Advanced lithium-ion batteries (LIBs) are central to electric transportation and grid storage, yet their internal electrochemical, thermal and mechanical states remain hard to observe. A new review published in the Journal of Zhejiang University–SCIENCE A examines how nondestructive sensing and failure diagnosis can turn hidden aging signals into actionable health information, potentially improving safety, reliability and lifetime across electric vehicles and stationary storage. The review, authored by researchers from Zhejiang University, links degradation mechanisms to measurable physical signatures and evaluates surface, implanted, in situ integrated and noncontact sensing strategies. It also assesses multisource feature fusion, physics-informed and data-driven algorithms, and smart battery management concepts. For next-generation batteries, this means moving from reactive protection to predictive, mechanism-informed care.
Lithium-ion battery aging is driven by tightly coupled chemical, mechanical and thermal processes. Growth of the solid electrolyte interphase (SEI) and cathode electrolyte interphase (CEI), electrolyte decomposition, particle cracking and lithium plating can progressively consume active lithium, increase impedance and degrade structural integrity, while severe degradation may increase the risk of internal short circuits and thermal runaway. Conventional battery management systems (BMSs) mainly track voltage, current and surface temperature, but these external signals are distorted by polarization, side reactions, spatial averaging and delays. Synchrotron and magnetic resonance imaging offer mechanistic insight, yet their scale, cost and speed limit real-time use. Based on these challenges, deeper research is needed into nondestructive sensing, multisource feature fusion and intelligent diagnosis for lithium-ion battery aging and safety.
The review organizes nondestructive diagnostics into four families. Surface-attached sensors, including thermocouples, thermistors, resistance temperature detectors (RTDs) and fiber Bragg gratings (FBGs), are low-cost and easy to deploy. However, the spatial and temporal limitations of physical signal transmission from the battery interior to the sensing surface can introduce significant delays and measurement discrepancies, particularly under high-rate operation. Implantable sensors, such as microelectromechanical systems (MEMS) devices, thin-film strain gauges and optical fibers, improve signal fidelity by measuring internal temperature, pressure, strain and electrolyte chemistry, though long-term stability and vibration fatigue remain barriers. In situ integrated designs embed sensing functions into current collectors, separators or packaging, offering minimal intrusion and compatibility. Noncontact methods—magnetic-field imaging, acoustic and ultrasonic probing, gas analysis and electrochemical impedance spectroscopy (EIS)—provide system-level insight without direct contact.
To make full use of these heterogeneous signals, the review further discusses feature extraction, multisource information fusion and model-based interpretation. Electrochemical and thermal features can be extracted using methods such as incremental capacity analysis (ICA), differential voltage analysis (DVA) and differential thermal voltammetry (DTV), and subsequently interpreted using physics-based models, including pseudo-two-dimensional (P2D) and single-particle models (SPM), as well as emerging data-driven approaches such as physics-informed neural networks (PINNs), transformers and convolutional neural network–long short-term memory (CNN-LSTM) architectures. It also stresses feature selection, dimensionality reduction, cloud-edge collaboration and standardized interfaces. Together, these tools aim to distinguish normal breathing from lithium plating, gas generation and microcracking, and to translate weak multiphysics signatures into mechanism-specific warnings. The full review is available at 10.1631/jzus.A2600100.
The authors said the field is shifting from external, indirect observation toward direct internal perception. No single sensing modality can fully capture the complex and coupled processes underlying battery degradation and safety failure; the real advance comes when surface, implanted, in situ and noncontact signals are fused with physics-informed algorithms. They therefore emphasize sensing–algorithm co-design, in which sensing capabilities, signal processing and diagnostic models are considered together rather than treated as separate stages. They said this shift could make batteries more predictable, reliable and durable across electric vehicles and grid storage.
Practically, the framework could support earlier thermal-runaway warnings, more accurate state-of-health (SOH) and remaining-useful-life (RUL) estimates, and smarter fast-charging. In electric vehicles, it may enable predictive maintenance and cell-to-pack safety monitoring. In grid storage, it could improve fleet-level reliability, second-life assessment and fire prevention. Low-cost strategies using existing voltage, current and temperature signals, combined with cloud-edge computing, could ease deployment. The review cautions that sensor stability, manufacturing compatibility, data standardization, bandwidth and cost remain key barriers. It calls for modular, standardized, minimally intrusive sensing and algorithm co-design to move laboratory advances into scalable battery systems. For industry, integrating sensing and diagnostic capabilities into battery design could improve manufacturing compatibility and long-term reliability while facilitating the practical deployment of advanced diagnostic technologies in future battery systems.
