Evaluating Fire Severity in Electric Vehicles and Internal Combustion Engine Vehicles: A Statistical Approach to Heat Release Rates

A total of 16 single EV fire experiments and 17 single ICE vehicle fire experiments have been collated, with details sourced from corresponding reference materials (Table 1). The gathered information for both EVs and ICE vehicles is systematically presented in a comprehensive table format. For detailed specifics on EVs, refer to Table 2, and for ICE vehicles, see Table 3.

These tables includes a unique identifier for each vehicle, specifying the types of EV or ICE vehicle, the battery capacity in the EVs during the test, the state of charge of the battery, and the cell type of the battery. Additionally, it provides details about the model and year of manufacture for all vehicles. The table also covers information regarding the fire condition of the vehicles, such as the method used to initiate the fire during experiments i.e., the locations and ignition source of the fires, and the conditions of the vehicles (for example, the state of window openings). Finally, the table describes how the HRRs were measured, a process referred to as ‘Heat release rate evaluation method’. This extensive compilation of data aims to provide a clear and detailed overview of each experiment’s specifics, encompassing both EVs and ICE vehicles.

Table 1 comprehensively explains each column, detailing critical specifications and conditions relevant to the study of EV and ICE vehicle fires. This includes categories such as EV types—Battery Electric Vehicle (BEV), Plug-in Hybrid Electric Vehicle (PHEV), and Fuel Cell Electric Vehicle (FCEV)—and key parameters like Battery Capacity, State-of-Charge, and Battery Cell Type. For ICE vehicles, specifics such as Fuel Type and Fuel Balance are outlined. Additionally, the table provides insights into the ‘Condition’ of vehicles prior to ignition, ‘Ignition source’, and ‘Ignition location’. Furthermore, the ‘Reference’ column contains information about the primary source of the experiment. The column displays the publication date of the resource.

The ‘Heat release rate evaluation method’ column clarifies how the heat release rate curve was derived from each experiment. However, it’s important to note that not all sources provided explicit details on the exact techniques used, so the information in this column is based on the interpretations made from the original publications. Additionally, it is important to acknowledge that the different techniques used to measure heat release rates, such as mass loss rate, convective calorimetry, and species-based calorimetry, may lead to variations in the measurements of heat release rates [14]. This variability can influence the analysis of fire severity. The term ‘Mass loss’ in this context means that the heat release rate was determined through measuring the mass loss of the vehicle during combustion. ‘Enthalpy of mass flow rate’ indicates that the energy released during the combustion process was calculated by assessing the enthalpy, or heat content, of the mass flow rate of gases upstream from the fire. This method involves measuring the changes in thermal energy of the gases as they move away from the fire source, providing a comprehensive view of the total energy output of the combustion. ‘Species-based calorimetry’ refers to methods where the heat release rate was calculated using the depletion of O₂, or the generation of CO₂ and/or CO. All of the methods mentioned above, while not without limitations, remain practical and effective for estimating the HRR of complex fire scenarios involving lithium-ion batteries. This method may overestimate HRR specifically for batteries due to biases in capturing the unique combustion characteristics of battery components [15, 16]. Nonetheless, it provides a comprehensive measure of heat release rate, total heat output and oxygen consumption across diverse material types, which is essential for assessing fire behaviour in applications like fires in car parks. In EV fire assessments, which involve a combination of batteries, plastics, metals, and other materials, oxygen-based calorimetry captures the integrated fire dynamics, including how the battery interacts with other components. Although it may not be ideal for isolated battery tests, its consistency and interpretability make it the most viable approach for probabilistic assessment applications, allowing reliable comparisons across complex fire vehicle scenarios. Next, key observations from each vehicle fire experiment, covering both EVs and ICE vehicles, are discussed. These observations were drawn from the relevant references, and where available, time markers for significant events during the experiments were also noted. However, not all references provided specific timing, so for some events, the time markers were estimated within a range based on available information. Table 4 presents the key observations and important time markers for all EV experiments, while Table 5 summarizes these details for the ICE vehicle experiments.

