Applied Energy 332 (2023) 120497

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Full-scale fire testing of battery electric vehicles

Sungwook Kang a, Minjae Kwon a, b, Joung Yoon Choi a, Sengkwan Choi c, *
a
Fire Testing & Research Centre, Fire & Disaster Management Division, Korea Conformity Laboratories, Republic of Korea
b
Advanced Functional Nanohybrid Material Laboratory, Department of Chemistry, Dongkuk University, Republic of Korea
c
School of the Built Environment, Ulster University, UK

H I G H L I G H T S

• The characteristics of BEV fires is comparable with those of conventional passenger cars.
• The new combustibles in the battery pack make a minor contribution to the whole magnitude of BEV fires.
• The jet flame, caused by thermal runaway, accelerates the fire spread to other combustibles of BEVs.
• The uncertainties, induced from unforeseen thermal runaway and reignition, make a major risk to first responders.

A R T I C L E I N F O A B S T R A C T

Keywords: The market share of electric vehicles, powered by lithium-ion batteries (LIB), has been expanding worldwide
Battery electric vehicle with the global momentum towards green technology and improving the driving range on one full-charge.
Thermal runaway Studies are, however, still required on the fire safety of the latest but unmatured technology due to a distinc­
Lithium-ion battery
tive phenomenon called thermal runaway. In this study, a series of full-scale fire experiments were conducted,
Full-scale fire testing
focusing on the understanding of thermal behaviours of battery electric vehicle (BEV) fires. To provide up-to-date
Thermal behaviour
information on BEV fires, the latest BEV model with a high electric-energy capacity (64 kWh) was selected. For
comparative analysis purposes, a LIB pack and a BEV body were tested individually after being physically dis­
assembled. An internal combustion engine vehicle and a hydrogen fuel cell electric vehicle were also tested.
During testing, the combustion of the BEV fires continued for approximately 70 min, resulting in critical mea­
sures of burning being determined; peak heat release rate (pHRR), total heat released (THR), fire growth
parameter, and the average effective heat of combustion were measured to be 6.51–7.25 MW, 8.45–9.03 GJ,
0.0085–0.020, and 29.8–30.5 MJ/kg, respectively. It was also observed that the pHRR and THR were governed
by the combustion characteristics of typical combustible materials in the passenger cabin, rather than by that of
particular contents in the LIB pack with thermal runaway. Instead, a jet fire intensively discharging from the LIB
pack led to a rapid flame spreading to adjacent combustible components of the BEV, thereby accelerating the fire
growth. The findings could contribute to the activities of the first responders to BEV fire accidents, fire safety
engineers, and structural member designers. This study also makes public the measured thermal quantities for
further studies on the fire safety of existing or designing car-parking related structures.

Electric vehicle sales, including those of plug-in hybrid electric vehicles

(PHEV), globally reached 6.75 million units in 2021, approximately
1. Introduction
10.1 % of the market share of car sales worldwide [2]. Concurrently, the
capacity of electric vehicle battery packs has been gradually improving
Greatest attention is currently focused on electric vehicles relying on
to meet the growing demand for a longer driving range on one full-
electric battery energy as a power source, although the fundamental
charge; for example, in the U.S. market, the average capacity
idea of these vehicles was invented in the 1800 s [1]. Their market share
increased from 41.2 kWh in 2015 to 70.5 kWh in 2020 [3].
has been rapidly expanding against that of the conventional internal
In 2006 a massive electronics industry recall occurred in the global
combustion engine vehicle (ICEV), as electric vehicles represent green
laptop industries and in 2016 the production of an electronic device
technology as well as providing extraordinary driving performance.

* Corresponding author at: Belfast School of Architecture & the Built Environment, Ulster University, 2-24 York Street, Belfast BT15 1AF, UK.
E-mail address: [email protected] (S. Choi).

https://doi.org/10.1016/j.apenergy.2022.120497
Received 8 October 2022; Received in revised form 20 November 2022; Accepted 3 December 2022
Available online 15 December 2022
0306-2619/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
S. Kang et al. Applied Energy 332 (2023) 120497

Nomenclature f fire
g gauge
A surface area [m2] ini initial
cp specific heat [J/(kg.K)] loss loss
F view factor s surface
h convection heat transfer coefficient [W/(m2.K)] ∞ surroundings
ṁ mass loss rate [kg/s]
Abbreviations
Q̇ thermal energy [W]
BEV battery electric vehicle
q̇′′ heat flux [W/m2]
FCC fuel consumption calorimetry
T temperature [K]
FCEV fuel cell electric vehicle
t time [s]
HRR heat release rate
Greek symbols ICEV internal combustion engine vehicle
α thermal absorptivity LIB lithium-ion battery
ε thermal emissivity OCC oxygen consumption calorimetry
θ growth parameter for t-square fires [MW/s2] PHEV plug-in hybrid electric vehicle
ρ density [kg/m3] pHRR peak heat release rate
Σ Stefan-Boltzmann constant[W/(m2.K4)] THR total heat released

