first author. Doctoral Student at GIFT-SMART Center and Global College, SJTU. His research focuses on battery modeling, health management, and storage system assessment. He has published 6 SCI papers in journals including Cell Reports Physical Science, Applied Energy, Journal of Energy Storage, Journal of Power Sources, and Engineering Applications of Artificial Intelligence over the past three years as the first and co-authors.
Yijie Wang
Co‑first author and corresponding author. Postdoctoral researcher at GIFT-SMART Center, SJTU. His research focuses on reliability analysis and safety design of battery storage systems. He has published 7 SCI papers as first/corresponding author in journals such as Applied Energy, Advanced Engineering Informatics, Journal of Building Engineering, Journal of Power Sources, and Cell Reports Physical Science, and serves as a reviewer for multiple international journals, including Applied Energy, IEEE Transactions on Transportation Electrification, and Cell Reports Physical Science.
Mian Li
Corresponding author. Tenured professor at Global Institute of Future Technology, SJTU, and ASME Fellow. His research focuses on system design and optimization, advanced control theory, data analysis and decision theory. He leads projects funded by the National Development and Reform Commission, the Ministry of Science and Technology, the National Natural Science Foundation of China, and the Shanghai Municipal Science and Technology Commission. He is the Deputy Director of the Shanghai Key Laboratory of Urban Complex Risk Control and Resilience Governance, and the Director of the SJTU/GIFT–Siemens Sustainable Technologies Research Center. He has received the Best Paper Award from the ASME Design Automation Conference, the Best Ph.D. Thesis Award from the Department of Mechanical Engineering, University of Maryland, the SJTU Teaching Excellence Award, and the First Prize of Shanghai Teaching Achievement.
Weihan Li
Corresponding author, Junior Professor in Artificial Intelligence and Digitalization for Batteries at RWTH Aachen University and heads the “Artificial Intelligence and Digitalization for Batteries” research group at the Center for Ageing, Reliability and Lifetime Prediction of Electrochemical and Power Electronic Systems (CARL). In 2021, he received his doctorate in Electrical Engineering and Information Technology from RWTH Aachen University with the highest distinction, summa cum laude. Prof. Li has conducted research at leading international institutions, including the Massachusetts Institute of Technology, Imperial College London, and Stanford University. He has been named a Clarivate Highly Cited Researcher for two consecutive years, in 2024 and 2025, and has received numerous prestigious honors, including the BMBF BattFutur Junior Research Group Grant of approximately €2.2 million, the German Thesis Award / Deutscher Studienpreis from the Körber Foundation, the Reichart Prize from the Academy of Public Sciences in Erfurt, the vgbe Innovation Award, the Battery Young Research Award, and the RWTH Innovation Award.
