Sub Topic: Fire

11th U. S. National Combustion Meeting

Organized by the Western States Section of the Combustion Institute
March 24–27, 2019
Pasadena, California

Comprehensive analysis of dynamics and hazards associated

with cascading failure in lithium ion cell arrays
Ahmed O. Said1, Christopher Lee1, Stanislav I. Stoliarov1,*
1
Fire Protection Engineering, University of Maryland, College Park, MD 20742, USA
*
Corresponding Author Email: stolia@umd.edu
Abstract: In lithium ion arrays, thermal runaway may propagate from a failing cell to neighboring cells and grow into
a large-scale fire in a phenomenon referred to as cascading failure. In this work, a new experimental setup was
developed to investigate cascading failure using 12-18 cell arrays constructed from lithium cobalt oxide cells of 18650
form factor and 2600 mA h nominal capacity. The arrays were mounted in a wind tunnel with well-controlled
environmental conditions. Thermal runaway was initiated in one cell using an electric heater and observed to
propagate through the array using temperature sensors attached to the bottom surface of each cell. Cascading failure
was studied in N2 or air environment to elucidate the impact of combustion. The cell temperature allowed calculation
of row-to-row propagation speed (SP) in different test conditions. In addition to the cell temperatures, production rates
of O2, total unburned hydrocarbons, CO, CO2 and H2 were measured, and heats generated in chemical reactions
between the battery materials and in flaming combustion were computed. Results showed that SP was 9 times faster
in air than in N2. Chemical and combustion heats of 56.6 ± 2.5 and 60.1 ± 17.5 kJ per cell were generated during
cascading failure.
Keywords: Lithium Ion Battery Fire; Thermal Runaway Propagation; Calorimetry
1. Introduction
Lithium ion batteries (LIBs) are considered to be one of the most promising technologies for
effective electrical energy storage due to their high energy density and efficiency, longevity, and
lack of a memory effect [1]. However, exposure of LIBs to abnormal operating circumstances,
such as mechanical damage or external heating, may trigger failure in LIBs [1, 2]. The
aforementioned failure causes an increase of the cell internal temperature. This increase in
temperature results in the vaporization of a portion of the electrolyte and the formation of gases,
which raises pressure inside the enclosure of the LIB cell. When the internal pressure reaches a
certain threshold, safety vent ports located on the cell casing open to eject the formed gases at
relatively slow rates, consequently reducing the internal pressure and preventing the cell rupture;
this phenomenon is referred to as safety venting (SV) [1, 2]. As the temperature of the cell
continues to increase, exothermic reactions within and between the cell’s components are initiated.
The cell’s increasing temperature accelerates the chemical reaction rates inside the cell, resulting
in a rapid self-heating [1, 2]. The heating process eventually causes the cell to reach its thermal
runaway (TR) stage during which the temperature and gas ejection rate of the cell increase
dramatically [1, 2].
The failure of an individual LIB cell presents thermal hazards because of the significant
increase in temperature of the cell’s body and the substantial energy generation. Liu et al. [3-5]
developed an experimental apparatus, copper slug battery calorimetry (CSBC), which allowed
measurement of energy generation inside an LIB cell. Experiments were conducted separately to
calculate the total heat released resulting from flaming combustion of ejected LIB materials (fire
hazards). Liu et al. [3, 4] adopted cone calorimetry for these measurements, but underestimated

