The rapid advancement of electrochemical energy storage power stations toward larger capacity and higher power density has brought the thermal safety issues of the energy storage battery to the forefront of academic and industrial research. My study focuses on the thermal runaway spatial propagation characteristics, order reduction simulation modeling, and thermal insulation protection strategies for the energy storage battery. The primary objective of my research is to establish a comprehensive understanding of the sideward and longitudinal propagation mechanisms of thermal runaway in large-format lithium iron phosphate (LFP) batteries, which are widely used in stationary energy storage applications. The secondary goal is to develop a computationally efficient reduced-order model that can predict the three-dimensional temperature distribution during thermal runaway propagation, thereby enabling faster design iterations for safer energy storage battery systems.
1. Introduction and Research Context
Lithium-ion batteries have become the dominant technology for electrochemical energy storage due to their high energy density, long cycle life, and low self-discharge rates. However, the increasing capacity of individual cells, which has grown from 100 Ah to over 500 Ah in recent years, has exacerbated the risks associated with thermal runaway. When an energy storage battery undergoes thermal runaway, the released heat can trigger adjacent cells to also undergo thermal runaway through a chain reaction known as thermal runaway propagation. This chain reaction can lead to catastrophic fires and explosions in energy storage power stations, resulting in significant property damage and threats to human safety.
The safety incidents in energy storage power stations worldwide, including the 2021 Beijing Dahongmen energy storage station fire and the 2023 Queensland energy storage project fire, underscore the urgency of understanding the thermal runaway propagation mechanisms in energy storage battery systems. Unlike automotive power batteries, stationary energy storage systems follow a multi-scale hierarchy of “cell–module–cluster–cabin” and exhibit unique spatial propagation characteristics. The thermal runaway propagation in energy storage power stations demonstrates both lateral (sideward) propagation within a module and longitudinal (vertical) flame propagation between stacked modules. Moreover, the massive number of cells in an energy storage battery system makes three-dimensional thermal runaway propagation simulations computationally prohibitive, necessitating the development of efficient reduced-order models.
2. Experimental Methodology
2.1 Battery Specifications
The research object of my study is a commercial 100 Ah LFP energy storage battery manufactured by a leading Chinese battery company. The battery specifications are summarized in the following table.
| Parameter | Value |
|---|---|
| Positive electrode material | LiFePO₄ |
| Negative electrode material | Graphite |
| Nominal capacity | 100 Ah |
| Nominal voltage | 3.2 V |
| Upper cut-off voltage | 3.65 V |
| Lower cut-off voltage | 2.5 V |
| Maximum charge rate | 2 C |
| Maximum discharge rate | 4 C |
| Mass | 2245±5 g |
| Dimensions (L×W×H) | 130×36×211.8 mm |
| State of charge | 100% |
The battery has a prismatic aluminum casing with bolt-type terminals. Through a detailed teardown process, I determined that the internal structure consists of a dual-stacked cell configuration. The battery housing walls are 1 mm thick, while the top and bottom caps are 2 mm thick. The two stacked cell groups each measure 128 mm × 17 mm × 177.6 mm, with a combined mass of 1865.4 g. The total electrolyte mass is approximately 81.4 g, accounting for 3.6% of the total battery mass.
2.2 Experimental Platforms
I employed the Extended Volume Adiabatic Reaction Calorimeter (EV-ARC) to measure the characteristic temperatures of thermal runaway. The EV-ARC operates in a “Heat-Wait-Seek” mode, providing an adiabatic environment to simulate the thermal runaway process. The calorimeter records the battery temperature evolution under conditions where no heat exchange occurs with the surroundings, enabling accurate determination of the self-heating onset temperature T₁, thermal runaway triggering temperature T₂, and maximum thermal runaway temperature T₃.
For gas analysis, I used a constant-volume pressure vessel with a volume of 324 L to measure the gas generation during thermal runaway. A gas chromatograph (Trace1300, Thermo Fisher Scientific) equipped with four detectors and eight columns was used to determine the gas composition. The sideward thermal propagation experiments were conducted in a combustion chamber constructed according to GB/T 25207-2010, equipped with a forced ventilation system. For the longitudinal flame propagation experiments, I used a double-layer stainless steel platform with a glass fiber tube and a pulse igniter to ignite the vented gases.
