Thermal runaway of commercial 18650 Li-ion batteries with LFP and NCA cathodes – impact of state of charge and overcharge

Thermal runaway characteristics of two types of commercially available 18650 cells, based on Li x FePO 4 and Li x (Ni 0.80 Co 0.15 Al 0.05 )O 2 were investigated in detail. The cells were preconditioned to state of charge (SOC) values in the range of 0% to 143%; this ensured that the working SOC window as well as overcharge conditions were covered in the experiments. Subsequently a series of temperature-ramp tests was performed with the preconditioned cells. Charged cells went into a thermal runaway, when heated above a critical temperature. The following thermal runaway parameters are provided for each experiment with the two cell types: temperature of a ﬁ rst detected exothermic reaction, maximum cell temperature, amount of produced ventgas and the composition of the ventgas. The dependence of those parameters with respect to the SOC is presented and a model of the major reactions during the thermal runaway is made.


Introduction
Li-ion batteries 1,2 excel in energy density and cycle life. Unfortunately those benets come with a price: when Li-ion batteries are mistreated with high over-temperature or strong overcharge, they can transit into a so-called thermal runaway. During the thermal runaway, the battery temperature increases due to exothermic reactions. In turn, the increased temperature accelerates those degradation reactions and the system destabilizes. At the end of the thermal runaway, battery temperatures higher than 1000 C can be reached and high amounts of burnable and harmful gases can be released.
Because Li-ion batteries are widely used, the possible hazards of Li-ion batteries are a key issue for automotive, aerospace and consumer electronics industries. The safety characteristics of Li-ion battery systems depend (a) on the used cell type (geometry, materials), (b) on the initial conditions before misuse (state of charge, ageing effects), (c) on the type of misuse (over-temperature, over-charge) and (d) on external measures (built-in safety devices, forced cooling, connement). [3][4][5] In the past, accelerated rate calorimetry (ARC) tests with limited maximum temperature [6][7][8][9][10][11][12] and without limitation 13 as well as re experiments and mechanical abuse [14][15][16][17] with complete Li-ion cells were done. Recently over-temperature and over-charge tests with large format cells (which may be used for automotive applications) were published. [18][19][20] It is known that the severity of the thermal runaway event in overtemperature experiments increases with increasing SOC. 4,[21][22][23][24][25][26][27] It is also known, that a thermal runaway can be triggered by strong overcharge beyond safe voltage limits of the cell. [28][29][30][31][32][33][34] Even if the overcharge condition does not trigger a thermal runaway, safety may be compromised by Li-plating on the anode. 35 In our previous publication 36 the safety characteristics of three different commercial Li-ion batteries charged to 100% SOC were investigated. It was demonstrated, that cells with cathodes based on iron-phosphate as well as on metal-oxide material exhibit a thermal runaway in thermal-ramp experiments. The severity of the thermal runaway showed a strong dependence on the material composition of the cells.
In this publication two cell types are introduced and the mass inventory of the cells is calculated based on tear down results. The thermal runaway testing method is explained and the outcomes of experiments with discharged, partially charged, fully charged and over-charged cells are presented. Possible chemical reactions are listed and quantitative calculations of ventgas generation are made for two cases.

Samples
The two types of commercially available Li-ion batteries, with the geometrical format 18650, were purchased from two well known manufacturers. The rst cell, rated to a nominal capacity of C nom ¼ 1.1 A h is based on a Li x FePO 4 (LFP) cathode. The LFP material is considered as relatively safe. Unfortunately commercial LFP-based cells have lower capacity and nominal voltage compared to metal-oxide based cells. According to the datasheet the LFP cell is designed for a maximum discharge current of 30 A and has a cycle life of >1000 full discharge cycles.
The second cell has a much higher nominal capacity C nom ¼ 3. 35 A h and is based on a Li x (Ni 0.80 Co 0.15 Al 0.05 )O 2 (NCA) cathode. To our knowledge, this mass produced cell has the highest energy density which is commercially available as of 2013. It is specied to a maximum discharge current of 6.7 A h and its cycle life is >300 cycles.
In the following, the two cell types will be denoted as LFP and NCA for easy reading.