Subscripts
end end

need to be preliminarily determined, in conditions of open and enclosed

Table 1 spaces respectively. The parameters are the rate of heat release (HRR in
Research objectives corresponding to specimen type from the fire safety W) of a burning car, amount of total heat released (THR in J), fire growth
viewpoint. coefficient (in W/s2), fire load (in J/m2), mass loss (i.e., fuel consump­
Level Type of specimen Research objectives tion), effective heat of combustion (in J/g), incident heat fluxes in the
vicinity of the heat source (in W/m2), flame spread throughout the
1 Component Thermochemical/electrochemical characteristics
materials of the unit
burning vehicle, and time–temperature profiles of diffusion flames and
2 Single cell Thermal characteristics of the unit under thermal/ adjacent structural members. The open space indicates a place where the
electrical/mechanical impact conditions, test vehicle is in the open without re-radiated from heated surrounding,
mechanism of gas venting and thermal runaway while the enclosed space denotes a place where flame behaviours are
initiation, and quality/quantity of vent gas
affected by surrounding structures, such as multistorey car parks and
3 Multiple cells Thermal characteristics of the unit under thermal/
/single module electrical/mechanical impact conditions, and cell- underground spaces. After determining these parameters, the next step
to-cell thermal runaway/heat propagation is to theoretically/numerically predict time-dependent temperature
4 Multiple modules Module-to-module thermal runaway/heat distributions throughout spaces, loadbearing members, or local areas
propagation in horizontal and vertical directions covered by a sprinkler system.
5 Pack Thermal characteristics of the unit under thermal/
electrical/mechanical impact conditions, and
Since the significant LIB recall in 2006, research on the safety of LIBs
pack-to-vehicle body fire spread has been being conducted worldwide in different levels and objectives,
6 Battery Electric Magnitude of BEV fire hazards, and BEV-to- as summarised in Table 1. The final goal of such studies would be to
Vehicle (BEV) adjacent object fire spread reach an integrated full-interpretation of the LIB behaviour from mate­
rial to product levels. As part of the goal, the purpose of this study is to
understand the thermal behaviour of BEV fires by full-scale fire testing
model was permanently ceased, due to thermal runaway incidents
on recent BEVs with high energy capacities (i.e., Level 6). This article
originating from the potentially faulty lithium-ion batteries (LIBs) [4–5].
therefore primarily discusses the quantification of the extent of BEV fire
Such occasions drew attention to research on LIB safety, as well as to
hazards in an open-space environment for the first experimental work,
relevant regulations and standards; several LIB safety related standards
as defined in the previous paragraph, to obtain the key parameters under
have been being developed and revised globally for cell-, module-, and
open space and to make public the measured data for further studies.
even system-level LIBs, such as IEC, ISO, UN, SAE, UL, SAND and GB/T.
As significant resources are expected to be required for conducting
As the technology on LIB system has not yet fully matured to date, un­
full-scale car fire experiments, such tests have been intermittently per­
foreseen fire patterns have been widely reported at various international
formed worldwide since the 1990 s, mostly on ICEV fires but rarely on
incidents associated with battery electric vehicles (BEVs). In South
both PHEV and BEV fires [7–16]; an analytical approach collated
Korea, 17 local BEV-fire accidents occurred for the period between 2018
experimental data from several literature of large-scale car fire tests
and 2021 [6]. In particular, 6 events happened in enclosed areas, such as
[12]. With respect to the methodologies, some works used oxygen
manufacturing factories and underground car parks in buildings. Under
consumption calorimetry (OCC) to physically measure the HRR of car
such circumstances, more sever fire behaviour and subsequent damage
fires [7–10,14,16]. For the measurement, the OCC includes the mea­
were evaluated compared to those events that happened in open spaces.
surement station which measures velocity, temperature, smoke density,
As a response to these new fire scenarios, comprehensive revision should
and concentration of combustion gas [17–18]. An early attempt to test
be carried out on several technical aspects of design conditions, such as
real-scale ICEVs was made using an open OCC with a 14.2-m2-cross-
the fire degradation of loadbearing structures (of multi-storey car parks,
section hood and a 0.2-m2-cross-section duct [7]. Another study con­
underground spaces, and tunnels), the capacity of current fire suppres­
nected a 9-m2-cross-section hood with another to collect a greater
sion systems, and the applicability of the existing fire-suppression stra­
quantity of the combustion products generated from ICEV fires [8].
tegies, before BEVs become the majority of vehicles in societies.
Since the 2000 s, larger OCC was introduced to test more recently
In order to carry out these revisions, a list of key input parameters

2
S. Kang et al. Applied Energy 332 (2023) 120497

Table 2 two cars were produced from the same manufacturing platform. This
Specimen features and test conditions. ICEV version and a hydrogen fuel-cell electric vehicle (FCEV) were
Features and Test 1 Test 2 Test 3 Test 4 Test Test additionally tested in the fifth and sixth experiments to examine the
conditions 5 6 thermal characteristic differences between BEV, ICEV, and FCEV. The
LIB pack of Body BEV_2 BEV_3 ICEV FCEV six full-scale car fire tests were conducted upon an instrumented rig
BEV_1 of
under a 10-MW-scale OCC of Korea Conformity Laboratories (KCL).
BEV_1