1
Sub Topic: Fire

the total heat release because of difficulties in maintaining flame on the high speed jets emanating
from the safety vent ports, particularly during TR. Said et al. [6, 7], therefore, worked on
modifying the manner by which the ejected materials were ignited to boost combustion efficiency
compared to Liu’s studies [3, 4]. Also, the material ejection introduces a chemical hazard, as a
considerable portion of those materials are toxic. Maloney [8, 9] measured the concentration of
the gases ejected from various commercial cells in N2 (to prevent any combustion of gases). The
ejected gases consisted mainly of H2, hydrocarbons (mostly CH4) and CO in addition to non-
combustible CO2 and O2. Other studies [10-12] focused on measuring concentration of the
combustion products. Measurements showed that toxic products, particularly CO and HF, were
detected when LIBs were tested at high states of charge (SOCs).
The aforementioned hazards (thermal, fire, and chemical) are intensified significantly when
LIBs are assembled into large LIB arrays to satisfy high power demands. The failure of a single
LIB cell may induce failure in neighboring cells and cause TR propagation through the entire pack
(cascading failure). A limited number of experimental have been dedicated to the phenomenon of
cascading failure in LIB arrays. Lamb et al. [13] examined TR propagation in small scale LIB
modules constructed with either wall-to-wall 18650 cylindrical or pouch cells. Arrays of
cylindrical cells were found to be less prone to propagation of TR due to limited cell-to-cell contact
area in comparison to pouch cells. Lopez et al. [14] determined that adding 2 mm gaps between
cylindrical cells in arrays prevented propagation and alleviated physical damage in the tested
arrays (this suggestion may not be applicable due to limited space in most practical applications).
Overall, previous studies did not expand careful quantification of ejected gases and released
energy from a single failing LIB cell to LIB cell arrays. Also, these studies did not qualitatively
monitor the dynamics of cascading failure across each individual LIB in the tested arrays. In the
current study, a new experimental setup was built to provide a comprehensive analysis of hazards
that LIB arrays pose. Combining analysis of the dynamics, gaseous emissions, and energetics of
cascading failure into a single experimental setup was the main objective. The setup included a
sectioned wind tunnel, which was employed to provide well-defined boundary conditions for the
studied LIB arrays. The cascading failure was investigated in N2, as well as in air to elucidate the
impact of flaming combustion. Tenergy lithium cobalt oxide (LCO) cells, with 18650 form factor
and 2600 mA h nominal capacity, were used to construct 12-18 wall-to-wall cell arrays, which
were studied in this work. The setup provided simultaneous measurements of temperature of each
cell, and temperatures and volumetric concentrations of gases ejected during cascading failure.
The cells temperature allowed calculation of TR propagation speed in different test conditions.
The gas concentration profiles, recorded in the N2 tests, enabled calculations of mass yields and
lower flammability limit of the ejected gases, as well as the rate of chemical heat generation due
to reactions between battery components inside and outside the cell casings. The air tests were
used to obtain the flaming combustion energy of ejected battery materials.
2. Experimental
Tested arrays consisted of Tenergy ICR18650 [15] LIB cells (cylindrical geometry of 18
mm diameter and 65 mm height) with nominal capacity and voltage of 2600 mA h and 3.7 V,
respectively. These cells contain lithium cobalt oxide (LCO) cathode and carbon anode. The
positive terminals of these cells are equipped with four safety vent ports. The tested cells were
arranged in rectangular arrays of either 18 or 12 cells without any spacing in between adjacent
cells (wall-to-wall); configurations of tested arrays are displayed in Figure 1. All cells of the tested
arrays were at 100% state of charge (SOC). The cells were fixed in a stainless steel cell holder
that was placed inside a stainless steel wind tunnel, which provided controlled environmental

2
Sub Topic: Fire

conditions. Figure 1 displays a three-dimensional rendering of the wind tunnel, which was built
using stainless steel ducting. The wind tunnel consisted of four main sections: mixing chamber,
pre-test section, test section (containing the cell holder and cell array), and diagnostics section.

Figure 1: Schematic of tested cell array configurations and experimental setup (wind tunnel is
drawn to scale)
During experiments, 186 l min-1 of N2 or 640 l min-1 of air was flowed through the wind
tunnel and across the cells. Cell 2 in the tested array was intentionally forced into TR through
external heating with a DC power supply, and the propagation of failure from cell-to-cell was
tracked by recording the temperature of each cell’s bottom surface. The diagnostics in the setup
included three thermocouples measuring the exhaust temperature at different positions in the cross
section, and an emitted-gas sampling system equipped with O2, THC, CO, CO2, and H2 sensors.
The electric heater power and data acquisition software were turned on simultaneously. All
thermocouple and gas sensors were digitally sampled at a frequency of 2 Hz with National
Instruments DAQ modules and LABVIEW software. The tests were stopped once the cells
returned to their initial temperatures.
A redesigned CSBC apparatus combined with oxygen consumption calorimetry [16] was
used to measure the heat of combustion of the materials ejected from individual fully charged
Tenergy cells upon TR. Details of the apparatus and measurement procedures can be found
elsewhere [7]. The heat of combustion obtained in the redesigned CSBC tests was compared with
the heat of generated in the current cascading failure tests.
In an N2 environment, tests on arrays of 18 and 12 cells were repeated five and four times,
respectively, to accumulate statistics. The purpose of the N2 experiments was to investigate the
effects of an anaerobic environment and array size (3×6 or 3×4 array) on the behavior of cascading
failure. The smaller size of arrays (3×4) was also examined in an air environment to study the
effect of combustion of ejected battery materials on the dynamics of cascading failure; tests were
repeated four times. Additionally, five tests were performed on five single cells using the CSBC
apparatus.

