Performance of Extinguishing Agents against Lithium-Ion Battery Fires
- Open Access
- 01.01.2026
Erschienen in:
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by (Link öffnet in neuem Fenster)
Abstract
1 Introduction
Lithium-ion batteries (LIBs) have revolutionized energy storage, offering high energy and power densities, long cycle lives, and low self-discharge rates. Their widespread adoption spans portable electronics, electric vehicles, and grid-scale energy storage. However, the very properties that make LIBs attractive also present safety challenges, most notably the risk of thermal runaway leading to fire and potential explosions [1, 2]. This phenomenon, triggered by abuse scenarios such as overcharging, over-discharging, short circuits, or external heat, involves a rapid and uncontrolled increase in temperature and pressure within the battery cell [3]. The presence of flammable organic liquid electrolytes further exacerbates fire hazards, facilitating rapid propagation throughout the battery pack [4].
The consequences of LIB fires can be devastating and endanger life, property, and the environment. High-profile incidents involving LIB fires in various applications underscore the urgent need for effective fire safety measures [5]. These incidents highlight the critical importance of understanding the mechanisms of LIB fires and developing strategies for their prevention, mitigation, and suppression.
Anzeige
A comprehensive approach for the fire safety of LIBs involves multiple layers of protection. This includes enhancing the inherent safety of battery key components (electrolyte, active materials, binders, separator, additives) and cell design, implementing robust battery management systems to prevent and detect thermal runaway, and developing effective fire suppression techniques. Although advancements in materials science and BMS technology have improved battery safety, effective fire suppression remains a critical area of ongoing research [6].
Traditional fire extinguishing agents, such as water and foam, may not be ideal for LIB fires owing to the unique characteristics of these batteries. Water, while effective in cooling some fires, can react with the electrolyte in LIBs (additional formation of HF), potentially exacerbating the situation. The high temperature and pressure generated during thermal runaway can also hinder the penetration and effectiveness of conventional agents. Moreover, any casing, i.e. BESS enclosure or electric vehicle pack and chassis on top, may hinder the penetration of the agent into the affected spot.
Consequently, specialized fire suppression agents and techniques tailored to LIB fires are essential. Researchers are actively exploring various approaches, including novel extinguishing agents, cooling methods, and fire-resistant materials [7]. Water mist systems, delivering finely atomized water droplets for effective cooling and flame suppression, and encapsulator agents, which form a barrier to prevent oxygen from reaching the fire, are among the most scrutinized solutions [1]. Several tunnel fire detection and fighting systems are currently available in the market, each with its own advantages and disadvantages. Although no single system is perfect, the water mist system is one of the top-performing conventional tunnel fire suppression systems available. Other approaches include the use of inert gases, such as nitrogen or argon, to displace oxygen and develop fire-resistant electrolytes and battery casings.
Recent advances in fire-suppressing agents for mitigating LIB fires have focused on the development of composite materials with enhanced properties [7]. These materials offer advantages such as environmental friendliness, high heat dissipation rates, electrical insulation, and prevention of re-ignition. A comprehensive review of various fire-extinguishing agents highlights their distinct characteristics and effectiveness in addressing LIB fires by considering factors such as insulation, toxicity, fire suppression speed, reactivity with fire sources, degradation capacity, and corrosion [8].
Anzeige
Despite these developments, a comprehensive understanding of suppression effectiveness across different battery configurations and failure modes is limited. This study addresses this gap by evaluating multiple suppression agents against lithium-ion battery fires, with particular focus on their ability to prevent thermal runaway propagation in an emulated domestic energy storage system.
The research objectives of this study were to:
1.
Evaluate the effectiveness of water mist, encapsulator agents, carbonate-based suppressants, and liquid nitrogen in extinguishing LIB fires.
2.
Assess these agents’ capability to prevent thermal runaway propagation.
3.
Quantify key performance metrics, including extinguishment time, cooling rate, and re-ignition prevention.
This paper presents experimental results from controlled tests, providing quantitative data to inform the development of more effective fire safety protocols for lithium-ion battery systems.
2 Materials and methods
2.1 Test Facility
The experiments were carried out within the basement garage of the Manchester Towers building, which is part of the GMFRS training facility located in Bury, UK. Figure 1 shows the aerial view of the facility, and inside the testing house with and without Battery Energy Storage System (BESS) fitted, as well as the extinguishers ready for the tests. The test site featured steel front doors that remained open during the trials, and a side door providing alternative access. Additionally, the building was equipped with an extraction, filtration, and scrubber system, which was operational for the duration of testing.
Fig. 1
The test facility at GMFRS, A Labelled overhead view of the Manchester Towers building, B A 3D view of the Manchester Towers building, C Image of the front of the garage in the Manchester Towers building, D Inside of the Manchester Towers garage, E Image of the 9-Litre extinguisher cannisters used to apply the agents, F An image of the assembled test article in the garage
2.2 Battery Modules
The specifications of the NMC 532 cells and modules employed in the tests are summarized in Table 1.
Table 1
Characteristics of the NMC 532 cells
Cell/module | Dimensions (length x width x height)/mm | Mass/kg | Voltage/V | Energy capacity/Wh | Volumetric energy density/Wh L-1 | Energy density/Wh kg-1 |
|---|---|---|---|---|---|---|
Cell | 261 × 216 × 7.9 | 0.914 | 3.65 | 209 | 469 | 229 |
Module | 300 × 222 × 68 | 8.5 | 14.6 | 1670 | 369 | 196 |
2.3 Test Methodology
Five modules were mounted in an open-fronted steel case and locked in position by steel rods threaded at both ends and nuts, Fig. 2 shows an image of the assembled BESS prior to mounting onto the test rig. The modules were located in the case such that there was a space around all four sides of the module assembly, with rather larger gaps on top and bottom. Once assembled, the BESS was placed on a steel frame, as shown in Figures S3(a–c) and S4 in Supplementary Information, and described in another paper [9], held in place by threaded steel bars, nuts, and a steel strap. The BESS in Figure S4 is thus not an exact representation of a domestic BESS, as the cells in the latter are usually completely enclosed in a steel case; hence, the design employed in our tests provides an advantage to the extinguishers, as the modules are accessible to the suppressants if the individual cells are not.
