Early warning of thermal runaway in lithium-ion batteries remains challenging because early internal chemical signals are difficult to access in sealed commercial cells. Here, we demonstrate an embedded Pd-FBG and reference FBG configuration for internal H2 sensing in a commercial 280 Ah prismatic LFP cell. A laser-welded hermetic optical feedthrough enabled sealed optical access, with a measured leakage rate of 2.6 × 10-6 Pa·m3/s under 1 MPa. Hierarchical simulations provided qualitative insight supporting the vent-adjacent region as the preferred sensor location under the investigated cell configuration. The embedded cell showed stable short-term cycling behavior during 100-cycle paired testing. During external-heating thermal-runaway testing, the corrected Pd-FBG signal crossed the 30 vol% operational H2-equivalent accumulation threshold at 996 s, preceding safety-vent opening by 84 s and the surface-temperature-rise-rate thermal-runaway criterion by 541 s in the tested cell. This work demonstrates a function-preserving route for hermetically integrated, electrically passive internal H2 monitoring in large-format prismatic cells.
Thermal runaway in lithium-ion batteries is a violent thermochemical reaction triggered by coupled mechanical–thermal–electrochemical abuses, the early warning remains a critical challenge. Conventional warning methods rely on external sensors, which suffer from limited response speed and spatial resolution. Consequently, the smart batteries with embedded flexible sensing is emerging as a consensus solution for early fault diagnosis. However, a systematic understanding of the design, fabrication, integration, and application of flexible embedded sensing in batteries is still lacking, hindering its practical deployment. Here, we systematically review the critical bottlenecks of embedded flexible sensors in terms of early-warning responsiveness, failure-mode adaptability, implantation compatibility, and scalable manufacturability. We provide an in-depth discussion across various sensing types, and highlights the pivotal role of flexible electronic materials in realizing ultrathin, low-intrusion, and high-stability sensing architectures. Furthermore, we summarize diverse manufacturing pathways, followed by integration strategies tailored to different cell types. For engineering applications, we propose an embedded sensing ultra-early warning system for large-format batteries, which integrates a sensor quantity and layout design scheme, a fault diagnosis strategy, and a hierarchical ultra-early warning model for thermal runaway. Finally, we envision the future convergence of embedded flexible sensors with artificial intelligence, digital twin modeling, and wireless communication technologies, emphasizing the need to develop multimodal, durable, and scalable in-cell sensing systems. Such advances will enable the evolution of battery safety management from “measurable” to “interpretable” and ultimately to “controllable”, providing a technological cornerstone for the next generation of high-safety, smart energy storage systems.
Inter-cell barriers can delay thermal-runaway propagation (TRP) but may impair normal-operation heat rejection. This study uses a three-dimensional structured thermal-resistance model to compare aluminum, mica laminate, silica aerogel, pure paraffin, and composite phase change material (PCM) barriers under identical module geometry and loading. Abuse performance is defined by the time at which the spatially averaged protected-cell temperature first reaches 150 °C; fast-charge performance is assessed using peak cell temperature and non-uniformity. In the conduction-dominant baseline, this threshold is reached at 848 s for aluminum, 873 s for composite PCM, 986 s for mica, and 990 s for pure paraffin. Silica aerogel does not reach the threshold by the end of the recorded observation window and is therefore reported as right-censored rather than assigned an exact delay. A Rosseland radiative-conductivity sensitivity and a generic bypass-conductance sensitivity indicate that the silica result depends strongly on heat-transfer paths omitted from the baseline; neither surrogate resolves view-factor radiation or vent-gas transport. In the modeled cases, stronger insulation delays protected-cell heating but raises fast-charge temperatures, whereas conductive barriers show the opposite trade-off. A hypothetical enhanced-PCM property target reaches the 150 °C threshold at 1191 s while maintaining a modeled fast-charge peak temperature near 30.0 °C. This case is an aspirational parametric point, not a demonstrated material. The results establish a within-model benchmark and identify the joint roles of thermal resistance, sensible and latent storage, and normal-operation cooling.