3
Sub Topic: Fire

3. Methods, Results, and Discussion

3.1 Dynamics of Cascading Failure
The onset time of safety venting (SV) was defined as 0.5 s (the measurement’s resolution)
before the derivative of the cell’s bottom temperature (dTLIB/dt) became negative. During
experiments, the onset of SV was observed by an audible clicking sound accompanied by the
appearance of gases at the wind tunnel outlet; the corresponding time was recorded to validate the
dTLIB/dt based criterion. Temperatures corresponding to the determined onset times are referred
to as SV onset temperatures. It is also important to mention that SV was not captured for many
cells. This was most likely due to simultaneous initiation of SV and thermal runaway (TR) when
the cells were rapidly heated during cascading failure.
To identify the onset time of TR, the maximum dTLIB/dt was first determined. The onset
of TR was subsequently defined as the point in time preceding the maximum when dTLIB/dt
became greater than 14 K s-1. This particular value for the derivative criterion was selected as it
produced TR onset times that closely corresponded to the times of audible explosions, which were
also accompanied by a significant increase in the exhaust flow, observed and recorded by the
operator during the experiments. Additionally, these derivative values pinpointed the start of the
sudden spike in the trend of each LIB’s temperature. The TR onset times and corresponding
temperatures were identified for every cell.
The TR end time was defined as the time that followed the maximum dTLIB/dt when the
derivative decreased below 6.5 K s-1. This value of dTLIB/dt was selected because the resulting
times closely corresponded to the times of return of the gas concentration signals to their respective
baselines for individual cells with failure durations that were clearly separated in time from the
rest of the cells in the array.
In all cascading failure experiments (conducted in N2 and air), all LIB cells underwent TR.
The TR onset times of individual cells were not reproducible despite carefully controlled heating
conditions. However, advancement of TR from one row to the next showed a reasonable degree
of reproducibility. Therefore, the dynamics of failure were analyzed on a row-to-row basis. The
TR onset time of each row was calculated by averaging the onset times of all cells in the row.
Subsequently, these times were converted to row-to-row propagation speeds (SP).
Figure 2 summarizes SP data obtained in this study. The error bars in the figure and in the
entire manuscript were computed from the scatter of the data as two standard deviations of the
mean. The data obtained in N2 (Figure 2(a)) show a relatively steady propagation through the
array. 18 and 12 cell arrays yield essentially the same results. The 18 cell data suggest that the SP
trend may be sinusoidal (an acceleration followed by deceleration), but it is difficult to establish
the presence of this trend with certainty given significant uncertainties in the data points. The
average rate of propagation for all rows and array sizes in N2 was found to be 0.080 ± 0.025 s-1.

Figure 2: TR propagation speeds of cell arrays in environments of (a) N2 and (b) air

