Preventing Cell-to-Cell Propagation of Thermal Runaway in Lithium-Ion Batteries

It is presumed that heat generated from a trigger cell under thermal runaway (TR) in multi-cell Li-ion batteries is transferred to adjacent cells mostly by convection of ejected hot matter (and to a lesser degree by direct contact and radiative heat transfer). Therefore, venting the energized materials (ejecta) from the battery compartment should prevent cell-to-cell TR propagation. However, engineering solutions to vent ejecta from TR of an individual cell fail to prevent TR propagation, subsequently causing battery ﬁ res. Real-time in situ FTIR spectroscopy of ejecta from a cell driven into TR demonstrates that large amounts of carbonate esters are already vented from the cell before it goes into TR. The vented hot gases cool down and condense on top of adjacent cells. Subsequently, when the trigger cell reaches TR, this condensate ignites, transferring heat and potentially driving the receiving cells into TR. Computational ﬂ uid dynamics and thermal simulations of this pathway support the experimental ﬁ ndings. Numerical results indicate that a fraction of the solvent vented from the trigger cell is suf ﬁ cient for ef ﬁ cient TR propagation. Our results shed new light on thermal propagation in multi-cell Li-ion batteries and suggest novel methods to prevent TR propagation. Our results shed new light and suggest novel methods

In a multi-cell lithium-ion (Li-ion) battery, cell-to-cell propagation of thermal runaway (TR) originating from a single cell, presents the highest risk for battery users and its operating environment. TR propagation can generate large amounts of heat and fire as well as toxic and corrosive materials, even if TR starts from only one ("trigger") cell. Cascading failures across multiple cells can also produce high-energy shrapnel. The most striking examples of such failures are in large (e.g., electric vehicle) batteries compared to failures in single-cell mobile phones (the latter due to poorlydesigned cells). The increasing market demands for electric vehicles, electric scooters, robots, aerial vehicles, power grids and entertainment electronics require large multi-cell Li-ion batteries in vast quantities. As the number of such batteries increases, so will the likelihood of more and more large battery failures that can cause fires and explosions.
Most of the experimental work and modeling of TR in large batteries has been focused on processes occurring inside an individual Li-ion cell. [1][2][3] So far processes of cell-to-cell propagation of TR are only discussed in terms of direct contact and radiative heat transfer as well as concurrent or subsequent convection of ejected hot gases and solids. Thus, currently existing measures for preventing TR propagation are limited to the use of thermal insulators and fire walls between cells, 4-6 natural convection and forced cooling that circulate coolants around the cells, 5,6 phase-changing chemicals, 7 and fire retardants mixed with the electrolyte. [8][9][10][11] A recent work by Lopez et al. suggests that increasing the spacing between adjacent cells has the propensity to reduce the risk of propagation. 12 Barring natural and forced cooling, most other techniques have not gained traction in the battery industry as measures in preventing TR propagation.
While the objectives of such studies 3,4,12 have been to prevent thermal propagation, the respective protocols suggest that they are better suited for delaying internally-generated TR. The main assumption in TR propagation is that both energy and material from the trigger cell are transported to "receiving" cells-all cells interacting with and/or affected by the trigger cell. Such TR propagation occurs mostly by heat convection (and to lesser degrees by heat conduction and radiation as indicated by Lamb et al., 13 and this heat transfer initiates TR in the receiving cells. Therefore, it is argued 5,6 that removing heat from the battery compartment by natural or forced convection (e.g., vent channels) should eliminate cell-to-cell propagation of TR. For example, natural or forced convection may help cooling the cells from the resistive heat generated during normal charging and discharging. 5,6 However, convective transport techniques found commonly in electric vehicles, are not implemented in other large Li-ion batteries, whose housing is sealed. Sealed battery housings are efficient traps of heat from TR in a single cell, thus promoting TR propagation to other cells in the battery.
