A cost-effective alternative to accelerating rate calorimetry: Analyzing thermal runaways of lithium-ion batteries through thermocouples

deviating within ~1 % compared to nail penetration tests. Moreover, six distinct stages during TR could be observed, in accordance with literature. This shows that the thermal propagation test using thermo-couples is able to align well with other methods such as accelerating rate calorimetry, but is considerably easier to employ.


Introduction
Large format prismatic lithium-ion battery (LiB) cells constitute a popular choice for battery electric vehicles (BEVs), partly due to their higher energy density and the limited manufacturing cost as compared to smaller cell formats.Nickel-rich nickel manganese cobalt oxide (NMC; e.g.NMC811 which is LiNi 0.8 Mn 0.1 Co 0.1 O 2 ) battery cells are favorable for traction batteries in heavy-duty BEVs, given their high power and energy density [1].However, there are safety concerns with using large format cells with nickel rich NMC that must be addressed before deployment [2][3][4].The thermal stability of this cathode material is a concern for battery safety due to the increased risk for a thermal runaway (TR) [5,6].
Determining the heat release from a cell during a TR event is necessary for assessing the risk of cell-to-cell thermal propagation, which is triggered by heat transmission.Inside the traction battery, the cells are typically arranged into modules, each housing multiple cells.The mechanical structure of the module could make the TR behavior different from cell level due to the internal heat transmission from one cell to another.Therefore, methods for characterizing the TR process and the corresponding heat release from a battery cell inside a battery module or a pack are desired for developing safe high-capacity batteries.
With the rapid expansion of battery technology in vehicle applications, the demand for studying the thermal properties is expected to rise.
The TR chain reactions involve self-sustaining mechanisms and a positive feedback loop once TR is triggered [7][8][9][10][11].In principle, the decomposition of the cathode materials produces oxygen that supplies the combustion of electrolyte and anode materials, thereby escalating the temperature of the system.Furthermore, a high temperature increases the reaction rate of the exothermic chain reactions, which in return generates heat at an increasing rate.
Feng et al. introduced a framework of characteristic temperatures {T 1 , T 2 , T 3 } to describe the TR process in a battery [8,[12][13][14][15][16][17][18][19].T 1 is the onset temperature for self-heating where a temperature increase exceeding the ambient temperature can be noticed.T 2 is the onset temperature for TR, where rapid chain reactions are triggered and massive heat is released.T 3 , in turn, represents the highest temperature during the TR process.The increment from T 2 to T 3 is directly related to the total heat release under adiabatic condition.
To achieve a deeper understanding of TR mechanisms, segmenting the process into distinct stages according to temperature ranges and their associated heat transfer characteristics have proven useful.Several studies have described the TR event as a four-stage process [20][21][22].Within the framework developed by the Feng et al., on the other hand, a more comprehensive depiction of TR comprises six stages (I-VI) [12,23,24].Stage I corresponds to the initiation of thermal event when the battery temperature rises, leading to slightly lower voltage and capacity under elevated temperatures.In stage II (often 57-190 • C), the battery undergoes self-heating driven by the decomposition of the solid electrolyte interphase (SEI), with concurrent SEI regeneration and consumption [25,26] and T 1 is reached.Then in stage III, the self-heating rate is slowed down when the separator undergoes melting.Polyethylene (PE) based separators typically have a melting point between 130 • C and 140 • C and polypropylene (PP) at around 170 • C [27,28].Stage IV is characterized by the release of the stored electrical energy as a consequence of internal short circuits (ISC) and combustion reactions consuming the negative electrode active materials.The inner pressure builds and the cell swells until venting.In Stage V, separation between the electrodes is lost when the ceramic coating breaks down, causing major ISC and a complete voltage drops.Massive heat is released from anode combustion, cathode decomposition, electrolyte decomposition, etc.The temperature increases drastically in this stage, corresponding to T 2 .Finally, in stage VI, the temperature rate is decelerated due to limitations of active materials or oxygen supply.The remaining reactions keep raising the temperature to its maximum, T 3 .
The rate of temperature increase (dT/dt) is another parameter that describes the intensity of the TR reactions.A high temperature rate predominantly resultis from low thermal instability of the battery materials.Assuming constant heat capacity and constant mass of the battery through TR, it also provides insights about the heating power from the TR chain reactions as described in Eq. ( 1): Accelerating rate calorimetry (ARC) is a standard technique for studying thermal events of batteries, such as TR, examples of such studies on NMC batteries are available in the literature [1,13,[20][21][22][29][30][31].ARC uses a heat-wait-seek (HWS) cyclic operation mode, where each cycle contains one step where the test object is heated and one step which is a waiting period to allow for temperature homogenization while seeking for the temperature at which self-heating starts.The cycle is repeated until the temperature increase rate exceeds a predefined threshold due to self-heating.When the threshold is found, the calorimeter switches to an adiabatic mode where the temperature evolution of the battery is monitored.Once the calorimeter has reached a predefined end temperature of a few hundred degrees, it switches to a non-adiabatic cooling mode in order to protect the test equipment and heater [12,22].ARC tests can clearly identify critical temperatures such as T 1 -T 3 .When the cooling mode is on, adiabatic conditions do not apply and heat dissipation occurs by forced convection around the vessel.However, the amount of heat loss is often neglected as the time span from cooling mode to T 3 is relatively short, typically seconds, and the cooling effect on the sample is limited.
For NMC batteries, standard parameters in ARC thermal runaway experiments include raising the temperature by 5 • C and then waiting for 30 min to observe when the self-heating threshold temperature is exceeded.These self-heating thresholds typically range from 0.01 to 0.02 • C per minute [12,13,[21][22][23]32,33]. The choice of this threshold rate affects the temperature at which self-heating is detected, with a higher exothermic threshold resulting in a higher detected self-heating temperature and a longer trigger time, as noted in Ref. [12].
The ARC methodology is limited by the size of the measurement chamber.For example, a standard ARC typically involves a cylindrical containment vessel of 9 cm diameter and 10 cm depth, limiting the sample to 0.64 L [34].Therefore, extended volume (EV + ARC) [34], yielding a space of ca.50 L, are often necessary, but involves extra costs.Another constrain is the that a test battery of too high capacity may damage the ARC facility.
The limitation of ARC is thereby partly the limitations in size of the test object, and partly to the availability of ARC testing facility due to the high cost.These can amount to above 0.2 million € [35].Thereby, it is desirable for relevant industries and research institutes to have a more accessible method to explore safety properties of batteries.
Here, we investigate the possibility to use thermal propagation test data to develop an alternative to ARC measurements with better availability.Thermal propagation tests are frequently employed in battery safety testing [11,36,37] in order to study safety properties of battery at module level.These tests usually include thermocouples at critical positions on the battery cells.However, the data from such tests have so far not being employed to find critical safety parameters such as T 2 , T 3 , temperature increase rate and heat release at cell level.In this study, we explore the possibility to utilize thermal propagation test data to determine these parameters.For the purpose of verification, published data of temperature increase rate obtained from ARC testing are compared in the discussion [1,13,[20][21][22][29][30][31].Some of these ARC studies also provide the possibility of estimating the heat release through the specific heat capacity of their experiment samples, and are also used for comparison.
Considering the challenges with ARC measurement, this study aims at developing an alternative and comprehensive method to study TR parameters, specifically employed on large format prismatic LiB cells and modules through the use of readily accessible instruments.To this end, a series of thermal propagation tests has been carried out at module level.The cell temperature is captured at several positions on the outer surfaces of the cells in order to follow the temperature distribution and evolution.A substitutional method is thereby proposed to investigate the TR processes, determining T 2 and T 3 , temperature rate, and total heat release during TR.