The discussion of the observations from the EV experiments focuses primarily on the involvement of batteries, as the powertrain is the key difference in combustible materials between EVs and ICE vehicles. Analysing the time markers of significant incidents revealed that, in 7 out of the 16 EV experiments, battery involvement occurred before the peak HRR was reached, while in 8 experiments, the batteries became involved only after the peak HRR. In one experiment, limited information was available regarding the timing of battery involvement.

The profiles for heat release rates derived from the experimental data are showcased in Fig. 1 and Fig. 2. These figures juxtapose the heat release rate curves from various experiments within each defined category on a unified axis, facilitating a direct comparison. This approach emphasises the validity of the classification system and highlights the variability present in the data. The determination of the peak heat release rate and the time to reach peak heat release rate are extracted from the profile curves if not mentioned clearly from the source. In this study, total heat released values are directly taken from explicit mentions in the source materials, avoiding calculations due to instances where experiments ended or data recording stopped prematurely, impacting the accuracy of Total Heat Released (THR) estimations.

Bild vergrößern

Bild vergrößern

Time averaging of the heat release rate curves was not applied in this study. To ensure consistency in timing, the incipient stage of the fire was excluded, and time was standardised to begin from the onset of significant fire growth. For cases where suppression efforts occurred, only one experiment involved active suppression, and this suppression took place after the fire had reached its peak HRR. As this study focuses on the growth phase, this experiment was retained in the dataset to capture the relevant fire dynamics during the critical early stages.

This work is expected to provide invaluable insights for fire engineers, particularly in designing safe car parks, by offering a detailed comparison of design fires in terms of HRR profiles in EV and ICE vehicle fires. As previously studied and assessed for its suitability to characterize design fire for a single passenger vehicle [2], these design fires are characterized by a combination of growth and decay curves, where the growth phase is described using a t-squared function;

In Eq. 1, \(\dot{Q}\left(t\right)\) represents the heat release rate, expressed as a function of time, where t, denotes the duration from ignition until the peak heat release rate is reached, and α, peak symbolizes the coefficient of fire growth. This approach presupposes a certain fire geometry, illustrating a time-squared proportional relationship for fires that ignite and spread radially. Such a model incorporates variables including the rate of flame spread, the rate of mass loss, and the combustion heat value [27]. Incorporating the decay phase of the fire, the analysis extends to utilize an exponential decay method, but now characterizing the reduction in the heat release rate post-peak. This exponential decay follows a specific equation that mirrors the gradual diminishment of the fire’s intensity over time. The exponential decay is such that;

By combining the growth method with this exponential decay approach, a complete design fire curve emerges, applicable to passenger vehicles as demonstrated in prior research [2]. Building upon the established methodologies, the design fires EVs and ICE vehicles can now be constructed, utilizing the available data points from the distributions. Such detailed modelling paves the way for a thorough comparison between the fire dynamics of EVs and ICE vehicles, with a particular emphasis on the growth phase of the fire.

However, to construct the proposed design fire, a method is required to capture the dynamics of the HRR over time from the HRR profiles of EV and ICE vehicles, representing both the fire’s development and decay phases. The first step involves examining the collected HRR profiles to assess their suitability for fitting. Each profile will be fitted individually, beginning with the t-squared fire growth model to represent the initial increase in HRR as the fire develops. This will be followed by applying the exponential decay model to describe the reduction in HRR as the combustion process slows.

This section focuses on transforming the collected experimental data into a meaningful resource that engineers can use to make informed decisions regarding appropriate design fires. The aim is to perform a statistical analysis of the collated data from previous section, with particular attention to identifying trends in the growth phase of vehicle fires. Given that the primary difference between EVs and ICE vehicles lies in their energy storage systems, analysing the growth phase is crucial. For EVs, batteries as energy storage can lead to rapid thermal runaway, potentially presenting unique fire dynamics compared to ICE vehicles.