2.3 Thermal Runaway Characteristic Temperature Measurement
Since the internal structure of the battery prevents direct insertion of thermocouples, I employed the two-cell sandwich method to measure the internal temperature. The main thermocouple was placed between the large surfaces of two batteries, and the assembly was fixed with Teflon tape and placed inside the EV-ARC cavity. The key characteristic temperatures were determined as follows:
The self-heating onset temperature T₁ was found to be 116.6 °C, occurring at 2069.40 minutes from the start of the test. The venting temperature Tv was detected at 158.4 °C after 3063.3 minutes. The thermal runaway initiation temperature T₂ was measured as 179.7 °C at 3187.85 minutes, and the maximum temperature T₃ reached 479.0 °C at 3204.09 minutes.
The obtained T₂ value of 179.7 °C is considerably lower than the typical 200–300 °C range reported for various battery chemistries in the literature. This indicates that the studied energy storage battery requires relatively low energy input to trigger thermal runaway, suggesting a higher propensity for thermal runaway propagation in practical applications. To obtain a more accurate representation of the internal temperature, I calibrated T₃ using a single-cell thermal runaway simulation model. The calibrated T₃ value was determined to be 672.4 °C, which is more than 150 °C higher than the measured surface temperature, consistent with observations reported for aluminum-cased batteries in the literature. The calibrated total heat release ΔH during thermal runaway was calculated as 1226.3 kJ using the following relationship:
$$\Delta H = cm(T_3 – T_1)$$
where c is the specific heat capacity of the battery (985.3 J/(kg·°C)) and m is the battery mass (2.245 kg).
2.4 Gas Generation Characteristics
The thermal runaway gas generation test was conducted in the constant-volume pressure vessel. The experimental setup included eight thermocouples to monitor battery body temperature and environmental temperature. The pressure sensor P₁ inside the vessel measured the pressure changes to calculate the gas generation. The battery was heated by the heating plate to trigger thermal runaway, and after the experiment, the gas was collected using gas sampling bags for compositional analysis.
The pressure inside the vessel changed from the initial 109.4 kPa to a peak of 136.6 kPa during thermal runaway, ultimately stabilizing at approximately 130.9 kPa. Using the ideal gas law, I calculated the gas generation volume as follows:
$$PV = nRT$$
$$\Delta n = n_x – n_0$$
The total gas generation was 2.1 mol, corresponding to a unit capacity gas generation of 0.021 mol/Ah. The compositional analysis revealed that hydrogen was the dominant component, accounting for 65% of the total gas. Carbon dioxide comprised 13%, methane 10%, carbon monoxide 6%, ethylene 5%, ethane 1%, and propane 0.06%. The flammability limits of the mixed gas were calculated using Le Chatelier’s law:
$$L_{mix} = \frac{1}{\sum_{i=1}^{n} \frac{x_i}{L_i}} \times 100\%$$
The calculated flammability range of the mixed gas was 4.8–41.4%, indicating a relatively low lower flammability limit and thus a high risk of explosion. This finding highlights the importance of considering gas explosion hazards in the design of energy storage battery systems.
3. Spatial Propagation Characteristics of Thermal Runaway
3.1 Sideward Thermal Propagation in Modules
I conducted sideward thermal propagation experiments on a 1P4S (1 parallel, 4 series) module configuration. Four batteries (1# through 4#) were assembled with a heating plate placed on the front surface of battery 1#. Mica plates and aluminum alloy fixtures were used to provide thermal insulation and mechanical compression. The experimental phenomena were recorded using a video camera, and temperature and voltage data were collected at a sampling frequency of 10 Hz.
During the experiment, all four batteries underwent thermal runaway sequentially. The time intervals between thermal runaway initiations were as follows: battery 1# initiated at 762 s, battery 2# at 1378 s, battery 3# at 1695 s, and battery 4# at 2200 s. The sideward propagation of thermal runaway in the energy storage battery module was clearly observed. The venting duration varied significantly between batteries, with values of 204 s, 399 s, 156 s, and 309 s for batteries 1#-4#, respectively. Interestingly, the longer venting times were associated with lower peak temperatures, which I attributed to the variable contact thermal resistance between batteries that changes during thermal runaway due to casing expansion.
