Journal of Energy Storage 68 (2023) 107785

Contents lists available at ScienceDirect

Journal of Energy Storage

journal homepage: [Link]/locate/est

Research papers

Impact of different Li-ion cell test conditions on thermal runaway

characteristics and gas release measurements
Ola Willstrand a, b, *, Mohit Pushp a, Petra Andersson a, Daniel Brandell b
a
RISE Research Institutes of Sweden, Box 857, 501 15 Borås, Sweden
b
Department of Chemistry - Ångström Laboratory, Uppsala University, Box 358, 751 21 Uppsala, Sweden

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

Keywords: The increasing use of lithium-ion batteries requires further efforts in safety testing and evaluation. It is of utmost
Gas production importance that the effects of different test conditions are understood, particularly for validation of computer
Gas composition models. While plenty of data from thermal runaway tests are available in literature, few are from large test series.
Large format cell
The missing systematic approach to evaluate the impact of different test conditions implies uncertainty when
Trigger methods
comparing test results. In addition, the fast pace in cell development, including an increasing utilization of larger
Test apparatus
State of charge cells, necessitate the validation of previously published results. This work presents thermal runaway data from 37
tests on one type of large format prismatic lithium-ion cell (157 Ah). The tests are conducted in a closed pressure
vessel with inert atmosphere as well as in an open setup below an exhaust collector hood. Further, six different
thermal runaway trigger methods are employed as well as four different states of charge. Emphasis is put on the
gases produced, a key aspect for safety evaluation. The results are compared with literature data and a new
modified method is proposed for calculating the characteristic venting rate in a closed pressure vessel. It is
concluded that the trigger method affects the gas production rate, mass loss, and maximum temperature of the
cell as much as its state of charge. The large cell format potentially impacts the specific total gas production and
enhances the effects of different trigger methods, but has a small impact on other evaluation parameters. No
significant differences were observed in the test results due to the different test setups, apart from differences due
to potential combustion of the released gases in ambient atmosphere.

1. Introduction A key aspect for safety evaluation is to determine what gases are
evolved during TR. While this has indeed been researched in recent
The ongoing electrification and increased use of Li-ion batteries has years, compared to the total number of papers focusing on TR, relatively
put an emphasis on battery safety, particularly since battery fires can be few are available that include relevant data on gas composition and gas
dramatic. This is mainly due to the use of flammable organic electrolytes production during TR [14–34]. The gas components that have been
in these batteries and the potential for accelerating rate of exothermic targeted differ in these studies, but for battery vent gas at TR in inert
reactions that take place during thermal decomposition of the battery atmosphere, there is consensus that the main constituents comprise H2,
materials, called thermal runaway (TR). Several extensive reviews have CO2, CO, and some hydrocarbons (such as CH4, C2H4, C2H6 and C3H6).
in recent years outlined the basics of TR behaviour, potential causes and In wet gas, there will also be condensable components such as electro­
consequences, as well as factors affecting the TR development [1–13]. lyte solvent vapor and water. There will, however, additionally be
However, a more profound understanding of failure events requires hundreds of other gas components that can be of toxicity concern, even
more data, while standards and test methodologies require a continuous in small concentrations. Several studies have focused on the character­
development. The impacts on results from different test conditions, such ization of these toxic gases, mainly from battery fires, i.e. when most of
as test setup, test scenario, and size of test object, have not been fully the flammable gases were combusted, as opposed to in inert atmosphere
understood. It is important to investigate such impacts to ensure accu­ [13,35–46]. Due to the larger number of potential components to search
rate validation and to be able to develop good computer models, which for, many different species are reported in different studies; especially a
are essential to avoid many large-scale and costly tests. large range of organic compounds. The most often reported gas

* Corresponding author.
E-mail address: [Link]@[Link] (O. Willstrand).

[Link]
Received 9 February 2023; Received in revised form 3 May 2023; Accepted 20 May 2023
Available online 1 June 2023
2352-152X/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ([Link]
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

components from battery fires that are of toxicity concern are CO, HF, cells (20-100 Ah) [16,17,22,27,33–35,37,38,43,72–78]. Only a few tests
HCl, HCN, NOx, and SO2. Moreover, Diaz et al. [39] concluded that the have been found where a battery cell of >100 Ah in rated capacity has
TR vent gas in inert atmosphere is more toxic than after combustion of been used [20,26,46], which motivated this present work. This study
the gases, even though the amount of toxic gas is lesser without com­ therefore focus on whether the results and conclusions would be the
bustion. The most toxic compounds are according to them HF, COF2, same for the large format cells, when comparing with already published
acrolein, CO, formaldehyde, HCl, and electrolyte solvents. data for smaller cells.
A significant amount of data on produced flammable gases is derived This study focuses on the impact of test setup, SOC, and trigger
from a research group based in Austria (Golubkov, Essl and co-workers). method on TR characteristics and gas release. There are several previous
Overall, they have concluded that the cell type [14–16], the state of studies that have focused on the impact of SOC
charge (SOC) [15,17], and the trigger method [16] have great influence [1,15,17,24,38,48,54,59–61,64,74–76], but few have related this
on the amount of produced gas, venting rate, and gas composition, as impact to the influence of other parameters, other than cell type. To
well as on the onset temperature and maximum temperature increase. delimit the test series in this study, the cell type was not varied, neither
This is important since it affects the risk of thermal runaway propagation material composition nor format/size, and no aged cells were used.
and fire spread. Further, while test repeatability is generally good [16], However, a comparatively large format cell was used, which comple­
some variations can be large, for example the occurrence of internal ments previous studies on smaller cells. In this context, the aim was to
short circuits at different stages during heating [22]. An early internal address open questions such as: What are the effects of the atmosphere
short circuit can result in thermal runaway without the presence of a and boundary conditions during TR tests? How large are the effects of
smaller first venting, which is otherwise typical during the slow heating different TR trigger methods in comparison to e.g. different SOC?
process of a battery cell. Golubkov, Essl et al. [16,22] have also devel­
oped a methodology for gas production rate estimation based on the 2. Methodology
pressure curve in a closed setup that is used and further discussed in this
paper. Kennedy et al. [24], in turn, distinguish between wet and dry vent A total of 37 tests were conducted on one type of large format pris­
gas, where the former includes condensable components. From their test matic Li-ion cell (NMC811/C). The rated cell capacity was 157 Ah,
series, including arrays of cells, they conclude that the total amount of which is, to our best knowledge, larger than what has previously been
produced gas and the venting rate are affected by the cell SOC. Koch published for NMC cells. The tests were performed in two different
et al. [27] have presented an impressive test series with 51 tests on cells setups: a closed pressure vessel with inert atmosphere and an open setup
in the range 20–80 Ah. They observed that there is an independency below an exhaust collector hood. The two setups are further described
between gas composition and total gas volume, while gas volume, onset below, together with associated measurements. Further, six different TR
temperature and mass loss are correlated to cell capacity and energy trigger methods were employed as well as four different SOC levels.
density. The higher the energy density, the lower will the TR onset Many of the tests were repeated once to be able to separate parameter
temperature be and the higher will the mass loss be. Further, the cell variations from repeatability uncertainties.
capacity is proportional to the total amount of produced gas but will not
affect the gas composition. 2.1. Thermal runaway trigger methods
In the numerous papers published on the topic of thermal runaway
(TR) of Li-ion batteries, a substantial number of experiments have been The different trigger methods are visualized in Fig. 1 and further
performed. However, large test series are still rare and since the test specified in Table 1 where the total number of tests conducted for each
methodology differs significantly between different studies, it is hard to method is given. For the two heating ramps, the cell was placed in be­
make accurate comparisons to evaluate the influence of different con­ tween two large aluminium plates heated by several evenly embedded
ditions, especially since cell type and cell materials always differ, which cartridge heaters. Ceramic insulation was wrapped around the plates to
usually have a great impact on the results. Eliminating variations due to limit heat losses and the power needed to maintain the specified heating
different test setups and methodologies, the Battery Failure Databank ramp. The ramp was specified for the temperature of the aluminium
provided by NASA [47] is a great resource, where results from about 260 plates to provide fast temperature feedback and stable temperature in­
tests are available to date. The databank covers calorimetric data crease of the cell.
focusing on internal heat generation during TR, but lacks data on e.g. gas The local heater is a patented TRIM (Thermal Runaway Initiation
production and gas composition. Furthermore, the databank covers Methodology) device developed by the National Research Council of
today mainly small cylindrical cells. Duh et al. [6] have reviewed data Canada (NRC) [79]. By locally and rapidly increasing the temperature of
from tests performed only on 18,650 cylindrical format cells, which the cell's active material above a critical level, TR is initiated without
eliminates the influence of cell format. From this analysis, they could preheating the complete cell or neighbouring cells (in case of propaga­
present generic safety differences between LFP, NMC, LCO, and NCA tion tests). The device is ramped by pulsed heating to a temperature set-
type cells. The variations within these groups can still be substantial, point that ideally should be lower than the cell wall failure temperature.
especially if different generations of cells and material development are In this test series, a ramp of 50 ◦ C/s and a set-point temperature of
considered. In general, it is difficult to collect and compare tests from 500 ◦ C were used to be able to initiate TR as fast as possible and still be
different studies, when too many conditions are varied. It is unclear how below the melting temperature of aluminium. To minimize the time for
much different test parameters are affecting the results, even though it is heat transfer, the plastic covering of the cell was removed at the location
evident from literature that the cell chemistry together with the SOC of the heating element. In addition, the element was soaked in a nickel-
have a great impact. based heat transfer paste before being attached to the cell.
The fast-changing battery market with constantly updated material In the nail penetration tests, a pneumatically operated steel nail, 4
compositions and cell formats require new experimental data both to mm in diameter, was rapidly injected in centre of the cell side wall. The
validate findings from previous studies as well as to gain new knowl­ penetration depth was assured to be at least half of the cell thickness.
edge. For example, the published experimental work, to date, covers The nail was fed through a steel plate fixture and, closest to the cell,
mainly small cells, while the trend for some applications and product through ceramic insulation (in 4 out of 5 tests) to limit heat transfer to
suppliers is to use increasingly larger cells. Many tests have been per­ the pneumatic nail rig. In the overcharge tests, a constant charging
formed on cylindrical 18,650 format cells current of 13 A was used until TR occurred. This was the maximum
[14,15,18,25,28–31,40–42,47–67], and also on other small formats (< current, limited by the cable size available in the closed pressure vessel
20 Ah) [13,24,32,36,39,44,68–71]. Tests on larger format cells are at the time. The overcharge tests were only performed in the closed
however less frequent, and can generally be classified as medium-sized setup. Finally, for the external fire, a propane burner with a defined