Most technological approaches described above [4][5][6][7]12 have limited application in battery manufacturing. For example, fire walls and thermal insulators may help prevent TR propagation, but they will also prevent heat dissipation during normal charge and discharge, thus increasing the chances of heat-induced cell damage and TR. Addition of fire-retardant chemicals to the electrolyte increases a cell's internal resistance increasing the Joule heating (or i 2 R heating). [8][9][10][11] The ideal properties for using fire-retardants in electrolytes, as described more than a decade ago, continue to remain out-of-reach. 14 Phase change chemicals have not been widely used, probably due to added weight and volume inherently diminishing gravimetric and volumetric energy densities of the battery. Reducing the probability of TR propagation through increased spacing between cells suggested by Lopez et al. 12 appears most practical among all the proposed techniques. Increasing the spacing will increase battery volume with little increase in its weight, a worthwhile compromise for improving safety against fire and explosion. Thermal propagation modeling has also suggested that appropriate air gaps and mica insulation combined with a thermally conductive matrix can reduce the TR risks in Li-ion batteries. 15 Cooling through appropriately-designed mini-channels has been proposed as a method for suppressing cell-to-cell TR propagation. 16 Here we report another physical-chemical pathway that channels more than half of the thermal energy transferred from the trigger to the receiver cells through a mode that is unrelated to the three currently-described modes (conduction, convection and radiation). 4 We have found that a commonly observed, but mostly ignored phenomenon, namely, venting of gaseous matter out of a trigger cell z E-mail: Rengaswamy.Srinivasan@jhuapl.edu *Electrochemical Society Member.
as it progresses toward but before undergoing actual TR, appears to be primarily responsible for TR propagation. We use in situ highspeed hyperspectral and Fourier Transform Infrared (FTIR) imaging techniques, off-line chemical analysis by Gas Chromatography-Mass Spectrometry (GC-MS) of the vented gases, and gravimetry, to determine the temperature, chemical composition and mass of the pre-TR vented matter. In fact, before going into TR the trigger cell vents large amounts of organic carbonate esters that are highly flammable. Upon exiting the cell, the vented hot gases cool down and condense as a liquid on top of adjacent receiving cells. Subsequently, when the trigger cell reaches TR mode, the flame ignites the liquid solvent that burns on the surface of receiving cells, transferring heat into them, and potentially driving them into TR. Computational fluid dynamics (CFD) and thermal simulations of this four-step pathway-venting of solvents from the trigger cell prior to TR; condensation of the solvent on receiving cells; flow of hot ejecta from the TR to the receiving cells; and burning of the solvent on the surface of the receiving cells-complement the experimental studies. The numerical results indicate that a small fraction of the total solvent vented from the trigger cell is sufficient for successful TR propagation. Our results shed new light on the thermal propagation processes in multi-cell Li-ion batteries and suggest novel methods to prevent TR propagation.

Experimental
Cell and battery.-In our experiments we characterized model LG HG2 18650 cells (LG Corp, South Korea). The contents of the LG HG2 cells, including the carbon anode, carbonate-based organic solvents, the LiPF 6 salt in the electrolyte, the melting separators and the NMC cathode, are representative of most commonly used materials in most of the Li-ion cells produced today. There are a number of published studies discussing TR behavior in cells with similar contents. 17,18 Freshly-procured LG HG2 cells were condition-cycled by first discharging at 21°C, at C/4 rate to 2.7 V, then fully charging using constant current (C/4 rate)-constant voltage (CC-CV) protocol to 4.2 V. Forced thermal runaway was initiated using a 20-W thin-film heater wound around the cylindrical surface of the cell. A K-type thermocouple, firmly attached near the positive terminal monitored the cell's surface temperature (T surf ). Nickel tabs, welded to the cell terminals, were connected to a Solartron Frequency Response Analyzer, SI 1250 (France) through the Electrochemical Interface, SI 1287 (France). The heater and the wires were held in place by wrapping them in Kapton tape. Cell voltage, impedance and T surf were continuously monitored during heating. The assembled cell was held tightly inside a 10-cm long, 19.2-cm diameter FR4 tube, with the positive terminal end recessed roughly 0.5-cm from one end of the tube. In this cell assembly, when thermal runaway was initiated, the cell almost always ruptured at the positive terminal end, never at the negative terminal end, and the cell's side wall was rarely damaged. The positive terminal-end rupture generally resulted in a 0.6-cm diameter hole at the location of the positive terminal, without damaging the crimp. A highresolution X-ray CT scanner (North Starr X-50) with a voxel resolution of 12.8 μm × 12.8 μm × 12.8 μm was used to scan the interior of an intact cell before and after pre-TR venting ( Supplementary Fig. S1, available online at stacks.iop.org/JES/167/ 020559/mmedia). We also measured the temperature of the vented gas along a vent channel using K-type thermocouples positioned at different locations away from the vent valve of the cell. The cell was weighed before and after venting in order to determine the amount of material that vented.