Cell composition, methods
For the interpretation of the misuse experiment results it is benecial to know the mass split of the cell components. Unfortunately information regarding detailed cell composition is kept condential by the manufacturers. We had to make a tear down and an analysis of the cell components for both cell species by ourselves. The following parameters were measured directly using the same methods and equipment as in ref. 36: Mass of the anode and cathode coating, the electrolyte, the current collector foils, the separator and the housing material.
The mole-ratios of the different transition metals and phosphor in the cathode coating.
Additionally, separator foils were examined with differential scanning calorimetry coupled with thermal gravimetric analysis (DSC-TGA, NETZSCH STA 449 C). Separator samples were rinsed with diethyl carbonate and dried in a desiccator for 12 hours. During the test the DSC-TGA was ushed with and the heat ramp was set to 10 K min À1 .

Cell composition, results and discussion
It is not in the scope of this work to compile an exhaustive material inventory of the two commercial cell types. Nevertheless, to obtain some insight into chemical reactions taking place during cell misuse, it is helpful to make at least rough estimations for cell components that were not accessible to direct measurements (Table 1). Estimations for the amount and composition of active material, particle coating, binder, carbon black and the SEI in the electrode coatings as well as for the amount of salt, additives and soluble SEI in the electrolyte were discussed with our project partners. Effects of cell formation were considered. The compositions of the separators were estimated from DSC measurements.
2.2.1 Binder and conducting agent. The mass ratio of binder material and conducting agents in the electrode coatings was not measured. We assume that sodium carboxymethylcellulose (CMC) with a degree of carboxymethyl substitution (DS) of 0.7 is used as the anode binder 37 and polyvinylidene uoride (PVDF) is used as the cathode binder. 38 CMC is a cost effective binder material in the anode, but can not be used in the cathode. We suppose that 5% of anode coating and 2.5% (NCA) or 5% (LFP) of cathode coating is binder material.
Additionally a conducting agent is needed to improve the electrical conductivity between the cathode particles and cathode substrate-foil. We suppose that 2.5% (NCA) or 5% (LFP) of cathode coating consists of carbon black. We justify the increased amount of binder and conducting agent of the LFP cell with its higher power capability.
2.2.2 LFP particle coating. The active cathode material of the LFP cell consists of Li x FePO 4 . The Li x FePO 4 particles need to be nano structured and carbon coated to achieve good diffusion of Li-ions and good inter-particle electrical conductivity. 39 It is hard to tell which amount of carbon coating was actually used in the tested commercial battery. Optimum values of carbon coating found in the literature vary from 1.5% to 15%. 40 We assume that 10% of the LFP cathode consists of carbon coating. Please note, that this might be the upper estimate. One of the reviewers suggested, that the carbon coating of a commercial battery is probably in the range of 1% to 2%.
2.2.3 Electrolyte and SEI. The amount of salt in the electrolyte could not be measured as well, it is supposed that both cells use the traditional salt LiPF 6 with a concentration of 1.1 mol L À1 . The density of the electrolytes is estimated with 1.21 kg L À1 .
Vinylene carbonate (VC) is a common solid electrolyte interface (SEI) improving additive. 41 We assume that 2% of VC was added to the electrolyte. 42 During initial charging VC and EC undergo reduction reactions and form the SEI at the surface of the graphite particles of the anode. A fully developed SEI prevents further reduction of the electrolyte solvents. 43 The SEI composition and formation reactions can be complicated 41,44,45 and lie beyond the scope of this work. Instead, for further calculations, we treat the SEI as being made of only four components: (1) The polymerization product of VC 41,46 (1) (2) The organic Li-carbonate from EC reduction 47-49 Table 1 Mass split of the discharged NCA and LFP cell. Please note, that the mass ratios for the binder, carbon black, the SEI and the salt were not measured; instead rough estimates are given.
(4) And LiF which can be produced from decomposition of the salt and the Li-carbonate 53 We assume that all VC (2% of electrolyte) goes into polymerization (1) and that the additional SEI components (CH 2 OCO 2 Li) 2 : Li 2 CO 3 : LiF are in the ratio 1/2 : 1/4 : 1/4. 44 The components of the SEI are listed (Table 1) as a part of either anode or electrolyte depending on their solubility in the electrolyte solvent. 54 To calculate the actual amounts of lithium containing SEI we need to take the irreversible capacity loss into account.
2.2.4 Irreversible capacity loss. We think that the most economical anode material for both manufacturers is surface treated natural graphite. During cell assembly the graphite is in delithiated state and the cathode is in fully lithiated state. At the rst charging (cell formation) an amount of lithium n irr Li that is equivalent to $8% of the maximum anode-Li-capacity is trapped. 2 The associated charge C irr is called irreversible capacity loss: here F is the Faraday constant and n a C6 is the amount of graphite units C 6 in the anode (in mol). We assumed that all trapped lithium is integrated and immobilized in the SEI according to the chemical reactions (2)-(4). The calculated values for the NCA and LFP cell are n irr Li (NCA) ¼ 12.1 mmol and n irr Li (LFP) ¼ 5.4 mmol respectively. As a consequence, aer formation, the cathode can never again be fully lithiated. Even when the cell is fully discharged, n irr Li is missing, and the amount of Li per stoichiometric formula in the cathode is <1.
The effect of the missing lithium n irr Li (proportional to C irr ) in the cathode is taken into account in further stoichiometric calculations.
2.2.5 Residual capacity. Commercial Li-ion cells must not be discharged beyond their rated minimal operation voltage (V min (NCA) ¼ 2.5 V and V min (LFP) ¼ 2.0 V) during normal cycling. If cells are discharged to voltages lower than V min dissolution of the copper foil may occur, 55 because the anode potential may reach the oxidation potential 56 of Cu. Anodes of cells that are discharged to V min are not fully delithiated, instead a small amount of Li stays in the anodes and acts as a safety margin to keep the anode potentials below the copper dissolution potential. We assume that the residual capacity C res (which is proportional to the amount of residual Li n res Li ) equals to 1% of the nominal cell capacity: The amount of residual lithium is considered in further calculation of the lithiation states of both electrodes.

