In the context of global carbon neutrality goals, the integration of high-proportion renewable energy sources has positioned large-scale energy storage as a critical enabler for the secure operation of new power systems. Among various technologies, lithium-ion battery-based electrochemical energy storage, particularly using lithium iron phosphate (LiFePO4) chemistry, has emerged as a key solution due to its millisecond-level power response and flexible configuration capabilities. However, the operational safety of these energy storage cells remains a significant concern, as thermal runaway (TR) incidents triggered by mechanical, electrical, or thermal abuse can lead to catastrophic cascading failures within densely packed battery modules and cabinets. This study addresses this critical challenge by developing a comprehensive framework for thermal runaway propagation modeling and fault localization in large-format LiFePO4 energy storage cells.
The safety of energy storage cells is paramount in grid-scale applications. Recent years have witnessed numerous safety accidents in energy storage stations worldwide, highlighting the urgent need for advanced thermal management and fault diagnosis methodologies. Thermal runaway in energy storage cells represents a complex multiphysics phenomenon involving coupled electrochemical-thermal-mechanical processes that can propagate through battery modules and escalate to system-level failures. Understanding the triggering mechanisms and propagation characteristics of thermal runaway in energy storage cells is therefore essential for developing effective safety strategies.
This research employs a multi-scale approach, beginning with experimental characterization of 280 Ah LiFePO4 energy storage cells under thermal abuse conditions. The experimental investigation systematically analyzes key parameters including voltage collapse characteristics, temperature evolution, and thermal runaway propagation thresholds across different states of charge (SOC). The experimental data serves as the foundation for developing high-fidelity multiphysics models that accurately capture the complex behavior of energy storage cells during thermal runaway events.
The modeling framework incorporates a novel parameter identification approach using the Forgetting Factor Recursive Least Squares (FFRLS) method to establish time-variant mappings between SOC and thermal runaway parameters. This enables the development of an optimized thermal propagation model that accounts for the dynamic relationship between the operational state of energy storage cells and their thermal runaway characteristics. The model successfully predicts thermal runaway propagation paths and energy release patterns in battery modules under varying SOC conditions, providing valuable insights for safety design optimization.
Building upon the experimental and modeling results, this study proposes an advanced fault inversion localization methodology that combines thermal propagation path constraints with hybrid optimization algorithms. The approach integrates a Genetic Algorithm (GA) and Grey Wolf Optimizer (GWO) to enhance the Time Difference of Arrival (TDOA) based fault localization framework. This innovative methodology enables precise identification of thermal runaway initiation points within battery clusters, even under complex operational scenarios with varying sensor configurations and state conditions.
Experimental Investigation of Thermal Runaway in Energy Storage Cells
The experimental investigation focuses on characterizing the thermal runaway behavior of 280 Ah LiFePO4 energy storage cells, which represent typical configurations used in grid-scale energy storage applications. The cells feature a nominal voltage of 3.2 V and operate within a voltage window of 2.5 V to 3.65 V. Prior to testing, all energy storage cells underwent standardized conditioning procedures according to IEC 62660-1 standards to ensure consistent initial conditions.
Thermal abuse tests were conducted on energy storage cells at different SOC levels (25%, 50%, 75%, and 100%) to systematically evaluate the impact of energy state on thermal runaway characteristics. A constant power heating element of 1000 W was applied to trigger thermal runaway, while multiple K-type thermocouples monitored temperature evolution at critical locations on the cell surface. Voltage response was simultaneously recorded to capture the electrochemical behavior during thermal runaway progression.
The experimental results reveal significant dependencies between SOC and thermal runaway parameters in energy storage cells. Higher SOC levels correlate with earlier thermal runaway triggering, higher peak temperatures, and faster temperature rise rates. The critical temperature thresholds for thermal runaway initiation in energy storage cells were identified within the range of 190°C to 205°C, with full SOC cells exhibiting the most severe thermal behavior.