4
Sub Topic: Fire

Figure 2(b) compares the propagation speed data across 12 cell arrays for N2 and air.
Unlike in the case of N2, the data obtained in air show an acceleration in SP with the TR progression
through the array. This acceleration was attributed to the impact of flaming combustion, which
contributed to heating of the downstream cells. For row 1 to row 2 propagation, this impact was
insignificant but increased markedly as more and more cells underwent TR and intensity of the
flaming combustion increased. The average rates of SP for the late stages of failure propagation
(row 2 to row 3 and row 3 to row 4) were computed as shown in Figure 2(b) and found to increase
by a factor of 9 when arrays were examined in air instead of N2.
3.2 Failure Onset Temperatures
Table 1 summarizes the measured temperatures of SV and TR onsets as well as the
maximum temperatures achieved during TR. These temperatures showed no significant
dependence on the cell’s position in an array and, therefore, were averaged over all cells for the
experiments conducted under the same conditions. The SV and TR onset temperatures also did
not also depend on the presence of air or array size. When these temperatures were averaged over
all cascading failure experiments, values of 356 ± 12 and 370 ± 6 K were obtained for the SV and
TR onsets, respectively. The maximum temperatures did show slight dependence on the
experimental conditions. Both the presence of air and increase in the size of the array produced
relatively higher maximum temperature values, perhaps, due to the higher maximum overall heat
generation rates achieved in these experiments.
Table 1: Summary of the cell temperatures (temperatures probed at the cell’s bottom surface).
SV onset TR onset Maximum
Array
Atmosphere temperature temperature temperature
size
[K] [K] [K]
18 cells N2 353 ± 18 364 ± 7 740 ± 19
12 cells N2 359 ± 17 381 ± 12 686 ± 23
12 cells Air 350 ± 16 371 ± 11 727 ± 33
The SV and TR onset temperatures measured in the current study varied markedly when
compared to those ones reported in a previous work by Liu et al. [4] for the same LIB cells. For
fully charged cells, Liu obtained SV and TR onset temperatures of 451 ± 5 and 470 ± 4,
respectively. These significantly different values are attributed to differences in the experimental
setups and heating conditions. In Liu’s work, the cells were heated slowly and uniformly, but in
the current study, each cell in the array experienced rapid and non-uniform heating. This non-
uniformity resulted in uneven temperature distribution across the body of each cell, meaning that
the bottom surface temperature would be smaller than the temperature corresponding to the failure
initiation locations (side walls of the cell where most of the heat transfer occurred). This
observation suggests that temperature-based failure detection thresholds for real battery packs
should not be rely on experiments with slow, uniform heating rates because the temperatures of
the cells are often not spatially uniform in real failure scenarios.
3.3 Mass Loss and Mass Loss Rate of Ejected Gases
Each cell tested was weighed before and after each test. These initial and final mass values
were used to develop estimated time-resolved mass and mass loss rate trends for each cell. It was
assumed that mass was always lost linearly [4, 7]. The onset of SV was often indistinguishable,
so it was assumed that all of the mass loss occurred during the duration of TR. For each test, the
mass loss rate trends for each cell were added together in order to determine the total mass loss
rate of the cell array, ṁLIBs. Table 2 presents data of the average initial mass of individual LCO
cells (with plastic packaging removed) for each set of experimental conditions. The table also

5
Sub Topic: Fire

contains information on total mass loss per cell number. It is observed that the size of the examined
array had minor effects on total mass loss per cell number. The air environment, however, raised
the mass loss to approximately 41.5% of initial mass rather than 38.4% for the same sized arrays
tested in N2. This increase in mass loss percentage was attributed to the greater possibility of
having ruptured cells in air compared to N2.
Table 2: Summary of initial mass and total mass loss of LCO cells.
Initial mass Mass loss
Array
Atmosphere per cell number per cell number
size
[g] [g]
18 cells N2 43.40 ± 0.03 16.7 ± 0.9 (38.4%)
12 cells N2 43.50 ± 0.06 16.7 ± 1.0 (38.4%)
12 cells Air 43.60 ± 0.06 18.1 ± 1.5 (41.5%)
3.4 Yields, and Flammability Limits of Ejected Gases
Signals from the five gas sensors were used to estimate gas yields ejected from the cell
arrays. Due to the inert atmosphere, all gases were known to have been ejected from the cells
themselves, not produced through combustion. Methane was assumed to be the dominant THC
gas [8, 9]. Mass yields, m, for each of the five measured gases, i, were determined by integrating
the mass flow rates, ṁ, for each gas from beginning of test to end time of test (tend). These mass
flow rates were determined by equating the gas to N2 ratios for the mass flow rates and mass
fractions, Y. Mass fractions were calculated from the gas mole fractions, X, and molecular weights,
M. The mole fraction of N2 was estimated by assuming that it made up the remainder of the gas
stream that was not measured by the gas sensors. This process is shown in detail in the equations
below.
𝑡 𝑌 𝑌 𝑋 𝑀𝑖
𝑚𝑖 = ∫0 𝑒𝑛𝑑 𝑚̇𝑖 𝑑𝑡, where: 𝑚̇𝑖 = 𝑌 𝑖 𝑚̇𝑁2 , 𝑌 𝑖 = (1−∑ 𝑋𝑖 )𝑀 (1)
𝑁2 𝑁2 𝑖 𝑁2
The yields of gases are given in Table 3; the reported yields are normalized either by the
total initial mass of all cells or by the number of cells in the array. Within the computed
uncertainties, 12 and 18 cell arrays produced the same yields of gases. When averaged over both
array sizes and normalized by the total initial mass of all cells, the O2, THC, CO, CO2 and H2
yields become 59.0E-5 ± 23.0E-5, 36.2E-3 ± 10.7E-3, 40.7E-3 ± 4.9E-3, 32.4E-3 ± 3.8E-3 and
20.1E-4 ± 3.9E-4, respectively.
Table 3: Summary of estimated gas yields.
Gas production in 18 cell arrays Gas production in 12 cell arrays
Normalized per Normalized Normalized per Normalized
initial cell mass per cell initial cell mass per cell
[-] number [-] number
[g] [g]
O2 64.0E-5 ± 41.0E-5 0.028 ± 0.018 52.0E-5 ± 15.0E-5 0.023 ± 0.007
THC[CH4] 36.8E-3 ± 12.9E-3 1.60 ± 0.56 35.5E-3 ± 20.3E-3 1.55 ± 0.89
CO 43.1E-3 ± 3.8E-3 1.87 ± 0.16 37.7 ± 9.8E-3 1.64 ± 0.43
CO2 34.8E-3 ± 2.8E-3 1.51 ± 0.12 29.3E-3 ± 7.1E-3 1.27 ± 0.31
H2 20.7E-4 ± 6.7E-4 0.090 ± 0.029 19.3E-4 ± 4.1E-4 0.084 ± 0.018
The mass yields of the flammable gases (THC, CO, and H2) used to calculate the lower
flammability limit in air, LFLmixture, of a mixture of these gases ejected from the cells. This
calculation process, following Le Chatelier’s mixing rule, is outlined in the equation below.