Fig. 2
The assembled BESS test article
The frame was located in the basement garage of the Manchester Towers facility. The garage had steel doors that were kept open, and the basement of the tower had a metal shutter that was kept closed (see Figure S5). The locations of the steel frame and CCTV, and GoPro cameras inside the garage are shown in Figures S6 – S8. The latter were 4 K 60 FPS Wi-Fi Dual Screen Waterproof Sports Action Cameras.
The cells in each module were arranged in two separate sets (top and bottom) of 2p2s. Initially, the voltage of each parallel pair of cells was monitored; however, for operational reasons, it was changed to the voltage across each 2p2s quartet of cells, as shown in Figs. 3a, b. The location and specification of the type-K thermocouples are shown in Fig. 3; Table 2 respectively. The supplementary information also provides details of the manufacturing of the steel case and the assembly of the modules.
Fig. 3
Thermocouple and voltage sensor locations for the experiments (A experiments where all parallel voltage terminals were measured, B experiments where the 2p2s cell quartet was measured.)
Table 2
Specification of the type K thermocouples used throughout the experiments
Temperature Range | Accuracy (−40 °C–+375 °C) | Accuracy (−40 °C–+375 °C) |
|---|---|---|
−200 °C–+1260 °C | ± 2.5 °C | ± 0.75 °C |
In all the tests, each module was precharged to 100% SOC and assembled into the system. Once in place, one parallel cell pair in module 2 (numbered from the bottom) was overcharged, as shown in Figure S9 at a constant current of 120 A, corresponding to a ca. 1 C overcharge. The power supplies employed to charge and overcharge the modules were ETPS LAB-DSP 012.5–120, 0–12.5, 0–120 A, 1500 W (Single Phase). Overcharge was employed as the abuse method as a number of our previous tests [9] showed that overcharging always initiated fire in the thermal runaway process.
When thermal runaway (TR) was initiated, the power supply was switched off. Flames were allowed to burn for 1 min before a fire officer, in a full breathing apparatus, deployed the extinguisher agent with as many cannisters as required.
2.4 Extinguishers
2.4.1 Water-Based Agents
To ensure a comprehensive evaluation, four distinct suppression agents were acquired. These agents were selected to encompass a variety of fire suppression mechanisms and to reflect practical usage scenarios, drawing upon prior research [10], delivery methods, availability, and extinguisher size (9 L).
The extinguisher agents used in the tests are as follows:
-
- Water mist.
-
- Encapsulator agent - water mist + encapsulator agent.
-
- Carbonate agent - water mist + ammonium bicarbonate-based additive.
-
- Mixed agent (12.5%) - water extinguisher + proprietary mixture/trade secret (contains natural boron compounds, food-grade wetting agents, and anionic detergents).
Three replicates were performed for each test using 9 L hand-held water mist canisters, up to five per test.
To evaluate the performance of different extinguishing agents, the time to extinguish the fire, the time to re-ignition, and temperature measurements were recorded.
Prior to the main tests, control tests were conducted, where the same method of abuse was used, but no extinguishing method was applied to assess a scenario where no intervention was applied.
2.4.2 Liquid Nitrogen
Additionally, three experiments were performed with liquid nitrogen as the extinguishing/cooling agent. The liquid nitrogen was released via a lance (under a pressure of 4 bar) from a 200 L dewar.
The BESS apparatus was identical to all other trials, with the exception of a custom lance which was used to release the liquid nitrogen.
The lance to project the stream of liquid nitrogen at the BESS was manufactured from 9.5 mm bore (3/8” BSP size) stock steel tubing, a nominal outside diameter of 16 mm, with a length of 1.6 m. One end of the lance was threaded with a 3/8” BSP to allow the attachment of a shut-off valve and connection to a commercial braided stainless steel cryogenic hose. The hose was attached to the 200 L liquid nitrogen Dewar, which was set back through the side door of the test chamber for safe operation by the fire officer wearing full protective PPE and breathing apparatus. The lance was held in position using lab stands and clamps, the former stabilized using concrete slab weights, as the pressure from the release of liquid nitrogen would otherwise dislodge the lance.
In the first test, the lance was positioned at a 56 °angle to the side of the frame, 380 mm away from the BESS, at a height of approximately 435 mm, as shown in Fig. 4A. The nitrogen stream was directed at the gap between the overcharged module, second from the bottom, and the edge of the steel case.
In the second and third tests, an addition was made to position it for better gas flow in and around the metal casing to increase the cooling area. The changes to the lance were made by using 22 mm copper tubing, with two straight lengths and an elbow, to extend the tube at a right angle of approximately 340 mm to direct the nitrogen stream directly at the gap between the steel case and the overcharged module, as shown in Fig. 4B. The tip of the end of the copper pipe was flattened to create a fan aperture approximately 6 mm wide × 30 mm high. The distance between the module and the aperture was approximately 265 mm.
Fig. 4
Diagrams of the liquid nitrogen lance apparatus. A 1 st test apparatus with the lance at a 56° angle to the test frame, B 2nd and 3rd test apparatus with the elbowed lance
3 Results
3.1 Control Test (CT) - No Suppression
This section presents the results of the control test, in which no extinguisher agent was applied, allowing the (BESS) to burn freely. The control test results serve as a baseline for comparing the performance of various fire suppression agents.
The overcharged (OC) cell pair voltage collapsed, and the module ignited 1700 s after overcharging was initiated, with peak voltage and temperature before ignition reaching 5.5 V and 21 °C, respectively. The overcharge characteristics of the modules employed in this work are discussed in [6], and various electrochemical processes taking place have been well described in the literature [11‐14].