High-energy-density lithium-ion batteries (LIBs) are limited by strong kinetic constraints that make fast charging difficult and restrict their use in weight-sensitive transportation systems, such as low-altitude electric mobility. To address this challenge, we propose a fast-charging optimization method for high-energy-density NCM811/graphite LIBs. The method couples an electrochemical–thermal model with a genetic algorithm to design multi-stage constant-current (MSCC) fast-charging protocols. These protocols minimize charging time under model-predicted anode potential and thermal safety limits, and a four-stage constant-current (4SCC) configuration is selected as the preferred trade-off between charging speed and implementation complexity. The optimized 4SCC protocols were evaluated on commercial cells with an energy density of approximately 280 Wh kg−1 at 0, 25, and 45 °C through charging-speed comparison and lithium-plating diagnostics. The optimized protocols reduced the full-charge time by approximately 26% and 59% relative to the manufacturer-recommended strategies at 0 °C and 45 °C, respectively, and by 69% relative to the standard C/3 protocol at 25 °C. The measured cell–surface temperature remained below the prescribed safety threshold throughout charging. Furthermore, voltage-relaxation and post-mortem SEM diagnostics did not detect lithium-plating evidence under the tested conditions. In a 500-cycle single-cell comparison at approximately 25 °C, the optimized 4SCC cell retained capacity close to that of the standard-charged cell. Open-circuit voltage (OCV) shift analysis, incremental capacity analysis (ICA), and differential voltage analysis (DVA) revealed no obvious additional degradation signature in this single-cell comparison. These results suggest that physics-based modeling combined with intelligent optimization provides a practical route for designing fast-charging protocols for high-energy-density LIBs.
Lithium-ion batteries have become the core power source and are playing a pivotal role in the field of electric aviation. However, the understanding of the thermal runaway (TR) characteristics of aircraft power battery modules under unique low-pressure conditions is obviously lagging behind. Here, the systematic battery-to-module analysis of the TR and propagation behaviors of 40 Ah LiNi1/3Co1/3Mn1/3O2/graphite prismatic power battery modules is conducted under typical low-pressure conditions at 95, 70 and 35 kPa. Compared with previous studies, the innovation of this work lies in that the TR characteristic parameters including temperature, voltage, expansion force, pressure, mass loss and heat generation under low pressure were monitored. The evolution laws of multi-characteristic parameters process were comprehensively analyzed during TR. Two important results are presented: (i) With the decrease of ambient pressure, the TR eruption pressure of 100% SOC batteries gradually increases, while the expansion force decreases. Moreover, the temperature of eruptions rises as ambient pressure decreases, resulting in more severe thermal-pressure impact hazards. (ii) A multi-parameter TR hazard assessment model is constructed to qualitatively analyze the TR hazards of single batteries and battery modules under low pressure. Generally, under lower ambient pressure, the thermal hazards of batteries during TR decreased, whereas the ejection impact hazards and TR propagation risks of the module grew continuously. This work paves the way for further investigations into the safety of electric aircraft power battery modules.
Stack pressure is a critical mechanical boundary condition in lithium-ion battery (LIB) modules and packs, applied to maintain interfacial contact and control dimensional changes during operation. Despite its widespread use, its influence on battery performance and degradation remains difficult to quantify, as reported trends show large variability across chemistries, cell formats, and testing conditions. This review synthesizes experimental evidence describing how external compression affects discharge capacity, direct current internal resistance (DCIR), and aging behavior in LIBs. To complement the qualitative synthesis, a literature-derived dataset was compiled from studies reporting quantitative pressure-dependent performance data. The dataset was analyzed through visual correlation analysis and Pearson correlations to explore relationships between stack pressure, operating conditions, and electrochemical response. Although a universal quantitative relationship could not be established due to the large variability across studies, consistent qualitative influences emerge. In particular, the effect of compression varies with testing conditions and becomes more pronounced at higher C-rates, indicating a strong coupling between mechanical boundary conditions and transport limitations. These findings highlight the need for standardized experimental methodologies and systematic datasets to establish reliable pressure management strategies in lithium-ion battery systems.