The thermal runaway process in energy storage cells follows a characteristic sequence of events: initial heating leads to SEI decomposition, followed by separator melting and internal short circuit formation. Subsequent reactions between electrode materials and electrolyte generate substantial heat, leading to temperature escalation and eventual thermal runaway. The voltage response of energy storage cells during this process shows distinct patterns, with sudden voltage drops serving as early indicators of impending thermal runaway.
| Parameter | 25% SOC | 50% SOC | 75% SOC | 100% SOC |
|---|---|---|---|---|
| Safety Vent Opening Time (s) | 1407 | 916 | 378 | 338 |
| Voltage Drop Time (s) | 1496 | 1430 | 984 | 327 |
| Thermal Runaway Trigger Time (s) | 2683 | 1698 | 1010 | 469 |
| Peak Temperature Time (s) | 2721 | 1799 | 1219 | 847 |
| Voltage Drop Temperature (°C) | 136.5 | 136.6 | 114.9 | 61.2 |
| Safety Vent Opening Temperature (°C) | 107.9 | 101.7 | 70.3 | 73.8 |
| Thermal Runaway Trigger Temperature (°C) | 201.8 | 157.2 | 120.1 | 79.4 |
| Peak Temperature (°C) | 236.8 | 238.4 | 350.9 | 378.3 |
| Maximum Temperature Rise Rate (°C/s) | 2.2 | 3.9 | 16.4 | 26.5 |
The experimental data provides crucial validation benchmarks for the development of computational models of energy storage cells. The comprehensive dataset captures the dynamic evolution of thermal runaway in energy storage cells, enabling detailed analysis of the underlying mechanisms and facilitating the development of accurate predictive models for thermal behavior in practical energy storage applications.
Multiphysics Modeling of Thermal Runaway in Energy Storage Cells
The development of accurate computational models for energy storage cells requires careful consideration of the coupled electrochemical-thermal processes during thermal runaway. Traditional modeling approaches based on static parameters and linear superposition assumptions often fail to capture the complex nonlinear behavior of energy storage cells, particularly under dynamic SOC conditions. This research addresses these limitations through a novel data-model driven approach that incorporates SOC-dependent parameter variations.
The thermal runaway process in energy storage cells is governed by multiple exothermic side reactions that follow Arrhenius kinetics. The key reactions include SEI decomposition, negative electrode-electrolyte reactions, positive electrode-electrolyte reactions, and electrolyte decomposition. The heat generation from these reactions can be described by the following equations:
$$Q_{SEI} = H_{SEI} \cdot W_{SEI} \cdot R_{SEI}$$
$$Q_{ne} = H_{ne} \cdot W_{ne} \cdot R_{ne}$$
$$Q_{pe} = H_{pe} \cdot W_{pe} \cdot R_{pe}$$
$$Q_{ele} = H_{ele} \cdot W_{ele} \cdot R_{ele}$$
where $Q_i$ represents the heat generation rate for each reaction ($W/m^3$), $H_i$ denotes the heat of reaction per unit mass ($J/kg$), $W_i$ is the reactant mass concentration ($kg/m^3$), and $R_i$ represents the reaction rate ($s^{-1}$). The total side reaction heat generation is given by:
$$\Sigma Q_{side-tot} = Q_{SEI} + Q_{ne} + Q_{pe} + Q_{ele}$$
The reaction rates follow Arrhenius-type expressions:
$$R_i = A_i \cdot c^m \cdot \exp\left(-\frac{E_{a,i}}{RT}\right)$$
where $A_i$ is the pre-exponential factor ($s^{-1}$), $E_{a,i}$ is the activation energy ($J/mol$), $R$ is the universal gas constant (8.314 J/(mol·K)), $T$ is temperature (K), $c$ is the concentration of unstable lithium, and $m$ is the reaction order.
A critical innovation in this modeling approach is the incorporation of SOC-dependent activation energies for the side reactions in energy storage cells. The relationship is expressed as:
$$E_{a,i}(SOC) = k_{i,0} + k_{i,1} \cdot \exp(k_{i,2} \cdot SOC)$$
where $k_{i,0}$, $k_{i,1}$, and $k_{i,2}$ are parameters identified using the FFRLS method. This formulation captures the reduction in activation energy barriers with increasing SOC, which significantly affects the thermal runaway behavior of energy storage cells.
The FFRLS parameter identification method employs the following recursive equations:
$$y(k+1) = \phi^T(k+1) \hat{\theta}(k) + e(k+1)$$
$$\hat{\theta}(k+1) = \hat{\theta}(k) + K(k+1)e(k+1)$$
$$K(k+1) = \frac{P(k)\phi(k+1)}{\lambda + \phi^T(k+1)P(k)\phi(k+1)}$$
$$P(k+1) = \frac{1}{\lambda}[P(k) – K(k+1)\phi^T(k+1)P(k)]$$
where $y(k+1)$ is the system output, $\phi(k+1)$ is the input vector, $\hat{\theta}(k)$ is the parameter estimate vector, $K(k+1)$ is the algorithm gain, $P(k)$ is the covariance matrix, and $\lambda$ is the forgetting factor (set to 0.98 in this study).