Figure 5(A)–(B) presents graphs illustrating the cell-pair voltages and thermocouple temperatures recorded during the control test. Thermocouple T5 displayed a rapid increase to a maximum temperature of 372 °C, reflecting the initial rapid flame propagation, before decreasing and oscillating at approximately 300 °C. In contrast, T6 showed a more gradual and steadier rise before levelling out at approximately 210 °C. Thermocouple T1, located directly below the overcharged cell pair (V5), showed a steady temperature increase, which commenced with the collapse of V5. As the last cell pair entered thermal runaway, the temperature of T2 increased as it was located at the top of module 2, above the topmost cell pair V7.
Fig. 5
Temperature (A) and Voltage sensors (B) plots from the overcharge failure of the control test
The protected thermocouples (i.e., those located between modules) showed similar 2-stage responses: a slower rise before levelling out, then a very rapid increase (≥ 4 °C s-1) taken as thermal runaway. The relative times for stage 1 differed, with stage 1 of the response of T4 being much shorter and the increase in temperature during stage 1 being much lower (stage 2 commenced at 400 °C for T1, 450 °C for T2 & 3, and 175 °C for T4) compared to the responses of the other three thermocouples. Interestingly, before stage one, all four thermocouples showed a similar gradual increase in the temperature of ca. 0.08 °C/s.
The voltage data indicated that thermal runaway initially propagated to the other cells in the second module before spreading to the lower module 1. Thermal runaway then sequentially affected modules 3, 4, and 5. The corresponding thermocouple readings showed that the temperature of each module increased, coinciding with the voltage collapse and subsequent thermal runaway of the respective module. Thermal runaway of the modules led to a rapid rise in temperature, reaching a maximum of approximately 890 °C. TC6 revealed that the temperature of the rear of the container peaked at 355 °C. The duration from ignition to final module collapse was approximately 700 s.
The control test, without any intervention or suppression effort, demonstrated a rapid and intense burn (< 700 s from initial ignition) of the entire BESS. The temperature within the system exceeded 890 °C during the uncontrolled fire scenario, highlighting the potentially dangerous and rapidly escalating situation that first responders may face when arriving at a BESS fire scene.
3.2 Water Mist (WM)
This section presents the results of tests using water mist as the suppression agent. In addition to assessing the performance of standard water, these tests allow for comparing the effectiveness of water-based agents with additives.
In test 1 (WM1), ignition occurred at 1709 s with peak voltage and temperature during the overcharge period of 5.5 V and 29 °C, respectively. For test 2 (WM2), ignition occurred at 1717 s, with peak voltage and temperature being 5.5 V and 26 °C, respectively. In test 3 (WM3), ignition occurred at 1673 s with peak voltage and temperature of 9.8 V and 73 °C, respectively. In all three tests, the voltage collapse resulted in flames.
During the burning period, prior to water mist application, the exposed thermocouples TC1, TC5, and TC6 exhibited substantial increases, reaching peaks at 180 °C, 215 °C, and 60 °C, respectively, in WM1. This temperature increase was attributed to direct flame contact, as observed in previous studies [9, 11, 13]. The voltage data revealed that thermal runaway propagation was limited, with only module 2 cell pair voltages collapsing.
The application of water mist immediately suppressed the fire, causing TC5 and TC6 to cool below 100 °C. TC1 initially increased to 270 °C before rapidly cooling to 105 °C. Towards the end of the first water mist application, TC2 began to rise rapidly, indicating significant heat transfer to module 3.
The water mist application was discontinued owing to excessive vapour cloud production from the modules. Immediately after the application stopped, TC2 rapidly increased to 500 °C within 300 s. At this temperature, the voltage data showed that the voltages in module 3 collapsed, and thermal runaway occurred, coinciding with the re-ignition of the fire.
The water mist was reapplied to the reignited fire and was immediately suppressed. However, no cooling effect was observed for TC2 when TC1 was cooled to below 100 °C. The application was stopped again owing to vapour cloud production, resulting in a rapid rise in TC3, which eventually led to the collapse of the voltages in module 4 and the re-ignition of the fire. The fire was then allowed to burn out, causing the remaining modules 1 and 5 to enter thermal runaway at 2355 s and 1625 s, respectively.
Similar patterns were observed in WM2 and WM3, with water mist effectively suppressing flames but failing to prevent thermal runaway propagation. In WM2, three full extinguishers (27 L) were used, while in WM3, three and a half extinguishers (22 L) were deployed.
The most striking finding was the delay in thermal propagation when water mist was applied. Without suppression, the battery modules experienced a rapid voltage collapse in only 697 s. In contrast, water mist tests extended this critical time to between 1800 s and 2425 s, potentially providing crucial extra time for emergency response and evacuation. The mist effectively knocked down the fire, which significantly reduced the risk of flame propagation to any adjacent modules and buildings.
However, water mist application alone is insufficient to prevent the propagation of thermal runaway reactions once they are initiated [1, 15, 16]. The water mist deployment created a substantial vapour cloud, which forced fire officers to withdraw owing to deflagration risks, highlighting the nuanced challenges of battery fire suppression.
Figure 6 illustrates voltage and temperature responses for exemplary WM test (WA1). Other testes are presented on Figures S9 −12 in Supplementary Materials.
Fig. 6
Water Mists test (WM2) – A temperature and B voltage readings
3.3 Encapsulator Agent (EA)
The encapsulator agent is an aqueous mixture with a proprietary composition that encapsulates fuels, rendering them non-flammable and non-ignitable, and interrupts the free radical chain reaction.
As in the case of previous tests (control and water mist), the overcharge period was similar, indicating good uniformity across the module range. The first test, EA1, suffered several drawbacks during execution and was therefore deemed not fully representative from a replication point of view. However, owing to some observations, we believe it is worth highlighting some key findings from this test.
In EA2, thermocouples TC1, TC5, TC6, and TC7 exhibited notable responses to ignition, registering temperatures of 158 °C, 383 °C, 281 °C, and 387 °C, respectively. The EA immediately suppressed the flames and cooled some elevated thermocouples to 100 °C. Interestingly, TC1 continued to increase to 800 °C, seemingly unaffected by the cooling properties of the agent. The voltage data revealed that thermal runaway propagated to module 1 during the application, resulting in the excessive production of vapour cloud. Towards the end of the application, TC2 also began to rise and stabilize at 100 °C. Due to the excessive vapour cloud generation, which was accelerated by the thermal runaway of module 1, the application of the agent was halted after one and a half extinguishers (ca. 13 L volume).