The second-life utilization of retired lithium-ion batteries (LIBs) is critical to achieving global sustainability goals, yet prevailing industrial sorting strategies suffer from inherent limitations. Specifically, current methods rely on full-cell features, causing poor electrode-level aging consistency in reassembled packs and lacking fast, low-cost testing for economic viability. This study proposes a rapid sorting method tailored to the electrode-level aging pathways of retired LIBs, with the aim of guiding rational sorting based on electrode-level aging diagnosis. Concretely, LIB aging pathways are determined by estimating full-cell capacity, lithium inventory, and cathode/anode capacity via voltage curve fitting. Also, a mapping model library consisting of 1260 models correlates rapid-test resistance features with the four aforementioned capacity metrics. Experiments show mechanism-derived features play a dominant role in capacity estimation, yielding high precision with an average R2 of 0.9 and an optimal R2 exceeding 0.95. Furthermore, the method enables targeted and accurate diagnosis of cathode and anode aging states that are consistent with the actual operational conditions of LIBs. Compared with conventional sorting strategies, the proposed approach significantly improves the intra-pack capacity consistency, reducing the cell-to-cell capacity variation by 19.9%, 16.1%, and 13.2% at discharge rates of 0.2C, 1C, and 2C, respectively. This work realizes rapid and accurate estimation of the electrode-level aging states of retired LIBs, enhances the consistency and reliability of recombined battery packs, and thus provides technical support for the industrialization of LIB second-life utilization.
Accurate State-of-Health (SOH) estimation is critical for the safety and economic management of large-scale battery systems. However, random and incomplete charging behaviors in the field degrade the accuracy of conventional models, which typically require fixed voltage windows or full charging cycles. To resolve this limitation, we present a universal SOH estimation framework based on a synergistic sequence–feature fusion architecture, by integrating a boundary-independent feature extraction protocol with a dual-stream neural network. Evaluated on a diverse dataset comprising 320,000 operational segments across six commercial cathode chemistries, including LFP, NMC, and LCO, the proposed method achieves a mean absolute error (MAE) of 1.69% and a minimum error of 1.06%, outperforming all the representative baseline models. Moreover, the error variation across different datasets is less than 1.6%, demonstrating robust cross-chemistry adaptability. This methodology provides a potential solution for intelligent battery management in real-world environments.
The widespread application of lithium-ion batteries (LIBs) is constrained by thermal safety issues, particularly thermal runaway (TR), driving the demand for advanced thermal protection strategies. Inorganic phase change materials (IPCMs), recognised for their inherent nonflammability and high energy storage density, have emerged as a promising cooling medium with higher safety for battery thermal management (BTM) and TR mitigation. The rapid evolution of IPCMs has created a growing need for summarising and assessing the emerging strategies for IPCM-based battery thermal protection. To fill this gap, this work provides a comprehensive and focused review of IPCM-based strategies for LIB, spanning from material designs to battery system validation and practical application. It first introduces a series of material-level modification approaches, including encapsulation, structural support, additive enhancement, and functional flexible design. It focuses on their roles in improving various properties and stability of IPCM, as well as evaluating their applicable scenarios and limitations. Subsequently, the practical application of IPCM-based BTM systems during normal battery operation is reviewed, highlighting their cooling performance in battery temperature and temperature differences. Meanwhile, some horizontal comparisons are also conducted to explore the relative advantages of IPCM. After that, a dual-stage thermal storage mechanism of IPCMs is deeply analysed and discussed. Their impressive performances in suppressing TR initiation and propagation are further reviewed. Finally, current challenges and future directions of IPCM's practical deployment and system selection are discussed to guide the development of next-generation IPCM material modification and advanced battery thermal protection systems.
Accurate remaining useful life (RUL) prognostics for proton exchange membrane fuel cells (PEMFCs) are essential for reliable operation and cost-effective maintenance. Existing long-term prognostics approaches often improve accuracy by relying on large datasets and black-box neural networks, but provide limited interpretability of the predicted voltage degradation curve. This study develops a physics-constrained symbolic regression-physics-informed neural network (PhySR-PINN) framework that, under dimensional consistency constraints, identifies a compact and physically interpretable dual-time-constant exponential degradation representation from voltage time-series data within a physics-constrained symbolic search space. The extracted fast and slow time constants are associated with recoverable voltage loss at the scale of hundreds of hours and irreversible voltage-loss accumulation at the scale of thousands of hours, respectively, supporting an interpretable description of PEMFC voltage degradation. The identified degradation representation is then embedded into the PINN as an explicit structural constraint to strengthen degradation-consistent learning for long-term prognostics. Experiments with different activation functions and training-data proportions demonstrate sustained long-term prognostic performance under limited-data conditions. Cross-dataset validation under multiple training-data proportions further indicates stable long-term prognostic behavior, with the best tested case reaching 98.97% RUL prediction accuracy.