2
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

Fig. 1. The different TR trigger setups used in the tests.

Table 1
Specification of TR trigger methods.
TR trigger Closed Open Fixture
setup setup

1 ◦ C/min 6 tests 2 tests 2 electrically heated aluminium plates

heating ramp on each side of the cell, ceramic
insulation on the outside of the plates
20 ◦ C/min 3 tests 7 tests 2 electrically heated aluminium plates
heating ramp on each side of the cell, ceramic
insulation on the outside of the plates
Local heater 2 tests 2 tests 21.5 cm2 heating element on one side,
(50 ◦ C/s → steel plate fixture, ceramic insulation
500 ◦ C) between element and steel plate
Nail penetration 2 tests 3 tests Pneumatic nail rig, steel plate fixture,
ceramic insulation on one side between
cell and steel plate (4/5 tests)
Overcharge 2 tests Charging cables, no extra fixture
External fire 8 tests Propane burner underneath, cell fixed
in between two small brackets on top of
a steel net grating

output of 30 kW was located underneath the cell. For obvious reasons,

the external fire was only used in the open setup.

2.2. Closed/inert setup

Fig. 2. Pressure vessel system setup used for TR tests in inert atmosphere.

The closed setup consisted of an 80-litre pressure vessel, seen in

Fig. 2, rated for up to 32 bar at 200 ◦ C. Once closed, the vessel was 101.325 kPa. The amount of moles of gas released from the cell (ncell ) is
purged with a steady supply of nitrogen, resulting in an inert environ­ obtained by subtracting the moles of nitrogen at the start of the test (nN2 )
ment with <1 % oxygen. Any gases released by the cell during testing from the total number of moles after venting/thermal runaway (n2 ).
accumulated inside the vessel. The pressure inside the vessel was That is
monitored by two pressure sensors, calibrated for 0–4 bar and 0–25 bar, P2 V
( [ ])
ϕO2
respectively. Gas temperature inside the vessel was monitored by two ncell = n2 − nN2 = − n1 ⋅ 1 − , (2)
RT2 100
type K thermocouples. Using the pressure and temperature recordings
within the vessel, the volume of gas released by the cell could be where P2 , V, T2 and n1 correspond to the final pressure within the vessel,
calculated. From the ideal gas law, the gas volume at normal tempera­ the inner free volume of the vessel, the final temperature within the
ture and pressure (NTP) becomes vessel, and the moles of gas at initiation of the test, respectively. Note
ncell RTNTP that ϕO2 corresponds to the oxygen concentration at initiation of the test,
Vcell = , (1) which was <1 %. By subtracting the initial oxygen, and assuming that all
PNTP
oxygen is consumed in the test, all reaction products are included in the
where R is the universal gas constant, TNTP = 293.15 ◦ C and PNTP = battery vent gas calculation.

3
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

Golubkov et al. [22] introduced a characteristic venting rate (V̇ ch ),

defined by the minimal time duration when the pressure rises by 50 %,
assuming that 50 % of the gas is produced during this time. In other
words, it is assumed in this calculation that TR venting only occurs
between the start of the pressure increase and until the maximum
pressure in the vessel is reached. This could be somewhat coarse, since
gas production generally extends for a longer time but at a slower pro­
duction rate, whereby pressure is decreasing due to cooling of the gases
inside the vessel. Fig. 3 shows a simplified result graph from one of the
tests conducted. The cumulative gas volume curve (at NTP) is calculated
using Eq. (1) at every time step, and the total volume of gas (Vcell ) is
obtained from the flat steady state value after TR. In the early phase, the
curve is distorted due to an inhomogeneous temperature distribution
inside the vessel at high venting rates, why real gas volume and actual
venting rate cannot be determined in this time period. This is the reason
for defining the characteristic venting rate. However, as indicated by red
in the graph, it is possible to make a more realistic estimation of the total
volume of gas that has been produced at the time when maximal pres­ Fig. 4. Placement of thermocouples: on main side and short side surfaces, on
sure is obtained. In this paper, that estimated gas volume (VP,max ) is used negative terminal, and near the safety vent of the cell (gas temperature).
in the calculation of the characteristic venting rate:
/ cell were weighed using a scale. In addition, the remaining particulate
V̇ ch =
VP,max 2
, (3) matter in the vessel was collected and weighed after the tests.
minΔt

where minΔt is the minimal time duration when the pressure rises by 50 2.3. Open setup
%.
After TR, gases were extracted from the vessel into gas bags. The The open setup consists of a steel net cage centred underneath a gas
contents of the bags were analysed regarding their composition of per­ collector hood and exhaust duct, as seen in Fig. 5. The cell was posi­
manent gases and hydrocarbons. The gases were analysed by either an tioned inside the cage with the safety vent directed upwards. Three sides
Agilent 490 Micro-GC gas analyser, or X-STREAM and BINOS 100 in­ of the steel net cage were covered by non-combustible insulation boards,
struments including nondispersive infrared (NDIR) sensor for CO, CO2, to better direct gases upwards in case of side venting from the cell. The
and CH4, and an electrochemical oxygen analyser. Both instruments flow rate in the duct system was between 0.5 and 0.9 m3/s for the
were calibrated using a synthetic mixture of gases similar to the gases different tests. An almost laminar flow was rendered by a set of guide
expected from TR; 2 % H2, 2 % CH4, 4 % CO, 12 % CO2, and 80 % N2. vanes and by employing sufficient length of the duct ahead of mea­
The uncertainty was about 2 % of absolute values. Further, a HY- surements. Fourier Transform Infrared (FTIR) spectroscopy was used for
ALERTA 500 hydrogen detector was used calibrated with the same time resolved gas content analysis of the effluents. The equipment used
synthetic mixture as above (2 % H2). A flame ionization detector (FID) was a Bomem MB100 FTIR instrument with a 0.9 L heated gas cell and a
was used to measure total hydrocarbons (THC) and was calibrated
against a known concentration of 9000 ppm propane.
Each battery cell was instrumented with four type K thermocouples,
as shown in Fig. 4. Three of them were taped to the surfaces and one was
placed near the safety vent of the cell. The voltage level of the cell was
also measured.
At the end of the tests, the remaining gas was evacuated from the
pressure vessel, which was then cleaned. The remainders of the battery
Total gas prod. / Pressure

0 1 2 3
Time (min)
Gas volume Pressure
Fig. 3. Simplified result graph showing pressure inside the vessel and cumu­
lative produced gas volume. It highlights the estimated total gas volume at
maximum pressure in the vessel (VP,max ), which was used instead of the total
gas volume produced from the cell (Vcell ) when calculating the characteristic
venting rate. Fig. 5. The open system setup used for TR tests in normal atmosphere.