| Battery | Pre-test mass (g) | Post-test mass (g) | Mass loss (g) | Mass loss rate |
|---|---|---|---|---|
| 1# | 2243.3 | 1800.5 | 442.8 | 19.74% |
| 2# | 2247.4 | 1796.0 | 451.4 | 20.09% |
| 3# | 2242.4 | 1818.5 | 423.9 | 18.90% |
| 4# | 2240.7 | 1804.5 | 436.2 | 19.47% |
The surface peak temperatures during sideward propagation ranged from 452.4 °C to 539.2 °C for the non-heated surfaces. I observed a significant internal thermal propagation phenomenon, where the front and back surface temperature jumps were separated by intervals of 77–118 s. This internal propagation occurs because the thermal runaway front advances relatively slowly through the LFP cell stack. The maximum temperature difference (MTD) between front and back surfaces varied from 303.1 °C to 532.9 °C, with battery 2# exhibiting the shortest internal propagation time due to prolonged low-rate heat transfer during its 399 s venting period, which resulted in more uniform heating of the internal cell stack.
The thermal runaway propagation speed was calculated using:
$$v_i = \frac{\delta}{\Delta t_i}$$
where δ is the battery thickness (36 mm) and Δtᵢ is the thermal runaway duration. The individual battery propagation speeds ranged from 0.106 to 0.128 mm/s, while the module average propagation speed was determined to be 0.081 mm/s.
3.2 Energy Flow Distribution in Sideward Propagation
To quantitatively understand the energy distribution during thermal runaway propagation, I performed a detailed energy balance analysis for each battery in the module. The energy flow paths include: heat used to self-heat the battery (Qi), heat transferred to the next battery (Qi→i+1), heat transferred back to the previous battery (Qi→i−1), heat dissipated through radiation (Qrad), heat dissipated through convection (Qcov), and heat lost through venting and ejection (Qelse). The conductive heat transfer was calculated using Fourier’s law:
$$Q_{i \rightarrow i+1} = \int_{t_{i-start}}^{t_{i-end}} \frac{\lambda_{cell} A_1}{\sigma} [T_{(i+1)f} – T_{(i+1)b}] dt$$
$$Q_{i \rightarrow i-1} = \int_{t_{i-start}}^{t_{i-end}} \frac{\lambda_{cell} A_1}{\sigma} [T_{(i-1)b} – T_{(i-1)f}] dt$$
The equivalent axial thermal conductivity of the battery was calculated as:
$$\lambda_{cell} = \frac{L_{cell}}{\frac{L_{shell}}{\lambda_{shell}} + \frac{L_{stack}}{\lambda_{stack}}}$$
The heat used for self-heating was determined by:
$$Q_i = cm(T_{ave} – T_0) – Q_{heater} – Q_{i \rightarrow i-1}$$
The convective and radiative heat losses were calculated using:
$$Q_{conv} = \int_{t_0}^{t_1} h A_2 (T_s – T_\infty) dt$$
$$Q_{rad} = \int_{t_0}^{t_1} \varepsilon_{rad} \sigma_{rad} A_2 (T_s^4 – T_\infty^4) dt$$
My calculations revealed the following energy distribution pattern for interior batteries in the module: 73.42% of the total energy was used for self-heating, 2.13% was transferred to the next cell, 0.92% was transferred back to the previous cell, 0.73% was lost through convection, 0.27% through radiation, and 22.54% was dissipated through venting and ejection. This energy distribution highlights that the majority of the thermal runaway energy remains within the originating energy storage battery itself, while a relatively small fraction is available to trigger propagation to adjacent cells.
3.3 Longitudinal Flame Propagation in Modules
To investigate the vertical propagation of fire in energy storage battery systems, I designed three sets of experiments with different numbers of batteries in the upper and lower layers, as summarized in the following table.
| Experiment | Configuration | Upper-layer outcome | Classification |
|---|---|---|---|
| a | 1 upper, 1 lower | No thermal runaway | Non-propagation group |
| b | 2 upper, 2 lower | No thermal runaway | Non-propagation group |
| c | 3 upper, 3 lower | Simultaneous thermal runaway | Fire propagation group |
In the experiments, the bottom-layer batteries were heated from their front surfaces. After the first battery vented, the pulse igniter was used to ignite the vented gases. In experiment a, the flame from battery 1# caused the upper battery temperature to reach only 120.9 °C, insufficient to trigger venting or thermal runaway. In experiment b, the combined effect of two lower-layer batteries’ flames raised the upper batteries’ temperature to 162.6 °C, causing the upper batteries to vent but not to undergo thermal runaway. In experiment c, with three lower-layer batteries, the upper-layer three batteries underwent simultaneous thermal runaway, demonstrating the phenomenon of synchronized flame propagation.