6
Sub Topic: Fire

𝑚𝑇𝐻𝐶 /𝑀𝑇𝐻𝐶 +𝑚𝐶𝑂 /𝑀𝐶𝑂 +𝑚𝐻2 / 𝑀𝐻2

𝐿𝐹𝐿𝑚𝑖𝑥𝑡𝑢𝑟𝑒 = 𝑚𝑇𝐻𝐶 /𝑀𝑇𝐻𝐶 𝑚𝐶𝑂 /𝑀𝐶𝑂 𝑚𝐻2 / 𝑀𝐻2
(2)
+ +
𝐿𝐹𝐿𝑇𝐻𝐶 𝐿𝐹𝐿𝐶𝑂 𝐿𝐹𝐿𝐻
2
In this equation, CO2 presence is neglected in order to provide a conservative estimate of LFLmixture.
Additionally, one key parameter that is important to be assessed while designing enclosures for
commercial LIB packs is the maximum volume of enclosure where gas ejection from a single cell
becomes a flammable mixture in an air environment. Quantification of this volume was performed
using the equation of state for ideal gases, assuming standard temperature and pressure conditions.
Mass yields of THC, CO, and H2 (reported in Table 3) were utilized to compute the
LFLmixture of the flammable mixture via Eqn. 2. On average, both arrays of 18 and 12 cells released
flammable mixtures with LFLmixture of 5.79 ± 0.12 vol. % in air. The volume of an enclosure,
where the gas ejection from a single cell produces a flammable mixture in an air environment, was
determined to be less than or equal to 0.087 ± 0.017 m3.
3.5 Energetics of Cascading Failure
3.5.1 Chemical Heat Generation
Testing LIB cell arrays in an inert medium (N2) allowed determination of the chemical heat
generated from the reactions between cell components, ECHG, during cascading failure. ECHG was
computed from the changes in the enthalpy of the flow out of the test section using Eqn. 3:
𝑡
𝐸𝐶𝐻𝐺 = ∫0 𝑒𝑛𝑑 {[𝑚̇𝑁2 𝑐̅𝑝 𝑁 + ∑ (𝑚̇𝑖 𝑐̅𝑝 𝑖 ) + 𝑚̇𝐶 𝑐̅𝑝 𝐶 ] (𝑇𝑒𝑥 − 𝑇0 ) + 𝑃𝑙𝑜𝑠𝑠 − 𝑃ℎ𝑒𝑎𝑡𝑒𝑟 } 𝑑𝑡 (3)
2
This energy balance consists of five main terms: the heat carried by the N2 flow, the heat carried
by the gases ejected (i) from the cells, the heat carried by the solid graphite (c) ejected from the
cells, the heat lost to the surroundings, and the heat supplied by the electric heater. The N2, ejected
gas, and graphite terms are expressed as functions of mass flow rate, average specific heat, 𝑐̅𝑝 ,
average reading of the exhaust thermocouples, Tex, and initial average reading of the exhaust
thermocouples, T0. All specific heats in Eqn. 3 were taken at an average temperature and
calculated with temperature-dependent polynomial equations found in literature [17, 18]. The N2
mass flow rate was predetermined, and the measured gas mass flow rates were calculated
identically to the mass yield analysis. The graphite mass flow rate was estimated by subtracting
the sum of the measured gas mass flow rates from ṁLIBs, as shown in Eqn. 4.
𝑚̇𝐶 = 𝑚̇𝐿𝐼𝐵𝑠 − ∑ 𝑚̇𝑖 (4)
The power loss term, Ploss, took into account conduction loss to the wind tunnel insulation
and steel support struts of the cell holder. These losses, which were relatively small due to the
presence of the thermal insulation, were estimated using the steady-state version of Fourier’s law.
The input power trend, Pheater, was developed by monitoring the DC power supply during
experiments; Pheater would initially be fixed to approximately 115 W. This power was attempted
to be maintained throughout each test, but the violent nature of the cell failure often caused the
heating wire to break, resulting in zero input power. The integral of Pheater was in a range of 16-
52 kJ, and this range represented 1.5–7.6% of the total chemical energy produced by the entire cell
arrays during cascading failure. Therefore, the variation in input failure power did not have a
significant effect on the overall dynamics of failure or the total chemical energy produced from
the cell array.
Table 4 provides information on ECHG values normalized by the initial mass of all cells,
number of cells, and cell nominal electrical capacity. Differences between 18 and 12 cell results
are within each other’s uncertainties. When averaged over both array sizes, ECHG becomes 56.6 ±
2.5 kJ per cell number, 1.30 ± 0.06 kJ per g of initial cell mass, or 21.76 ± 0.97 kJ per unit nominal
electrical capacity in A h.