4
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

spectral resolution of 4 cm− 1. The sampling for the FTIR was done using sampling in these tests. The analysis focuses on the effect of different test
a heated ceramic particle filter and a 10 m heated PTFE sampling line for setups, different TR trigger methods, and different SOC levels, as well as
a sampling flow rate of 4 L/min. The sampling equipment and the cell if there are differences compared to previous work due to the large cell
were heated to 150 ◦ C. In addition, carbon dioxide and carbon monoxide format used. In addition, since the overcharge TR trigger method
concentrations were also analysed using a X-STREAM NDIR sensor in a changes the SOC, it is difficult to separate impact from change in SOC
parallel sampling line. and impact from overcharge degradation reactions leading to TR. In the
Each battery cell was instrumented with four type K thermocouples, discussion of impact from different TR trigger methods, focus is on
as shown in Fig. 4, and the voltage level of the cell was measured. The comparing the different TR trigger methods that can be associated with a
battery cell was weighed before and after the test using a scale, and for certain SOC level.
all TR trigger methods except the external fire the cell was positioned on
top of a scale during the tests for time-resolved mass loss data.
3.1. Gas production
3. Results
The total gas volume, calculated based on temperature and pressure
An overview of the tests performed and some important evaluation measurements in the closed setup, divided by the rated capacity of the
parameters are shown in Table 2. Note, for the heating ramp method, tested cell is shown in Fig. 6a. As seen, the repeatability is good and
some differences in heating time of the cell were caused by the differ­ there is no significant influence from the TR trigger method. In contrast,
ences in contact area between the plates and the cell, e.g. due to swelling the effect of SOC is clear with a linear trend of increasing amount of gas
of the cell during the tests. It would be recommended to use a heat produced with increasing SOC.
transfer paste, as was done for the local heater method, to increase The characteristic gas production rate during venting is shown in
repeatability. A couple of tests resulted in cell case explosion, affecting Fig. 6b; this is defined by Golubkov et al. [22] but modified according to
some of the evaluation parameters, such as temperature recordings, the discussion above in the description of the setup. With the modifi­
mass loss, and calculation of gas production rate. This explains why a cation used in Eq. (3), the characteristic gas production rate becomes
few tests have been excluded in some of the result graphs. Note that the 20–40 % lower if VP,max is used instead of Vcell in the equation. Due to the
cell case explosions could be due to mechanical restrictions caused by estimation of VP,max the uncertainty is increased, but the characteristic
the fixtures in the test setup. In addition, with the external fire as TR venting rate should be closer to the actual gas production rate.
trigger method, most cells ventilated through the side wall or bottom Compared to the unmodified characteristic venting rate, the variation
due to weakening of the aluminium casing. This affected the gas for each TR trigger method is relatively low. With a 1 ◦ C/min heating
ramp, the modified venting rate was 67 ± 6 % (one standard deviation,

Table 2
Test series and results overview, including time until thermal runaway, cell temperatures (at start of thermal runaway and maximum), mass loss, and specific total gas
production.
Test nr. Trigger method Setup SOC Time to TR Cell temp. at start of TR (◦ C) Max. temp. (◦ C) Mass loss Spec. total gas prod. (L/Ah)

5 1 ◦ C/min Closed 25 % 2.7 h 185 425 18 % 0.35

2 1 ◦ C/min Closed 50 % 2.6 h 179 515 43 % 0.71
1 1 ◦ C/min Closed 75 % 2.3 h 162 571 68 % 1.38
11 1 ◦ C/min Closed 75 % 2.4 h 170 508 71 % 1.44
6 1 ◦ C/min Closed 100 % 2.1 h 150 612 78 % 2.17
8 1 ◦ C/min Closed 100 % 2.1 h 148 614 75 % 2.10
22 1 ◦ C/min Open 100 % 2.2 h 160 451 77 % N/A
29 1 ◦ C/min Open 100 % 2.1 h 151 610 72 % N/A
16 20 ◦ C/min Closed 50 % 13.4 min 187 591 21 % 0.73
3 20 ◦ C/min Closed 100 % 14.0 min 139 786 53 % 2.12
13 20 ◦ C/min Closed 100 % 10.0 min 154 716 54 % 2.08
19 20 ◦ C/min Open 25 % 14.3 min 178 449 18 % N/A
35 20 ◦ C/min Open 25 % 17.2 min 206 482 20 % N/A
26 20 ◦ C/min Open 50 % 14.3 min 191 572 26 % N/A
34 20 ◦ C/min Open 50 % 16.5 min 181 619 23 % N/A
24 20 ◦ C/min Open 75 % 14.1 min 166 627 34 % N/A
30 20 ◦ C/min Open 75 % 16.1 min 168 642 33 % N/A
28 20 ◦ C/min Open 100 % 11.8 min 136 617 50 % N/A
9 30 kW burner Open 25 % 4.3 min N/A 639 21 % N/A
15 30 kW burner Open 25 % 6.0 min N/A 711 20 % N/A
10 30 kW burner Open 50 % 4.5 min N/A 744 24 % N/A
12 30 kW burner Open 50 % 4.0 min N/A 788 23 % N/A
4 30 kW burner Open 75 % 1.6 min N/A 814 32 % N/A
14 30 kW burner Open 75 % 2.7 min N/A N/A 31 % N/A
7 30 kW burner Open 100 % 2.5 min N/A 684 62 % N/A
17 30 kW burner Open 100 % 2.5 min N/A N/A N/A N/A
33 Local heater Closed 100 % 10 s Ambient 624 45 % 2.21
36 Local heater Closed 100 % 12 s Ambient 696 33 % 2.11
37 Local heater Open 25 % 18 s Ambient 442 17 % N/A
32 Local heater Open 100 % 12 s Ambient 889 35 % N/A
27 Nail penetration Closed 100 % <1 s Ambient 838 42 % 2.06
31 Nail penetration Closed 100 % <1 s Ambient 857 41 % 2.09
25 Nail penetration Open 25 % 11 s Ambient 395 21 % N/A
20 Nail penetration Open 100 % <1 s Ambient 665 54 % N/A
23 Nail penetration Open 100 % <1 s Ambient 725 43 % N/A
21 Overcharge Closed 128 % N/A 104 641 77 % 2.57
18 Overcharge Closed 131 % N/A 80 N/A N/A 2.64

N/A means not applicable, or where reliable results were not available.

5
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

Fig. 6. a) Specific total gas production at normal temperature and pressure (NTP), b) characteristic (max) gas production rate, and c) 30 s average gas production
rate calculated for tests performed in the closed setup.

6
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

1 SD) of the unmodified rate, with a 20 ◦ C/min heating ramp, the value measurements, where the latter are presented in Fig. 8. As seen, the tests
was 80 ± 2 %, and with the local heater, nail penetration, and over­ in the open setup where no ignition of the released gases occurred, at 25
charge, the combined value was 63 ± 2 %. This means that the effect on % and 50 % SOC, displayed good correlation with the gas analysis of CO
venting rate from different TR trigger methods was not diminished by and CO2 in the closed and inert setup. When combustion of the gases
the modification, due to estimation of VP,max . occurs, most of the CO is turned into CO2, as expected. CO2 is also
As shown in Fig. 6b, the effect of different TR trigger methods can be produced from combustion of present hydrocarbons. The proportion of
almost as large as different SOC on the gas production rate. The more the CO relative to CO2 is a measure of the combustion efficiency, and the
cell is preheated before reaching TR, as when using slow heating ramps, results show the lowest ratio (i.e. the highest combustion efficiency) for
the higher the characteristic gas production rate will be. The charac­ the external fire trigger method as well as for the 20 ◦ C/min heating
teristic rate will be close to the maximum gas production rate, especially ramp, at 75 % SOC. There are several explanations for this; the higher
at high SOC, which in this case was approximately 10 times higher than temperatures gained by both higher SOC and certain TR trigger methods
the 30 s average gas production rate, shown in Fig. 6c. The effect of will generally increase combustion efficiency, while a high venting rate
preheating the cell before TR is less significant for the 30 s average gas and thereby availability of oxygen to the hottest regions may also
production rate. Reproducibility was good, except for the two tests decrease the combustion efficiency. The 1 ◦ C/min heating ramp had the
performed with the local heater device. However, the times until TR lowest combustion efficiency as compared to the other trigger methods,
from the start of the heating ramp were 10 s and 12 s, respectively, and due to the high venting rate at 100 % SOC (see the characteristic venting
the times from start of TR to the maximum cell temperature were 90 s rate presented in Fig. 6b for tests in the closed setup).
and 95 s, respectively. This shows good repeatability for the triggering Other toxic gases which the FTIR was calibrated for, such as SO2, NO,
time and for the heat increase rate during TR for these two tests. Dif­ NO2, HCl, and HCN were either not detected or not quantifiable in this
ferences in the gas venting rate, as well as particle mass loss (Fig. 9), can study, but have been reported in other battery cell tests [36,46,80]. It is
possibly be attributed to differences in venting area from the cell, since
parts of the aluminium cell casing were melted at 100 % SOC.
600