Battery and thermocouples
Prismatic NMC811/graphite LiBs were used for the test.The cell has been co-developed by Scania CV AB and Northvolt AB, and will be employed in an upcoming commercial vehicle model [38].Every prismatic LiB cell featured a capacity of 157 Ah, rendering this one of the largest existing commercial cells [39][40][41].The cells have a nominal voltage of 3.6 V, and were fully charged to 100 % state of charge (SoC) for the test.The heat capacity of the complete cell was estimated by the manufacturer as nearly 2000 J/K, which was utilized in the estimations of heat release.
The battery module (Fig. 1) comprised 17 cells and a heating element that took up the space of one extra cell in the module.Temperatures of the first five cells in the module were measured through installed thermocouples, thereby rendering it possible to follow the process of thermal propagation.Only the first five battery cells were measured, due to the limit of data logging channels, but this was considered sufficient to extract the main thermal parameters.As illustrated in Fig. 1, each cell had 5 thermocouples at different positions to capture the temperature evolution of a cell.
The thermocouples utilized were of K type, i.e. temperature sensors made of Chromel and Alumel.These generally have a wide operational temperature range, reasonable accuracy, low cost, and were primarily selected for their good accessibility.It should be noted, however, that it is challenging to obtain high-quality data using these sensors in a thermal propagation test, possibly due to that the thermocouples may experience short circuits in an electric field.The measurements resulted in a large extent of signal noise and unexpected missed logging of data, especially during the TR event.Therefore, a specific data postprocessing routine was necessary to implement (section 2.5).