Before conducting statistical analysis, however, it is necessary to make engineering decisions to determine the suitability of each dataset for inclusion. While this section will offer recommendations on dataset relevance, ultimately, engineers must decide which data are most applicable to their specific use cases. Therefore, the section will present all findings, allowing users to select the information that best aligns with their requirements while benefiting from the insights and trends identified here.

Based on a dataset of 16 EV and 17 ICE experiments gathered from the literature (as detailed in Tables 1, 2, 3, and 4), this subsection outlines the process of filtering experimental data for statistical analysis. The initial step involves reviewing each dataset to identify any prominent issues with the experimental setup that could introduce bias.

During this review, several tests were identified where the HRR of the ignition source reached or exceeded 2 MW, which could skew the analysis. These tests, with fires starting at such a large scale, do not accurately represent early-stage fire growth. Including these data points could bias the statistical analysis, as the accelerated growth phase in these cases would misrepresent typical fire development. Consequently, tests EV3, EV4, EV5, EV6, EV7, ICE2, and ICE3 were excluded from the analysis.

Additionally, test EV14, involving a fuel cell electric vehicle (FCEV) with a small battery, was removed. Since the study focuses on the impact of different energy storage systems within vehicles, the inclusion of a fuel cell EV, which does not rely solely on battery energy storage, would not be relevant to the primary research objective. Thus, EV14 was excluded to maintain consistency in comparing EV and ICE vehicles based on battery involvement.

This subsection presents the approach for analysing key metrics to draw meaningful insights on the fire characteristics of EV and ICE vehicles, supporting the development of representative design fire scenarios. The analysis centres on several important parameters, including Peak Heat Release Rate (PHRR), Time to Peak Heat Release Rate (TPHRR), and THR. These metrics provide valuable information about fire intensity, growth and decay rate, and overall energy output during combustion. Here, “peak” refers to the highest heat release rate observed, representing the maximum fire intensity reached during each experiment.

The primary focus of the data analysis is on calculating the fire growth coefficient, particularly within the growth phase of vehicle fires. This phase is essential for understanding potential hazards, as faster fire growth poses a greater threat due to rapid escalation in fire intensity. By comparing EV and ICE vehicles, the analysis aims to reveal whether EV fires, potentially influenced by battery involvement, exhibit a faster or different growth pattern than ICE vehicle fires.

Table 6 and Table 7 provide details on the PHRR, TPHRR, THR, growth and decay coefficients for both EVs and ICE vehicles. The growth and decay coefficients for both EVs and ICE vehicles were obtained using the fitting method explained in the Sect. 2.2.1. The data analysis only includes datasets that meet the filtering criteria explained in the previous subsection, ensuring a focused and reliable analysis. However, to provide comprehensive access to all collated data, Appendix 1 includes the full dataset, encompassing all vehicle experiments gathered.

Table 8 and Table 9 present comprehensive statistical analyses of the PHRR, TPHRR, THR, and growth coefficient for EV and ICE vehicles, respectively. These tables include the mode, median, mean, standard deviation, as well as the minimum and maximum readings for each variable. The statistical calculations derive from a filtrated dataset comprising 10 experiments for EVs and 15 for ICE vehicles. Notably, the mode is calculated under the assumption that the variables are discrete, an approach that facilitates understanding of the most frequent occurrence within the dataset.

The inclusion of mode, median, mean, and standard deviation is pivotal in statistical analysis, as each measure provides unique insights into the distribution of data. The mean offers a central value, while the median provides a midpoint that divides the dataset, helping to understand data skewness. The mode reveals the most common value, offering insights into the dataset’s concentration. Standard deviation measures the dispersion around the mean, indicating the variability within the dataset. Collectively, these metrics allow for a nuanced interpretation of the data, aiding in the identification of patterns, variability, and central tendencies within the statistical distribution.