Length N/A 4,180 4,165 4,670

2. Experimental
(mm)
Width (mm) N/A 1,800 1,800 1,860
Height (mm) N/A 1,570 1,550 1,630 2.1. Calorimetry
Weight (kg) 449 1,206 1,540 1,685 1,320 1,820
Nominal 64.092 N/A 39.240 64.092 N/A 1.560
In fire safety engineering, the HRR and THR are critical measures
total
energy
enabling one to evaluate the magnitude of fire events, which are
capacity conventionally obtained from calorimeter tests based on the principle of
of LIB oxygen depletion during combustion developed in the 1970–80 s [20].
(kWh) These parameters for car fires have been rarely measured worldwide as
State of 100 N/A 100 100 N/A 20
the greater mass that specimens have the larger calorimetry with rele­
charge of
LIB (%) vant facilities is needed. Significant laboratory, financial, technical, and
Thermal Heating a Heating the Heating a Heating manpower resources are therefore required for such tests.
runaway surface of vehicles’ bottom surface of combustible In the present tests, the combustion products released from burning
initiation single cell boundaries using single cell contents of
vehicles were collected by means of a 51.5-m2-cross-section conical
or flaming using a a propane burner using a vehicles using a
ignition heating heating pan of heptane
hood suspended approximately 8.6 m above the test floor, as shown in
sheet sheet Fig. 1(a). The hood was coupled with a 3.8-m2-cross-section exhaust
duct which was connected to draft fans capable of delivering approxi­
mately 3,200 m3/min, as shown in Fig. 1(b). As a certain amount of
manufactured cars which would release greater amounts of heat once oxygen involved in the incoming air flow is expected to be consumed in
ignited [9–10,14,16]. There existed an alternative method of estimating the course of combustion of the carbon-based object (e.g., vehicles), a
the HRR by measuring the weight loss rate of specimens using load cells reduction in the oxygen concentration relative to the ambient air occurs
and adopting a known constant (namely the heat of combustion) in the as a function of time [21–22]. This variation is detected by a para­
HRR calculation [11]. This technique uses fuel consumption calorimetry magnetic oxygen analyser. The changed oxygen-concentration in mole
(FCC) without the measurement station, thereby facilitating obtaining fraction was used to calculate the factor of oxygen depletion. This factor
HRRs. However, the constant adopted needs to be firmly based on was subsequently applied to calculate the HRR and THR of tested ob­
experimental data. In case of BEVs, as less information is available yet, jects, based on the fact that the average heat release per unit mass of
the HRR measurement using the OCC should be preceded with the HRR oxygen appears to be a constant of 13.1 MJ/kg with a standard deviation
calculation utilising the FCC only. In this study, both the OCC and FCC of 0.35 MJ/kg [20].
were therefore used, thereby measuring HRRs and estimating the
average effective heat of combustion of BEV fires for further research.
Attempts to experiment on real-scale BEVs began to be made from 2.2. Specimen
the early 2010 s [12,14,16]. However, the BEVs in the previous studies
were still in their infancy, which involved smaller amounts of low Test vehicles were the five-door models with similar dimensions and
energy-capacity LIBs and flammable plastics than the recent models weights, except FCEV, as listed in Table 2. The BEV model was reported
with at least 30-kWh-LIBs and luxurious interiors [19]. For up-to-date as developed from the identical manufacturing platform to that used for
information on BEV fires, three BEVs manufactured in 2020 were the existing ICEV version. The flame spreading behaviours in the bodies
experimentally examined in this study. For the first and second tests, as of the two types of vehicles were therefore assumed to be comparable.
tabulated in Table 2, a unit of BEVs was disassembled into its body and a The BEV model had two trim-levels in terms of LIB energy capacity,
LIB pack. This allowed examination of individual contributions of the which involved 39- and 64-kWh LIB packs for a normal and a long
parts to the global BEV fire. Fires generated from two fully charged BEVs driving-ranges on one full-charge respectively, as illustrated in Fig. 2(a)
with different energy capacities were studied in the third and fourth and 2(b). A LIB pack was also mounted on the FCEV trunk. Its energy
tests. The BEV model is indistinguishable from its ICEV version as the capacity (1.56 kWh) was much lower than those of BEVs.
The LIB packs of the drivable BEVs were fully charged in operating

Fig. 1. Large-scale cone calorimetry in KCL.

3
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 2. The layouts of LIB modules in packs with two options.

fourth tests, the instrumented LIB-packs and BEV-bodies were reas­

Table 3
sembled. Although the reassembled BEVs were operatable, the speci­
Specifications of LIB cell.
mens were moved upwards and away by a forklift truck for safety
Model Nominal Nominal Size (mm) Weight purposes as the BEVs’ cooling system was disconnected. After the
capacity (Ah) voltage (V) (g)
preparations, all the specimens were covered with waterproof polyvinyl
E63B 60 3.63 (L)310×(W) 887.8 chloride (PVC) blankets and conditioned indoors for at least 2 weeks
(NCM 114×(t)15
before the experiments.
622)

2.3. Instrumented test rig

conditions before disassembled from the vehicle bodies. The isolated LIB
packs were subsequently opened to equip heating sheets, thermocou­
To facilitate the measurement of multiple parameters during the
ples, and voltmeters on the internal LIB cells, as shown in Fig. 2 and 3(a)-
large-scale fire testing, a test rig was designed. The rig was composed of
(c). The heating sheets were intended to initiate the thermal runaway of
a mounting platform, structural steel members (columns and beams),
a predetermined cell. The thermocouples were used to examine the heat
and one-way water-cooled pipes, as shown in Fig. 4(a). As it was ex­
propagation from the ignited to the surrounding modules inside the
pected that fire plumes with high temperature over 800 ℃ would
packs and to the pack’s metal housings. The voltmeters were used to
maintain for at least 10 min during each of the tests, tap water at room
monitor the voltage drops of modules to confirm the thermal runaway.
temperature was allowed to flow through the ceiling and column pipes
The specifications of the pouch-type LIB cells installed in the BEVs are
of the rig to prevent its thermal failure. The rig dimensions were
listed in Table 3. Fig. 3(d) shows the distribution of thermocouples in the
determined based on the Korean regulations for the minimum size of a
passenger cabin, motor compartment, trunk, and tyres. For the third and
standard parking space with a 2.5 m width, a 5.0 m length, and a 2.3 m

4
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 3. Instrumented LIB packs.

height. Three types of sensors were installed in the rig, which were the 2.4. Procedure
thermocouple, heat flux gauge, and load cell, as illustrated in Fig. 4:
A series of instrumented specimens were placed at the centre of the
• To measure the temperature of combustion gas above different parts mounting platform in order. For smooth combustion, the front- and rear-
of the vehicle, thermocouples (Model: 24-AWG k-type dual-wall seat windows of vehicles’ left doors were half-opened to allow air flow in
insulated wire) were aligned with the vehicle’s longitudinal mid­ to the passenger cabin, wherein the front, rear, left and right indicate the
plane on the upper water-cooled pipes (TC_GB, TC_GW, TC_GR, and ordinary directions of driving. Air in the tyres was deliberately released
TC_GCRW), as illustrated in Fig. 4(b). Thermocouples were also so as to prevent unnecessary explosions. The thermal runaway of LIB
positioned at the mid-height of the vehicle’s front- and rear-seat packs was initiated by either heating a single LIB-cell inside the pack or
windows (TC_FW and TC_RW at z = 1,270 mm) and at the same x- heating the lower boundaries of the pack, as listed in Table 2:
and y-coordinates but on the upper pipes (TC_GFW and TC_GRW at z
= 2,270 mm); • A 575-W heating sheet with 90 × 65 mm in dimensions was taped to
• Four water-cooled Schmidt-Boelter gauges (Model: SBG01, Hukse­ the surface of a mid-cell of the LIB-pack’s central module (i.e.,
flux) were aligned with the rig edges at z = 1,270 mm, facing the Module_9), as demonstrated in Fig. 5(a). The heater was powered by
front-bonnet, trunk, left front- and rear-seat windows of vehicles. an electric power supply enabling one to manually control voltage
These positions were intended to estimate the amount of irradiance and current. A constant amount of electrical energy was intended to
on a vehicle parked adjacent to a burning BEV; be consistently supplied to the heater. Once thermal runaway was
• Four load cells composed of 16 strain gauge transducers that range triggered, the electricity was cut-off due to the electrical short-circuit
from 0.5 to 1,000 kg each (Model: R-1000, CAS) were assembled failure of the heating element. This setup was intended to simulate a
with the mounting platform. The sensors measured the variation in typical scenario of thermal runaway originating from abuses or de­
mass of the specimen during the combustion of vehicles. fects in the LIB cell inside the pack;
• A 300-kW sand-box burner supplied with propane was placed un­
All time-dependent data of temperature, heat flux, and weight were derneath the LIB pack, as illustrated in Fig. 5(b), which was planned
acquired at intervals of 0.1 s using a DAQ system (Model: CompactDAQ, to simulate a worst-case scenario of car accidents in which a quantity
National Instruments). Finally, the specimen was surrounded by eight of fuel has spilled, spread, and ignited underneath a BEV.
video cameras to observe its thermal behaviour and fire events.
The experiments aimed to collect physical quantities in the condition
of the BEV combustion. The test setup therefore focused on triggering a