7
Sub Topic: Fire

Table 4: Chemical heat generation during cascading failure of LCO cell arrays in N2.
ECHG
Normalized per Normalized per
Normalized per
unit initial cell unit nominal
Array size cell number
mass capacity
-1 [kJ]
[kJ g ] [kJ A-1 h-1]
18 cells 1.34 ± 0.09 58.0 ± 4.1 22.3 ± 1.7
12 cells 1.26 ± 0.04 54.9 ± 1.8 21.1 ± 0.7
To accurately estimate chemical heat generation with respect to available electrical energy
inside the cell, ECHG was normalized by electrical stored energy and estimated to be 1.57 ± 0.18 kJ
per unit electrical stored energy (kJ). ECHG was found to be within 14% of that reported by Lyon
and Walters [19], 65.7 kJ per cell number, for the same LIB cells. Liu et al. [4] reported a chemical
heat generation value of 37.3 kJ for LCO cells at 100% SOC. This estimate [4] represented only
the energy generation inside the cell enclosure and did not include energy released by reactions
occurring between ejected materials outside the cell body. However, Liu et al. [4] extrapolated a
chemical heat generation of 59 kJ based on the assumption that no materials were lost from the
cell and all possible chemical reactions occurred inside the body of the cell; this value compares
favorably with the current results.
3.5.2 Flaming Combustion Energy Generation
The rate of heat release (PFlaming) associated with combustion of ejected materials was
estimated using a different technique of oxygen consumption calorimetry. This technique was
based on Huggett’s empirical observation [20] that most of combustibles released a nearly constant
amount of heat per unit mass of consumed oxygen, as described in Eqn. 5.
𝑃𝐹𝑙𝑎𝑚𝑖𝑛𝑔 = 𝐸 (𝑚̇𝑂2 |𝑖𝑛 − 𝑚̇𝑂2 |𝑜𝑢𝑡 ) (5)
-1
E is the heat release per unit mass of oxygen (13.1 kJ g of O2); this value of E is an empirically
derived constant [20]. 𝑚̇𝑂2 |𝑖𝑛 is the mass flow rate of oxygen at the inlet of the wind tunnel; this
flow rate was assumed to be constant and calculated from the mass flow controller setting and the
air composition (21 vol. % of O2 and 79% vol. % of N2). While 𝑚̇𝑂2 |𝑜𝑢𝑡 is the mass flow rates of
oxygen at the outlet of the wind tunnel, and was computed using the same method followed in
Eqn. 1. The integral of PFlaming over the duration of the experiment provided an estimate of the
effective heat produced from the combustion of ejected battery materials (EFlaming). The resulting
integral values normalized by the initial mass of all cells, number of cells, and nominal electrical
capacity are summarized in Table 5.
Table 5: Heat generation in flaming combustion of ejected LIB materials.
EFlaming
Normalized
Normalized
per unit Normalized per unit
per cell
Test type initial cell nominal capacity
number
mass [kJ A-1 h-1]
-1 [kJ]
[kJ g ]
Cascading failure tests (12 cell arrays) 1.38 ± 0.4 60.1 ± 17.5 23.1 ± 6.7
CSBC test (single cells) 2.5 ± 0.4 107 ± 18 41.2 ± 7.0
In air tests, the concentrations of exhaust gases (O2, THC [CH4], CO, and CO2) were
measured during cascading failure of the 12 cell arrays. The H2 analyzer was disconnected as it is
recommended not to be exposed to oxygen. The measurements showed that when cascading