FTIR spectroscopy.-Thermal runaway tests were conducted in open air, allowing spectral measurements of the ejecta by properly placing hyperspectral and Fourier Transform Infrared (FTIR) imagers. The hyperspectral imager is a Starlight Xpress spectrograph containing a Lodestar CCD guide camera for visual alignment, and attached to a QHYCCD Monochrome Cooled Astronomy Camera (Model QHY163M) used for spectral light collection. Measurement capabilities of the instrument include wavelength range 350 to 900 nm (in ∼300 nm steps), 0.25 nm spectral resolution, with capture rate of 400 fps. Wavelength is calibrated with a Hg-Ar low pressure lamp source. The FTIR instrument is a ABB Bomem FTIR Spectroradiometer (Model MR304) containing both MCT and InSb detectors for simultaneous collection of mid-wave IR (3-5 μm) and long-wave IR (8-12 μm) spectral radiation, and calibrated with a NIST traceable Blackbody source. The two imagers were kept away from the cell, focused on the path of the ejecta. Two high-speed (1-frame ms −1 ) video cameras (Photron, Model SA4) captured the pre-TR and TR events. Time-resolved emission and absorption IR spectra of the ejecta were recorded at the rate of 2.5-ms per frame, from a distance of 170 cm away from the FR4 tube (holding the cell assembly), while focusing over a 12-cm × 12-cm cross-sectional area in front of the tube where the ejecta came out. The cell assembly, the wires leading to the impedance meter, the high-speed video camera and hyperspectral and FTIR imagers are shown in Supplementary Fig. S2.
GC/MS analysis.-The solvents vented by the cell before it experienced TR were characterized using Gas Chromatography-Mass Spectrometry (GC-MS). The instrument used was Agilent Technologies, USA, 7890 Gas Chromatograph with a 5977 A Mass Spectrometer. The vented solvents were first collected in a glass bottle, extracted and subjected to GC-MS analysis ( Supplementary  Fig. S3).
Simulations.-The CFD++ package (Metacomp Technologies Inc., Agoura Hills, CA) was used to conduct Computational Fluid Dynamics (CFD) simulations of the vented gas and ejecta matter. ATLAS (Aero Thermal Loads and Stresses: a simulation program developed at the Johns Hopkins Applied Physics Laboratory) was used for thermal modeling to estimate heat transfer from hot ejecta matter to receiving cells. The CAD drawing of the vent channel geometry for a simulated 3S15P battery is shown in Supplementary  Fig. S4. The input parameters used in CFD simulation of ejecta flow, following TR in the trigger cell are listed in Supplementary Table I. The thermal properties of cell and battery materials as well as parameters used in CFD simulation of the pre-TR venting of dimethyl carbonate (DMC) are all listed in the Supplementary Tables II and III, respectively.