5
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 4. Location of sensors in the testing rig.

thermal runaway rather than on examining the parameters representing 3. Results and discussion
a thermal runaway initiation or a flaming ignition, such as the activation
energy of thermal runaway, and the time to the onset of the phenome­ 3.1. HRR and THR of car fires
non. All specimens were allowed to be fully burnt out during the tests.
Fig. 6 shows the overall HRR and THR profiles in the six tests,
physically measured in the open-space condition using the large-scale

6
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 5. Initiation of thermal runaway.

Fig. 6. Time-dependent profiles of HRR and THR.

full burnt-out in the open-space environment. In the very early stage of

Table 4
fires (approximately 1–2 min), the rapid increases in HRR were observed
Measures of thermal hazards of vehicle fires.
in the BEV-body, BEV_2, and ICEV tests. These prompt growths were
Measures Test 1 Test 2 Test 3 Test 4 Test Test caused by the activation of the ignition tools (i.e., the propane burner or
5 6
heptane pans, as introduced in Table 2), rather than any rapid com­
LIB pack Body of BEV_2 BEV_3 ICEV FCEV
of BEV_1 BEV_1 bustions of the specimens. In the fire growth stage, the ignited vehicles
generated gradually-increasing heats and reached peak values (i.e.,
Peak heat release 1.54 7.81 6.51 7.25 7.66 5.99
rate, pHRR
pHRR) in approximately 11–17 mins. The fully developed fire stages
(MW) during the tests, were not maintained for a long period, except the FCEV.
Total heat 1.30 7.53 8.45 9.03 8.08 10.82 The fires subsequently turned into the fire decay stage with gradual
released, THR decreases in HRR. In this stage, fluctuations in HRR were observed in the
(GJ)
BEV_2 and BEV_3 tests due to a series of thermal runaway events. The
HRR and THR of the LIB pack were smaller than those of the vehicles,
cone calorimetry. The solid and dashed lines denote the HRR on the and fluctuations were also observed due to the thermal runaway
scale of the primary vertical axis and the THR on that of the secondary phenomena.
vertical axis, respectively. In the figure, to facilitate the comparison The HRRs and THRs of all the specimens are tabulated in Table 4.
among the data, all the onsets of HRR growths were adjusted at zero The parts of the HRRs produced from the ignition burner or fuel pans
min. Overall, personal vehicle fires remained for at least 70 min until the were excluded in the THR calculation. The LIB pack (64 kWh) and BEV-

7
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 7. Chronological data of pHRR and THR for vehicle fires.

simultaneously combusted. In the extent of vehicle fires, as the simul­

Table 5
taneous combustions generally depends on several aspects, such as the
Categories of t-square fires [23].
location of ignition spot, order of flame spread, and ventilation condi­
Growth t1 θ(MW/ Typical real fires tion, it is challenging to find a correlation between the pHRR and
rate (s) s2 )
weight.
Slow 600 0.0028 Densely packed wood products Fig. 7 shows the chronological data of pHRR and THR for the vehicle
Medium 300 0.0111 Solid wood furniture (desks), or individual fires in the existing studies [7–16] and present paper. The pink and
furniture items with small amounts of plastic
Fast 150 0.0444 High stacked wood pallets, cartons on pallets, or
green dashed boxes indicate the respective data that were physically
some upholstered furniture measured using the OCC and that theoretically estimated using an
Ultrafast 75 0.1778 Upholstered furniture, high stacked plastic average effective heat of combustion with the measured mass loss rate
materials, or thin wood furniture (wardrobes) utilising the FCC. There is a lack of information on the rest. Most of the
data are associated with ICEV fires, while a few papers demonstrate the
parameters for BEV fires as pointed out with the numbers in the figure
body fires generated 1.54 and 7.81 MW of pHRR and 1.30 and 7.53 GJ of
and tabulated in the legend table, except No. 7, 11 and 12; the asterisk
THR, respectively. The BEV_3 with the 64-kWh pack resulted in heat
denotes the vehicles comparable to the present specimen in terms of the
with 7.25 MW of pHRR and 9.03 GJ of THR, respectively. The THR of the
size and curb weight. From the data, summaries can be made, as follows:
BEV_3 fire was comparable with the sum of the individual THRs of the
(1) The present measurements remained in the pHRR and THR ranges
pack and body (8.83 GJ). This observation led to a summary that a larger
for the models with the asterisks. (2) The data for the recent BEV model
contribution of BEV fires came from the BEV-body than the LIB-pack in
(No. 9 and 10) were greater than those for the earlier models (No. 1–6)
terms of pHRR and THR. Accordingly, the combustion that occurred in
due to the comparatively large quantities in BEV mass and LIB energy
the passenger cabin governed the two measures in the BEV fires, despite
capacity. (3) It was challenging to generalise about the tendencies on the
the vigorous thermal runaway generated from the LIB packs. This is
pHRR and THR along year or mass for the extent of car fires due to the
based on the presumption that greater amounts of combustible materials
differences in car-model, -size, the materials applied, and the HRR
exist in the body, particularly in the cabin, than in the pack. The initial
measurement method.
weights of the vehicle-body and pack were 1,206 and 449 kg, respec­
tively. Although no information on such amounts were available, the
3.2. The degree of fire growth
proportions of the combustible materials used in the body and pack
could be estimated by measuring the amounts of mass lost after full
The degrees of the BEV fire growth were quantified adopting a
combustions, which are discussed in the following sections.
generic fire-growth curve (i.e., t-square fire) used in fire protection an­
A gap existed between the THR values for the BEV_2 and BEV_3 fires
alyses. A growth of the time-dependent HRR can be characterised by the
due to the difference between the energy capacities of the two vehicles.
time to reach 1 MW of HRR (t1 in s), which is generally categorised into
The THRs of the BEV fires were higher than that of the ICEV but smaller
four levels based on the growths of typical real fires, as listed in Table 5.
than that of the FCEV. This tendency was parallel to their weights. It
should be note that such comparative works are acceptable with respect The relationships among HRR (Q̇f ), t1, and time (t) are defined as follows
to THR but not as regards pHRR. The pHRR represents a phenomenon of [23]:
burning at the limited instant at which the most volatiles are
Q̇f = 1, 000(t/t1 )2 = θt 2 (1)