After the application was stopped, TC2 started to rise rapidly and then stabilized at 450 °C, indicating significant heat transfer to module 3. The voltage eventually collapsed at 636 s. No re-ignition occurred, and the module produced large amounts of vapour cloud. The heat propagation continued to module 4, with TC3 rising to 400 °C and then to 890 °C, and the voltage collapsed at 969 s. Finally, module 5 entered thermal runaway at 1275 s, leading to re-ignition of the fire at 1457 s.
As in the previous tests, EA3 immediately knocked down the fire, causing the exposed thermocouples to decrease in temperature to 100 °C. TC1 was also affected by the application, decreasing to 100 °C. During the application, TC2 rose to 345 °C but was briefly cooled to 220 °C by the end of the suppression. The knockdown of the flames resulted in excessive vapour cloud production, leading to the encapsulator agent application being stopped after approximately two and a half extinguishers were used (ca. 22 L in total).
Although EA1 was deemed unrepresentative of the encapsulator agent’s capabilities and was thus excluded from performance assessments, it may reflect real-world scenarios where not everything functions as intended. In EA2 and EA3, all five modules experienced thermal runaway, with final times of 1305 and 2427 s, respectively. This suggests that, likewise for the water mist suppressant, the encapsulator agent delayed but did not prevent thermal runaway propagation. A consistent pattern across the three EA tests was prolonged re-ignition, where all propagated thermal runaway events resulted in vapour cloud production rather than flames until the final module voltage collapsed and entered thermal runaway.
Figure 7 illustrates voltage and temperature responses for exemplary EA test (EA2). Other testes are presented on Figures S13–16 in Supplementary Materials.
Fig. 7
Encapsulator Agent test (EA2) – A temperature and B voltage readings
3.4 Carbonate Agent (CA)
The carbonate agent is a water-based suppressant containing an ammonium bicarbonate additive, making up 5% of the total mixture. Carbonates, commonly used in dry powder extinguishers, function by starving the fire of oxygen through the production of CO2, resulting from the endothermic decomposition of carbonates occurring during the suppression process.
During the initial burning period of the first test, CA1, TC1, and the exposed thermocouples TC5, TC6, and TC7 all showed a significant response to ignition, with them rising to peaks at 295 °C, 349 °C, 139 °C, and 250 °C, respectively.
The fire was quickly extinguished by the application of a carbonate agent, resulting in the exposed thermocouples dropping to approximately 150 °C. TC1 continued to increase, reaching a maximum of 785 °C during the application with no apparent cooling. Interestingly, TC6 (back of the metal casing) continued to increase even with the application, indicating that the heat was transferred from the overcharged module. Thermal runaway propagated downwards to module 1 as the voltage of the upper half of module 1 collapsed at 165 s. After 150 s, TC2 increased to 100 °C, indicating significant heat transfer to module 3. The application of the agent was stopped (after two full extinguishers, 18 L in total) because of the excess vapour cloud production from the modules.
Similar patterns were observed in CA2 and CA3, with the carbonate agent effectively suppressing flames but failing to prevent thermal runaway propagation. Conducted tests revealed that the carbonate agent was effective in suppressing the fire but was unable to prevent the subsequent propagation of thermal runaway. The CA1 test had the shortest time to final ignition, with a 37-second increase compared to the control test. The CA2 and CA3 tests demonstrated longer final burn values of 909 and 1247 s, respectively, indicating that the carbonate agent prolonged the time to final ignition, although this increase was considered insignificant when compared to the performance of the water mist and encapsulator agents.
Across all three tests, re-ignition occurred only after all modules had entered thermal runaway, with extended periods of flame-free thermal runaway propagation following the initial fire elimination. Interestingly, CA1 was the sole instance where re-ignition occured immediately after thermal runaway propagation, allowing the firefighter to re-enter the container and apply the agent a second time, as the present vapour cloud was consumed by the flame, hence lower risk of explosion and toxic contamination.
Notably, large volumes of vapour cloud were produced following the fire suppression. Thermal Runaway propagation continued without flames, suggesting significant convective heat transfer. Furthermore, the thermocouple temperatures were considerably higher during the period between thermal runaway propagation events compared to the previous tests, indicating a lack of cooling capacity from the extinguisher.
Figure 8 illustrates voltage and temperature responses for exemplary CA test (CA2). Other testes are presented on Figures S17–20 in Supplementary Materials.
Fig. 8
Carbonate Agent test (CA2) – A temperature and B voltage readings
3.5 Mixed Agent (MA)
The mixed agent is a water-based extinguisher with a proprietary mix of compounds (12.5%), including natural-born compounds, food-grade wetting agents, and anionic detergents, contained in a standard 9 L compressed canister. The addition of detergents suggests that it may operate using a similar method to the encapsulator agent by aiding the penetration of the system’s geometry and fuel.
The overcharge behaviour seen in most previous tests occurred in MA1 and MA2, with the voltage reaching a maximum of 5.5 V before dropping and collapsing to 0 V, triggering the ignition. However, in MA3, an alternative behaviour, similar to WM3, occurred when the voltage spikes by 9 V before collapsing. Despite the alternate overcharge behaviour of MA3, it was MA2, which gave the unexpected ignition behaviour with the module releasing a large amount of vapour cloud for 270 s after the voltage collapse before it ignited. Tests MA1 and MA3 behaved as expected, with the modules igniting immediately with the collapse of the cell voltage. The temperatures before their voltage collapse were 21, 40, and 76 °C, respectively.
In contrast to the tests carried out on water mist and other additive suppressants, MA was less effective at extinguishing the flames; the times taken to do so in the three tests were 92 s, 100 s and 76s. This should be compared with a few seconds or less for other water-based suppressants.