Total mass of CO 2 (g)

3.2. Gas composition
500
The heating ramp TR trigger method (1 C/min and 20 C/min)
◦ ◦

applied in the closed setup provided quantifiable gas composition data,

400
and the average gas composition from these tests is presented in Fig. 7. It
can then be seen that hydrogen (H2) and carbon monoxide (CO) increase
300
with SOC, while carbon dioxide (CO2) decreases. The total hydrocarbons
(THC) detected by the FID showed no clear trend with respect to SOC,
200
and the same applied to the methane (CH4) component, for which
100
approximately 50–60 wt% was detected by the BINOS 100 instrument.
Micro-GC analysis performed at a later stage showed the same trends
0
with respect to CO, CO2, and THC with respect to SOC, and displayed no
significant differences depending on TR trigger method. However, H2 25% 50% 75% 100%
had most likely leaked out of the gas bags, why this data is not presented
in further detail. State of charge
In the open setup, only CO, CO2 and hydrogen fluoride (HF) were
detected by the FTIR in the smoke exhaust channel. HF was only Closed/Inert Open (no fire) Open (fire)
detected in tests with external fire as trigger method, as well as in two
tests with the heating ramp, both at 25 % SOC. The concentrations of HF
were comparatively low, and the maximum total amount detected in a 200
single test was 4.2 mg/Wh. This can be compared to the 6 and 16 mg/
Total mass of CO (g)

Wh detected for two different cells by Sturk et al. [44], approximately

40–70 mg/Wh detected in cell tests by Ribiére et al. [36], or 20–200 mg/ 150
Wh detected in tests with several different cells by Larsson [13]. The
measurements of CO and CO2 by the FTIR correlated well with the NDIR
100
100%
15% 14% 21% 22%
80% 10% 12% 50
13% 15% 12%
60% 26%
36%
0
40% 44%
58% 25% 50% 75% 100%
20% 48%
28% 22% State of charge
0%
25% 50% 75% 100% Closed/Inert Open (no fire) Open (fire)
State of charge Fig. 8. Total mass of carbon dioxide and carbon monoxide (averages from all
CO2 CO THC H2 tests) measured by NDIR in the closed and open setup, respectively. At 25 %
and 50 % SOC some of the tests in the open setup resulted in ignition and
Fig. 7. Average gas composition from tests in the closed setup and heating combustion of the released gases and some did not, which is separated in the
ramp as TR trigger method. graphs (fire/no fire).

7
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

possible that some of these gases were present, but there were unspec­ method. The external fire method triggered TR within 1.6–6 min (from
ified compounds overlapping in the spectral response such that no high to low SOC), however, the heat flux to the cell was in this method
quantifiable data could be retrieved for these gases. larger than with the heating ramp method.
At 25 % SOC, no particulate matter was found around the cell after
the test, indicating that approximately 20 % of the cell weight was
3.3. Mass loss turned into gases and vapor. In fact, subtracting the particulate matter
from the total mass loss (possible in the closed setup where particulate
The cell mass loss is shown in Fig. 9, including both gas and particle matter was vacuumed and weighed separately) shows that the mass loss
mass loss. Note, for the highest mass losses (>70 %) there were sub­ due to vaporisation and gas production were 23 ± 2 % (1 SD), inde­
stantial melting of the aluminium casing, allowing most of the cell pendent on the SOC. Based on this, what can explain the large differ­
content to be ejected. There were effects on the results from both SOC ences in total gas production due to different SOC, shown in Fig. 6a? The
and TR trigger method, but there was no significant influence from the total gas production is calculated based on the ideal gas law, which
different test setups, including atmosphere, pressure, and temperature means that one mole of gas will have the same volume no matter the
differences of the surroundings. At 100 % SOC, there were some vari­ molecular size and weight. The total gas production dependence on SOC,
ances with the same TR trigger method that can be attributed to dif­ presented in Fig. 6a, should therefore be directly related to differences in
ferences in venting area, as also mentioned regarding the gas molecule sizes and the extent of reactions. Using the gas composition of
production. However, it should also be noted that for the nail penetra­ the main reactions' end products, presented in Fig. 7, and assuming that
tion tests there was a difference in application of ceramic insulation for the THC consists of only methane, the resulting mass losses would be 3
one of the tests, resulting in a difference in heat spread for this test. This %, 7 %, 11 %, and 16 % for 25, 50, 75 and 100 % SOC, respectively. The
was the test with the highest mass loss at 100 % SOC, deviating from the differences between these percentages and the actual mass loss of
other tests. approximately 23 % must thereby originate in larger and heavier mol­
For a given state of charge, the variation in mass loss for the three ecules present in the gas. The relative amount of these other molecules
heating ramp methods correlated with the variation in heat energy decreases with SOC and total gas volume as the extent of reactions in­
input. Considering conductive heat transfer, the input of heat energy is creases. Note that with 25 % and 50 % SOC in the closed setup,
proportional to the temperature gradient and time. The average tem­ condensed liquid was detected in the setup when cooled down to room
perature gradient is approximately the same while the time until TR temperature.
where approximately 10 s for the local heater, 10–17 min for the 20 ◦ C/
min heating ramp, and 2–2.7 h for the 1 ◦ C/min heating ramp. The
3.4. Temperatures
higher the input energy, the higher the mass loss. In addition, the fast
response of the local heater, which limits heat transfer to other parts of
TR onset temperatures from the heating ramp tests are shown in
the cell, means that the results would be similar to the nail penetration
Fig. 10a. The thermocouple on the cell terminal was used to have a good
estimation of the cell internal temperature. The criterion for TR was set
80% to when the temperature rate of rise exceeded 1 ◦ C/s, as defined in UN
GTR 20 [81] (another limit sometimes used is >100 ◦ C/min [6,76]). The
repeatability was good and there was no significant influence from the
70% different test setups or the different heating rates. Note, some variances
in onset temperature may have resulted from differences in contact area
between the heating plates and the cell during the tests, resulting in
60%
different temperature gradients. The larger the temperature gradient,
the lower average cell temperature is expected at onset of TR. As seen in
50% Table 2, the local heater and nail penetration method rendered a TR due
to locally high temperature which did not affect the measured average
Mass loss (%)

cell temperature at all at the TR triggering time. The slower heating

40% ramps had much smaller temperature gradients and the small differ­
ences in measured onset temperatures using either 1 ◦ C/min or 20 ◦ C/
min heating ramp, indicate that TR onset temperatures for an adiabatic
30% and perfectly uniform temperature distribution test would be similar to
presented results in Fig. 10a. At least, the TR onset temperature was
clearly decreasing with higher SOC, which is in line with previously
20% published data [1,54,59,61,75,76]. A linear fit to the data gives 18 ◦ C
decrease of TR onset temperature per 25 % SOC increase.
Fig. 10b shows the maximum surface temperatures during TR. This is
10%
the maximum temperature from any of the thermocouples. The varia­
tion is larger compared to TR onset temperatures, but the trends are
0% evident: increasing temperatures with increasing SOC, and an obvious
0% 25% 50% 75% 100% 125% 150% influence from the TR trigger method. A linear fit to the data gives
approximately 90 ◦ C increase in maximum temperature per 25 % SOC
State of charge increase. When comparing the maximum temperatures in Fig. 10b with
1 °C/min, Closed 1 °C/min, Open the mass loss and gas production rate, it seems that a more violent and
20 °C/min, Closed 20 °C/min, Open severe TR (for a certain SOC) gives a lower maximum temperature of the
Local heater, Closed Local heater, Open cell casing. More heat is lost to the surroundings due to the higher mass
Nail penetra on, Closed Nail penetra on, Open loss, and potentially, more of the reactions are happening outside of the
Overcharge, Closed External fire, Open cell casing due to the higher venting rate. Cooling of gases due to the
Joule-Thomson effect when going from higher to lower pressure might
Fig. 9. Cell mass loss percentage of original weight, including both gas and have a small effect as well, but should be limited to a few degrees
particle mass loss. considering e.g. carbon dioxide at reasonable pressures [82].