8
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 8. The quantification of BEV fire growths in Tests 3 and 4.

where θ denotes the parameter of growth. growth rates listed in Table 5. The time–temperature profiles demon­
Fig. 8(a) and (b) show the changes in the HRR at the fire growth stage strate the temperature variations of the LIB packs at the same stage (i.e.,
in the third and fourth tests, respectively, with design fire scenarios; TC_PTC, TC_PBR, TC_PBC, TC_M2_C and TC_M9_C, as illustrated in
dashed lines denote the calculated t-square curves with the typical Figs. 2 and 3(c)). Important events are described on the top of the figures

9
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 9. Temperature-time profiles of BEVs in Tests 3 and 4.

Fig. 10. Photos of the specimen status at the important events in Test 4.

10
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 11. Time-temperature profiles of LIB-packs.

with arrows marking the temporal points at which these incidents of the energy capacities of the LIB packs. The difference in the growth
occurred during the tests. Note that as explained in the previous section rate between the two car fires is closely related with the difference in the
of procedure, two different heating methods were applied in the third major source spreading flames to adjacent combustible materials. The
and fourth tests; thus, the temporal points at which the HRR started jet flame discharging from the BEV_3 LIB pack heated neighbouring
rising were not identical in the two tests. components more vigorously than the 300-kW-burner did in the BEV_2
The growth parameters (θ) for the BEV_2 and BEV_3 fires were fire.
estimated to be 0.0085 and 0.020 respectively by plotting best-fit curves In the BEV_2 test, the 300-kW-burner continuously heated the bot­
to the measurements, as expressed with crimson dashed lines in Fig. 8(a) tom of the LIB pack for 11 min and 6 s, as evidenced by the blue solid line
and 8(b). The quantities indicate that these fires developed with the in Fig. 8(a). This heating approach resulted in a rapid increase in the
rates reaching 1 MW of HRR approximately 343 and 224 s after their pack housing (TC_PTC and PBC) and early ignitions of tyres (TC_WF and
ignitions, respectively. These rates can be qualified by ‘slow medium’ WR), as shown in Fig. 8(a) and 9(a). Although the ignited flames were
and ‘fast medium’, respectively, based on the categories in Table 5. subsequently spread to the trunk and cabin in order, the flame spread
Particularly, the BEV_3 fire exhibited a faster rate than BEV_2, as well as failed to trigger any thermal runaway of LIB modules until the pHRR at
the swiftest rise to the peak over other test vehicles as shown in Fig. 5, approximately 15 min, as shown in Fig. 8(a); the first thermal runaway
even though the two cars were identical BEV models, with the exception of modules was observed at 22 min and 22 s (TC_M2_C). On the other

11
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 12. The time-dependent heat flux variations in BEV_3 and ICEV fires.

hand, in the case of BEV_3, a mid-cell in the central LIB module in the From the observation on the measurements, several findings can be
pack (M9) was heated by an electric heating sheet until the onset of the summarised, as follows: (1) The jet flame released from the LIB pack led
first thermal runaway at 21 min and 20 s, as shown with the red solid to the rapid growth of the BEV fire. (2) Although the jet fire originating
line (TC_M9_C) in Fig. 8(b). As this heating approach raised a local from the LIB pack accelerated the ignition and combustion of other BEV
temperature inside the pack-housing, other components of the car had components, the jet flame was not the primary contributing factor to the
maintained the ambient temperature until the onset of thermal HRR and THR. Rather, the majority of the heat released at an instant (i.
runaway, as illustrated in Fig. 9(b). The list on the right of the figures e., pHRR) was governed by the flames created from the conventional
indicates the critical events observed during tests; the moments in time combustion of the organic-polymer components of the car. (3) The BEV
for these events are marked with numbered arrows in the graphs. Once fire originating from the LIB pack was more rapidly developed than that
triggered, grey vent-gas and jet flames were sequentially emitted, as derived from elsewhere, as well as that in the ICEV and FCEV fires as
shown in Fig. 10(a) and 10(b), respectively; the photos captured at each evidenced in Fig. 6. (4) The first responders should be more cautious
of the events are shown in Figs. A1 and A2. These flames heated the rear when approaching an accident site of BEVs. This is because the damage
tyres and other components vigorously, which resulted in rapid of LIBs inside the pack housing is unobservable from outside the vehicle
spreading to the passenger cabin in which many combustibles of the car- and the activation of thermal runaway is unpredictable. In addition,
components existed in the condition of ambient temperature. once triggered the fire is rapidly developed.