The temperature characteristics revealed that the upper-layer batteries reached a maximum surface temperature of 640.2 °C, which was 22.1% higher than the lower-layer batteries. The maximum temperature rise rate in the upper-layer batteries was 14.0 °C/s, 86.7% higher than the maximum rate of 7.5 °C/s observed in the lower-layer batteries. This asymmetry in thermal hazards arises because the upper batteries experience combined heating from their own thermal runaway reactions and the external flame heating from below.
I further analyzed the temperature rise patterns of the upper-layer batteries, revealing three distinct temperature rise staircases preceding their synchronized thermal runaway. Each staircase consisted of a gentle “flame baking” phase (when the lower battery vents) and a steep “flame jet” phase (when the lower battery undergoes violent thermal runaway). The temperature rise rates during the flame jet phase (13.9–17.6 °C/min) were approximately double those during the flame baking phase (6.4–7.6 °C/min), demonstrating the higher heat transfer efficiency of the jet flame.
3.4 Energy Transfer in Fire Propagation
The energy required to trigger thermal runaway in the upper-layer battery was calculated using the characteristic temperature data. Since the self-heating onset temperature T₁ was 116.6 °C, the triggering temperature T₂ was 179.7 °C, and the venting temperature Tv was 158.4 °C, the required energy for triggering was determined as:
$$Q = cm\Delta T$$
The energy required to trigger venting was 294.3 kJ, and the energy required to trigger thermal runaway was 341.8 kJ. However, the energy that actually triggered fire propagation in experiment c was determined to be 379.7 kJ, which was close to the calculated thermal runaway triggering energy, confirming the validity of my analysis.
By decoupling the heat transfer paths, I determined that the heat transferred through the bottom surface was 180.5 kJ, accounting for 47.6% of the total triggering energy, while the heat transferred through the side surfaces was 198.4 kJ, accounting for 52.4%. This near-equal distribution between bottom and side heat transfer paths underscores the importance of comprehensive thermal insulation that covers both surfaces to effectively prevent fire propagation.
Regarding the heating power in different stages, the flame jet phase delivered significantly higher heating power (511.5–648.7 W) compared to the flame baking phase (248.6–279.5 W). Each individual temperature staircase provided 111.2–130.4 kJ of energy to the upper-layer battery. The physical mechanism of synchronized flame propagation was revealed as follows: the lower-layer batteries underwent sequential thermal runaway from bottom to top, and each event provided a similar energy increment to all upper-layer batteries simultaneously. After accumulating sufficient energy through three staircase increments, the upper-layer batteries simultaneously reached the thermal runaway threshold, causing synchronized fire propagation.
4. Reduced-Order Thermal Runaway Propagation Model
4.1 Model Framework
I developed a reduced-order thermal runaway propagation model based on the Krylov subspace method. The model was constructed by integrating a heat transfer model in state-space form with a zero-dimensional thermal runaway model. The three-dimensional heat conduction equation was converted into a state-space formulation:
$$[A]\dot{T}(t) + [C]T(t) = [L]u(t)$$
$$y = [R_f]T(t)$$
The coefficient matrices were derived from the finite element discretization in ANSYS Transient Thermal software. Matrix [A] captures the thermal inertia of the system, matrix [C] represents heat diffusion within the system, matrix [L] is the input distribution matrix, and matrix [Rf] extracts the output temperatures from the nodal values.
The Krylov subspace method was employed to reduce the order of the system. A r-dimensional Krylov subspace is defined as:
$$K_r(E; b) = span\{b, Eb, E^2b, …, E^{r-1}b\}$$
The Arnoldi algorithm was used for generating an orthonormal basis [V] of the Krylov subspace, enabling the transformation:
$$\dot{T}_r(t) = V^T E V T_r(t) + V^T b u(t)$$
$$y_r(t) = R_f V T_r(t)$$
The reduced-order system preserves the first r moments of the original system’s transfer function, guaranteeing accurate approximation of the thermal dynamics. I used the “MOR for ANSYS” plugin developed by Rudnyi to perform the order reduction within ANSYS.
The thermal runaway model was built in Simulink using data from EV-ARC measurements:
$$q(t) = \frac{\Delta H}{\Delta t} \int_0^t q(\tau) d\tau \quad for \quad T(t) > T_2$$
$$Q(t) = \int\int\int q(t) dV$$
where q(t) is the heat generation power, ΔH is the total energy release (1226.3 kJ), Δt is the energy release duration (40 s), and T₂ is the thermal runaway trigger temperature (179.7 °C).