The experimental results demonstrate that the Mixed Agent, like the other suppressants tested, is unable to prevent thermal propagation between cells within a module. However, the data from MA1 suggest that the agent may delay thermal propagation between modules under certain conditions. This would require the suppressant to have direct access to the modules, which is not the case for commercial domestic battery energy storage systems that are fully enclosed. Additionally, the continuous suppression of the fire and the resulting significant vapour cloud production necessitated the withdrawal of the firefighter due to the explosion hazard.
In the MA3 test, thermal runaway spread to all five modules, indicating the agent’s ineffectiveness in preventing this phenomenon. Nevertheless, the agent delayed the onset of thermal runaway propagation, as the final thermal runaway event in MA3 occurred at 1177 s, representing a 477-second improvement compared to the control test without any suppression. The application of the agent also caused the thermocouples to exhibit a significant cooling effect, similar to that observed with water mist and the encapsulator agent.
Figure 9 illustrates voltage and temperature responses for exemplary MA test (MA3). Other testes are presented on Figures S21 −24 in Supplementary Materials.
Fig. 9
Mixed Agent test (MA3) – A temperature and B voltage readings
3.6 Liquid Nitrogen (LN)
Liquid nitrogen was used to determine whether thermal runaway could be prevented by enhanced cooling. It must be noted that liquid nitrogen is released from the tank as a gas, but still approximately at −195 °C, which in theory should be able to rapidly cool the modules.
The liquid nitrogen rapidly cooled the surrounding air below its dew point, forming large amounts of water vapour [17, 18] which obscured the cameras, rendering the observation of events challenging; hence, to determine when the liquid nitrogen was deployed and when it stopped, the GoPro affixed to the fire officer’s helmet was employed in the first two tests only as it failed in test 3.
In the LN1 test, a vapour cloud was vented from module 2 until the module ignited, which coincided with the voltage collapse of the overcharged cell pair. During the burning period, TC1 and the exposed thermocouples TC5, TC6, and TC7 showed a significant response to ignition, with them rising to peaks of 451 °C, 201 °C, 218 °C, and 300 °C, respectively. The voltage data showed that no thermal runaway propagation occurred in the burning period when only module 2 voltages collapsed.
Liquid nitrogen was applied to the fire, and it was immediately knocked down. This resulted in TC5 and TC7 dropping below 0 °C. TC1 continued to increase, reaching 842 °C during application. The constant application of LN created a mixture of a vapour cloud and condensed water vapour in the air. This led to almost zero visibility in the container. However, LN can still be applied as it was remotely activated.
The liquid nitrogen stream was directed directly at the overcharge module and into the casing, and had little or no positive effect: the gas stream initially seemed to fan the flames rather than extinguish them; in the LN3 test, the final burnt time was 621 s, 80 s shorter than that in the control test. In LN1 and LN2, liquid nitrogen appeared effective at delaying thermal runaway propagation final times of 1272 and 1017 s, respectively. Under the conditions of the tests described in this report, liquid nitrogen can extinguish a fire from lithium-ion batteries; however, there is no convincing evidence that it can prevent or delay thermal propagation and hence stop the production of the toxic and flammable vapour cloud.
The low temperature of liquid nitrogen was apparent in all three tests, with the exposed thermocouples dropping below 0 °C; however, it did not appear to have a significant cooling effect on the modules, as the thermocouple temperatures rapidly increased once the application was stopped. A possible reason for the lack of cooling is that liquid nitrogen was only applied to one specific region of the system, as it was delivered through a fixed lance. Additionally, when LN is applied to the TR casing, it not only absorbs heat from the pack but also from the environment. Consequently, the actual heat absorbed by LN from the TR casing is smaller than the theoretical heat absorbed during gasification [17].
Figure 10 illustrates voltage and temperature responses for exemplary LN test (LN3). Other testes are presented on Figures S25-28… in Supplementary Materials.
Fig. 10
Liquid Nitrogen test (LN3) – A temperature and B voltage readings
3.7 Statistical Analysis
Each analysis supported the superior performance of water mist and encapsulator agents while revealing complex interactions between variables and highlighting the multifaceted nature of suppression effectiveness. The analyses consistently demonstrated significant method-dependent differences in both primary outcomes and temporal characteristics. Relevant Figures and Tables are presented in the Supplementary Information.
3.7.1 ANOVA and Post-hoc Analysis
The ANOVA and post-hoc analysis transform qualitative observations into quantitative evidence, controlling for Type I error across multiple comparisons to ensure observed differences represent genuine performance variations. This framework establishes standardized methodology for evaluating new suppression agents, enabling systematic comparison against established benchmarks and building cumulative knowledge for the field.
One-way ANOVA revealed significant differences between suppression methods (F(5,12) = 42.31, p < 0.001, η² = 0.82). Post-hoc Tukey’s HSD tests showed that water mist and encapsulator agent significantly outperformed other methods in both maximum temperature reduction and propagation delay (p < 0.001). The control group exhibited the highest mean heating rate (3.90 ± 0.14 °C/s), while water mist and encapsulator agent showed notably lower rates (2.65 ± 0.78 °C/s and 2.65 ± 0.49 °C/s respectively, p < 0.01).
3.7.2 Variance Components Analysis
This analysis quantifies the relative importance of experimental factors, revealing that method selection is the dominant factor while temporal factors and method-time interactions remain significant. The approach provides a roadmap for research prioritization, indicating that optimizing application protocols could yield substantial performance gains with existing technologies.
Method differences accounted for the largest portion of total variance (42.5%), followed by temporal variation (18.2%), method-time interaction (13.5%), and residual variance (25.8%). The between-group variance ratio was notably high for suppression duration (29.45) but lower for effectiveness measures (1.61). Environmental and operational factors contributed to the remaining variability, with temperature control showing intermediate variance ratios (15.32).
3.7.3 Correlation and Regression Analysis
These analyses establish mechanistic relationships underlying suppression effectiveness, with high R² values indicating cooling rate serves as a reliable predictor of performance. This enables researchers to focus on optimizing cooling rates rather than extensive trial-and-error testing, while suggesting hybrid approaches could achieve synergistic effects.