Test setup
The tests were carried out on a module that was enclosed in a box, mimicking the use condition in a battery pack.The module included one trigger cell (the right cell in Fig. 1) comprising several heating elements and 17 battery cells, where thermal pads were inserted in between each cell.The thermal pads in the module are normally applied to hinder heat transfer from a thermal event, and thereby to prevent or stop thermal propagation.In practice, this allows the vehicle user significantly more time to escape from an observed event than if no thermal isolation exists.
The front measurement point (F) on a cell was located at the geometric center of the wide lateral surface facing the direction of heat transmission, and the back point (B) was the center of the opposite lateral surface.The left (L) and right (R) side measurement points, named according to the direction of heat transmission, were positioned at two thirds of the height on the narrow lateral surfaces.The top measurement point (T) was on the top surface next to the safety vent.
The lateral measurement positions (F, B, L and R) rendered information on the temperature distribution within the battery module, and the top measurement positions uncovered the venting behaviors of the cells.
The TR was triggered by direct contact of a block heating device according to the recommended triggering method by Chinese national standard GB 38031-2020 [42].This heated up the first neighboring cell at the front side (F) on its wide lateral surface until TR was observed by either exponential temperature increase or smoke.This, in turn, resulted in a similar triggering event for the next battery cell, and so on.

Determination of characteristic temperatures
The front (F) and back (B) side measurement points, compared to the left (L) and right (R) side measurement points, are closer to the core of the jelly rolls inside the cells and experience less heat dissipation to the module frame.Therefore, to determine the characteristic temperatures, a more trustworthy representation of the cell temperature was conducted from front (F) and back (B) side measurement data.The maximum temperature, T 3 , was by estimated from the average of the front and back measurement data, thus accounting for the uneven heat distribution due to the large geometry of the battery cell.Moreover, T 2 was derived solely from the back side measurement data, since the measurement at the front side was affected by the previous cell that had been heated through TR.

Estimation of heat release
The heat transmission within a battery module causes heat transfer from a high temperature cell to the other cells, its surrounding air and module frame.The heat transfer to the surroundings was neglected in all estimations.However, the heat transfer to the other cells could be significant considering high thermal mass of the battery cells.The accumulated heat release was estimated by summing up the contributions of all five the measured cells in the module through Eq. (2); this approach thus includes heat transfer between the cells: Eq. ( 2) quantifies the total heat release of the first five measured cells as a function of time, and the heat release from TR of each individual cells can then be separated in the time domain.In order to perform the calculation, the temperature of each cell during TR (T t,n ) needs to be defined.Here, the average of front (F) and back (B) measurement data was applied.
A similar method using temperature measurement has been employed previously to calculate heat release from adiabatic ARC tests [20,43,44].Moreover, the propagation test setup was open to atmosphere, but considering the short time span from the onset of TR to the end of heat release of the cell, typically around 15 s, the heat dissipation from the module frame should be negligible.
For example, Fig. 7 (discussed below) shows explicitly how the total heat release of the first five battery cells developed by each individual TR event.Additionally, the decay of the curve between each TR event indicates heat dissipation to the module frame and ambient air environment, and the rate of heat dissipation can be also assessed from the slope of the decay.