4.2 Model Validation
The model geometry included the battery casing, two cell stacks, positive and negative terminals, top cap, heating plate, mica plate, and fixture. The material properties used in the model are summarized in the following table.
| Component | Material | Density (kg/m³) | Cp (J/(kg·°C)) | λ (W/(m·°C)) |
|---|---|---|---|---|
| Casing | Aluminum alloy | 2770 | 875 | 170 |
| Cell stack | Stack material | 2407.7 | 985.3 | λx=λy=24.93; λz=0.98 |
| Positive tab | Aluminum | 2689 | 951 | 237.5 |
| Negative tab | Copper | 8933 | 385 | 400 |
| Heating plate | Stainless steel | 7750 | 480 | 15.1 |
| Mica plate | Mica | 2500 | 500 | 0.34 |
| Fixture | Aluminum alloy | 2770 | 875 | 170 |
The boundary conditions included thermal conduction between components, convective heat transfer with the environment (h = 15 W/(m²·°C)), and radiative heat transfer (surface emissivity ε = 0.95). The environmental temperature was set to 17.5 °C. The model was validated against the 1P4S sideward thermal propagation experiment data. The error analysis for all key feature points is shown in the table below.
| Location | Symbol | Simulation trigger time (s) | Experiment trigger time (s) | Error | Simulation peak temp (°C) | Experiment peak temp (°C) | Error |
|---|---|---|---|---|---|---|---|
| Front | T1f | 763.2 | 762.5 | 0.09% | 644.7 | 640.1 | 0.72% |
| Front | T2f | 1263.6 | 1337.9 | 5.55% | 475.0 | 452.4 | 5.00% |
| Front | T3f | 1684.3 | 1690.0 | 0.34% | 500.0 | 539.2 | 7.27% |
| Front | T4f | 2221.5 | 2200.2 | 0.97% | 493.8 | 497.9 | 0.82% |
| Back | T1b | 848.2 | 827.9 | 2.45% | 475.1 | 451.4 | 5.25% |
| Back | T2b | 1455.3 | 1398.5 | 4.06% | 500.3 | 529.8 | 5.57% |
| Back | T3b | 1807.3 | 1797.4 | 0.55% | 493.7 | 490.8 | 0.59% |
| Back | T4b | 2318.6 | 2308.2 | 0.45% | 402.1 | 401.5 | 0.15% |
All key feature point errors were within 10%, confirming that the reduced-order model accurately predicts the temperature evolution during thermal runaway propagation in energy storage battery systems.
4.3 Temperature Prediction
I used the validated reduced-order model to predict the internal core temperatures and tab temperatures of the batteries, which could not be directly measured in experiments. The predicted internal core temperatures revealed that the first cell’s internal temperature increased sharply at 747.9 s, reaching a peak of 722.3 °C at 968.6 s. The second cell experienced thermal runaway initiation at 1263.3 s with an initial temperature of 110.6 °C, reaching 687.2 °C at 1494.3 s. The third cell initiated at 1597.4 s with an initial temperature of 78.8 °C, peaking at 668.4 °C at 1850.4 s. The fourth cell underwent initiation at 2098.1 s and reached 711.3 °C at 2308.1 s.
The internal thermal runaway onset temperatures ranged from 78.8 °C to 110.6 °C, while the peak internal temperatures were between 668.4 °C and 722.3 °C. I observed that the internal thermal runaway initiation time was slightly earlier than the surface initiation time, and the internal temperature was approximately 200 °C higher than the surface temperature. This temperature difference is attributed to the heat transfer resistance between the internal cell stack and the aluminum casing.
The tab temperature predictions showed a wave-like increasing trend during the propagation process. Each battery’s terminal experienced the largest temperature rise (82.1–102.7 °C) during its own thermal runaway event, while terminal temperature rises from remote battery thermal runaway events were significantly smaller. The maximum terminal temperature reached approximately 335 °C for both positive and negative terminals. The slight difference between positive and negative tab temperatures was due to the different thermal properties of aluminum (positive) and copper (negative). The negative tab temperature for cell 3 reached a peak of 334.9 °C.