Available capacities in the electrodes
With identied amount of active cathode material n a cat and with known C irr and C res the theoretically usable capacity of the cathode aer cell formation can be calculated and compared to the nominal capacity as given in the data sheet.
In the case of LFP cell C u cat ¼ 1.16 A h. In theory, LFP material can be fully delithiated, and C u cat should be equal to C nom . In our work, the calculated C u cat exceeded C nom . According to the data sheet the LFP cell is rated to C nom ¼ 1.1 A h and the measured capacities in the allowed voltage range were even smaller ( Fig. 1). The discrepancy may be caused by incomplete utilization of the LFP material of a real cell or by ageing effects of the cathode. It is noteworthy that the available capacity of the LFP anode exceeds the C u cat by 50%. In other words, the anode of the LFP cell is overbalanced. This makes sense for a high power cell, as it allows high charging currents with reduced risk of Li-plating.
In contrast to LFP, the NCA cathodes should not be fully delithiated during normal operation. Correspondingly, the theoretically available capacity of the NCA cell of 4.42 A h was higher than the nominal capacity 3.35 A h. The calculated capacity of the active material in the anode was 4.06 A h. That means the NCA anode was slightly overbalanced by 21%.
2.3.1 Separator. The composition of the separator materials was deduced from DSC-TGA measurements. The separator of the LFP cells showed endothermic (melting) peaks at 132 C and 159 C which are typical for a 3-layered laminate with a polyethylene (PE) core between two polypropylene (PP) skin layers (PP/PE/PP). We assume that the LFP separator consists of 2/3PP and 1/3PE.
The separator of the NCA cell showed only one indistinct endothermic peak at $130 C. We assume that the NCA separator consists of ultra-high molecular weight polyethylene (UHMWPE) membrane. 57,58

Experimental
In this work a total of 23 thermal ramp experiments with the two cell types were done at different SOC. Each experiment consisted of the following steps; the cell underwent a open circuit voltage (OCV) check, was charged to the selected SOC and inserted into the sample holder. The sample holder was attached inside a sealed reactor and the thermal ramp experiment was started (the test-rig and thermal ramp method is described in ref. 36). Aer the thermal ramp experiment gas samples were taken and analysed.