The energy conservation equation for energy storage cells incorporates the heat generation from side reactions and considers both convective and radiative heat transfer:
$$\rho C_p \frac{\partial T}{\partial t} = \lambda \nabla^2 T + \frac{V_{cell}}{V_{batt}} \Sigma Q_{side-tot} – \frac{Q_{diss-tot}}{V_{batt}}$$
where $\rho$ is density ($kg/m^3$), $C_p$ is specific heat capacity ($J/(kg·K)$), $\lambda$ is thermal conductivity ($W/(m·K)$), $V_{cell}$ is the cell volume, and $V_{batt}$ is the battery volume. The dissipation term $Q_{diss-tot}$ includes both convective and radiative components:
$$Q_{diss-tot} = Q_{conv} + Q_{rad} = hA_{batt}(T – T_{amb}) + \epsilon \sigma_B A_{batt}(T^4 – T_{amb}^4)$$
where $h$ is the convective heat transfer coefficient ($W/(m^2·K)$), $A_{batt}$ is the surface area ($m^2$), $T_{amb}$ is ambient temperature (K), $\epsilon$ is emissivity, and $\sigma_B$ is the Stefan-Boltzmann constant ($5.67 \times 10^{-8} W/(m^2·K^4)$).
| Parameter | 25% SOC | 50% SOC | 75% SOC | 100% SOC |
|---|---|---|---|---|
| SEI Decomposition Ea,SEI (J/mol) | 1.36×105 | 1.28×105 | 1.19×105 | 1.14×105 |
| Negative Electrode-Electrolyte Ea,ne (J/mol) | 1.42×105 | 1.35×105 | 1.25×105 | 1.17×105 |
| Positive Electrode-Electrolyte Ea,pe (J/mol) | 1.80×105 | 1.71×105 | 1.62×105 | 1.56×105 |
| Electrolyte Decomposition Ea,ele (J/mol) | 1.86×105 | 1.78×105 | 1.69×105 | 1.64×105 |
The three-dimensional multiphysics model was implemented using finite element analysis, with careful attention to mesh independence and numerical convergence. The model successfully predicts the thermal behavior of energy storage cells across different SOC conditions, with peak temperature deviations of less than 3% compared to experimental measurements. This high-fidelity modeling approach provides a robust foundation for analyzing thermal runaway propagation in battery modules and developing effective fault localization strategies for energy storage systems.
Thermal Runaway Propagation in Energy Storage Modules
The propagation of thermal runaway in energy storage modules represents a critical safety concern for large-scale battery systems. This study investigates the thermal runaway propagation characteristics in modules comprising 52 series-connected 280 Ah LiFePO4 energy storage cells arranged in a 4×13 configuration. The module design incorporates practical engineering features such as thermal insulation pads and cooling plates to accurately represent real-world energy storage applications.
Thermal runaway propagation in energy storage modules exhibits distinct spatial and temporal patterns that are strongly influenced by SOC levels. The analysis reveals a critical SOC threshold between 50% and 75%, below which thermal runaway propagation does not occur despite single-cell triggering. This finding has significant implications for the safety management of energy storage systems, suggesting that operational strategies maintaining SOC below critical levels could effectively prevent cascading failures.
The propagation mechanism in energy storage modules follows a characteristic path constrained by thermal conduction pathways and material properties. The process can be divided into four distinct phases: single-cell instability, neighborhood diffusion, spatial propagation, and global failure. The thermal propagation velocity shows significant anisotropy, with faster propagation along the layer direction (y-axis) compared to the through-layer direction (x-axis) due to differences in thermal conductivity ($\lambda_y = \lambda_z = 18$ W/(m·K) vs $\lambda_x = 1.5$ W/(m·K)).
The thermal propagation time intervals between adjacent energy storage cells in the module vary with SOC and direction. For 100% SOC modules, the propagation interval along the x-axis is approximately 73 seconds, while along the y-axis it reduces to 66 seconds. For 75% SOC modules, these intervals increase to 98.2 seconds and 89.5 seconds respectively, indicating the significant influence of energy state on thermal runaway dynamics in energy storage cells.
The voltage response of energy storage cells during thermal runaway propagation provides complementary information for fault detection and localization. The sequential voltage collapse of energy storage cells follows the thermal propagation path, with time delays corresponding to the thermal propagation intervals. This correlation enables the development of voltage-based monitoring strategies for early warning of thermal runaway propagation in energy storage systems.