4.4 Parametric Studies
I conducted parametric studies by adjusting key parameters in the reduced-order model to explore their effects on thermal runaway propagation characteristics. Four parameters were investigated: thermal runaway trigger temperature T₂, total heat release ΔH, convective heat transfer coefficient h, and heating power P.
Effect of thermal runaway trigger temperature T₂: I simulated four scenarios with T₂ values of 180 °C (baseline), 190 °C, 199 °C, and 200 °C. The results showed that increasing T₂ from 180 °C to 190 °C extended the first cell’s thermal runaway onset time by 1.1% and the module propagation time interval by 16.8%. When T₂ was increased to 199 °C, the onset time was delayed by 16.3% and the propagation interval by 41.1%. Importantly, when T₂ was increased to 200 °C, the thermal runaway propagation was completely suppressed. This finding indicates that selecting separator materials with melting temperatures above 200 °C would effectively prevent thermal runaway propagation in this energy storage battery.
Effect of total heat release ΔH: I varied ΔH from 100% to 95%, 93%, and 90% of the baseline value. Reducing ΔH to 95% extended the propagation interval by 17.1% and decreased the average peak temperature by 3.0%. At 93% ΔH, the propagation interval was extended by 29.6%. At 90% ΔH, complete suppression of thermal runaway propagation was achieved. The thermal runaway onset time of the first cell remained essentially unchanged since this parameter affects only the energy released after trigger.
Effect of convective heat transfer coefficient h: Increasing h from 15 to 25 W/(m²·°C) extended the propagation interval by 25.3%. At h = 30 W/(m²·°C), the interval was extended by 29.5%. Complete suppression occurred when h was increased to 35 W/(m²·°C). These results demonstrate that enhanced thermal management with forced convection can effectively mitigate thermal runaway propagation risk in energy storage battery systems.
Effect of heating power P: I investigated six heating power levels from 400 W to 1400 W. The thermal runaway onset time decreased from 2169.5 s at 400 W to 389.8 s at 1400 W, following an exponential decay pattern with the fitted function:
$$y = 9657.96686 \times exp(-x/235.19275) + 398.22434$$
Interestingly, the propagation time interval exhibited a non-monotonic relationship with heating power, first increasing from 1212.3 s at 400 W to a maximum of 1463.2 s at 1000 W, then decreasing to 1317.0 s at 1400 W. This non-monotonic behavior suggests an optimal heating power range for safety testing that maximizes the observable propagation characteristics.
4.5 Computational Efficiency Analysis
To evaluate the computational efficiency improvement of the reduced-order model, I built a three-dimensional FVM model using the CFD software Ansys Fluent. The FVM model used the same mesh file as the reduced-order model before order reduction to ensure a fair comparison. I also extended the module configuration to a 1P13S structure to investigate the efficiency gains for larger energy storage battery modules. The simulation time comparison is summarized in the following table.
| Configuration | FVM model time | Reduced-order model time | Time reduction |
|---|---|---|---|
| 1P4S module | 328 min 49 s | 8 min 45 s | 97.3% |
| 1P13S module | 53 h 56 min 14 s | 1 h 10 min 4 s | 97.8% |
The reduced-order model achieved over 40 times speedup compared to the conventional FVM model while maintaining prediction accuracy within 10%. The computational efficiency gain is consistent across different module sizes, demonstrating the scalability of the reduced-order approach for large-scale energy storage battery system simulation.
5. Thermal Insulation Protection Strategies
5.1 Thermal Barrier Model
I developed a reduced-order thermal barrier model based on the Krylov subspace method. The model geometry included the fixture, heating plate, battery 1, and battery 2, with the insulation material simplified as a virtual thermal resistance layer. This simplification is justified because pure insulation materials typically have small specific heat capacities and thin geometries, making their heat absorption capacity negligible compared to their thermal resistance effect. The thermal resistance of the virtual insulation layer was calculated as:
$$R = \frac{d}{\lambda}$$
where d is the thickness and λ is the thermal conductivity of the insulation material.
The model was validated against two-cell thermal propagation experiments without insulation material. All key feature point errors were within 5%, confirming the model’s accuracy.