8
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

Fig. 10. a) Thermal runaway onset temperatures for heating ramp tests, measured on the terminal, and b) maximum cell surface temperature during ther­
mal runaway.

4. Discussion total gas production is one of the highest, which is not surprising since
the NMC811 cathode material used in this study is typically more
The results on total gas production are in line with previously pub­ reactive as compared to other types, for example,NMC111 cathodes
lished data for Li-ion cells, as seen in Fig. 11, presenting the specific total having less nickel. The large cell format may also have contributed to the
gas production from several studies. From the literature data, there are high specific total gas production. A large cell will have a higher total
no clear trends with respect to differences in cell type except for the LFP heat production, and feedback from one part of the cell to other parts
type cells, which for this reason are highlighted in the figure. The total may further prolong the reactions taking place or raise the temperature
gas production is partly dependent on the available material and partly even further. Note that the maximum specific total gas production in
on the extent of reactions, which in turn is dependent on the tempera­ Fig. 11 (3.25 L/Ah) is from a large prismatic cell of 134 Ah [20]. Further,
ture. LFP type cells are generally more thermally stable, having lower Koch et al. [27] present a linear fit of 1.96 L/Ah for 51 cells at 100 %
electrode potentials and a flatter voltage profile, meaning that electrode SOC, ranging from 20 to 81 Ah in cell capacity, which is close but
potentials change very little with SOC. Since the reactivity of the ma­ slightly lower as compared to the result in this study. Both these studies
terials is related to the electrode potentials, the temperature will also be indicate that the cell size contributes to the total gas production.
related to these, and thereby the total gas production. In fact, plotting When it comes to the effect of different TR trigger methods on gas
the total gas production versus cell potential, instead of SOC, renders a production, there is not much to compare with except work by Essl et al.
more similar linear increase for LFP as for the other types. Further, [16]. In their tests, three different TR trigger methods were compared:
Fig. 11 indicates that LFP cells have a higher specific total gas produc­ heating (~2 ◦ C/min), overcharge (1C rate), and nail penetration,
tion at low SOC (0 % and 25 %) compared to the other types. This means applied on two different types of cells; pouch and hard case prismatic.
that there should be more available electrolyte and active material in The characteristic venting rate varied from about 30–250 L/s in these
these cells, which could be attributed to the typically more power tests, affected by both TR trigger method and cell format. It was then
optimized LFP cells leading to larger cells for the same rated capacity concluded that overcharge-triggered TRs can be harsher with respect to
[83]. Heating at low SOC will result in evaporation of the electrolyte, e.g. total gas amount and venting rate, which agrees well with this study
and the extent of reactions will typically be less compared to higher SOC as well as other works [3]. However, the heating ramp trigger method,
where higher reactivity and higher temperatures result in a more severe in those tests by Essl et al., resulted in the lowest total amount of gas and
TR. Note that tests at 25 % SOC in this present study resulted in a typical venting rate, which was in contrast to the results by Larsson et al. [84].
TR event, though less violent when comparing with higher SOC tests. At They found that overcharge and heating can end up in more severe
higher SOC, it seems that the higher electrode potentials and more abuse, due to the input of electric power or heat. The results of this study
reactive material cells have a large impact on the total gas production rather corroborate these early findings by Larsson et al., while they are
and plays a greater role as compared to the total amount of material/ not in full agreement with those from Essl et al. Further, based on the
electrolyte available. In the present study, the SOC dependence on the theory that the input of electric or heat energy prior to TR would result

9
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

sorted on cell chemistry, cell size and format, TR trigger method, test
setup, or total gas production.
The TR onset temperatures presented in this study are in general
lower when comparing with previous studies [1,6,27,54,59,75,76], but
e.g. in [61,63] they report onset temperatures below 150 ◦ C at 100 %
SOC for NMC cylindrical cells. In the compilation done by Duh et al. [6]
on 18,650 cells at 100 % SOC, the average TR onset temperatures are
close to 200 ◦ C for both LCO, NMC, and NCA cells. The onset tempera­
ture should be related to the reactivity of the electrodes, which, in
general, is related to the energy density. Koch et al. [27] have in their
study presented a correlation between onset temperature and volu­
metric energy density of a cell. Although they have not measured onset
temperatures as low as 150 ◦ C, their trend gives that this temperature
correlates well with a high cell energy density of approximately 650
Wh/L, as was the case in the present study. Further, according to Koch
et al., the low onset temperature should also result in high mass loss
(>50 %), which is true in this case for the heating ramp TR trigger
method (Fig. 9). When it comes to maximum temperatures, the compi­
lation done by Duh et al. gives average maximum temperatures of
approximately 700–800 ◦ C for the chemistries mentioned above. There
are large variations, as in the present study, but the averages are similar.
As mentioned in chapter 3.4, a linear fit to the data gives approximately
a 90 ◦ C increase of maximum temperature per 25 % SOC increase. This is
a very similar trend as that presented in [54]. In [74] the corresponding
increase is estimated to be 125 ◦ C per 25 % SOC increase, and in [61,76]
even higher.
As shown for the temperature measurements, mass losses, and CO/
CO2 measurements, the variations in results due to the different test
setups are small, if excluding differences due to ignition of the released
gases. The test setups include differences in atmosphere and surrounding
temperature and pressure during the tests. The largest differences in
surrounding temperature and pressure arose during TR, but there was
Fig. 11. Average values for each SOC level of the specific total gas production also an overpressure of about 20 kPa in the pressure vessel from start of
(total gas volume divided by rated cell capacity) from the present study, in the tests which occurred when purging the vessel with nitrogen. The
comparison with literature data. Non-filled markings are from Golubkov, Essl, effect of pressure differences has been studied by others [62,86,87]
et al. [14–17], and filled markings from other studies [18,20,21]. Tests on LFP focusing on low pressures encountered during e.g. flight transport. Chen
type cells are provided with a different colour. et al. [62] conclude that an absolute pressure of 64 kPa has little impact
on the mass loss, but significant impact on lowering the combustion heat
in a more violent and severe scenario, one would expect, for a given state compared to normal pressure. Wang et al. [86] conclude that low
of charge, that the mass loss would correlate with the input of energy. pressures affect TR onset temperatures, maximum surface temperatures,
This was elaborated in chapter 3.3 and correlates well with the results. mass loss, and toxicity of gases. Contrary, the overpressure differences in
Literature data on the gas composition, sorted by SOC for compari­ this study show no such trends.
son with this study, is summarized in Table 3. The variation between Finally, in general it is reasonable that a large cell format enhances
different tests is large, indicated by the large standard deviations noted. the effects of different TR trigger methods since the total amounts of heat
The found data show the same trends as the present study when SOC is and electrical energy that can be stored are larger. The difference be­
increased, i.e. increasing H2 and CO, decreasing CO2, and approximately tween e.g. a slow heating ramp and nail penetration becomes more
constant THC. However, the present study also has values that are significant if (1) the heat storage capacity of the cell increases, and (2)
outside the intervals indicated in Table 3, e.g. H2 is higher at 25 % SOC the TR initiated by nail penetration is more local in comparison to the
but lower at higher SOC compared to the table, and the CO/CO2 ratio is total volume of the cell. Comparing the characteristic gas production
higher overall. In addition, the average gas composition presented in the rate, the mass loss, and the maximum temperatures, for which the effects
2022 Battery Scorecard by DNV (maritime classification society) [85] of different TR trigger methods were the largest, with results from Essl
shows a much higher percentage of H2 (43 %) compared to the present et al. [16] one can conclude that mass loss and maximum temperatures
study. For the data set presented in Table 3, no trends were found when vary more with the TR trigger method in the present study. The char­
acteristic venting rate is not calculated in the same way but varies
equally. Essl et al. tested medium size cells of 60 Ah and there is a lack of
Table 3
Literature data from [14–21,23–29,34,37,38]. Presented are average values
other studies in open literature where different TR trigger methods are
(±1SD) from 63 tests in total. The majority of tests are conducted at 100 % SOC, compared, especially where gas production characteristics are evalu­
and for lower SOC the values are based on 3–7 tests at each SOC level. ated. Note, when comparing results with published literature data, many
parameters are varied, why comparative experimental studies are rec­
25/30 % 50/60 % 75 % 90/100 % Overcharge
SOC SOC SOC SOC ommended to conclude on the effects of specific differences, e.g. dif­
ferences in cell size. It can, however, be seen that average results
H2 (%) 9 (±4) 19 (±12) 27 (±4) 25 (±14) 28 (±3)
THC 17 (±11) 23 (±15) 18 (±8) 21 (±13) 12 (±7)
normalized with cell capacity in this study do not differ significantly
(%) from previous data on small cells, why the influence of cell size does not
CO (%) 5 (±1) 12 (±12) 16 (±16) 23 (±14) 34 (±16) seem to be larger than the influence from other differences between the
CO2 69 (±12) 46 (±17) 39 (±15) 32 (±17) 27 (±14) cells in different studies.
(%)