Accurate state of health (SoH) estimation in high-voltage storage (HVS) systems for battery electric vehicles (BEVs) can significantly reduce production and ownership costs. This cost-reduction potential arises because more precise SoH information enables manufacturers to minimize excess aging reserves installed to compensate for underestimated SoH, while still meeting regulatory warranty obligations over vehicle lifetime. However, conventional SoH estimation methods require external power supplies and lengthy measurement procedures. This study proposes a reconfigurable HVS architecture concept that enables autonomous SoH estimation during BEV standby periods without external power sources. By conducting all measurements during vehicle standby periods, the vehicle executes scheduled, reproducible diagnostic events throughout the vehicle lifetime. This increases estimation frequency, reduces SoH changes between diagnostic events, and thereby improves estimation accuracy and numerical convergence. The proposed HVS architecture concept dynamically switches between series, parallel, and split-series–parallel configurations of high-voltage modules (HVMs) to generate controlled voltage gradients. Generated inter-module currents are used for SoH estimation, while a DC/DC converter regulates current amplitudes. A developed proof-of-concept hardware test bench validates autonomous capacity-based SoH (SoHC) estimation using degradation mode analysis methods, as well as resistance-based SoH (SoHR) estimation via hybrid pulse power characterization (HPPC) and electrochemical impedance spectroscopy (EIS). Results reveal strong agreement between the test bench and potentiostat reference measurements. SoHC estimations at 100%, 85%, and 70% show a mean absolute error (MAE) difference of 0.24 percentage points between the two systems. Corresponding SoHR estimations yield MAE deviations of 1.65 percentage points for HPPC and 2.55 percentage points for EIS. Beyond diagnostics, the proposed concept also enables battery heating and balancing. A cost analysis indicates economic plausibility. If affected hardware costs increase by 6.5% while the SoH estimation error decreases by 50%, total HVS costs may be reduced by roughly $190 per vehicle over 160,000km.
Machine-learning-based state of health (SOH) estimation methods often suffer from poor generalization owing to the scarcity of labels from field data, leading to inaccurate results and improper management of battery health when implemented in a battery management system. To address this challenge, this study proposes a weakly supervised learning (WSL) framework for battery SOH estimation that requires only a minimal number of labels. First, numerous SOH-related weak labels are generated from unlabeled dQ/dV curves by employing an electrochemical feature that characterizes the intensity of the predominant electrochemical reaction. Next, a weakly supervised pre-training technique is conducted on a deep neural network (DNN) to learn the battery aging information existed in these weak labeled data. Finally, the DNN is fine-tuned by using a limited number of labeled data to achieve accurate and robust SOH estimation. The proposed WSL framework demonstrates superior generalization capability in SOH estimation using only six labeled samples across diverse scenarios, covering 176 batteries and 320 electric vehicles with variations in cathode materials, capacity configurations, and operating conditions. Experimental validation shows the average root mean squared errors (RMSEs) of SOH estimations in within- and cross-dataset scenarios range from 0.51% to 1.86% and 0.5% to 2.7%, respectively. These results underscore the strong potential of the proposed WSL to overcome labeling bottlenecks across a wide range of battery materials and operating conditions.
