Battery fault detection is crucial for maintaining the safety and reliability of large-scale lithium-ion battery systems, especially in demanding applications like electric vehicles and energy storage power stations. However, existing research primarily addresses either temporal patterns or spatial variations in isolation. This paper presents a comprehensive review of fault detection from a spatio-temporal perspective, with a specific focus on AI-driven methods that integrate temporal dynamics with spatial sensor data. The contributions of this review include an in-depth analysis of advanced techniques such as transfer learning, foundation models, and physics-informed neural networks, emphasizing their potential for modeling complex spatio-temporal dependencies. On the engineering side, this review surveys the practical application of these methods for early fault detection and diagnostics in large-scale battery systems, supported by case studies and real-world deployment examples. The findings of this review provide a unified perspective to guide the development of robust and scalable spatio-temporal fault detection methods for EV batteries, highlighting key challenges, promising solutions, and future research directions.

In metallurgy, the quenching process often induces changes in certain material properties, such as hardness and ductility, through the rapid cooling of a workpiece in water, gas, oil, polymer, air, or other fluids. Given that lithium-ion batteries operate under relatively benign conditions, conventional rapid cooling does not significantly affect the property variations in their internal electrode materials during normal use. However, thermal runaway presents an exception due to its dramatic temperature fluctuations from room temperature to several hundred degrees Celsius. In this study, we investigated NCM811 cathodes in 18,650 batteries subjected to transient thermal runaway followed by rapid cooling using several advanced analytical techniques. The results reveal a phenomenon characterized by intergranular cracking within NCM811 cathode materials when exposed to rapid cooling during transient thermal runaway. Furthermore, lithium-ion cells utilizing reused NCM-182.4 electrodes in fresh electrolyte demonstrate a reversible capacity of 231.4 mAh/g after 30 cycles at 0.1 C, highlighting the potential for reusing NCM811 cathodes in the lithium-ion battery recycling process. These findings not only illustrate that NCM811 particles may experience intergranular cracking when subjected to rapid cooling during transient thermal runaway, but also the rapidly cooled NCM811 electrodes exhibit potential for reuse.

Lithium-ion batteries are widely used in energy storage systems and consumer electronics; however, long-term usage leads to capacity degradation, which affects system efficiency and safety. Existing studies have largely focused on individual cells and their SOH and SOC metrics, with less attention paid to larger battery packs, while the SOP more effectively reflects the overall operational characteristics of a battery pack. Therefore, this study proposes a parameter-optimized prediction method based on PSO and SVM, using SOP as a key health indicator for life prediction. In this method, the global optimal solution is obtained by simulating the collaborative search behavior of the particle swarm, dynamically updating particle positions and velocities; this solution determines the SVM’s critical parameters, namely the kernel parameter g and the penalty coefficient c, which are then used to train the SVM model to enhance its generalization ability and prediction accuracy. The results indicate that the PSO-SVM model can effectively capture the degradation characteristics of battery packs, achieving high-precision SOP prediction, with most prediction errors below 5% and the maximum error reaching 10.91%.

This study proposes a novel pneumatically driven mechanically prestressed rhombic bistable composite laminate with asymmetric cylindrical curvature, which exhibits two weakly-coupled cylindrical shapes where each shape is influenced by planform and geometry parameters. A reduced-order analytical model is developed to predict the laminate’s quasi-static equilibrium shapes and snap-through transitions of the laminate under pneumatic work loading, which is triggered by the internal pressure applied to the fluidic channels. A sensitivity study based on the model investigates the influence of key planform and geometric parameters (the internal angle α and aspect ratio E) on the laminate’s out-of-plane deflection and snap-through pressure. The results show that increasing α reduces the critical prestrain required to achieve bistability and amplifies the out-of-plane deflection, while excessive α may lead to monostable curvature. Variations in aspect ratio modify the coupling stiffness between orthogonal PEMC layers, thereby influencing the bistable domain and critical snap-through pressure. These findings provide methods for the design of bistable composite structures.

This study investigates the coupled quasi-static response and stable-state switching behavior of mechanically prestressed rhombic bistable composite laminates under internal pneumatic pressurization and external mechanical loading. A rhombic bistable composite laminate with embedded fluidic channels is proposed, where pneumatic pressurization is employed to reconfigure the deformation state and modulate the coupling between the laminate morphology and external actuation loads. An efficient reduced-order analytical model is developed to capture the interactions among geometric configuration, prestrain distribution, internal pressure, and external mechanical loading, enabling the rapid prediction of the deformation evolution and load–deflection response under coupled loading conditions. The main innovation of this work is integrating rhombic geometric tailoring, intrinsic pneumatic actuation, and multimode external loading into a unified analytical framework. The results demonstrate that the interior angle, prestrain distribution, and loading mode can effectively regulate equilibrium morphology, snap-through energy, and actuation efficiency. Parametric analyses reveal that the rhombic geometry introduces pronounced shear–bending coupling, providing an additional geometric degree of freedom for tailoring bistable configurations and energy barriers. In particular, a smaller interior angle generally reduces the snap-through energy barrier, whereas front-side prestrain increases the energy required for stable-state switching by enhancing the initial curvature. Comparisons among different loading modes further show that transverse point loading provides the highest energy conversion efficiency, in-plane loading requires the largest input energy, and pressure-assisted actuation exhibits intermediate efficiency. These findings provide fundamental insights and practical design guidelines for programmable morphing and load-efficient stable-state switching for rhombic composite laminates operating under coupled internal–external loading environments.

Power allocation in wind-powered multi-stack proton exchange membrane (PEM) electrolysis depends on both load-dependent efficiency and thermal response. This study evaluates coordinated allocation, stack rotation and temperature control for four parallel 200 kW stacks. A control-oriented model couples electrochemical and thermal dynamics with rectifier losses and hydrogen compression power. Particle swarm optimisation determines the allocation, an equivalent health index guides stack rotation, and model predictive control (MPC) tracks temperature references from an offline stack-efficiency map. Predicted polarisation voltages differ from experimental data by no more than 1.4% at 313 and 333 K. In annual simulations, rotation reduces the cross-stack index standard deviation from 725.10 to 2.92 h. MPC achieves a simulated temperature-tracking root mean square error of 0.019 K, 91.12% below that of fuzzy proportional–integral–derivative control. Across three representative wind power scenarios, the probability-weighted daily hydrogen production is 199.85 kg, 0.59% above optimal allocation at 333 K. The corresponding increases over equal and sequential allocation are 1.26% and 2.43%, respectively. These results quantify the additional hydrogen yield from load-dependent temperature control and the operating-load balance obtained through stack rotation under the evaluated conditions.