The thermal propagation model successfully captures the complex interactions between energy storage cells in the module, including heat transfer through conduction, convection, and radiation. The simulation results show excellent agreement with experimental observations, validating the model’s capability to predict thermal runaway propagation in practical energy storage configurations. This understanding of propagation mechanisms forms the basis for developing effective containment strategies and fault localization methods for energy storage systems.
Fault Inversion Localization Methodology for Energy Storage Clusters
The development of accurate fault localization methods for energy storage clusters is essential for ensuring the safe operation of large-scale battery energy storage systems. This research proposes a novel fault inversion localization methodology that combines thermal propagation path constraints with hybrid optimization algorithms to achieve precise identification of thermal runaway initiation points in battery clusters.
The fault localization framework employs a two-stage approach: rapid preliminary localization followed by precise inversion positioning. In the first stage, temperature sensors deployed around the energy storage cluster capture the thermal signature of propagating thermal runaway. The Time Difference of Arrival (TDOA) algorithm processes the sensor activation time differences to identify the most probable fault region within the cluster. The TDOA positioning principle is based on hyperbolic equations derived from signal time differences:
$$\Delta t = \frac{d_A – d_B}{v}$$
where $\Delta t$ is the time difference between signals arriving at reference points A and B, $d_A$ and $d_B$ are the distances from the target point P to A and B, and $v$ is the signal propagation velocity. The target location satisfies the hyperbolic equation:
$$\frac{(x – x_A)^2}{a^2} – \frac{(y – y_B)^2}{b^2} = 1$$
where $(x_A, y_A)$ and $(x_B, y_B)$ are coordinates of reference nodes, and $a$, $b$ are hyperbolic parameters related to $\Delta t$ and $v$.
The preliminary localization uses a matching approximation approach to identify the most likely fault region. For each candidate region $i$, the approximation degree $L_i$ is calculated as:
$$L_i = \frac{1}{n} \sqrt{(\tau_{i1} – \tau_{01})^2 + \cdots + (\tau_{ij} – \tau_{0j})^2 + \cdots + (\tau_{in} – \tau_{0n})^2}$$
where $\tau_{ij}$ is the time delay of the $j$-th sensor for region $i$, $\tau_{0j}$ is the measured time delay of the $j$-th sensor, and $n$ is the total number of sensors.
In the second stage, a hybrid Genetic Algorithm-Grey Wolf Optimizer (GA-GWO) refines the fault location within the identified region. The GA-GWO algorithm combines the global search capability of genetic algorithms with the efficient convergence of grey wolf optimization. The algorithm maintains a population of candidate solutions representing possible fault locations and states of energy storage cells, with the fitness function defined as:
$$\text{Fitness}(t_i) = \frac{1}{n} \sum_{j=1}^{n} (t_{i,j} – t_{1,j})^2$$
where $t_{i,j}$ is the time difference between the first and $j$-th sensors for candidate $i$, and $t_{1,j}$ is the corresponding measured time difference.
The hybrid optimization process involves several key steps. First, the GA operations (selection, crossover, mutation) generate a diverse population of candidate solutions. Then, the GWO leadership hierarchy (alpha, beta, delta wolves) guides the population toward promising regions of the search space. The position update in GWO follows:
$$\vec{X}(t+1) = \vec{X}_p(t) – \vec{A} \cdot \vec{D}$$
where $\vec{X}$ is the position vector, $\vec{X}_p$ is the position of the prey, $\vec{A}$ is a coefficient vector, and $\vec{D}$ is the distance vector calculated as $\vec{D} = |\vec{C} \cdot \vec{X}_p(t) – \vec{X}(t)|$.
The performance of the fault localization methodology is evaluated using a directional dual-threshold error metric (DD-TEM) that accounts for the anisotropic thermal propagation characteristics in energy storage clusters:
$$DD(p) = \sum_{i \in \{x,y,z\}} \left( \frac{|p_i – a_i|}{\tau_i} \right)^2 + \alpha \prod_{i: |p_i – a_i| > \tau_i} \lambda$$
where $a = (a_x, a_y, a_z)$ is the actual battery center coordinate, $p = (p_x, p_y, p_z)$ is the predicted coordinate, $\tau_i$ are directional thresholds representing energy storage cluster dimensions, $\lambda$ is a penalty coefficient for cross-layer errors, and $\alpha$ is a weighting factor.