Over-discharge is a prevalent electrical abuse scenario in lithium-ion batteries, yet the associated safety risks remain insufficiently evaluated. In this work, we conduct a comprehensive analysis of battery performance across the full depths of over-discharge (DoODs) and establish quantitative evaluation metrics to enable safe operation. Initially, the over-discharge mechanism is comprehensively elucidated based on post-mortem and half-cell testing analyses, according to which the over-discharge severity is classified into five levels. Furthermore, cyclic over-discharge behavior and safety boundaries are systematically analyzed, with a focus on internal short circuit (ISC) onset. Additionally, multiple in-situ methods are employed to decouple and quantify the effects of different degradation modes on battery performance. Finally, the correlation between the over-discharge-induced performance degradation and the cell thermal stability is analyzed. Quantitative analysis reveals that the increase in internal resistance (IIR) has the least impact, with the 0 V cutoff threshold serving as a definitive boundary for the degradation pathways. Prior to reaching this 0 V threshold, degradation is primarily driven by loss of lithium inventory (LLI) and anode-related loss of active material (LAM), with LLI playing a more prominent role. Conversely, beyond the 0 V threshold, the anode potential rises to 3.5 V vs. Li/Li+, triggering copper current collector dissolution, subsequent Cu2+ migration, and redeposition; this sequence escalates ISC risks and shifts the dominant mechanism from LLI to LAM, accompanied by intensified cathode deterioration. Over-discharge significantly compromises battery thermal stability, reducing the self-heating onset temperature by 25 °C. When operating beyond the 0 V threshold, separator collapse temperature decreases by 12 °C.
The rapid expansion of renewable generation and electric vehicles places unprecedented stress on grid infrastructure, yet operators still struggle to continuous, real-time visibility into the true useable capacity of battery energy storage system (BESS). Existing state-of-health (SoH) estimation paradigms depend on manually engineered statistical features—techniques tailored to single-cell or electric-vehicle contexts—and therefore collapse under the extreme variability of renewables-coupled BESS charging processes and the amplified noise or data gaps inherent to large series–parallel arrays. We introduce a field-ready, end-to-end deep learning framework that leverages explicit semantic embeddings to fuse 24 h of raw current, voltage, and temperature data at both global and segment levels and estimates SoH via a lightweight Transformer encoder with learnable weighted fusion. By delivering second-level, high-precision SoH estimates from minimal (∼102 kB), locally stored measurements, our approach achieves state-of-the-art performance (MAE = 1.81 %) on an eight-year, ten-system field dataset spanning four chemistries and five series–parallel configurations, which enables a new class of real-time grid intelligence—feeding dynamic capacity constraints into power-flow solvers, degradation-aware control schemes, and automated market bids to extend asset life, reduce costs, and accelerate renewable integration. This work signals a paradigm shift from feature-driven SoH checks to continuous, feature-agnostic, edge-ready SoH estimation for residential PV-coupled battery energy storage systems, providing a practical foundation for future extension to broader grid-storage applications.
An essential objective of modern battery management systems is to ensure safe battery operation. The system shall issue a warning when it detects a fault condition—the earlier, the better. Reliable fault detection requires a sensitive but equally robust detection method. In battery systems, a conspicuous change in the cell-to-cell variation can be an adequate fault indicator. To detect anomalies like this, a recurring method in the scientific literature is principal component analysis (PCA) or one of its variants. Its effectiveness depends heavily on input data preprocessing, though, a topic that is often disregarded. To raise awareness of proper preprocessing, we investigate common techniques in terms of their impact on PCA-based fault detection. We take thermal faults and internal short circuits in battery systems as prototypical examples for our investigation. Our results highlight the importance of selecting appropriate preprocessing for the problem at hand. Well-matched preprocessing shapes the input data such that the theoretical requirements of PCA are met to a greater extent, which enhances overall detection performance. Based on this understanding, we propose a combination of outlier-robust sample studentization and Pareto scaling that improves fault detection over alternatives from the literature in our exemplary application. The detection sensitivity and robustness are greatly enhanced, particularly for minor faults, without adding significant computational complexity.