Strong negative correlations were observed between cooling rate and maximum temperature (−0.89), and positive correlations between propagation time and effectiveness (0.92). Regression analysis revealed significant relationships between cooling rate and maximum temperature (R² = 0.89, p < 0.001) and between propagation time and effectiveness (R² = 0.92, p < 0.001). Heating rate showed a moderate negative correlation with effectiveness (−0.71, p < 0.005).
3.7.4 Time-to-Event Analysis
This analysis accounts for censored thermal runaway data, providing accurate risk estimates and standardized metrics for comparing effectiveness across studies. The hazard ratios provide evidence-based guidance for emergency response planning and support development of adaptive protocols tailored to specific scenarios.
Survival analysis (defined as reduction in thermal propagation) showed significantly different hazard rates between methods (log-rank test, p < 0.001). Water mist demonstrated a 71% reduction in hazard (HR = 0.29, 95% CI: 0.19–0.44), and encapsulator showed a 69% reduction (HR = 0.31, 95% CI: 0.20–0.47). Median survival times ranged from 697 s (control) to 2427 s (encapsulator), with water mist achieving 2353s.
3.7.5 Frequency Domain Analysis
This analysis transforms time-domain data into frequency-space representations revealing system dynamics, with higher damping ratios providing quantitative evidence of stabilizing effects. The approach enables identification of optimal application timing patterns and opens possibilities for real-time monitoring systems using characteristic frequency signatures.
Power spectral density analysis revealed distinct frequency signatures for each suppression method, with dominant frequencies at 0.001 Hz. Water mist and encapsulator showed the lowest power across all frequencies (32.1 and 30.5 dB, respectively), while control exhibited the highest (45.2 dB). Bandwidth analysis showed increased damping ratios for water-based methods (0.18–0.19) compared to the control (0.12).
3.7.6 Uncertainty Quantification
This framework enables distinction between genuine performance differences and measurement noise, while identifying critical variables for experimental control. The approach guides experimental protocol optimization and provides statistical foundation for establishing performance standards with appropriate confidence levels.
Temperature measurement uncertainties ranged from 3.2% (encapsulator) to 7.2% (liquid nitrogen). Propagation time uncertainties increased with longer times, from 5.7% (control) to 15.3% (liquid nitrogen). Sensitivity analysis identified battery state of charge (SOC) as the most influential parameter (sensitivity coefficients: temperature 0.88, propagation 0.91), followed by initial temperature (0.82, 0.75).
3.7.7 Multivariate Analysis
This analysis reveals multi-dimensional suppression effectiveness patterns, with distinct clustering suggesting different methods operate through fundamentally different mechanisms. The approach enables development of comprehensive performance indices integrating multiple criteria and understanding trade-offs between characteristics like vapour production and cooling effectiveness.
Principal Component Analysis extracted three significant components explaining 86.5% of total variance (PC1: 45.2%, PC2: 25.8%, PC3: 15.5%). PC1 showed strong loadings for effectiveness (0.85) and propagation time (0.88), while PC2 captured variance in cooling rates (−0.42). Cluster analysis revealed a clear separation between water-based methods and others, with parallel coordinates confirming distinct performance patterns across multiple variables.
4 Discussion
The systematic evaluation revealed critical insights about suppression mechanisms and performance trade-offs. A crucial finding relates to the correlation between vapour cloud production and cooling effectiveness (r = 0.87, p < 0.001). While higher vapour production correlates with improved cooling, it also increases explosion risks, particularly in unventilated settings. This creates a fundamental trade-off between suppression effectiveness and safety that must be carefully managed in real-world applications.
Temperature profile analysis identified four distinct phases: initial heating, suppression cooling, vapour evolution, and thermal propagation. The reduction in heating rate strongly correlates with increased time to final thermal runaway propagation (r = 0.89, p < 0.001), suggesting a direct mechanistic link between cooling capability and propagation prevention. These findings align with Liu et al. [19], who reported that water mist could achieve cooling rates exceeding 100 K/s, with surface temperatures rapidly decreasing to 373.15 K within seconds of application.
Our findings regarding suppression effectiveness align with Willstrand et al. [20], who demonstrated that internal activation of water-based fire suppression systems inside battery packs has significant potential for lasting cooling effects and increased chances of mitigating thermal runaway propagation. This is particularly evident in our temperature profile analysis, which identified four distinct phases: initial heating, suppression cooling, vapour evolution, and thermal propagation.
Table 3; Fig. 11 visualises a comparative analysis of suppression method effectiveness through temperature evolution and propagation patterns, and reveals distinct performance clusters.
Fig. 11
Effectiveness of tested suppression methods
Table 3
Suppression agent effectiveness summary
Rank | Suppression Method | Time to Final Burn (s) | Improvement vs. Control (%) | Mean Propagation Delay (s) | Cooling Rate (°C/s) | Max Temperature Reduction | |
|---|---|---|---|---|---|---|---|
1 | Control | ~ 650 | 0% (baseline) | 756 ± 89 | 1.87 | N/A | |
2 | Water Mist | ~ 2,100 | ~ 180% | 1,456 ± 204 | 2.65 | Highest | |
3 | Encapsulator Agent | ~ 2,000 | ~ 170% | 1,398 ± 186 | 2.65 | High | |
4 | Mixed Agent | ~ 950 | ~ 45% | 1,089 ± 298 | 2.12 | Moderate | |
5 | Carbonate Agent | ~ 900 | ~ 45% | 1,034 ± 187 | 1.98 | Moderate | |
6 | Liquid N2 | ~ 900 | ~ 30% | 1,067 ± 445 | 2.43 | Moderate | |
Water mist and encapsulator agents demonstrate enhanced performance with the longest propagation delay times (2353s and 2427 s, respectively) and most effective temperature control, maintaining peak temperatures below 800 °C. Maximum temperatures varied significantly across methods (Control: 890 °C, Water Mist: 800 °C, Encapsulator Agent: 780 °C, Others: 840–870 °C, p < 0.01), supporting another findings [21] about the critical role of cooling capacity in suppression effectiveness.