It is crucial to acknowledge the inherently mathematical nature of these statistical calculations. The authors recognize that fire dynamics encompass a range of complex behaviours and interactions far beyond what can be directly captured through these statistics alone. However, the influence of probabilistic analysis in this context is significantly enhanced by the volume of available data. Although probabilistic methods might not yield exact predictions, they are instrumental in providing a general sense of the potential fire behaviour. By leveraging a substantial dataset, probabilistic analysis offers valuable insights into the characteristics of fire scenarios in terms of fire severity in the form of heat release rates, even amidst the inherent uncertainties and complexities of fire dynamics.

One important consideration is that, for the ICE vehicle experiments, data on the total energy released were limited, with only four data points available. As a result, the statistical analysis for this parameter may not be sufficiently extensive for broader application, and users of this analysis should take this limitation into account.

A nuanced interpretation of these findings is required, especially when considering the role of battery involvement in EV fires. When the battery does not contribute to the initial HRR, EV and ICE vehicles exhibit comparable fire characteristics due to similarities in materials used, such as interior fabrics and plastics. However, once the battery becomes involved, the fire dynamics of EVs shift considerably, with a tendency for more rapid-fire growth and higher peak intensities. This distinction highlights the importance of considering battery involvement in any fire scenario analysis, as it significantly affects the overall fire behaviour. Therefore, while general patterns between EVs and ICE vehicles are observable, the presence and timing of battery engagement in EVs are crucial variables that impact fire intensity and duration, underscoring the complexity and variability inherent in fire dynamics across vehicle types.

The process known as distribution fitting is pivotal for accurately capturing the diverse behaviours of fire within a dataset. It involves selecting a statistical distribution that best represents the span of data relevant to different fire behaviours, particularly critical when analysing fire severity characteristics across the varied landscape of EVs and ICE vehicles. This study utilizes the JASP software, which is a free, open-source statistics software developed by the University of Amsterdam that offers both standard and advanced statistical techniques with a user-friendly drag-and-drop interface and real-time results display [28, 29]. The software offers three methodologies for determining the best-fit distributions for fire severity characteristics: the Kolmogorov–Smirnov, Cramer-von Mises, and Anderson–Darling methods. Among these, the Kolmogorov–Smirnov method is selected due to its emphasis on the central part of the distribution, where most outcomes are concentrated, rather than focusing on the distribution’s tails.

To conduct this fitting, several distributions are considered, including Gamma, Weibull, Exponential, Lognormal, and Normal, all with the lower bound set to zero to reflect the data’s nature accurately. The selection process not only involves evaluating the fitting statistics—where a lower value indicates a more accurate fit—but also considers the frequency with which distribution shapes are used and their availability in other software for subsequent research.

The results of this analytical approach, showcasing the ranked order distributions PHRR, TPHRR and growth coefficient are meticulously compiled in Table 10 and Table 11. It is important to note that the analysis prioritizes these specific metrics as they are necessary in developing design fire scenarios, while the data for THR is currently not considered pivotal for this purpose. Additionally, a significant portion of the tests compiled lacks comprehensive information on THR, underlining our focused approach. For the decay coefficients, distribution fitting was not conducted, as both datasets consistently produced a fixed value of -0.01. Therefore, this value was established as the fixed decay coefficient. Accompanying these tables are the fitting statistics, providing a transparent and comprehensive view of how well each distribution fits the data, thereby maintaining the original intent of illustrating the methodological rigor applied in analysing fire severity characteristics within EV and ICE vehicle classifications.

In this statistical analysis, the distribution shape was chosen based on the lowest distribution fit value obtained for each dataset. For instance, when analysing the PHRR for both EV and ICE datasets, the Log Normal distribution provided the lowest distribution fit value for each. To streamline future distribution shape selections for PHRR data, the Log Normal distribution was therefore selected as a standard for this metric, given its consistent fit across both datasets.

However, determining the appropriate distribution shapes for TPHRR and growth coefficient was less straightforward, as the lowest distribution fit values did not correspond to a single distribution type for these parameters. For TPHRR the Gamma distribution was selected, as it consistently demonstrated low fit values across both EV and ICE datasets. For the growth coefficient, the Weibull distribution was chosen due to its consistently low fit values for both datasets.