Initial OCV check
We applied the same OCV measurement procedure as in ref. 36. Each sample was fully discharged to 0% SOC (2.5 V) and then fully charged to 100% SOC (LFP: 3.5 V, NCA: 4.2 V). The health status of the cells was checked by comparing the measured capacities with the nominal capacity from the manufacturer. Typical OCV proles are given in Fig. 1. BaSyTec CTS cell test system and Heiden Power DC-source-load were used for battery cycling.

Sample preparation
Aer the OCV check the insulation foil was stripped from the cell and the sample was weighed. Three K-type thermocouples were spot-welded to the cell housing. Then the sample was wrapped in a thermal insulation layer and inserted into the heating sleeve of the sample holder. Finally, the sample holder was installed inside the reactor, the electrical connections were made and the reactor was sealed.

SOC set-point
The cell was brought to the desired SOC by charging or discharging, starting from 100% SOC. The coulomb counting method was used for SOC calculation and the charge/discharge was stopped when the required SOC was reached. For experiments with SOC < 100% the cell was discharged outside of the reactor. For SOC > 100% the cell was overcharged inside the reactor, for safety reasons. In order to prevent cell heating, the overcharge current was set to very low values. The SOC setpoints of all experiments are marked in Fig. 1.

Thermal-ramp experiment
The sealed reactor was evacuated and ushed with inert gas. The heaters were turned on. The sample inside the reactor was heated slowly with a rate of 2 C min À1 (NCA) or 4 C min À1 (LFP). Cell temperatures, gas temperatures and the pressure in the sealed reactor were recorded. At some point the cell transited into thermal runaway and ventgas was released in the reactor. The amount of gas inside the reactor n ideal sum was calculated using the ideal gas law Here p denotes the pressure in the reactor, V ¼ 0.0027 m 3 is the reactor volume, R is the gas constant, q gas is the gas temperature in the reactor (in K) and n 0 is the initial amount of gas in the reactor at the start of the experiment.
The eqn (11) is only valid, when q gas is equal to the mean gas temperature in the reactor. During the thermal runaway a violent cell venting may take place and hot gases are released into the pressure vessel. In the rst seconds aer venting, when the gas temperature inside the reactor is not homogeneous, n ideal sum may be over or underestimated. Thus, given n ideal sum values were calculated when the gas temperature was in equilibrium.

Ventgas analysis
Gas samples were taken aer the thermal runaway reaction. If no thermal runaway occurred, then the gas samples were taken aer the cell temperature exceeded 250 C. The gas was analysed with a gas chromatograph system (GC, Agilent Technologies 3000 Micro GC, two columns, Mol Sieve and PLOTU). A thermal conductivity detector (TCD) was used to detect permanent gases. The GC was calibrated for H 2 , O 2 , N 2 , CO, CO 2 , CH 4 , C 2 H 2 , C 2 H 4 and C 2 H 6 . The GC used Ar and He as carrier gases.

Role of the inert gas
Before each experiment, the reactor was lled with inert gas to prevent reactions of the vent-gas with the reactor atmosphere. We used either N 2 or Ar as inert gas. Both gases have advantages and disadvantages.
Advantages of using Ar as inert gas: in this case N 2 is not present in the reactor. There are no reactions which can produce N 2 during thermal runaway. The only possible source of N 2 in a ventgas sample is leakage from ambient air. Therefore, the presence of N 2 (accompanied by O 2 ) in the GC results indicates gas leakage. The amount of Ar in the samples could not be quantied, because it was used as a carrier gas in the GC setup.
Advantages of using N 2 as inert gas: in this case N 2 fulls two functions. It serves as inert gas and also as an internal standard. Since the amount of N 2 in the reactor is known (V N 2 ¼ 0.0027 m 3 ), absolute amounts of other detected gas components can be derived from their relative GC results r GC The absolute amount of vent-gas n GC sum can be calculated from the GC results.
The amount of ventgas calculated with the ideal gas eqn (11) can be compared with the total amount of gas from GC results (13). If n ideal sum ¼ n GC sum than it is likely, that all formed gases were detected by the GC.
However, there is also one strong disadvantage of using N 2 . If leaks from ambient air occur, leaked N 2 falsies the internal standard. Therefore, for the most experiments we used Ar as inert gas in the reactor and detected N 2 indicated gas leaks.
Only in the last three experiments, aer enough experience was gained, we were condent to use N 2 as the inert gas.