12
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. 13. The time–temperature profiles of fire plumes in BEV_2 and BEV_3 fires.

3.3. Observation of LIB-pack temperatures capacity, up to approximately 800 ℃. In the BEV_3 test, the thermal
energy generated from the first thermal runaway of the nineth module
Fig. 11(a) and 11(b) show the time-dependent temperature varia­ directly contributed to the temperature rise in the upper housing from
tions inside and outside the LIB packs of BEV_2 and BEV_3, respectively. the pack inside, due to the buoyancy of hot vent-gas and flames. Thermal
The thermocouples locations in the legend are described in Fig. 2. In the runaway was extended at M9_C, M6_F, M6_C, M10_C, M1_C, and M2_C in
BEV_2 test, the external heat source initially raised the surface temper­ order. The lower area temperature of the housing was gradually
ature of the lower alloy-housing (TC_PBF, PBC and PBR). Although heat increased.
was transferred through the metal to the housing’s upper area (TC_PTF, From the heat transfer viewpoint, the external heat originating from
PTC and PTR) in conduction, it is likely that the temperature rise of the the propane burner underneath the LIB pack is supposed to be distrib­
upper part was mainly caused by the heat generated from the flames uted to the alloy-metal housing, internal LIBs, BEV body frame, and
spread to the tyres and passenger cabin. The first thermal runaway was surroundings. The amount of thermal energy transferring to the LIBs
identified at the eleventh event, which was approximately 7 min after could be reduced, provided that a large quantity of heat is dissipated to
the HRR peak (pHRR). Since then, a series of thermal runaways occurred the others. The metal housing that directly faces the external flame has a
in the LIB modules at M2, M1, M9, and M10 in order, which resulted in relatively high heat inertia. This mass absorbed a certain amount of
the rapid temperature rise of the metal-housing with a high heat thermal energy to raise its temperature, thereby retarding the increase in

13
S. Kang et al. Applied Energy 332 (2023) 120497

Table 6 temperature development, Ts (t), can be obtained using the global con­
Gross combustion efficiency of vehicles. servation equation for energy, as follows:
Specimen, year mini [kg] mend [kg] Δm/mini [%] ΔHc,eff [MJ/ [ ]
dT α ( ) ( )
kg] ρs cp,s Vs s = As s q̇′′g − εs σ Ts4 − T∞4 + hs Tf − Ts (4)
dt αg
ICEV(Peugeot 406), 2002 1,454 1,192 18.0 26.0
[9] Fig. 13(a) and 13(b) show the time–temperature profiles of buoyant
ICEV(Peugeot 406), 2002 1,362 1,107 18.7 27.5 fire plumes measured at a 2,270-mm height and generated from BEV_2
[9]
and BEV_3, respectively. The two fires resulted in peak temperatures
ICEV(C2), 2013 [13] 1,303 1,029 21.0 24.4
BEV(SOC100), 2014 [16] 1,448 1,115 23.0 N.A beyond 900 ℃ which were mainly contained contributions from the fire
BEV(SOC85), 2014 [16] 1,475 1,180 20.0 16.6 plumes above the bonnet and front window of the cars (TC_GB and GW).
ICEV, 2014 [16] 1,344 1,008 25.0 14.7 The peak moments of the HRRs were similar to those of the temperatures
BEV(SOC100), 2014 [16] 1,650 1,287 22.0 12.8
in the tests. The black dashed line in the figures indicates a nominal
PHEV(SOC85), 2014 [16] 1,467 1,159 21.0 15.0
PHEV(SOC100), 2014 1,712 1,267 26.0 13.1
temperature–time curve based on the early tests in the 1900 s, which has
[16] been most widely used in building fire engineering [24]. The BEV fires
LIB pack, 2020–2021 449 421 6.3 45.9 are likely to endanger the durability of the structural members above the
[Present] cars for at least 10 min, in which the gas temperatures exceeded that of
BEV body, 2020–2021 1,206 944 21.7 28.8
the standard fire curve. In the BEV_3 case, its combustion is expected to
[Present]
BEV_2, 2020–2021 1,540 1,256 18.4 29.8 be more hazardous to the structural members than that of BEV_2, due to
[Present] its rapid growth and intensive heat released.
BEV_3, 2020–2021 1,685 1,389 17.6 30.5 The concept of average effective heat of combustion (ΔHc,eff in MJ/
[Present]
kg) that represents the gross combustion efficiency of a substance was
ICEV, 2020–2021 1,320 1,026 22.3 27.4
[Present]
adopted to evaluate the present BEV fire hazards. This approach could
FCEV, 2020–2021 1,820 1,423 21.8 27.3 generate date for further studies on the safety measures. The parameter
[Present] for the fire generated from the BEV combustion was defined to be a
constant from the engineering application viewpoint, as follows:
∫
the internal LIB temperature. It was observed in Fig. 11(a) that the tend
Q̇f (t) THR
temperatures of the thermocouples positioned on the exposed bottom- ΔHc,eff = = (5)
tini ṁloss (t) mini − mend
surface of the housing (i.e., TC_PBF, PBC, and PBR) increased rapidly
in the direct heating condition, while the temperature increases of the Table 6 shows the calorific values, ΔHc,eff , and the relevant mass
LIB modules (i.e., TC_M_1, 2, 9 and 10) were relatively slow. It is well measurements of the previous and present specimens; the prior samples
known that the alloy-metal material allows heat to rapidly transfer to were introduced with asterisks in Fig. 7. As a result, mass in the range of
the internal components due to its high thermal conductivity. It was 17.6 to 26 % of the total weight was generally burnt during car fires.
additionally realised that this massive solid partially contributed to the Approximately 30 MJ/kg of ΔHc,eff was recorded from the present BEV
thermal protection of the internal LIBs against external heating for a fires, which was greater than the values for the ICEV and FCEV fires, as
limited period of time by absorbing heat. On the other way, it could be well as for the individual combustion of the BEV-body. This was because
challenging to lose heat from the alloy cover and to prevent the tem­ the total heat generated from the LIB pack (1.3 GJ) was relatively large
perature increases of internal LIBs once this mass reaches a threshold as compared to the amount of mass consumed (28 kg). The overall heat
temperature. intensity released per unit mass therefore increased in the BEV fires. For
the individual LIB pack, ΔHc,eff was estimated to be 45.9 MJ/kg, which
was comparable with those of flammable fuels and highly combustible
3.4. Thermal hazard to adjacent objects polymers, such as n-Pentane (45.69 MJ/kg [25]) and polypropylene
(42.66 MJ/kg [26]). This finding is related to the fact that major flam­
Fig. 12(a) and 12(b) show the time-dependent variation in total mable components of a LIB cell are typically polypropylene and poly­
exposure heat fluxes arriving at the predetermined four points sur­ ethylene (44.60 MJ/kg [26]) for separators and mixed carbonate
rounding and distanced from the burning cars in the fourth and fifth solvents (12.7–24.7 MJ/kg [27]) for electrolytes. Additionally, it turned
tests. Both the car fires recorded their peaks at the points mostly within a out that the ΔHc,eff calculation was highly dependent on the applied test
range of 40 to 60 kW/m2 in the fully developed fire stage. Although one methods for measuring the HRR and relevant test environments, which
of the values exceptionally reached up to 110 kW/m2 at the BEV front, made the generalisation of a correlation between mass and ΔHc,eff
more tests should be repeated to justify this phenomenon. In further challenging.
studies, the exposure heat fluxes measured at the points using heat flux
gauges (q̇′′g ) could be used to evaluate the possibility of an adjacent car 4. Conclusions
ignition by predicting the car’s surface-temperature (Ts ). The incident
To quantitatively evaluate the hazards caused by battery electric
heat flux on the measuring area of the gauge (Q̇g ) is equal to the heat
vehicle (BEV) fires, a series of real-scale fire tests were conducted on the
released from the fire source (Q̇f ) multiplied by the absorptivity of the BEVs and the separated parts of lithium-ion battery (LIB) pack and BEV-
probe’s surface (αg ) and the configuration factor from the fire to the body. Other types of cars, such as internal combustion engine vehicle
gauge (Ff− g ), as follows: (ICEV) and hydrogen fuel cell electric vehicle (FCEV) were additionally
tested for comparative purposes.
Q̇g = αg Ff − g Q̇f (2)
Although several full-scale car-fire tests have been performed in
The dash in the subscript of F implies “to”. The thermal energy, Q̇f , prior studies, limited studies focused on BEVs. Existing literature is for
early BEV models. In markets, personal hybrid vehicles have been
has a relationship with the measured heat flux, q̇′′g , as follow:
shifting to be larger and more luxurious, and have larger energy capacity
Af of LIBs in case of BEVs. To cope with this trend, an innovative test
Q̇f = q̇′′g (3)
αg Fg− f scheme was designed on the instrumented test rig under the 10-MW
calorimeter system. Thermocouples were additionally installed inside
where Ff− g = Ag /Af Fg− f . Accordingly, the time-dependent surface-