The advantage of distribution fitting becomes more apparent when considering the variability and uncertainty inherent in fire behaviour. Summary statistics such as the mean, median, and mode provide central tendencies but fail to capture the full range of possible fire behaviours, particularly the likelihood of extreme events. In contrast, distribution fitting allows for the generation of percentile-based design fire inputs, enabling fire engineers to account for both typical and rare fire scenarios. This approach is particularly relevant in the context of EV fires, where variability in fire dynamics is influenced by factors such as battery involvement, ignition source, and vehicle configuration.

Distribution fitting also plays a critical role in addressing uncertainties associated with fire measurements. Even under controlled conditions, repeated fire tests of the same vehicle may yield slightly different HRR curves due to random variations in material properties, ignition points, or fire progression. By fitting distributions to the data, this analysis accommodates those natural fluctuations, allowing for more robust and adaptable fire safety assessments. Summary statistics alone may overlook these nuances, resulting in overly conservative or overly optimistic fire scenarios.

Furthermore, distribution fitting provides greater flexibility in tailoring design fires to specific risk tolerances. Fire engineers can select design fire inputs based on a desired percentile value, balancing conservatism and practicality in performance-based fire safety design. For instance, a design fire constructed using the 90th percentile value will capture more extreme fire behaviours, ensuring that infrastructure designs account for higher-risk scenarios. This flexibility allows engineers to make informed decisions based on the specific safety objectives of their projects, rather than relying solely on average values that may not reflect the range of possible fire outcomes.

These distributional fits not only provide a structured way to comprehend the fire dynamics of EVs and ICE vehicles but also emphasise the differences in fire growth and intensity characteristics between the two vehicle types. By employing Log Normal for PHRR, Gamma for TPHRR, and Weibull for Growth coefficient, the analysis adeptly accommodates the skewed nature of these data, while the parameters μ and σ offer a detailed understanding of their distributions. The distribution parameters identified for EVs, and ICE vehicles are essential for creating probabilistic design fires for both types of vehicles for the use of simulation which are shown in Table 12.

This section examines the influence of battery size on the PHRR and growth coefficient in EVs, using data from the collated fire experiments. The objective is to determine whether there is a direct correlation between battery size and fire magnitude, as well as to understand how battery size affects fire growth dynamics. With larger batteries increasingly integrated into modern vehicles to support longer range and better performance, understanding this relationship is crucial.

For the EV fire experiments, a subset of 10 experiments—filtered from the total of 16—was analysed specifically to assess the impact of battery size on PHRR. The analysis was therefore limited to these 10 vehicles, for which battery sizes were documented. Figure 3 presents a log-scale graph illustrating the correlation between battery size and PHRR, revealing a logarithmic correlation with a coefficient of determination (R²) of 0.73 based on a simple regression analysis. This significant correlation suggests a notable relationship, indicating that larger battery sizes are associated with higher PHRR values, which points to a quantifiable impact of battery capacity on fire intensity. However, given that this analysis was based on only 10 vehicles, it is important to acknowledge that conclusions drawn from this limited dataset may not fully capture the broader dynamics of EV fires.

Bild vergrößern

The correlation between battery size and PHRR in EV fires has also been explored in recent studies, with Sun et al. [6] highlighting a similar relationship. While the current study found a logarithmic correlation, Sun et al. [6] established a power-law relationship and based on a larger dataset with various battery types and configurations, Sun et al.’s findings affirm the importance of battery size in influencing fire severity but also emphasize the roles of battery arrangement, cooling conditions, and air supply in fire behaviour. In contrast, the current study, with a smaller dataset, points to the need for further data to refine the observed correlation and suggests that battery size is just one of multiple factors affecting fire risks in EVs.