8
Sub Topic: Fire

failure initiated, the gas products contained discernible amounts of THC and CO in addition to a
near complete consumption of oxygen, indicative of incomplete combustion.
In separate experiments, the CSBC technique was used to measure the complete heat of
combustion of materials ejected during TR of individual cells. Table 5 lists the heat of combustion
obtained in these CSBC experiments. A comparison of the cascading failure and CSBC heats of
combustion indicates that the combustion observed in the cascading failure experiments is highly
incomplete (56% combustion efficiency with respect to the CSBC combustion), which is
consistent with the large amounts of THC and CO detected in these experiments. Although the
utilized air flow in the tunnel was on the high side of what is used in the actual LIB systems, this
air flow was also insufficient to sustain a complete combustion during cascading failure
experiments. Thus, the highly incomplete combustion is most probably a feature of real cascading
failure. Normalizing the cascading failure and CSBC heats of combustion by the electrical stored
energy yields 1.8 ± 0.5 and 3.2 ± 0.5 kJ per unit electrical stored energy (kJ), respectively.
4. Conclusions
A new experimental setup was developed to allow for detailed analysis of the dynamics of
failure propagation in LIB cell arrays. LIB cell arrays, of two different sizes, were constructed
from 18650 LCO LIB cells and examined in N2 and air environments. The examined cell arrays
experienced cascading failure when one of the cells was intentionally forced to undergo thermal
runaway. Time-resolved measurements of cells’ bottom temperatures were analyzed to provide
information on onset times and temperatures of safety venting and thermal runaway. The onset
times of thermal runaway were utilized to evaluate row-to-row propagation speed of thermal
runaway (SP) throughout the arrays. SP showed negligible dependence on the size of the arrays
and was estimated to be 0.08 s-1 in an N2 environment. When LCO cell arrays were tested in air,
the propagation speed increased by about 9 times the N2 value. The results showed that CO (toxic
gas) had the largest mass yields among all the gases ejected from a single LCO cell. The mass
yields of THC and CO2 were comparable but slightly lower than the CO mass yields. The gas
measurements also showed relatively small mass yields of H2 and confirmed speculations found
in literature [1, 21] regarding formation of O2 during TR of LIB cells. Upon failure, the chemical
heat generation by an LCO cell was 56.6 ± 2.5 kJ per cell number, 1.30 ± 0.06 kJ per g of initial
cell mass, or 21.76 ± 0.97 kJ per unit nominal electrical capacity in A h. The total energy release
associated with flaming combustion was computed based on oxygen consumption in the wind
tunnel and found to be 60.1 ± 17.5 kJ per cell number; this estimate represented 58% complete
combustion efficiency. Lastly, under well ventilated conditions, a combination of chemical heat
generation and flaming combustion energy of an LCO cell during TR is five times the estimated
electrical stored energy of the cell. The results of this study provide a quantitative understanding
of hazards associated with cascading failure in LIB packs and lay foundation for effective
methodologies for early detection and mitigation of fires in LIB-based systems.
5. Acknowledgements
This research was supported by United Technologies Corporation (UTC) [Grant #
4318161] and is gratefully acknowledged. Authors would also like to show gratitude to the UTC
technical team for their insightful feedback that greatly assisted the research.
6. References
[1] Q. Wang, P. Ping, X. Zhao, G. Chu, J. Sun, C. Chen, Thermal runaway caused fire and
explosion of lithium ion battery, J. Power Sources, 208 (2012) 210-224.
[2] P.G. Balakrishnan, R. Ramesh, T.P. Kumar, Safety mechanisms in lithium-ion batteries, J.
Power Sources, 155 (2006) 401-414.