Lithium-ion batteries (LIBs) release large amounts of high-temperature flammable gases during thermal runaway (TR). However, the influence of high-temperature airflow on thermal runaway propagation (TRP) within confined environments has not been systematically investigated. To elucidate this effect, comparative experiments were conducted on 46,950 LIB modules (4 x 5 array) under standard and isolated exhaust packaging. Under standard exhaust packaging, the module featured a single exhaust channel that enabled high-temperature airflow to propagate freely throughout the module. In contrast, the isolated exhaust packaging utilized multiple exhaust channels to effectively mitigate the influence of high-temperature airflow on adjacent cells. Two significant findings were obtained: 1) The TRP velocities under standard and isolated exhaust configurations were 0.172 f 0.026 and 0.098 f 0.015 cells/min, respectively, corresponding to a 43.0% reduction. Moreover, the average maximum module temperature decreased from 720.1 f 33.1 degrees C to 655.6 f 29.5 degrees C, corresponding to a reduction of 64.5 degrees C. These results demonstrate that a multi-channel exhaust configuration effectively mitigates the influence of high-temperature airflow. 2) The diffusion pathway of high-temperature airflow exerts a pronounced influence on the TRP trajectory. Under standard exhaust packaging, high-temperature airflow spreads uniformly toward adjacent cells, thereby promoting diagonal TRP. Conversely, under isolated exhaust packaging, high-temperature airflow primarily affects cells within the same channel, thereby suppressing diagonal TRP. This study highlights the pivotal role of high-temperature airflow in governing TRP and provides valuable insights for the safe design of next-generation energy storage systems.
Accurate state-of-health (SOH) estimation is a prerequisite for ensuring the safety and performance of lithium-ion batteries (LIBs). Among available approaches, machine learning (ML) has attracted significant attention for SOH estimation but its performance is constrained by its reliance on the quality of data. This constraint hinders its real-world deployment. To address this challenge, recent studies have introduced various strategies to mitigate the impact of data dependence from the different perspectives. This review represents the first systematic analysis dedicated to mitigating the impact of data dependence in ML-based SOH estimation, distinguishing it from prior reviews focused on models or feature selection. Its key innovations reside in discussing the challenges and strategies for mitigating the impact from three aspects: data, learning paradigms, and models. Additionally, it summarizes the challenges of current strategies and provides an outlook on future research directions. By emphasizing the deployment implications of different strategies in electric vehicles, this review provides valuable insights for engineers and researchers facing data quality challenges. Ultimately, the impact of data dependence in real-world deployments can be further addressed through advances in strategies, bridging the gap between academic research and industrial applications.
Innovation in battery formats and chemistries has been one of the most significant revolutions in commercial battery technology over the past decade. However, the potential thermal challenges accompanying such advances remain insufficiently unexplored. Herein, we implant fibre Bragg grating sensors in a variety of cylindrical battery formats to monitor the temperature and heat evolutions of lithium-and sodium-ion chemistries operando. Combining experiments with modelling, we demonstrated that NaMO2/hard carbon cells had lower thermal dissipation than LiFePO4/graphite cells. Furthermore, axial heat dissipation becomes increasingly important with larger cell sizes. Moreover, analysis of heat generation reveals different heat sources in the two chemistries and emphasizes the equal importance of tab number and spatial distribution in thermal design. This work provides a comprehensive comparison of thermal behaviours across cell chemistries and sizes, offering new insights into battery thermal design.
The adoption of vehicle-to-grid (V2G) technology, enabling the integration of electric vehicles (EVs) with the power grid, holds substantial potential for enhancing energy management. Battery electric buses (BEBs) are a key subset of EVs, and exploring their role in V2G offers an additional pathway for urban load shifting and improving grid flexibility. This study develops a city-scale BEB V2G optimization model to quantify the citywide peak-shaving potential of BEBs using real-world operational data from Shenzhen, China. Electricity demand over a week-long horizon is optimized by coordinating BEB charging and discharging at a 5-minute temporal resolution. Results show that charging-only optimization applied to 15,000 sampled BEBs, nearly the entire Shenzhen BEB fleet, reduces the average peak-valley difference ratio (PVDR) by 13.5% during a high-load week and by 14.5% during a low-load week. Under V2G participation, these reductions increase to 35.4% and 37.5%, indicating that discharging activities are essential for substantial peak-shaving gains. From an economic perspective, V2G service also improves profit. Each additional 1000 BEBs in V2G reduces PVDR by approximately 0.010, indicating strong scalability. Sensitivity analysis further shows that increasing charging/discharging power is effective when only a small number of BEBs engage in V2G; however, when more than 10,000 BEBs are available, further increases offer negligible additional benefit. Comparison across non-service time ratios reveals that BEBs with ratios below 40% contribute minimally to peak shaving, whereas those above 60% provide significantly greater reductions in PVDR. These findings suggest that BEBs with extended non-service periods should be prioritized for V2G participation.