Further insight into the relationship between battery size and PHRR comes from Willstrand et al. [17], whose study focused on isolated battery fire tests rather than full EV fires. Although Willstrand et al. observed a positive correlation between battery energy capacity and PHRR, their findings were limited to individual battery cells and packs, without the additional materials and components present in an EV. This distinction is crucial, as EV fires involve other combustibles such as plastics, metals, and interior materials that can significantly affect fire dynamics. Despite these differences, Willstrand’s findings align with the general trend observed in the current study, reinforcing the notion that battery size correlates with fire intensity. However, the complexity of full-scale EV fires suggests that extrapolating from battery-only tests may not capture the complete fire behaviour in EV scenarios.

The investigation further extends to analysing the relationship between battery size and the corresponding growth coefficient across the same subset of 10 EV experiments, as shown in Fig. 4. This phase of the analysis aims to determine whether battery size has a measurable effect on the rate of fire growth—a critical factor in understanding fire dynamics and safety implications in EVs. Using a simple regression analysis with a logarithmic trend line, a coefficient of determination (R²) of 0.31 was obtained. While this value indicates some correlation, it suggests that battery size does influence the rate of fire growth, albeit to a moderate extent.

Bild vergrößern

In interpreting these results, it is essential to consider additional factors that may affect the growth coefficient, such as the location of ignition (interior or exterior), ignition power, timing of battery involvement, and the specific battery types and technologies used. Each of these factors can significantly influence fire dynamics—yet in real-world scenarios, they are often beyond control and inherently random. This analysis takes a broad, inclusive approach, acknowledging the variability and unpredictability of these elements. The goal is to establish a foundational understanding that can accommodate this randomness. As more data becomes available, we anticipate that clearer trends will emerge, even amidst the variability of these uncontrolled factors, ultimately allowing for a more reliable prediction of fire growth in EVs.

Building on these initial findings, it is essential to approach the observed correlation between battery size and fire growth coefficient in EVs with caution. EVs are highly complex systems and cannot be considered as simple, homogenous objects where straightforward correlations might apply. They involve multiple dimensions, varying materials in both interior and exterior construction, and diverse battery placement configurations, all of which add to their complexity. Additionally, numerous other factors affect fire dynamics, including the ignition location and the types of materials inside the vehicle. For example, different ignition points can significantly alter fire behaviour and potentially impact fire magnitude.

One potential application of the analysed data is the construction of a design fire model that fire engineers can use to assess fire behaviour in various scenarios. In this section, an example design fire is developed based on the data presented in Table 6 and Table 7, utilizing the mean (average) values derived from the statistical distributions to create a representative fire scenario. The primary objective behind formulating these mean design fires for EVs and ICE vehicles is to facilitate a comparative visualization of the potential hazards associated with fires in these vehicle categories. This visual comparison is essential for understanding the distinct fire behaviour and risks presented by each type of vehicle during fire incidents. By constructing design fires for both EV and ICE vehicles, this model allows for a straightforward comparison.

It is important to highlight that the ICE vehicle data included in this analysis represents the most recent information gathered in the current study. Furthermore, to provide a comprehensive perspective on the evolution of vehicle fire hazards, a design fire model based on data collated in previous research, specifically for ICE vehicles up to the year 2013 as documented by Tohir et al. [9], has also been developed. This historical comparison allows the authors to juxtapose the design fire characteristics of ICE vehicles across different time frames (up to 2013 and 2014–2023) against those of EVs.

The comparative study of design fire profiles using the mean (average) distribution over time among EVs, ICE vehicles up to 2013, and ICE vehicles from 2014 to 2023 reveals distinct fire development behaviours across these vehicle categories as shown in Fig. 5. For EV), the design fire reaches its PHRR at approximately 19 min, while the ICE vehicles from 2014–2023 reach their PHRR around 20 min. Notably, the PHRR for EV fires is approximately 2000 kW higher than that of the ICE2014–2023 fires. Comparing fire growth is more meaningful when we consider the time required to reach a particular HRR. By choosing a common HRR target of 4000 kW, we observe that the EV design fire reaches this level at about 15 min, whereas the ICE2014–2023 design fire reaches it around 19 min. This difference indicates that EV fires grow slightly faster, and the 4-min lead time is critical. A quicker growth rate means that EV fires have a greater potential for rapid spread, increasing the urgency for effective suppression or containment. Looking at older ICE vehicles (pre-2013), the design fire reaches 4000 kW around 27 min, significantly slower than ICE2014–2023. This quicker escalation in newer ICE vehicles may reflect the increased use of combustible materials, such as polymeric or plastic components, which accelerate the growth rate of fires [10, 27].