Results
We did 23 thermal-ramp experiments with NCA and LFP cells set to different SOC. The results are summarized in Tables 2 and  3. Typical experiment runs are shown in Fig. 2 and 4. The dependence of the thermal runaway parameters on the SOC is visualised in Fig. 3 and 5.

NCA cells
We tested the thermal stabilities of discharged as well as partially charged, fully charged and over charged NCA cells.
Discharged NCA cells (Experiment 1-5) showed no pronounced thermal runaway characteristics. Only small unremarkable exothermic peaks were observed between 150 C and 300 C. The amount of gas depended on the timespan which the cells spend at increased temperature: aer the initial burst plate opening of the cell housing the vent-gas was released from the cell into the reactor with an uniform rate. There was no sudden gas liberation and no violent chemical reaction. CO 2 was the major identied component of the vent-gas. Interestingly, the mass loss of the discharged cells of 4.4 g equalled to the mass of electrolyte in the cells (Table 1).
In Experiment 1 we used N 2 as internal standard. The GC detected n GC sum ¼ 23.2 mmol of produced gas (Table 3). In Table 2 Results of thermal ramp experiments with NCA and LFP cells. Here SOC is the state of charge, q o is the onset temperature, q m is the maximum cell temperature during the experiment, Dm is the mass loss of the cell, n ideal sum is the measured amount of produced vent-gas (11) and the chemical components are those species that were detected by the GC system. Missing values could not be measured or detected. The ratios of the detected gases are given in mol% contrast, the amount of ventgas inside the reactor (11) was much higher n ideal sum ¼ 65.4 mmol. We conclude that the GC could not identify the missing 42.2 mmol of gas, because its setup was optimized for a limited set of permanent gases.
The cells with SOC $ 25% displayed an unmistakable thermal runaway behaviour. When (partially) charged NCA cells were heated beyond a critical temperature, self accelerating exothermic reactions started and the cell temperatures suddenly increased up to maximum values in the range of 739 C and 1075 C.
The onsets of the exothermic reactions were obtained from the rate plot: the temperature, where a rst clear deviation towards increased temperature rate was detected, was dened as the onset temperature q o . For NCA cells with SOC # 100% q o was in the range between 136 C and 160 C. Overcharged NCA cells (SOC > 100%) showed much lower onset temperatures between 65 C and 80 C. It is an important nding, that overcharged NCA cells can proceed straight into thermal runaway when heated above 65 C.
The thermal runaway reactions were accompanied by abrupt vent-gas releases. Cells with higher SOC produced more vent-gas. Up to 317 mmol of gas were recorded. The gas composition depended on the SOC as well: the fractions of CO 2 decreased and the fractions of CO and H 2 increased with rising SOC. A clear trend for other detected gases (CH 4 , C 2 H 4 and C 2 H 6 ) was not observed.
We used N 2 as inert gas in the Experiments 13 and 14 (overcharged NCA) in the same way as in Experiment 1. The calculated amounts of gas n ideal sum and n GC sum were in good agreement, indicating that all produced gases were detected by the GC. In other words, it is likely that the quantitative GC results (Table 3) represent the major vent-gas components for over-charged cells and that only smaller amounts of gas may be missing.  Table 3 Thermal runaway parameters of experiments with NCA cells. Experiments with N 2 as internal standard were selected and the amounts of measured gases are given in absolute units (12). The amount of vent-gas n ideal sum and n GC sum was calculated with ideal gas eqn (11) and with results of the GC (13)