10
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

5. Conclusions Data availability

This work presents thermal runaway data from 37 tests on one type The authors do not have permission to share data.
of large format prismatic Li-ion cell. Different TR trigger methods,
different SOC levels, as well as different cell size formats and test setups Acknowledgements
were discussed, and compared to existing literature data. In addition, for
the estimation of gas production rate, a new modified method was We thank David Raymand, Scania CV AB, and Tomas Verhallen and
proposed for calculating the characteristic venting rate in a closed Seungbok Lee, Northvolt Labs, for input on the manuscript. This work is
pressure vessel. The modification renders a more realistic production part of a project funded by the Swedish Energy Agency (project no.
rate, while still enabling good comparison of the venting characteristics 51787-1). Partners within the project comprise of RISE Research In­
affected by e.g. different TR trigger methods or SOC levels. stitutes of Sweden, Northvolt, Scania, and Uppsala University. We also
It was seen that different SOC levels have a large impact on the TR acknowledge support from Batteries Sweden (grant no. Vinnova-2019-
behaviour, ranging from total gas volume, gas production rate, gas 00064), and the StandUp for Energy consortium.
composition, and mass loss, to onset and maximum temperatures. For
example, the specific total gas production went from 0.35 L/Ah to 2.1 L/ References
Ah in the range 25–100 % SOC, TR onset temperature had a decrease of
18 ◦ C per 25 % SOC increase, and the maximum temperature increased [1] P. Jindal, J. Bhattacharya, Review—understanding the thermal runaway behavior
of Li-ion batteries through experimental techniques, J. Electrochem. Soc. 166 (10)
approximately 90 ◦ C per 25 % SOC increase. These trends are in line (Jun. 2019) A2165–A2193, [Link]
with previous studies. The large effect of different TR trigger methods [2] D. Ouyang, M. Chen, Q. Huang, J. Weng, Z. Wang, J. Wang, A review on the
has previously not been shown as clear as in this study, especially when thermal hazards of the lithium-ion battery and the corresponding countermeasures,
Appl. Sci. (Switzerland) 9 (12) (Jun. 2019), [Link]
it comes to gas production rate, mass loss, and maximum temperatures, [3] X. Feng, M. Ouyang, X. Liu, L. Lu, Y. Xia, X. He, Thermal runaway mechanism of
where the effect of different TR trigger methods is as large as the effect of lithium ion battery for electric vehicles: a review, Energy Stor. Mater. 10 (Jan.
different SOC levels. For example, the mass loss at 100 % SOC varied 2018) 246–267, [Link]
[4] Q. Wang, B. Mao, S.I. Stoliarov, J. Sun, A review of lithium ion battery failure
from 35 % to 75 % with different TR trigger methods, while the average mechanisms and fire prevention strategies, Prog. Energy Combust. Sci. 73 (Jul.
increase for 25–100 % SOC was 35 percentage points. Similarly, the 2019) 95–131, [Link]
maximum temperature at 100 % SOC varied almost 400 ◦ C with [5] X. Feng, D. Ren, X. He, M. Ouyang, Mitigating thermal runaway of lithium-ion
batteries, Joule 4 (4) (Apr. 15, 2020) 743–770, [Link]
different TR trigger methods, while the temperature increase between
joule.2020.02.010. Cell Press.
25 and 100 % SOC for a specific TR trigger method was generally less. It [6] Y.S. Duh, et al., Characterization on thermal runaway of commercial 18650
should also be noted that the TR trigger methods did not seem to affect lithium-ion batteries used in electric vehicles: a review, J. Energy Storage 41 (Sep.
the total gas volume produced or the gas composition. Neither did the 2021), [Link]
[7] M. Ghiji, S. Edmonds, K. Moinuddin, A review of experimental and numerical
two different heating rates affect the TR onset temperature. Further, studies of lithium ion battery fires, Appl. Sci. (Switzerland) 11 (3) (Feb. 2021)
there were no significant differences in the test results due to the 1–29, [Link]
different test setups, if excluding differences due to ignition of the [8] J.S. Edge, et al., Lithium ion battery degradation: what you need to know, Phys.
Chem. Chem. Phys. 23 (14) (Apr. 2021) 8200–8221, [Link]
released gases, showing that the atmosphere as well as small differences D1CP00359C.
in surrounding pressure and temperature have a low impact on the [9] L. Bravo Diaz, et al., Review—meta-review of fire safety of lithium-ion batteries:
characterization of a TR event. Moreover, the repeatability was good, industry challenges and research contributions, J. Electrochem. Soc. 167 (090559)
(Jan. 2020), [Link]
both concerning test repetitions without changing parameters and rep­ [10] J. Sun, B. Mao, Q. Wang, Progress on the research of fire behavior and fire
etitions with different test setups. The number of repetitions did not protection of lithium ion battery, Fire Saf. J. 120 (Mar. 2021), [Link]
allow uncertainties to be quantified but they seemed to be much less 10.1016/[Link].2020.103119.
[11] Y. Chen, et al., A review of lithium-ion battery safety concerns: the issues,
compared to the effects of certain parameter variations. strategies, and testing standards, J. Energy Chem. 59 (Aug. 01, 2021) 83–99,
It is reasonable that a large cell format enhances the effects of [Link] Elsevier B.V.
different TR trigger methods. However, this needs further comparative [12] P.V. Chombo, Y. Laoonual, A review of safety strategies of a Li-ion battery,
J. Power Sources 478 (Dec. 2020), 228649, [Link]
studies to be confirmed. It should although be noted that the average
JPOWSOUR.2020.228649.
results, disregarding the effects of different TR trigger methods, agreed [13] F. Larsson, Lithium-ion Battery Safety: Assessment by Abuse Testing, Fluoride Gas
well with previous studies on smaller cells. In this regard, there was no Emissions and Fire Propagation, PhD Thesis, Chalmers University of Technology,
significant influence from the large cell format, except for the specific Göteborg, Sweden, 2017.
[14] A.W. Golubkov, et al., Thermal-runaway experiments on consumer Li-ion batteries
total gas production. Investigating the effect on total gas production with metal-oxide and olivin-type cathodes, RSC Adv. 4 (7) (2014) 3633–3642, doi:
from different sizes of the same cell type would be motivated in future 10.1039%2FC3RA45748F.
studies . [15] A.W. Golubkov, et al., Thermal runaway of commercial 18650 Li-ion batteries with
LFP and NCA cathodes-impact of state of charge and overcharge, RSC Adv. 5 (7)
(2015) 57171–57186, [Link]
CRediT authorship contribution statement [16] C. Essl, A.W. Golubkov, A. Fuchs, Comparing different thermal runaway triggers
for two automotive lithium-ion battery cell types, J. Electrochem. Soc. 167 (13)
(Oct. 2020), [Link]
Ola Willstrand: Conceptualization, Methodology, Formal analysis, [17] C. Essl, et al., Comprehensive hazard analysis of failing automotive lithium-ion
Investigation, Writing – original draft. Mohit Pushp: Investigation, batteries in overtemperature experiments, Batteries 6 (2) (2020), [Link]
Writing – review & editing. Petra Andersson: Conceptualization, 10.3390/BATTERIES6020030.
[18] M. Lammer, A. Königseder, V. Hacker, Holistic methodology for characterisation of
Methodology, Writing – review & editing, Supervision. Daniel Bran­ the thermally induced failure of commercially available 18650 lithium ion cells,
dell: Conceptualization, Writing – review & editing, Supervision. RSC Adv. 7 (39) (May 2017) 24425–24429, [Link]
C7RA02635H.
[19] M. Karp, Heat Rate and Vent Gas Analysis, Federal Aviation Administration, 2019.
Declaration of competing interest
[20] W.Q. Walker, Characterization of the GS YUASA 134 Ah Cell Thermal Runaway
Events With Large Format Fractional Thermal Runaway Calorimetry (L-FTRC),
The authors declare that they have no known competing financial NASA Johnson Space Center, USA, 2020.
interests or personal relationships that could have appeared to influence [21] D.J. Robles, J. Jeevarajan, Fire and Gas Characterization Studies for Lithium-ion
Cells and Batteries, NASA Aerospace Battery Workshop, 2020.
the work reported in this paper. [22] A.W. Golubkov, et al., Thermal runaway of large automotive Li-ion batteries, RSC
Adv. 8 (7) (Nov. 2018) 4172–4186, [Link]