Results and Discussion
Pre-TR venting and TR event analysis.-The chemical constituents of a Li-ion cell can be quite reactive, although they are stable at temperatures below 80°C. 3 Instability starts when a cell is heated above 85°C, when the protective solid-electrolyte-interphase (SEI) layer on the carbon anode begins to break. 19 It is then followed by exothermic reactions between the anode and the electrolyte and the evaporation and degradation of the organic solvents between 85°C and 125°C. Most cells are designed with vent valves and a burst disc that is contained in a current interrupter device (CID). Temperature increase above 85°C is accompanied by generation of gases and an increase in internal pressure. When the pressure reaches a preset value (about 1224 kPa in most cells), the valves and burst disc will open to vent the gases and reduce the cell's internal pressure ( Supplementary  Fig. S1). Venting typically occurs when T surf is in the 125°C to 145°C range. The released gases frequently contain flammable organic chemicals whose boiling points may be below 125°C. There are multiple means to trigger TR in a Li-ion cell, 13 but in each circumstance, the occurrence of TR is preceded by an increase in temperature of the cell and cell venting. If the cell temperature rises only up to 145°C, it may not experience TR, but it will most likely vent. If the cell temperature exceeds 170°C, the cell may go into thermal runaway and deflagrate in the 190°C to 200°C range. The relationship between cell surface temperature (T surf ), cell voltage (E cv ), and the sequences of venting and TR are shown in Fig. 1.
All Li-ion cells contain mixtures of short linear aliphatic carbonates as electrolyte solvents. 2,3,20,21 The LG HG2 18650 cell contains a mixture of dimethyl carbonate (DMC), ethylene carbonate (EC) and propylene carbonate (PC). The boiling points of EC and PC at normal temperature and pressure are nearly identical (242°C). DMC boils at 90°C. By heating an over-discharged cell above 200 o C for over 3 h, we evaporated and vented off all solvents and estimated the total mass of the three solvents to be around 4.4 g. Heating the cell causes the vent valves to rupture ( Supplementary  Fig. S1), typically between 125°C and 145°C T surf . Presumably, the cell's internal temperature is lower than T surf but higher than the DMC boiling point. The duration of pre-TR venting is about 300 ms ( Supplementary Fig. S7b). The temperature of the vented gas along the vent channel decreases rapidly. At 0.25 cm from the valve, DMC was at 95°C, suggesting it is a gas, and at 8 cm and 20 cm it was 64°C, and 21°C (ambient temperature), respectively, suggesting fast condensation within short distances. We have characterized the gas that vented below 145°C using two different techniques: in situ time-resolved FTIR spectroscopy and off-line GC-MS analysis of material collected during venting. Most of the material that vented below 145°C is DMC (Fig. 2), independent of the state-of-charge of the venting cell. The FTIR emission spectrum indicates that DMC is venting out as a gas. Time-resolved FTIR emission spectra show that DMC gas concentration is decreasing, while FTIR absorption spectra suggest that subsequently it condenses as a liquid.
We have also analyzed the gas released during cell venting by collecting all the released gases in a clean glass bottle while heating the cell up to 141°C (Supplementary Fig. S3). A few minutes after venting, all gas that vented condenses as a liquid. We have also analyzed the condensed liquid using GC-MS, and the results confirm the presence of DMC as the dominant species, with EC and PC as minor components (Supplementary Fig. S3). Gravimetric measurements suggest that DMC is approximately 2.1 g.
During heating, there is a two-minute interval between pre-TR venting and TR, which allowed us to purge the cell of DMC solvent. If the DMC is not removed before TR, then igniting the vented gases in free space generates a flame in excess of 5 feet from the cell lasting from 250 to 900 ms, as inferred from high-speed video ( Supplementary Fig. S7a). Venting associated with TR results in 30 grams of solids, liquids, and gases being ejected from the cell over a period of 1 s or less. If DMC has been removed during the pre-TR venting, the flame during TR is much smaller. When DMC has been ignited by a sparker, it burns for approximately 300 ms, suggesting the pre-TR venting event of the flammable solvent is short-lived. After TR, the ejecta do not contain the three organic solvents (DMC, EC and PC). Present are CO 2 , CO and H 2 O, most likely resulting from the complete oxidation (burning) of DMC, EC and PC. Gaseous HF is also present in the ejected matter, and it could dissolve in H 2 O to form hydrofluoric acid, a powerful corrosive reagent. The initial temperature of all gases during TR is approximately 1500°C that drops in less than a second to about 600°C.