I surveyed commercially available insulation materials to inform the simulation matrix. The materials considered included aerogel (thermal conductivity 0.012–0.04 W/(m·°C), density 3.55 kg/m³, cost 1400 yuan/kg), ceramic fiber (thermal conductivity 0.03–0.07 W/(m·°C), density 96–128 kg/m³, cost 2–6 yuan/kg), polypropylene board (thermal conductivity 0.13–0.24 W/(m·°C), density 900–920 kg/m³, cost 7.5 yuan/kg), polyimide film (thermal conductivity 0.1–0.3 W/(m·°C), density 1300–1500 kg/m³, cost 200 yuan/kg), polyurethane foam (thermal conductivity 0.02–0.025 W/(m·°C), density 50–80 kg/m³, cost 12 yuan/kg), and ceramifying silicone rubber (thermal conductivity 0.2–0.3 W/(m·°C), density 1100–1500 kg/m³, cost 10 yuan/kg). Based on this survey, I designed 16 simulation schemes with four thickness values (1, 1.5, 2, and 2.5 mm) and four thermal conductivities appropriate for each thickness, covering the thermal resistance range from 0.0125 to 0.05 (m²·°C)/W.
5.2 Simulation Results
I evaluated the thermal barrier performance using three key indicators: the thermal runaway propagation time interval, the heat transfer power through the insulation material, and the total heat transferred during the critical time interval. The baseline propagation time interval without insulation was 303.4 s.
The simulation results for different insulation thicknesses and thermal conductivities are summarized as follows. For 1 mm thickness, the propagation time intervals were 450.5 s (λ=0.065 W/(m·°C)), 586.6 s (λ=0.045 W/(m·°C)), 995.0 s (λ=0.025 W/(m·°C)), and complete suppression (λ=0.02 W/(m·°C)). For 1.5 mm thickness, the intervals were 427.7 s (λ=0.105 W/(m·°C)), 570.7 s (λ=0.075 W/(m·°C)), 906.1 s (λ=0.045 W/(m·°C)), and complete suppression (λ=0.03 W/(m·°C)). For 2 mm thickness, the intervals were 408.0 s (λ=0.15 W/(m·°C)), 513.0 s (λ=0.11 W/(m·°C)), 752.7 s (λ=0.07 W/(m·°C)), and complete suppression (λ=0.04 W/(m·°C)). For 2.5 mm thickness, the intervals were 397.9 s (λ=0.2 W/(m·°C)), 482.2 s (λ=0.15 W/(m·°C)), 667.3 s (λ=0.1 W/(m·°C)), and complete suppression (λ=0.05 W/(m·°C)).
I further analyzed the relationship between the reciprocal of thermal resistance (1/R) and the propagation time interval, revealing an exponential decay relationship:
$$y = 2116.74374 \times exp(-x/21.53282) + 346.26916$$
Based on this relationship, I determined design guidelines for thermal insulation selection. Materials with thermal resistance R ≥ 0.022 (m²·°C)/W achieve 10-minute delay of propagation, R ≥ 0.035 (m²·°C)/W achieve 15-minute delay, and R ≥ 0.05 (m²·°C)/W achieve complete suppression of thermal runaway propagation.
The critical heat transfer power analysis showed that the peak heat transfer power was 556.3 W without insulation. Complete suppression of propagation was achieved when the peak heat transfer power was reduced to 382.2 W, representing a 31.3% reduction. The critical heat transfer during the time interval Δt was determined by integrating the heat transfer power curve. Without insulation, the heat transferred during 303.4 s was 105.92 kJ. The critical heat transfer for complete suppression was 83.89 kJ, representing a 20.8% reduction in energy transfer. These critical values provide quantitative design targets for thermal insulation materials in energy storage battery modules.
5.3 Experimental Verification
Based on the simulation results, I selected two commercially available insulation materials with a thickness of 2 mm for experimental verification. The first material was aerogel with a measured thermal conductivity of 0.036 W/(m·°C), and the second was a ceramic fiber-based composite phase change material (CPCM) with a phase change enthalpy of 1727 J/g. The CPCM had a thermal conductivity of 0.68 W/(m·°C) before phase change and 0.035 W/(m·°C) after phase change.
In the aerogel barrier experiment, battery 1# underwent thermal runaway at 751 s after the start of heating. The back surface peak temperature of battery 1# reached 476.1 °C at 996 s. Due to aerogel insulation, the front surface temperature of battery 2# never exceeded 130 °C, reaching a peak of only 125.9 °C at 1249 s. The maximum temperature difference across the aerogel was 373.9 °C at 973 s. Battery 2# remained completely unaffected, and its voltage remained stable throughout the experiment, demonstrating the effective suppression of thermal runaway propagation.