11
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

[23] M. Henriksen, K. Vaagsaether, J. Lundberg, S. Forseth, D. Bjerketvedt, Laminar [50] X. Liu, S.I. Stoliarov, M. Denlinger, A. Masias, K. Snyder, Comprehensive
burning velocity of gases vented from failed Li-ion batteries, J. Power Sources 506 calorimetry of the thermally-induced failure of a lithium ion battery, J. Power
(Sep. 2021), [Link] Sources 280 (Apr. 2015) 516–525, [Link]
[24] R.W. Kennedy, K.C. Marr, O.A. Ezekoye, Gas release rates and properties from JPOWSOUR.2015.01.125.
lithium cobalt oxide lithium ion battery arrays, J. Power Sources 487 (Mar. 2021), [51] M. Chen, et al., Investigation on the thermal hazards of 18650 lithium ion batteries
229388, [Link] by fire calorimeter, J. Therm. Anal. Calorim. 122 (2) (Nov. 2015) 755–763,
[25] L. Yuan, T. Dubaniewicz, I. Zlochower, R. Thomas, N. Rayyan, Experimental study [Link]
on thermal runaway and vented gases of lithium-ion cells, Process Saf. Environ. [52] Y. Fu, S. Lu, K. Li, C. Liu, X. Cheng, H. Zhang, An experimental study on burning
Prot. 144 (Dec. 2020) 186–192, [Link] behaviors of 18650 lithium ion batteries using a cone calorimeter, J. Power Sources
[26] Z. Huang, et al., Experimental investigation on thermal runaway propagation of 273 (Jan. 2015) 216–222, [Link]
large format lithium ion battery modules with two cathodes, Int. J. Heat Mass [53] M. Chen, O. Dongxu, J. Liu, J. Wang, Investigation on thermal and fire propagation
Transf. 172 (Jun. 2021), [Link] behaviors of multiple lithium-ion batteries within the package, Appl. Therm. Eng.
IJHEATMASSTRANSFER.2021.121077 p. undefined-undefined. 157 (Jul. 2019), [Link] p.
[27] S. Koch, A. Fill, K.P. Birke, Comprehensive gas analysis on large scale automotive undefined-undefined.
lithium-ion cells in thermal runaway, J. Power Sources 398 (Sep. 2018) 106–112, [54] C. Zhao, J. Sun, Q. Wang, Thermal runaway hazards investigation on 18650
[Link] lithium-ion battery using extended volume accelerating rate calorimeter, J. Energy
[28] S. Chen, Z. Wang, J. Wang, X. Tong, W. Yan, Lower explosion limit of the vented Storage 28 (Apr. 2020), [Link]
gases from Li-ion batteries thermal runaway in high temperature condition, J. Loss [55] Y.S. Duh, M.T. Tsai, C.S. Kao, Thermal runaway on 18650 lithium-ion batteries
Prev. Process Ind. 63 (Jan. 2020), [Link] containing cathode materials with and without the coating of self-terminated
[29] A.O. Said, C. Lee, S.I. Stoliarov, Experimental investigation of cascading failure in oligomers with hyper-branched architecture (STOBA) used in electric vehicles,
18650 lithium ion cell arrays: impact of cathode chemistry, J. Power Sources 446 J. Therm. Anal. Calorim. 129 (3) (Sep. 2017) 1935–1948, [Link]
(Jan. 2020), [Link] 10.1007/s10973-017-6356-7.
[30] D.P. Abraham, E.P. Roth, R. Kostecki, K. McCarthy, S. MacLaren, D.H. Doughty, [56] Z. Zhou, et al., Quantitative study on the thermal failure features of lithium iron
Diagnostic examination of thermally abused high-power lithium-ion cells, J. Power phosphate batteries under varied heating powers, Appl. Therm. Eng. 185 (Feb.
Sources 161 (1) (Oct. 2006) 648–657, [Link] 2021), [Link]
JPOWSOUR.2006.04.088. [57] P. Qin, J. Sun, Q. Wang, A new method to explore thermal and venting behavior of
[31] A.A. Abd-El-Latif, P. Sichler, M. Kasper, T. Waldmann, M. Wohlfahrt-Mehrens, lithium-ion battery thermal runaway, J. Power Sources 486 (Feb. 2021), https://
Insights into thermal runaway of Li–ion cells by accelerating rate calorimetry [Link]/10.1016/[Link].2020.229357.
coupled with external sensors and online gas analysis, Batter. Supercaps 4 (7) (Jul. [58] J. Zhao, S. Lu, Y. Fu, W. Ma, Y. Cheng, H. Zhang, Experimental study on thermal
2021) 1135–1144, [Link] runaway behaviors of 18650 li-ion battery under enclosed and ventilated
[32] Y. Fernandes, A. Bry, S. de Persis, Identification and quantification of gases emitted conditions, Fire Saf. J. 125 (2021), [Link]
during abuse tests by overcharge of a commercial Li-ion battery, J. Power Sources [59] B. Mao, P. Huang, H. Chen, Q. Wang, J. Sun, Self-heating reaction and thermal
389 (Jun. 2018) 106–119, [Link] runaway criticality of the lithium ion battery, Int. J. Heat Mass Transf. 149 (Mar.
[33] K. Zou, S. Lu, X. Chen, E. Gao, Y. Cao, Y. Bi, Thermal and gas characteristics of 2020), [Link]
large-format LiNi0.8Co0.1Mn0.1O2 pouch power cell during thermal runaway, [60] J.G. Quintiere, On methods to measure the energetics of a lithium ion battery in
J. Energy Storage 39 (Jul. 2021), [Link] p. thermal runaway, Fire Saf. J. 111 (Jan. 2020), [Link]
undefined-undefined. FIRESAF.2019.102911.
[34] Y. Zhang, H. Wang, W. Li, C. Li, Quantitative identification of emissions from [61] Z. Liao, et al., Hazard analysis of thermally abused lithium-ion batteries at different
abused prismatic Ni-rich lithium-ion batteries, eTransportation 2 (Nov. 2019), state of charges, J. Energy Storage 27 (Feb. 2020), [Link]
[Link] EST.2019.101065.
[35] O. Willstrand, R. Bisschop, P. Blomqvist, A. Temple, J. Anderson, Toxic gases from [62] M. Chen, D. Ouyang, J. Weng, J. Liu, J. Wang, Environmental pressure effects on
fire in electric vehicles, in: RISE Research Institutes of Sweden AB, RISE Report thermal runaway and fire behaviors of lithium-ion battery with different cathodes
2020, 2020, p. 90. and state of charge, Process Saf. Environ. Prot. 130 (Oct. 2019) 250–256, https://
[36] P. Ribière, S. Grugeon, M. Morcrette, S. Boyanov, S. Laruelle, G. Marlair, [Link]/10.1016/[Link].2019.08.023.
Investigation on the fire-induced hazards of Li-ion battery cells by fire calorimetry, [63] L. Liu, C. Lin, B. Fan, F. Wang, L. Lao, P. Yang, A new method to determine the
Energy Environ. Sci. 5 (5271) (2012), [Link] heating power of ternary cylindrical lithium ion batteries with highly repeatable
[37] Z. Huang, C. Zhao, H. Li, W. Peng, Z. Zhang, Q. Wang, Experimental study on thermal runaway test characteristics, J. Power Sources 472 (Oct. 2020), https://
thermal runaway and its propagation in the large format lithium ion battery [Link]/10.1016/[Link].2020.228503.
module with two electrical connection modes, Energy 205 (Aug. 2020), https:// [64] Q. Zhang, T. Liu, Q. Wang, Experimental study on the influence of different heating
[Link]/10.1016/[Link].2020.117906. methods on thermal runaway of lithium-ion battery, J. Energy Storage 42 (Oct.
[38] P. Liu, et al., Thermal runaway and fire behaviors of lithium iron phosphate battery 2021), [Link]
induced by over heating, J. Energy Storage 31 (Oct. 2020), [Link] [65] H. Li, H. Chen, G. Zhong, Y. Wang, Q. Wang, Experimental study on thermal
10.1016/[Link].2020.101714. runaway risk of 18650 lithium ion battery under side-heating condition, J. Loss
[39] F. Diaz, Y. Wang, R. Weyhe, B. Friedrich, Gas generation measurement and Prev. Process Ind. 61 (Sep. 2019) 122–129, [Link]
evaluation during mechanical processing and thermal treatment of spent Li-ion JLP.2019.06.012.
batteries, Waste Manag. (Elmsford) 84 (Feb. 2019) 102–111, [Link] [66] N.E. Galushkin, N.N. Yazvinskaya, D.N. Galushkin, Mechanism of thermal runaway
10.1016/[Link].2018.11.029. in lithium-ion cells, J. Electrochem. Soc. 165 (7) (2018) A1303–A1308, https://
[40] Y. Yang, et al., Carbon oxides emissions from lithium-ion batteries under thermal [Link]/10.1149/2.0611807JES.
runaway from measurements and predictive model, J. Energy Storage 33 (Jan. [67] A. Perea, A. Paolella, J. Dubé, D. Champagne, A. Mauger, K. Zaghib, State of charge
2021), [Link] influence on thermal reactions and abuse tests in commercial lithium-ion cells,
[41] N.S. Spinner, et al., Physical and chemical analysis of lithium-ion battery cell-to- J. Power Sources 399 (Sep. 2018) 392–397, [Link]
cell failure events inside custom fire chamber, J. Power Sources 279 (Apr. 2015) JPOWSOUR.2018.07.112.
713–721, [Link] [68] A.O. Said, C. Lee, X. Liu, Z. Wu, S.I. Stoliarov, Simultaneous measurement of
[42] J. Sun, et al., Toxicity, a serious concern of thermal runaway from commercial Li- multiple thermal hazards associated with a failure of prismatic lithium ion battery,
ion battery, Nano Energy 27 (Sep. 2016) 313–319, [Link] Proc. Combust. Inst. 37 (3) (Jan. 2019) 4173–4180, [Link]
NANOEN.2016.06.031. PROCI.2018.05.066.
[43] V. Premnath, Y. Wang, N. Wright, I. Khalek, S. Uribe, Detailed characterization of [69] D. Sturk, L. Hoffmann, A.A. Tidblad, Fire tests on E-vehicle battery cells and packs,
particle emissions from battery fires, Aerosol Sci. Technol. 56 (4) (2022) 337–354, Traffic Inj. Prev. 16 (2015) 159–164, [Link]
[Link] 15389588.2015.1015117.
[44] D. Sturk, L. Rosell, P. Blomqvist, A.A. Tidblad, Analysis of li-ion battery gases [70] V. Somandepalli, H. Biteau, Cone calorimetry as a tool for thermal hazard
vented in an inert atmosphere thermal test chamber, Batteries 5 (3) (Sep. 2019), assessment of Li-ion cells, SAE Int. J. Altern. Powertrains 3 (2) (2014) 222–233,
[Link] [Link]
[45] C. di Bari, et al., Recent safety focused overall analysis, testing and accident [71] C.M.R. Vendra, et al., Numerical and experimental characterisation of high energy
reviews towards safer e-mobility and energy storage, in: Proceedings From density 21700 lithium-ion battery fires, Process Saf. Environ. Prot. 160 (Apr. 2022)
International Conference on Fires in Vehicles (FIVE 2020), 2020. 153–165, [Link]
[46] P. Liu, Y. Li, B. Mao, M. Chen, Z. Huang, Q. Wang, Experimental study on thermal [72] Z. Wang, H. Yang, Y. Li, G. Wang, J. Wang, Thermal runaway and fire behaviors of
runaway and fire behaviors of large format lithium iron phosphate battery, Appl. large-scale lithium ion batteries with different heating methods, J. Hazard. Mater.
Therm. Eng. 192 (Jun. 2021), [Link] 379 (Nov. 2019), 120730, [Link]
applthermaleng.2021.116949. [73] Z. Huang, P. Liu, Q. Duan, C. Zhao, Q. Wang, Experimental investigation on the
[47] “Battery Failure Databank | NASA.” [Link] (accessed Aug. 09, cooling and suppression effects of liquid nitrogen on the thermal runaway of
2022). lithium ion battery, J. Power Sources 495 (May 2021), [Link]
[48] R.E. Lyon, R.N. Walters, Energetics of lithium ion battery failure, J. Hazard. Mater. jpowsour.2021.229795.
318 (Nov. 2016) 164–172, [Link] [74] X. Feng, et al., Investigating the thermal runaway mechanisms of lithium-ion
[49] P. Ping, et al., Study of the fire behavior of high-energy lithium-ion batteries with batteries based on thermal analysis database, Appl. Energy 246 (Jul. 2019) 53–64,
full-scale burning test, J. Power Sources 285 (Jul. 2015) 80–89, [Link] [Link]
10.1016/[Link].2015.03.035.