14
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. A1. Photos of the specimen status at the critical events in Test 3.

the LIB packs to observe the heat progress among internal LIB modules. was estimated to be 0.020 (fast medium) which was greater than that for
To initiate the thermal runaway of LIBs, internal and external heating another BEV fire unassociated with jet flames (0.0085, slow medium).
methods were employed by using the heating sheet and 300-kW-pro­ The findings discussed in this study could contribute mainly to the
pane-burner, respectively. These setups represent a typical scenario of activities of first responders to BEV fire accidents and secondarily to
thermal runaway accidents and a worst-case scenario of fuel-spilled car revisiting the safety of existing or new designs of car-parking related
accidents, respectively. structures. For the first responders, BEV fires originating from thermal
All the specimens were burnt out completely under the calorimetry runaway of LIB packs would be more hazardous than those derived from
hood without any enclosure structure additionally installed (i.e., open- elsewhere, due to late human awareness of flames and their rapid
space environment). Under the condition, the BEV fires continued up development once ignited. For the reconsideration of safety, approxi­
until 70 min. Their peak heat release rates (pHRR) were measured to be mately 30 MJ/kg of the average effective heat of combustion can be used
in the range of 6.51 to 7.25 MW, which were slightly lower than ICEV’s to quantify BEV fires. In the context of this, a LIB pack has a high
(7.66 MW) but higher than FCEV’s (5.99 MW). The total heats released property (approximately 45.9 MJ/kg) which is comparable to that of
(THR) from the BEV fires were measured to be in a range of 8.45 to 9.03 combustible fuels (e.g., n-Pentane with 45.69 MJ/kg).
GJ, which were also located in between than smaller ICEVs’ (8.08 GJ) Funding
and greater FCEV’s (10.82 GJ). The author(s) disclosed receipt of the following financial support for
In the BEV fires, the major contribution to the quantity of heat the research, authorship, and/or publication of this article: This project
release rate was determined by the combustion of the conventional is funded by NFA(National Fire Agency) and KEIT(Korea Evaluation
materials of the BEV body (pHRR: 7.81 MW, THR: 7.53 GJ), rather than Institute of Industrial Technology) through R&D programme on Devel­
by that in the LIB pack (pHRR: 1.54 MW, THR: 1.30 GJ). However, as a opment of Fire Safety Technologies for Emergency Response to Fire
jet fire intensively discharged from the LIB pack, it accelerated flame Hazards (No. 20008021).
spreading to adjacent combustible components, thereby leading to a
rapid growth of whole car-fire. The growth parameter for this BEV fire

15
S. Kang et al. Applied Energy 332 (2023) 120497

Fig. A2. Photos of the specimen status at the critical events in Test 4.