In the CPCM barrier experiment, battery 1# underwent thermal runaway at 787 s. The CPCM underwent phase change at 877 s when the back surface temperature of battery 1# was 120.4 °C. The back surface peak temperature of battery 1# reached only 261.0 °C, which was significantly lower than in the aerogel experiment due to the heat absorption by the phase change material. The front surface temperature of battery 2# peaked at 205.3 °C at 1469 s, which was still below the thermal runaway trigger temperature. The maximum temperature difference across the CPCM was 105.4 °C.
| Experimental group | Battery 1# back peak temp (°C) | Battery 2# front peak temp (°C) | Maximum temperature difference (°C) |
|---|---|---|---|
| Aerogel | 476.1 | 125.9 | 373.9 |
| CPCM | 261.0 | 205.3 | 105.4 |
The comparison revealed distinct advantages of each material. The CPCM effectively reduced the hazard level of the thermal runaway battery itself through the phase change heat absorption, reducing the back surface peak temperature of battery 1# by more than 200 °C compared to the aerogel experiment. However, the aerogel provided superior protection for the adjacent battery, as evidenced by the significantly lower peak temperature of battery 2# (125.9 °C vs. 205.3 °C). This difference is attributed to the structural collapse of the soft ceramic fiber matrix in the CPCM under the expansion pressure of the thermal runaway battery, which increases the effective thermal conductivity after phase change. The aerogel maintained its structural integrity throughout the experiment, preserving its low thermal conductivity.
Both materials successfully prevented thermal runaway propagation in the tested energy storage battery module, validating the simulation-based design approach. The experimental results align well with the simulation predictions that insulation materials with thermal resistance R ≥ 0.05 (m²·°C)/W can suppress thermal runaway propagation in this battery system.
6. Summary and Conclusions
My comprehensive study on the thermal runaway and thermal propagation characteristics of 100 Ah LFP energy storage battery systems, the development of efficient reduced-order simulation models, and the evaluation of thermal insulation protection strategies has yielded the following key findings. The studied battery exhibits a relatively low thermal runaway trigger temperature of 179.7 °C, indicating a higher susceptibility to thermal runaway compared to other LFP cells. The battery releases 2.1 mol of gas during thermal runaway, with hydrogen comprising 65% of the total gas volume. The flammable limit of the mixed gas is 4.8–41.4%, highlighting significant explosion risks.
In sideward thermal propagation, the surface peak temperatures range from 452.4 °C to 539.2 °C for non-heated surfaces, with a module average propagation speed of 0.081 mm/s. The energy flow analysis reveals that 73.42% of the total thermal runaway energy remains in the originating cell, 2.13% transfers to the next cell, and 22.54% dissipates through venting. In longitudinal flame propagation, the upper-layer batteries exhibit higher peak temperatures (640.2 °C, 22.1% higher) and faster temperature rise rates (14.0 °C/s, 86.7% higher) than the lower-layer batteries. The energy boundary for triggering fire propagation is 379.7 kJ, with 47.6% transferred through the bottom surface and 52.4% through the side surfaces. The flame jet phase provides approximately twice the heating power of the flame baking phase.
The developed reduced-order model based on Krylov subspace methods achieves prediction errors within 10%. The model accurately predicts internal core temperatures (668.4–722.3 °C) and tab temperatures (up to 335 °C). Increasing T₂ to 200 °C, reducing ΔH to 90%, or increasing h to 35 W/(m²·°C) can suppress thermal runaway propagation. The model achieves computational efficiency improvements of more than 40 times compared to conventional FVM models.
For thermal insulation protection, materials with thermal resistance R ≥ 0.05 (m²·°C)/W can completely suppress thermal runaway propagation in the studied energy storage battery system. The critical peak heat transfer power for suppression is 382.2 W (31.3% lower than without insulation), and the critical heat transfer during the propagation interval is 83.89 kJ (20.8% lower). Both 2 mm aerogel and 2 mm CPCM experimentally validated the suppression capability, with aerogel providing superior protection for the adjacent cell and CPCM excelling in reducing the hazard level of the thermal runaway cell itself.
My research contributes to the fundamental understanding of thermal runaway propagation in large-format LFP energy storage batteries and provides practical tools and strategies for enhancing the safety of energy storage battery systems. The reduced-order simulation approach enables rapid design iterations for thermal safety optimization, while the identified critical insulation parameters provide quantitative guidance for material selection and system design.