Bild vergrößern

The faster growth dynamics observed in EV fires hold critical implications for fire engineering, as the quicker a peak HRR is reached, the more rapidly heat is radiated to combustibles in the vicinity, leading to a faster spread of fire. The data further reveal that EVs, on average, exhibit a peak HRR approximately 2000 kW higher than that of ICE vehicles. Such a significant difference in PHRR highlights the heightened risk of fire spread in scenarios involving EVs, particularly in environments like car parks. This suggests that fires involving EVs not only have the potential to grow larger but also to spread more quickly. Given that this analysis is based on a limited dataset, if these findings hold true, they will pose considerable challenges to existing fire safety strategies and necessitate a re-evaluation of response protocols to better address the unique risks presented by EVs.

Further analysis has been conducted to integrate the generated mean design fire curves, constructed using average HRR values, onto the family of HRR curves for EV and ICE vehicles, as shown in Fig. 6 and Fig. 7, respectively. This approach aims to illustrate how the mean design fire curves align within the broader set of observed HRR curves. Both plots indicate that the generated design fire curves follow a conservative and reasonable trajectory relative to the family of curves, supporting their relevance as representative fire scenarios. It is noted, however, that these design fires are based on average values. To provide a more conservative and practical option for fire safety applications, an 80th percentile design fire curve has now been included in Figs. 6 and 7. The 80th percentile curve represents a more extreme fire scenario compared to the mean design fire, aligning more closely with the approach typically taken by fire engineers when selecting conservative design fires. The use of probabilistic inputs—such as percentile values—could yield design fires that occupy different positions within the family of curves, potentially offering a more nuanced perspective on fire growth and severity by accommodating a range of HRR outcomes. The addition of the 80th percentile curve highlights how percentile-based design fires can better account for variability, providing fire engineers with flexibility to choose design fires that reflect different levels of risk tolerance. This variability shows the importance of statistical analysis in performance-based fire safety design, where accurately representing the HRR data is essential for developing reliable design fire models [30].

Bild vergrößern

Bild vergrößern

In another analysis, Fig. 8 shows HRR curves for various passenger vehicle design fires, which are typically utilized by fire engineers when designing buildings involving vehicles, such as parking structures or tunnels. These design fires serve as critical input parameters for performance-based fire safety engineering, influencing decisions related to fire suppression systems, structural integrity, and ventilation. The figure juxtaposes the 80th percentile heat release rate curves for both EVs and ICEs against several commonly used reference design fires from literature, including Joyeux[31], the New Zealand Verification Method C/VM2[32], TNO[33] and Ingason’s models[34]. The comparison aims to assess whether the existing reference design fires remain adequate in representing the fire hazards posed by modern vehicles. From the figure, it can be observed that the 80th percentile curve for ICE vehicles lies somewhere in the middle of the spectrum compared to the reference plots. It is neither overly conservative nor excessively lenient. This positioning is expected, given that the curve is derived from statistical distribution rather than worst-case assumptions. Using a distribution-based design fire ensures that the approach is data-driven and avoids overly conservative assumptions that could lead to unnecessary construction costs. Ultimately, fire engineers can decide on the level of conservatism required based on the data presented, balancing safety with economic considerations.