Walker et al. measured the heat released by an LG HG2 18650 cell during TR using a "Thermal Runaway Calorimeter." The heat released is about 49.8 kJ. 18 Their setup did not have a vent channel for released gases to escape. It also demonstrated that the oxidizers needed for combustion did not come from the outside environment, but entirely from the cell. Our TR tests are conducted in open-air providing sufficient oxygen for burning. They suggested that the main component contributing to fire during TR is DMC. In all our tests, upon cell heating, the cells first vent DMC out around 130°C before experiencing TR around 190°C. More importantly, FTIR imaging shows that the ejected matter contains all cathode-active materials. They are chemically intact, although their physical state is changed. The in situ FTIR testing (data not shown) also revealed presence of cobalt oxide and manganese oxide in the vapor state, and nickel oxide emerging as a solid when the cell undergoes TR. These three metal oxides can fully oxidize all organic solvents without exposure to atmospheric oxygen, as has been well established. 22 Based on chemical reaction stoichiometry, we have determined that the quantity of the oxides present in each LG HG2 cell is sufficient to combust and set the organic solvents inside the cell on fire.
Three-dimensional CFD of pre-TR venting.-Our initial CFD results highlight the contours corresponding to the pressure, temperature, velocity and Mach number in a battery vent channel without explicit modeling the type of gas being vented ( Supplementary Fig. S5). These results suggest that for the selected vent channel geometry TR propagation should not occur. In reality, experiments show that even in this case TR propagation is occurring. Therefore, we expanded our CFD simulation to include explicitly pre-TR cell venting of DMC and its distribution along the channel. The input parameters used in this expanded CFD simulation are listed in Supplementary Table III. Since the pre-TR venting duration is around 300 ms, we used 10 μs as the time-step in the simulation. As already established, the temperature in the vent channel is <90°C, which is below the boiling point of DMC. Therefore, in the simulation, DMC inside the channel is assumed to be in liquid phase. Figure 3   . Both E cv and T surf are not sensitive to cell pre-TR venting. Therefore, cell voltage monitors and surface-mounted thermocouples do not detect this venting event. More than 30 s past venting, E cv oscillates before dropping to 0 V, presumably due to activating the current interrupter device in the cell.
ΔH c is approximately 32 kJ, representing more than 60% of the 49.8 kJ heat generated by the cell during TR without removing DMC. 18 The mass of DMC (M dmc ) deposited on a receiving cell, the heat (ΔH c ) generated on top of the cell while DMC is burning, and the fraction of that heat (ΔH f ) transferred into the cell will determine the rise in internal temperature (T int ) and the possibility of driving that cell into TR. Using CFD and thermal modeling techniques, we estimate M dmc , ΔH f and T int on receiving cells to identify the critical amount of M dmc necessary for driving a receiving cell into TR. We estimate the probability for TR initiation in a receiving cell under two different operating conditions. The first modeling iteration assumes that DMC deposited on a receiving cell is being ignited before the trigger cell goes into TR. In such a simplified scenario, combustion of the deposited solvent is the only source of heat. In the second more realistic modeling scenario, DMC deposited on a receiving cell is ignited at the same time the hot ejecta from the trigger cell flow on top of the receiving cells. In the second case, heat from the hot ejecta could add to the heat from the combusting solvent already deposited on the receiving cell, and part of the heat of combustion could be carried away by the flowing ejecta. The first case in which the ejecta flow is absent, poses a simpler scenario allowing to train and evaluate the performance of the modeling algorithms.