Data processing
The random signal noise mentioned above makes it difficult to derive accurate values from the sensors.Drastic oscillation and interruption of the signal in many channels were observed in the original dataset, especially after initiation of the TR when massive heat starts to be generated.Data processing, including cleaning, low-pass filtering and smoothing, was therefore necessary for both numerical and graphical analysis.
Outliers were screened for physical plausibility in the visual presentation of the results.This includes unfeasible data points, for example when the cell temperature during TR drops to negative values and shifts back to over 1000 • C in a time scale of seconds.
A Butter-Worth low-pass filter was implemented after data cleaning within the numerical differentiation framework developed by Floris van Breugel et al. [45,46].The Python-based library was published on Github [47].The level of noise was reduced by applying a filter with a cut-off frequency that was determined in the power spectra analysis of the dataset.

Results and discussion
In the following text, the TR reactions of the battery cells are characterized based on the thermal propagation through the module.An overview is given in Fig. 2, which displays the temperature profiles of the first five cells.As can be seen, the temperature increase in the module starts at Cell no. 1, which undergoes an exponential temperature burst from 136 • C to 863 • C.This event triggers a small rise in temperature in the adjacent cells, where Cell no. 2 displays an increase of about 65 • C. At the same time, the temperature declines in Cell no. 1 due to heat dissipation to Cell 2 and the module frame, until ca.1300 s.Then, both Cell no. 1 and 2 show a rapid increase in temperature due to the thermally triggered TR on Cell no. 2. A similar sequential phenomenon is then taking place along the series of cells.It is notable that the time intervals between these TR events become shorter from Cell no. 1 to 5.
It can be also seen from Fig. 2 that the heat transfer to the former cells where TRs have been triggered are much more significant than to the cells where TRs have not been triggered, e.g., when TR is triggered on Cell no. 3, there are substantial temperature peaks on Cell no. 1 and Cell no. 2, but only a comparatively minor increase on Cell no. 4 and 5.This might be due to that the aluminum cell casing melts during TR as the temperature reaches the melting point of 660 • C. Probably, the thermal pad also collapses under such high temperatures.The lack of thermal barriers makes heat transfer more significant to a battery cell that is already burnt by TR.
The TR behavior, including characteristic temperatures and different stages of TR were analyzed based on these temperature profiles.Moreover, some published ARC results are utilized for comparison.

Characteristic temperatures
The average T 2 for all five cells was 144 • C, i.e., somewhat higher than the PE melting point.This means that separation was sustained 10-15 • C above the melting point of PE before ISC occurred.In the literature there are examples of NMC batteries where separation is sustained until 243 • C [1].
Fig. 3 shows a decrease in the TR trigger time (t TR ) from Cell no. 1 to 5.This is likely due to that when the TR process propagates further into the module, e.g., to Cell no. 5, the cell was not only heated by its adjacent cell (Cell no.4), but also had been preheated by the module frame.In Fig. 4, the module frame displays an increasing temperature along with the TR propagation, which preheats the later cells in the module.The preheating phenomenon makes TR propagation in a battery module faster and faster.Preheating is observed on the other cells already when TR is triggered on Cell no. 1, as can be seen from Fig. 2.
Regarding T 3 , Cell no. 1 presented the highest value of 863 • C, while Cell no. 3 to 5 exhibited relatively low T 3 between 804 • C and 821 • C.This might be the result of extra heat input from the heater to Cell no. 1, as will be shown later, and can be attributed to the variation of TR trigger time.A long heating duration would lead to that more energy is accumulated by heating up more active materials of the cell before reaching the onset temperature of TR (T 2 ), and thereby generating significant exothermic chain reactions.profile, six stages of the TR development could be observed based on the temperature ranges and their corresponding temperature increase rates (with Stage I to V explicitly pointed out in Fig. 5).