CRediT authorship contribution statement References

Sungwook Kang: Methodology, Investigation, Data curation, Visu­ [1] R. Matulka, The history of the electric car. In: Department of Energy, https://www.
energy.gov/articles/history-electric-car, Assessed on 23 Mar 2022.
alization, Software, Writing – original draft. Minjae Kwon: Data cura­ [2] R. Irle, Global EV sales for 2021. In: EV Volumes.com, https://www.ev-volumes.
tion, Formal analysis, Project administration. Joung Yoon Choi: com/country/total-world-plug-in-vehicle-volumes/, Assessed on 23 Mar 2022.
Funding acquisition, Formal analysis, Resources, Project administration, [3] J. Edmondson, A. Holland, R. Collins, L. Gear, Electric car sales, models &
technologies database, IDTechEx, https://www.idtechex.com/en/research-report/
Supervision. Sengkwan Choi: Funding acquisition, Conceptualization, electric-car-sales-models-and-technologies-database/824, Assessed on 23 Mar
Methodology, Validation, Writing – review & editing. 2022.
[4] B. Johnson, Sony flamed for battery bungle, https://www.theguardian.com/
technology/2006/aug/31/sony.guardianweeklytechnologysection, Assessed on 11
Declaration of Competing Interest Nov 2022.
[5] S. Gibbs, Samsung suspends sales of Galaxy Note 7 after smartphones catch fire,
The authors declare that they have no known competing financial https://www.theguardian.com/technology/2016/sep/02/samsung-recall-galaxy-
note-7-reports-of-smartphones-catching-fire, Assessed on 11 Nov 2022.
interests or personal relationships that could have appeared to influence
[6] Y.H. Kim, How to resolve electric vehicle fires, National Fire Research Institute of
the work reported in this paper. Korea, In: Fire Protection News, https://www.fpn119.co.kr/171590, Assessed on
23 Mar 2022.
Data availability [7] Mangs J, Keski-Rahkonen O. Characterization of the fire behaviour of a burning
passenger car. Part I: Car fire experiments, Fire Safety Journal 1994;23:17–35.
[8] Shipp M, Spearpoint M. Measurements of the severity of fires involving private
Data will be made available on request. motor vehicles. Fire Mater 1995;19:143–51.
[9] D. Joyeux, J. Kruppa, L.-G. Cajot, J.-B. Schleich, P. van de Leur, L. Twilt,
Demonstration of real fire tests in car parks and high buildings, European
Appendix Commission, Technical Steel Research, Final Report, EUR-29466 EN, 2002.
[10] Y. Shintani, N. Kakae, K. Harada, H. Masuda, W. Takahashi, Experimental
See the Figs. A1 and A2. investigation of burning behavior of automobiles, 6th Asia-Oceania Symposium on
Fire Science and Technology, Daegu, Korea, 17-20 March 2004.

16
S. Kang et al. Applied Energy 332 (2023) 120497

[11] Okamoto K, Watanabe N, Hagimoto Y, Chigira T, Masano R, Miura H, et al. Burning [19] Sun P, Huang X, Bisschop R, Niu H. A review of battery fires in electric vehicles.
behavior of sedan passenger cars. Fire Saf J 2009;44:301–10. Fire Technol 2020;56:1361–410.
[12] N. Watanabe, O. Sugawa, T. Suwa, Y. Ogawa, M. Hiramatsu, H. Tomonori, H. [20] C. Huggett, Estimation of rate of heat release by means of oxygen-consumption
Miyanoto, K. Okamoto, M. Honma, Comparison of fire behaviors of an electric- measurements, Fire and Materials 4(2):61-65.
battery-powered vehicle and gasoline-powered vehicle in a real-scale fire test, 2nd [21] ISO 24473:2008, Fire tests – Open calorimetry – Measurement of the rate of
International Conference on Fires in Vehicles - FIVE 2012, pp.195-205, Chicago, production of heat and combustion products for fires of up to 40 MW, International
USA, September 27-28, 2012. Organization for Standardization, 2008.
[13] Tohir MZM, Spearpoint M. Distribution analysis of the fire severity characteristics [22] Bryant RA, Bundy MF. The NIST 20 MW calorimetry measurement system for
of single passenger road vehicles using heat release rate data. Fire Sci Rev 2013;2: large-fire research, NIST Technical Note 2077. National Institute of Standards and
5–30. Technology 2019. https://doi.org/10.6028/NIST.TN.2077.
[14] A. Lecocq, M. Bertana, B. Truchot, G. Marlair, Comparison of the fire consequences [23] Hadjisophocleous GV, Mehaffey JR. Fire Scenarios. Chap. 38, SFPE Handbook of
of an electric vehicle and an internal combustion engine vehicle, 2nd International Fire Protection Engineering. fifth ed.,. New York: Springer; 2016.
Conference on Fires in Vehicles - FIVE 2012, pp.183-194, Chicago, USA, September [24] Wang Y, Burgess I, Wald F, Gille M. “Introduction to enclosure fire dynamics” Chap
27-28, 2014. 3. Performance-Based Fire Engineering of Structures: CRC Press Taylor & Francis
[15] Spearpoint M, Tohir MZM, Abu AK, Xie P. Fire load energy densities for risk-based Group NW; 2013.
design of car parking buildings. Case Studies in Fire Safety 2015;3:44–50. [25] Drysdale D. “Fire Science and Combustion” Chap 1, An Introduction to Fire
[16] C. Lam, D. MacNeil, R. Kroeker, G. Lougheed, G. Lalime, Full-scale fire testing of Dynamics third edi. John Wiley & Sons West Sussex UK; 2021.
electric and internal combustion engine vehicles, 4th International Conference on [26] Walters RN, Hackett SM, Lyon RE. Heats of combustion of high temperature
Fire in Vehicles, Baltimore, USA, October 5-6, 2016. polymers. Fire Mater 2000;24(5):245–52.
[17] Dahlberg M. The SP industry calorimeter for rate of heat release measurements up [27] Mei J, Liu H, Chen M. Experimental study on combustion behaviour of mixed
to 10 MW. SP Report 1992;1992:43. carbonate solvents and separator used in lithium-ion batteries. J Therm Anal
[18] R.A. Bryant, M.F. Bundy, The NIST 20 MW calorimetry measurement system for Calorim 2020;139:1255–64.
large-fire research, NIST Technical Note 2077, 2019.

17