12
O. Willstrand et al. Journal of Energy Storage 68 (2023) 107785

[75] D. Lu, et al., Thermal behavior and failure mechanism of large format lithium-ion [81] “Global Technical Regulations (GTRs) | UNECE.” [Link]
battery, J. Solid State Electrochem. 25 (1) (Jan. 2021) 315–325, [Link] andards/transport/vehicle-regulations-wp29/global-technical-regulations-gtrs
10.1007/s10008-020-04810-z. (accessed Nov. 04, 2022).
[76] J. Chen, et al., Thermal runaway modeling of LiNi0.6Mn0.2C0.2O2/graphite [82] R.T. Balmer, Vapor and gas refrigeration cycles, in: Modern Engineering
batteries under different states of charge, J. Energy Storage 49 (2022), [Link] Thermodynamics, 2011, pp. 535–590, [Link]
org/10.1016/[Link].2022.104090. 374996-3.00014-2.
[77] Y. Wang, D. Ren, X. Feng, L. Wang, M. Ouyang, Thermal runaway modeling of [83] M.J. Lain, J. Brandon, E. Kendrick, Design Strategies for High Power Vs. High
large format high-nickel/silicon-graphite lithium-ion batteries based on reaction Energy Lithium Ion Cells, Batteries 5 (4) (2019), [Link]
sequence and kinetics, Appl. Energy 306 (Jan. 2022), [Link] batteries5040064.
APENERGY.2021.117943. [84] F. Larsson, B.-E. Mellander, Abuse by external heating, overcharge and short
[78] D. Ren, X. Feng, L. Lu, X. He, M. Ouyang, Overcharge behaviors and failure circuiting of commercial lithium-ion battery cells, J. Electrochem. Soc. 161 (10)
mechanism of lithium-ion batteries under different test conditions, Appl. Energy (2014) 1611–1617, [Link]
250 (Sep. 2019) 323–332, [Link] [85] DNV, 2022 Battery Scorecard. [Link]
[79] National Research Council of Canada, Thermal runaway initiation methodology ry-scorecard-228565?utm_campaign=ES_ADV_GLOB_22Q3_PROM_STOR_2022%
(TRIM): a leading-edge technology solution to improve the safety of battery 20Battery%20Scorecard%20-%20Full%20launch&utm_medium=email&utm_sou
systems. [Link] rce=Eloqua, 2022. (Accessed 7 September 2022).
-53e8-422e-9cf5-2845bbe92f33. (Accessed 29 November 2022). [86] H. Wang, et al., Study on the thermal runaway and its propagation of lithium-ion
[80] DNV, Technical Reference for Li-ion Battery Explosion Risk and Fire Suppression batteries under low pressure, Fire. Technol 56 (6) (Nov. 2020) 2427–2440, https://
[Online]. Available: [Link] [Link]/10.1007/S10694-020-00963-5.
-ion-battery-explosion-risk-and-fire-suppression-165062, 2019. (Accessed 3 [87] S. Xie, J. Sun, X. Chen, Y. He, Thermal runaway behavior of lithium-ion batteries in
November 2022). different charging states under low pressure, in: International Journal of Energy
Research, vol. 45, no. 4, John Wiley and Sons Ltd, Mar. 25, 2021, pp. 5795–5805,
doi: 10.1002%2Fer.6200.

13