For the first scenario (DMC deposited on a receiving cell being ignited before the trigger cell goes into TR), the following simplifying assumptions have been made. DMC is assumed to be uniformly deposited on top of the cell and layer thickness is derived from data in Supplementary Table IV. Heat and mass flux results from the CFD simulation of the ejecta channel (Fig. 3) are not superimposed on the unidirectional heat flux from the combustion of DMC on top of the receiving cell. The heat-releasing chemical interactions between the solvents and the SEI layer, at the anode and  the cathode, constitute thermal runaway reactions. Thermal property values of the jellyroll electrodes, liquid electrolyte, steel can wall, and thermal runaway reactions are taken from Hatchard et al. 23 Heat flux from the combusting DMC is propagating only in one direction -toward the cell's interior. All reactions are assumed to go to completion in 10 s from start. The thermal simulation grid for the single LG HG2 18650 receiving cell is shown in Fig. 4a. Simulated heat flux results after 10 s, for two extreme cases of deposited DMC (22 mg and 103 mg) are shown in Figs. 4b and 4c, respectively. Figure 5 shows a graphical representation of the T surf evolution resulting from the heat flux and possible exothermic reactions inside the cell initiated by the heat flux. Heat from 103 mg DMC when burning can cause TR, evidenced by the sudden increase in temperature starting at about 245°C. Heat from 22 mg DMC increases the temperature to about 50°C, but is insufficient to initiate exothermic reactions leading to TR. Heat from 95 mg or less does not initiate TR either. However, if T surf reaches 150°C potentially those cells can start venting and adding DMC on top of other cells, aiding more heat generation and driving more cells into TR (this scenario has not been modeled here).
In the more realistic second scenario (combined thermal effect of ejecta from the trigger cell and DMC combustion) we assume that not all heat from combusting DMC is transferred into the receiving cell. In reality, heat propagates in multiple directions; additionally, heat originating from the hot ejecta matter flow also through the channel. Furthermore, DMC combusts only when the trigger cell reaches TR and the metal oxides in the ejecta (oxides of cobalt, nickel and manganese) are available to sustain combustion. Therefore, the two sources of heat would be acting simultaneously, and both should be accounted in estimating the heat flux. In addition, the direction of heat flux should be both toward the cell and away from the cell, for example, along the ejecta channel. Running the thermal simulation using the above assumptions generates the data shown in Fig. 6. As seen from Fig. 5, 103 mg of burning DMC increases the temperature of the receiving cell's surface to about 200°C. On the other hand, simulation results shown in Fig. 6 suggest that even though the ejecta matter is much hotter (1500°C), this heat flux appears to be partially countered by the gas flow dynamics. Heat from only 200 mg burning DMC causes TR propagation, evidenced by the sudden increase in temperature starting at about 2 s after TR starts in the trigger cell. Therefore, approximately 200 mg of deposited DMC would be sufficient to initiate TR in a receiving cell in this case as well. As noted, CFD simulations demonstrate (Fig. 3) that several cells receive more than 200 mg DMC during pre-TR venting. These results suggest that functional ejecta channels may not prevent TR propagation.
TR propagation is a multistep phenomenon.-TR propagation includes several sequential processes, some of which actually precede TR in the trigger cell. First, the trigger cell vents gaseous organic solvents even before it undergoes TR. Next, the vented solvents condense on the outer surface of receiving cells. After the trigger cell goes into TR, it generates heat and provides the oxidizers to ignite the solvent deposited previously on top of the receiving cells. While the combusting solvent starts heating the receiving cells, hot ejecta from the trigger cell continue to pass over them. Part of the heat from these two independent sources-burning solvent and  hot ejecta-is transferred into the receiving cells. Our results show that if the solvent deposited on a receiving cell is on the order of 200 mg the receiving cell will itself go into TR, facilitating TR propagation inside the battery.

Conclusions
Propagation of TR is, perhaps, the major reason for fire and conflagration in large multi-cell Li-ion batteries. While the probability of spontaneous TR in an individual Li-ion cell is low-about one-in-tens of millions-such a TR event leading to battery fire and explosion increases with the number of cells in a battery as well as the number of batteries deployed world-wide. Therefore, preventing TR propagation is of paramount importance. It is commonly assumed that hot ejecta from a trigger cell in TR when vented through a properly-designed vent channels will prevent TR propagation. Our experiments and CFD and thermal modeling results suggest that TR propagation in a multi-cell Li-ion battery is a complex process. In particular, organic solvents venting out before a cell experiences TR play a major role in cell-to-cell TR propagation. Our finding ultimately has implications for innovative technical solutions aimed at preventing large multi-cell battery fires and explosions.