Stages of thermal runaway
The stages seen here are similar to those identified by Feng et al. and described in the introduction.However, the temperature ranges are slightly different due to the differences in battery chemistry and format.
• In Stage I, the temperature of a cell increases with a high rate by means of heat transfer from its adjacent cell that has already been heated.At these high temperatures, it is known that metal ions in the cathode materials can dissolve to the electrolyte, with corresponding battery degradation in terms of both capacity and voltage [12,23,24].• In Stage II, the temperature increase of each cell slowed down at around 90 • C. The reason for this is likely that the dimethyl carbonate (DMC) in the electrolyte solvent started boiling [48].The lower temperature increase rate is thereby caused by vaporization, which consumes significant amount of heat.• In Stage III, the temperature increase rate started to rise again at around 120 • C.This is likely due to the coexisting exothermic reactions of SEI decomposition and regeneration.The decomposition of the SEI generally starts at around 100 • C with reactions of metastable complex compounds, thereby generating porous SEI layer structures that expose anode materials to the electrolyte.New SEI is consequentially formed upon contact between the electrode and the electrolyte [26].Y. Yang et al.
cathode, electrolyte, binder, anode, etc. Extensive fire and smoke were observed in this stage.
Fig. 5 also displays some differences between the different cells in the module; for example, different temperature intervals for the identified stages.The initiate temperatures of Stage III and Stage IV increased from Cell no. 2 to 5.This may be because of an increasing ambient temperature.
In this TR propagation setup, the distinct heat transfer through the battery module enables explicit illustration of these stages.This further strengthens this method for investigation of the thermal processes.For other methods, e.g., cell lateral heating, it is difficult to generate this type of analysis.

Temperature increase rate
In Fig. 6, the results for temperature increase rate from present study is compared to the state of art ARC studies on NMC batteries found in literature [1,20,22,29].The thick black curve in Fig. 6 illustrates the computed average temperature increase rate from Cell no. 2 to 5. The maximum temperature rate highlights the intensity of the TR reactions under thermal propagation in the module.The average temperature increase rate from Cell no. 2 to 5 reached a maximum of about 35 • C/s, corresponding to a peak heat release power of 70 kW according to Eq. ( 1).
As can be seen from Fig. 6, a similar exothermic behavior is observed from the published ARC data and the thermal propagation test.The temperature increase rate exhibits an exponential growth at T 2 , and is sustained around a high temperature rate until reaching close to the maximum temperature.Then it decreases, most probably due to the lack of reactants, e.g., active material.The tested 157 Ah prismatic cell batteries present an immediate increase when reaching a T 2 around 144 • C.This is a comparatively low temperature, considering the NMC batteries in the referred studies exhibited a similar growth between 140 • C and 260 • C.This might be due to that the ceramic coatings of the separators in the tested cells lost their separating capability earlier than in the other studies, but has yet to be confirmed.It is noteworthy that Cell no. 1 has higher temperature increase rate than the average temperature increase rate from Cell no. 2 to 5, indicating more drastic TR reactions for Cell no. 1.The intensity of TR could be caused by the heat input from the heater to the first battery cell, as well as higher oxygen content in the testing enclosure for the TR event of the first cell.
Moreover, it is noteworthy that the TR propagation test in this study displayed a comparatively low maximum temperature increase rate in contrast to the ARC testing that had been conducted at cell level.One contributing factor may lie in that this study utilized battery cells with higher capacity, resulting in a higher concentration of active material in the testing enclosure that is competing for the oxygen in the ambient air.