The carbonate agent, mixed agent, and liquid nitrogen form an intermediate cluster with propagation times ranging from 909 s to 1279 s, showing moderate effectiveness but higher variability in performance. Temperature profile analysis shows these methods achieve initial cooling but struggle with sustained temperature control, evidenced by steeper temperature gradients during the propagation phase.
Notably, different studies [22] demonstrated that even high concentrations (15.2 vol%) of clean agents like Novec 1230 were required for effective suppression, achieving prevention of propagation in 67% of cases. This contrasts with our water-based methods’ results, where we observed more consistent suppression effects. The water mist’s superior performance aligns with [19] finding that only 1.95 × 10−4 kg/Wh−1 of water was needed for effective prevention, making it the most efficient among the compared methods.
A crucial finding from our research, supported by Willstrand et al. [20], relates to the correlation between vapour cloud production and cooling effectiveness (r = 0.87, p < 0.001). While higher vapour production correlates with improved cooling, it also increases explosion risks, particularly in unventilated settings. This creates a fundamental trade-off between suppression effectiveness and safety that must be carefully managed in real-world applications. Although the full explosion risk was not studied during this work (e.g. lack of working gas sensors), but will be part of the future tests.
Among the tested methods, water mist demonstrated the highest sustained cooling rate (2.53 ± 0.25 °C/s), followed by liquid nitrogen (2.43 ± 0.93 °C/s) and the encapsulator agent (2.40 ± 0.20 °C/s). However, liquid nitrogen showed concerning variability with the highest coefficient of variation (38.18%), indicating potential challenges in practical application.
The variance components analysis revealed significant between-group variance in suppression duration (ratio: 29.45) but a lower variance ratio for effectiveness (1.61). This aligns with other observations [21] about the importance of both agent selection and application timing. Our findings that water mist could control thermal runaway, to some degree, even with severe propagation, are supported by the demonstration of water mist effectiveness even when multiple cells enter thermal runaway [19].
The relationship between suppression duration and effectiveness revealed complex, method-specific patterns that challenge current knowledge in the subject. Liquid nitrogen demonstrated a strong positive correlation (r = 1.000) between application duration and effectiveness, while water mist showed a weak negative correlation (r = −0.188). This unexpected finding suggests that longer application periods don’t necessarily translate to better outcomes for all methods, particularly for water-based suppressants, as there is also the actual mode of the suppression mechanism important.
The comprehensive statistical analysis reveals several important patterns in suppression effectiveness. ANOVA results (F(5,12) = 42.31, p < 0.001, η² = 0.82) confirmed significant differences between methods, with water mist and encapsulator agent showing superior performance. Variance Components Analysis identified method differences as the primary source of variation (42.5%), indicating that the choice of suppression method is the dominant factor in determining outcomes.
The strong negative correlation between cooling rate and maximum temperature (r = −0.89, p < 0.001), and positive correlation between propagation time and effectiveness (r = 0.92, p < 0.001) provide quantitative support for the mechanistic relationship between cooling capability and suppression success. Time-to-event analysis revealed significant reductions in hazard ratio for water mist (HR = 0.29, 95% CI: 0.19–0.44) and encapsulator (HR = 0.31, 95% CI: 0.20–0.47) compared to control, translating to a greater than 70% reduction in thermal runaway risk.
Principal Component Analysis identified three significant components explaining 86.5% of total variance, with PC1 (45.2% variance) showing strong loadings for effectiveness (0.85) and propagation time (0.88). This multivariate analysis revealed distinct clustering of water-based methods separate from other approaches, suggesting fundamental differences in their suppression mechanisms.
Uncertainty quantification revealed increasing measurement uncertainty with longer suppression times, ranging from 5.7% (control) to 15.3% (liquid nitrogen). Sensitivity analysis identified battery state of charge as the most influential parameter (sensitivity coefficients: temperature 0.88, propagation 0.91), highlighting the importance of initial conditions in determining suppression outcomes.
The propagation pattern analysis revealed method-specific characteristics in the timing and progression of failure across modules. Water mist and encapsulator agent showed similar patterns of delayed propagation (mean delay times of 1951 ± 395 s and 1866 ± 1100 s, respectively, p < 0.001), outperforming other methods while exhibiting higher variability. This variability, while concerning, must be balanced against the overall effectiveness of these methods, particularly given Liu et al.‘s [19] identification of critical temperature thresholds and cooling factors for predicting control effectiveness.
Our testing configuration, with open module access, likely presents a best-case scenario compared to typical enclosed commercial installations, as demonstrated by Cui et al. [23] in their full-scale experimental study. The implications of this limitation are particularly significant when considering real-world applications where access to battery modules is typically restricted, on-site staff are less equipped than first responders (e.g. lack of PPE), and extinguishers are of a small volume to make a significant impact on the fire/thermal runaway.
Recent advances in fire-suppressing agents, as reviewed by Majeed et al. [7], have focused on composite materials with enhanced properties. While our testing didn’t include these newer materials, our statistical framework provides a robust methodology for evaluating such innovations, particularly in quantifying their consistency of performance and effectiveness across different phases of thermal runaway.
Yuan et al. [8] highlighted the importance of considering multiple factors in fire-extinguishing agent evaluation, including insulation, toxicity, fire suppression speed, reactivity, degradation capacity, and corrosion. Our analysis adds quantitative weight to these considerations, particularly regarding the trade-off between suppression effectiveness and vapour cloud production.
These findings have significant implications for the design and implementation of battery fire suppression systems. The data suggests that optimal suppression strategies should focus on:
1.
Maintaining consistent cooling rates rather than maximizing initial cooling.
2.
Balancing vapour production management with cooling effectiveness.
3.
Ensuring adequate ventilation strategies, particularly for enclosed spaces.
4.
Implementing early detection, rapid response capabilities and possibly remote activation.
Zhang et al. [3] emphasize the importance of next-generation fire-safe batteries. Our findings suggest that while current suppression methods can delay thermal runaway propagation (up to 179% improvement over control), the fundamental challenge of prevention remains unsolved. The statistical evidence strongly suggests that while current suppression methods can delay thermal runaway propagation, none can prevent it under all tested conditions – thus, proper risk assessment of the use of such systems is mandatory.