Bild vergrößern

However, the 80th percentile curve for EVs appears to be among the most conservative curves shown in the plot. This raises a critical question: does this suggest that the existing reference design fires are no longer conservative enough for modern EV fire scenarios? This observation warrants further investigation. It could indicate that EVs may present higher fire risks compared to traditional vehicles, especially since existing reference plots, such as Ingason’s from 2006, were developed before the widespread adoption of modern EV technology. Alternatively, it may reflect inherent differences in the fire dynamics of EVs, as battery fires can release energy suddenly, though this does not necessarily result in larger or quicker fires compared to ICE scenarios. While the current reference plots appear sufficient for ICE vehicle scenarios, the EV design fire curve’s positioning suggests that fire engineers may need to reconsider the conservatism of reference plots when accounting for EVs in building fire safety designs.

A key implication of reconsidering reference plots for modern fire scenarios, particularly those involving EVs, is the need to revisit how design fires are characterised in fire risk assessments. The design fire, typically represented by the HRR as a function of time, serves as a primary input for analyses ranging from simplified hand calculations to complex computational fluid dynamics (CFD) simulations. HRR data is generally derived from experimental methods, such as oxygen consumption calorimetry, to ensure accuracy in representing fire behaviour. Nevertheless, the inherent variability of fire dynamics means that repeated tests under identical conditions often yield HRR curves with minor discrepancies. This variability highlights the importance of using statistical methods to generate a representative reference curve that accommodates natural fluctuations in fire behaviour. By statistically analysing multiple datasets, a more robust and generalisable design fire curve can be developed, enhancing the reliability of performance-based designs in accommodating the unpredictable nature of real-world fire scenarios. This approach ensures that fire safety assessments are not only grounded in empirical data but also account for the uncertainties and complexities inherent in fire dynamics, thus supporting a more resilient and adaptable design framework.

In conclusion, this study provides a comprehensive analysis of heat release rate profiles between EVs and ICE vehicles, highlighting key differences in fire behaviour, particularly in terms of growth and peak heat release rates. Through detailed statistical analysis, this research demonstrates the influence of battery involvement in EV fires, with findings indicating that larger battery sizes correlate with higher peak heat release rates. The design fire models developed from this data offer valuable insights, equipping fire engineers with an alternative, probabilistic approach for simulating vehicle fire scenarios, beyond traditional deterministic methods for performance-based design. By incorporating probabilistic design fire inputs, this work broadens the toolkit for fire engineers, allowing for more informed and flexible decision-making in fire risk assessments and design.

This probabilistic approach differs from that of Hodges et al. [35], who established a model aimed at conservatively bounding design fires for passenger vehicles. Their model uses vehicle parameters such as battery energy capacity, curb weight, and gas tank capacity and was validated with full-scale fire test data to offer consistently conservative predictions of HRR and THR. This makes their model suitable for assessing fire hazards in a range of vehicle types, especially larger, high-energy-capacity vehicles. In contrast, the current study’s probabilistic framework provides a flexible alternative, allowing fire engineers to simulate a range of potential scenarios that can be refined as more data becomes available. While Hodges et al. advocate for a bounding, deterministic model, this study’s probabilistic design approach offers adaptability in assessing fire risks, particularly as vehicle technologies evolve.

As fire engineers work towards creating safer infrastructure and response measures for both vehicle types, this research provides essential guidance for designing targeted fire scenarios. The comparison between EV and ICE vehicles emphasises the need for continuous research and adaptation of fire safety standards to address the unique risks introduced by technological advancements. It is important to note, however, that this analysis is based on a limited dataset, and while informative, future studies should seek to expand the dataset with more varied experimental results to further refine our understanding of vehicle fire dynamics.

Future research efforts should focus on incorporating diverse experimental data to capture a wider spectrum of vehicle fire behaviours, particularly as new EV models and configurations are introduced. This expanded dataset will better equip the fire safety community to anticipate and mitigate fire risks across different environments. This foundational work paves the way for improvements in performance-based design of infrastructure, enabling fire safety professionals to make more informed decisions that address the unique challenges posed by electric mobility and evolving vehicle fire risks.