Heat release
The total heat release profile of the first five battery cells (see Fig. 7 (a)), calculated through Eq. ( 2), shows a step-like progression, with each "step" corresponding to the TR event of an individual cell (propagating from Cell no. 1 to 5).The vertical dimension of these "steps" quantify the heat release of the respective cell.The cell number is indicated by the timing of the heat increase step, as the input parameters are timedependent temperature data.
As can be seen in Fig. 7 (b), the first 5 battery cells within the module released a similar amount of heat, between 1.45 MJ and 1.74 MJ, yielding an average of 1.59 MJ.Consistently, the heat release amount assessed from a nail penetration test conducted on the same specification of the cell was 1.60 MJ [49].Noticeably, Cell no. 1 released the highest amount of heat, most likely because there was more energy input from the heater to the first battery cell compared to the other cells.Cell no. 5, in turn, released the lowest amount of heat, but also due to the absence of estimating the heat loss to Cell no.6 (where the temperature was not measured).
The average heat release was 1.59 MJ per cell, including heat transfer to other cells in the module, but not to the module frame and the ambient environment.This is still lower than the stored electric energy in a 100 % SoC cell, which is approximately 2.07 MJ.Besides the heat release, there is also gas emission during TR.This means that there is a large amount of energy released also by gas emissions.
The calculation of heat release using Eq. ( 2) neglects the change of specific heat capacity due to the mass loss during TR, which could be over 40 % for a high SoC level [50].This results in an overestimation of heat release.However, this is compensated by underestimations of heat release from e.g., undermeasurement of the cell temperature by not directly measuring the jellyrolls and through heat loss to the module frames.
Fig. 8 plots heat release of TR reactions from published ARC testing results and the research of this study.A linear relationship between heat release amount and the capacity of an NMC battery can be observed, displaying a slope of 10.65 kJ/Ah [1,13,[20][21][22][29][30][31].The findings from the thermal propagation test demonstrates a nominal heat release of 10.13 kJ/Ah, thus exhibiting an average heat release of 1.59 MJ per battery cell.As can be seen in Fig. 8, this value is closely aligned with the ARC-based regression line.This fact makes the thermal propagation test a credible approach for assessing heat release during TR.
The main difference of a non-adiabatic thermal propagation test and an ARC test is the test condition.While rapid heat dissipation can be expected in a thermal propagation test due to air turbulence, ARC offers an adiabatic experimental environment.However, in the ARC test vessel, the adiabatic condition is not ideal since the cooling mode is on at e.g., 300 • C, and there is also a heat exchange along the test vessel boundaries towards the calorimeter.The comparison above instead shows that considering the short time interval from the start to the end of a TR event, heat loss in both setups appears to be still minor.
As the heat capacity of the entire battery cell is in linear relationship with the heat release according to Eq. ( 2), is it of great importance to accurately determine this parameter.Nonetheless, the heat capacities of the specific cells are not always available from previous studies in literature.On the one hand, it is not straight-forward to measure heat capacity of such complex multi-materials objects as batteries and generalize from the obtained values.
On the other hand, there are more interests in TR behaviors than the heat release amount.Some studies examined heat capacity for the jelly rolls only, ending up at 1100 J kg − 1 K − 1 in the case of NMC532.This could exaggerate the heat release, because of an overestimated heat release for the aluminum cell container [12,13,20,21,29].
Some of the deviations seen in Fig. 8 can be explained from

Fig. 1 .
Fig. 1.The test setup of thermal propagation and thermocouple locations.F: middle of font surface facing the heating source; B: middle of back surface towards heat flux direction; L: 2/3 height at the left surface facing the heating source; R: 2/3 height at the right surface facing the heating source; T: top surface next to the vent.

Fig. 5 illustratesFig. 2 .
Fig. 5 illustrates Stage I to V of the mechanisms of TR propagation in the respective cells within the battery module.In this temperature

Fig. 3 .
Fig. 3.The onset temperature T 2 , the maximum temperature T 3 , and the trigger time t TR (green line) for the first five battery cells in the module.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4 .Fig. 5 .
Fig. 4. Temperature of the module frame on left and right sides of Cell no. 5.The temperature increases indicate the onset of TR from Cell no. 1 to Cell no. 5.

Fig. 6 .
Fig. 6.Temperature rate and heat release rate during TR for the cells tested here (black thick line) and from ARC data in literature [1,20,22,29].

Fig. 7 .
Fig. 7. (a).Total heat release of the first five cells; (b).Heat release for each cell derived from the curve of Q tot (t).

Fig. 8 .
Fig. 8. Correlation between heat production during TR and the battery capacity.Regression line slop = 10.65 kJ/Ah.The data is based on information from literature [1,13,20-22,29-31], apart from the thermal propagation test data from this current study.
total heat release of the measured cells as a function of time, MJ n the sequence of a cell in the module T t,n temperature of Cell no.n as a function of time, • C T t=0,n initial temperature of Cell no.n, • C T 1 the onset temperature for self-heating, • C (continued on next page) Y. Yang et al.