Time-series analysis of temperature profiles revealed method-specific signatures in the progression of thermal events. Analysis of temperature fluctuations during suppression showed distinct frequency components associated with different suppression methods (p < 0.001), suggesting potential for real-time monitoring and adaptive suppression strategies.
This understanding, combined with insights from previous studies [19, 20, 22, 24], indicates that future development should focus on integrated approaches combining efficient cooling methods with sophisticated monitoring and ventilation systems. The lack of complete suppression capabilities in any tested method suggests that the optimal approach may be to thoroughly consider both fire and vapour cloud explosion hazards before deploying BESS systems.
5 Conclusion
The comprehensive statistical analysis of various fire suppression methods for lithium-ion battery fires reveals a complex landscape of effectiveness, limitations, and safety considerations. While current techniques can significantly delay thermal runaway propagation, none can completely prevent it under the tested conditions. Water mist and encapsulator agent emerge as the most effective methods, demonstrating statistically significant improvements in propagation delay times (179% and 167% improvement over control, respectively, p < 0.001). Notably, the encapsulator agent did not achieve its claimed 6–10 times improvement in heat absorption compared to standard water mist, suggesting that water mist may be a more cost-effective and readily available option for practical applications.
The systematic evaluation of temperature profiles, propagation patterns, and suppression effectiveness reveals critical insights about temperature control dynamics and method performance. Initial cooling rates show method-specific patterns with high variability, while sustained cooling effectiveness varies significantly between methods. Maximum temperature reduction correlates strongly with propagation delay, and thermal gradient management proves crucial for propagation prevention. Water-based methods demonstrate superior overall performance despite high variability, while liquid nitrogen shows excellent initial cooling but poor sustainability. Carbonate and mixed agents maintain moderate, consistent performance, though all methods ultimately fail to prevent thermal runaway propagation.
Operational considerations emerged as crucial factors in suppression effectiveness. Application timing significantly impacts effectiveness (p < 0.01), with multiple shorter applications often outperforming single long applications. Vapour cloud management presents a critical safety concern, while system accessibility significantly affects suppression effectiveness. A significant finding was the production of large quantities of flammable vapour clouds during suppression efforts, potentially shifting the risk profile from fire to vapour cloud explosion hazards. This risk is particularly acute in unventilated settings, such as residential BESS installations, where vapour clouds could accumulate to explosive concentrations.
The experimental evidence suggests that the optimal balance lies not just in achieving 2.4–2.5 °C/s cooling rates, but in maintaining consistent cooling rates within this range. Water mist and encapsulator agent achieved similar cooling rates (2.65 °C/s) with much better consistency (CV = 18.7% and 19.2% respectively) compared to liquid nitrogen’s high variability, which explains their superior performance despite slightly higher cooling rates. The data indicates that sustained, consistent cooling in the 2.4–2.7 °C/s range is more effective than achieving the theoretical optimal rate inconsistently.
Yet these values are system specific and should be treated with caution when applied to any other system or experiments.
Statistical analysis of performance variability reveals that the coefficient of variation ranges from 18.68% to 38.18% across methods, with performance consistency correlating inversely with maximum effectiveness. Environmental conditions significantly impact suppression effectiveness, while system scale affects suppression efficiency non-linearly. These findings necessitate comprehensive technological development, including intelligent suppression systems with real-time monitoring, adaptive application strategies, and hybrid suppression approaches combining multiple methods.
Testing and standardization emerge as critical areas for improvement, requiring implementation of standardized module-level testing protocols, development of comprehensive performance metrics, and establishment of minimum performance standards for different applications. The complexity of BESS safety demands enhanced cooperation between battery manufacturers and fire safety experts, and the creation of advanced detection and early warning systems, all integrated within a framework of regulatory oversight.
While this study demonstrates that the water mist and encapsulator (water-based) agent had the best performance in the experiments, it is important to contextualise their application. Water is known to react with the electrolyte in the batteries to produce HF [25]. The introduction of such a harmful chemical presents a potential new hazard which must be considered in firefighting scenarios. In outdoor applications, the dispersion of HF to the environment significantly reduces the risks of HF contamination to nearby people, however in indoor applications, there is potential for the concentration to reach harmful levels. As such, the optimal choice of extinguisher agent is scenario dependent: water-based agents may provide effective cooling and fire knock down capability in outdoor applications, whereas alternative solutions could be investigated to approach enclosed environments where HF contamination must be avoided.
Future development should focus on reducing variability in suppression performance while maintaining cooling capabilities, improving vapour management strategies and ventilation systems, and developing more sophisticated control and monitoring systems. Long-term solutions require continued development of inherently safer battery technologies, implementation of more effective suppression strategies based on empirical data, and integration of advanced monitoring and control systems, all supported by comprehensive safety standards and protocols. The findings emphasize the need for mandatory ventilation requirements, stringent installation standards.
Enhanced collaboration between stakeholders in the battery and fire safety industries stands as a crucial requirement for advancing safety measures. This collaboration should focus on developing new extinguishing agents designed to address both fire suppression and vapour control, as well as creating advanced detection systems and proper automatic extinguisher systems for potential thermal runaway events. The complexity of BESS safety underscores the need for ongoing research, innovation, and regulatory oversight to ensure safe integration of these systems across various applications, from residential settings to large-scale energy storage facilities. This comprehensive approach, combining technological advancement with rigorous safety protocols, represents the most promising path forward for addressing the challenges revealed in this study.
Acknowledgements
This work was supported by the Faraday Institution as part of its “SafeBatt – Science of Battery Safety” project (FIRG061) and the Fire Industry Association. The authors are grateful to the Greater Manchester Fire and Rescue Service Training Centre in Bury, UK, for the opportunity to use the testing facility and the help of its firefighting instructors to execute the tests.
Declarations
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical Approval
This study was conducted using experiments and calculations and did not require ethics approval.
Informed Consent
This study did not involve human participants and, therefore, did not require informed consent.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.