9
Sub Topic: Fire

[3] X. Liu, S.I. Stoliarov, M. Denlinger, A. Masias, K. Snyder, Comprehensive calorimetry of the
thermally-induced failure of a lithium ion battery, J. Power Sources, 280 (2015) 516-525.
[4] X. Liu, Z.B. Wu, S.I. Stoliarov, M. Denlinger, A. Masias, K. Snyder, Heat release during
thermally-induced failure of a lithium ion battery: Impact of cathode composition, Fire Safety J.,
85 (2016) 10-22.
[5] X. Liu, Z. Wu, S.I. Stoliarov, M. Denlinger, A. Masias, K. Snyder, A thermo-kinetic model of
thermally-induced failure of a lithium ion battery: development, validation and application, J.
Electrochem. Soc., 165 (2018) 10.
[6] A.O. Said, X. Liu, Z.B. Wu, C. Lee, S.I. Stoliarov, Time-resolved analysis of thermal failure
of prismatic lithium ion batteries, 10th U.S. National Combustion meeting, College Park,
Maryland (2017), pp. 6, paper ID: 2FI-0437.
[7] A.O. Said, C. Lee, X. Liu, Z. Wu, S.I. Stoliarov, Simultaneous measurement of multiple
thermal hazards associated with a failure of prismatic lithium ion battery, Proc. Comb. Inst., 37 (3)
(2019) 4173-4180.
[8] T. Maloney, Lithium battery thermal runaway vent gas analysis , Report no. DOT/FAA/TC-
15/59, U.S. Department of Transportation, 2016.
[9] T. Maloney, Impact of lithium battery vent gas ignition on cargo compartment fire protection
, Report no. DOT/FAA/TC-16/34, U.S. Department of Transportation, 2016.
[10] F. Larsson, P. Andersson, P. Blomqvist, A. Lorén, B.-E. Mellander, Characteristics of lithium-
ion batteries during fire tests, J. Power Sources, 271 (2014) 414-420.
[11] P. Andersson, P. Blomqvist, A. Lorén, F. Larsson, Using Fourier transform infrared
spectroscopy to determine toxic gases in fires with lithium-ion batteries, Fire and Mater., 40 (2016)
999-1015.
[12] P. Ribiere, S. Grugeon, M. Morcrette, S. Boyanov, S. Laruelle, G. Marlair, Investigation on
the fire-induced hazards of Li-ion battery cells by fire calorimetry, Energ Environ. Sci., 5 (2012)
5271-5280.
[13] J. Lamb, C.J. Orendorff, L.A.M. Steele, S.W. Spangler, Failure propagation in multi-cell
lithium ion batteries, J. Power Sources, 283 (2015) 517-523.
[14] C.F. Lopez, J.A. Jeevarajan, P.P. Mukherjee, Experimental analysis of thermal runaway and
propagation in lithium-ion battery modules, J. Electrochem. Soc., 162 (2015) A1905-A1915.
[15] Tenergy Co. Specification Approval Sheet: T-Energy ICR18650 Lithium Ion Battery, Model:
30005–0. 2010; Available from: http://www.all-battery.com/datasheet/30005-0_datasheet.pdf.
[16] S. ASTM, Standard test method for heat and visible smoke release rates for materials and
products using an oxygen consumption calorimeter, E1354 − 16a, (2016).
[17] C. Borgnakke, R.E. Sonntag, Fundamentals of thermodynamics, 7th ed., John Wiley & Sons,
Inc., New York, 2009.
[18] A.T.D. Butland, R.J. Maddison, The specific heat of graphite: An evaluation of
measurements, J. Nucl. Mater., 49 (1973) 45-56.
[19] R.E. Lyon, R.N. Walters, Energetics of lithium ion battery failure, J. Hazard. Mater., 318
(2016) 164-172.
[20] M.J. Hurley, SFPE handbook of fire protection engineering, Springer, New York, 2016.
[21] T.M. Bandhauer, S. Garimella, T.F. Fuller, A critical review of thermal issues in lithium-ion
batteries, J. Electrochem. Soc., 158 (2011) R1.

10