| Battery State | Localization Method | Average DD Error Metric |
|---|---|---|
| SOC=25% SOH=100% | TDOA | 0.127 |
| PSO | 0.053 | |
| Single-dimension GWO | 0.079 | |
| Proposed Method | 0.027 | |
| SOC=50% SOH=100% | TDOA | 0.111 |
| PSO | 0.051 | |
| Single-dimension GWO | 0.094 | |
| Proposed Method | 0.023 | |
| SOC=75% SOH=100% | TDOA | 0.107 |
| PSO | 0.043 | |
| Single-dimension GWO | 0.040 | |
| Proposed Method | 0.021 | |
| SOC=100% SOH=100% | TDOA | 0.088 |
| PSO | 0.048 | |
| Single-dimension GWO | 0.057 | |
| Proposed Method | 0.019 |
The proposed fault localization methodology demonstrates robust performance across different SOC conditions and sensor configurations. The incorporation of SOC as an optimization parameter enables accurate localization even when thermal propagation characteristics vary with the energy state of storage cells. The method achieves localization accuracy exceeding 94% with DD error metrics below 0.03, significantly outperforming conventional approaches such as standard TDOA, Particle Swarm Optimization (PSO), and single-dimension GWO methods.
The effectiveness of the fault localization system was further evaluated under various sensor deployment scenarios. Results indicate that an 8-sensor configuration distributed around the energy storage cluster provides the optimal balance between localization accuracy and implementation cost. The system maintains reliable performance even with reduced sensor coverage or single-side sensor failure, demonstrating strong fault tolerance for practical energy storage applications.
| Number of Sensors | Distribution | Localization Accuracy | Average DD Error Metric |
|---|---|---|---|
| 2 | Left and Front Sides | 80% | 0.057 |
| 4 | Around Cabinet | 90% | 0.025 |
| 4 | Left and Front Sides | 90% | 0.041 |
| 8 | Around Cabinet | 95% | 0.021 |
| 16 | Around Cabinet | 95% | 0.020 |
Conclusion and Future Perspectives
This research has established a comprehensive framework for understanding and addressing thermal runaway propagation in LiFePO4 energy storage cells. Through systematic experimental investigation, advanced multiphysics modeling, and innovative fault localization methodologies, significant advancements have been achieved in the safety assessment and management of large-scale energy storage systems.
The experimental characterization of 280 Ah energy storage cells under thermal abuse conditions has provided valuable insights into the SOC-dependent nature of thermal runaway behavior. The identified critical temperature thresholds, triggering times, and propagation characteristics form an essential knowledge base for designing safer energy storage systems. The experimental results clearly demonstrate that higher SOC levels in energy storage cells lead to more severe thermal runaway consequences, highlighting the importance of operational strategies that maintain appropriate SOC levels for safety-critical applications.
The development of the SOC-adaptive thermal runaway model represents a significant improvement over traditional modeling approaches. By incorporating FFRLS-based parameter identification and establishing dynamic relationships between SOC and thermal parameters, the model achieves high accuracy in predicting thermal behavior across different operating conditions. This modeling capability enables proactive safety assessment and design optimization for energy storage systems containing numerous energy storage cells.
The proposed hybrid GA-GWO enhanced TDOA fault localization methodology demonstrates exceptional performance in identifying thermal runaway initiation points within battery clusters. The method’s ability to maintain high localization accuracy under varying SOC conditions and sensor configurations makes it particularly valuable for practical energy storage applications. The robust performance of this approach addresses a critical need in the industry for reliable fault diagnosis in large-scale battery energy storage systems.
Future research will focus on several important directions. First, the investigation of thermal runaway propagation under more complex operational scenarios, including combined electrical and thermal abuse conditions, will enhance the understanding of failure mechanisms in energy storage cells. Second, the development of advanced thermal management strategies, incorporating optimized insulation materials and cooling systems, will provide practical solutions for mitigating thermal runaway risks in energy storage systems. Finally, the integration of real-time monitoring and prognostic capabilities based on the developed models and algorithms will enable predictive safety management for grid-scale energy storage applications.
The methodologies and findings presented in this research contribute significantly to the safety enhancement of lithium-ion battery energy storage systems. By addressing the fundamental challenges of thermal runaway propagation and fault localization in energy storage cells, this work supports the continued deployment and safe operation of large-scale energy storage infrastructure essential for the global transition to sustainable energy systems.