LFP cells
In addition to the experiments with NCA cells, we did 7 thermal-ramp experiments with LFP cells at different SOC ( Table 2).
The discharged LFP cell (Experiment 17) showed a behaviour similar to discharged NCA cells. Exothermic reactions could not be detected. Aer the initial burst plate opening of the cell housing, the amount of gas increased evenly over time as the cell  was heated. For the discharged cell, the GC registered essentially only CO 2 . We suspect that the GC could not detect all gas components that were produced by the discharged cell: similar to Experiment 1 with a discharged NCA cell, signicant amounts of gas may be missing in the GC results, simply because the used GC equipment was not capable of detecting them. First mild exothermic reactions were seen for a cell that was charged to 25% SOC. The reactions were not strong enough to evolve into a distinct thermal runaway. Vent gas was produced continuously with time, likewise to the experiments with discharged cells.
LFP cells charged to SOC $ 50% showed pronounced thermal runaway reactions. Increasing SOC caused increasing maximum temperatures during thermal runaway. The maximum temperatures q m ranged from 283 C to 448 C.
The onset temperature q o was $140 C for cells between 50% SOC and 100% SOC. The cell overcharged to 130% SOC showed a exothermic reaction already at 80 C. In contrast to overcharged NCA cells, the initial exothermic reaction of the overcharged LFP cell could not sustain a full thermal runaway. The overcharged LFP cell proceeded into thermal runaway only aer it was heated by the heating sleeve beyond 140 C.
The amount of gas n ideal sum ranged between 31 mmol and 61 mmol and showed no clear dependence on the SOC. With increasing SOC the relative composition of the detected gases changed to lower CO 2 and higher H 2 fractions. The fractions of CO (max. 9.1%) were lower than for NCA cells.
The mass loss of the LFP cells ranged from 6.1 g to 7.1 g and is comparable to the amount of electrolyte (6.5 g) in this cells.

Discussion
It is tempting to pinpoint the main contributors of heat and gas release during the thermal runaway reactions. Can the amount of produced gas and its components be explained with a set of chemical equations?
Material n a j that is available for the reaction system is listed in Table 1. In addition lithium n a O2 and oxygen n a Li may be released in heated cells. Part of the material is consumed (by becoming a reactant n r j of the reaction system).
The reaction products may consist of gases, uids and solids. A measurable subset of the resulting gaseous products n GC i and the sum of ventgas n ideal sum is given in Table 3. The challenge is to nd the right set A of equations and to nd the utilisation number b for each equation (how oen is each equation applied) so that the calculated amounts of products n p i match the measured values: and In other words, the difference of calculated and measured amounts of products is dened as the cost function and the system is restricted by the amounts of reactants and products. The algorithm should minimise the cost function and respect the restrictions.
The mathematical problem was solved using the LIPSOL linear programming toolbox in Matlab. The set of chemical equations and two explicit calculations (discharged NCA cell, Experiment 1 and over-charged NCA cell, Experiment 13) are disclosed in the next subsections: The lithiation state x can be calculated using The amount of liberated O 2 is  64,65 If assumed that the partly lithiated LFP cathode in a Li-ion battery consists of a mix of lithiated (LiFePO 4 ) and delithiated (FePO 4 ) particles 66 then the oxygen release of a partially charged cathode is given by: The absolute amount of O 2 from the LFP cell can be calculated with equations similar to (18) and (19).
Both cathodes materials NCA and LFP can contribute O 2 (Fig. 6) which in turn can take part in further exothermic degradation reactions. The amount of O 2 is higher for delithiated cathodes (battery is charged). Note, that because of the irreversible capacity loss during formation of actual cells, the cathode can not be fully lithiated by discharge of the cell: even at 0% SOC (battery is discharged) the lithiation factor x < 1 and a small amount of O 2 may be released.

Exposure of lithium by the anode
On the anode side graphite particles can defoliate and expose intercalated Li at temperatures above 230 C. 67,68 The amount of released Li depends on SOC of the battery: The NCA cell can release n a Li (SOC ¼ 100%) ¼ 126.2 mmol in the fully charged state and n a Li (SOC ¼ 0%) ¼ 1.24 mmol in discharged state.

Typical chemical reactions
In this section we compile a list of probable degradation reactions which may take place during thermal runaway. The most signicant chemical reactions may be reactions with O 2 and Li: partially delithiated cathodes release O 2 and partially lithiated anodes release Li at elevated temperatures (17), (20) and (21). Both released materials are highly reactive and promote a number of reactions that are summarized in a previous publication. 36 Additionally, following reactions are considered: Combustion of the carbon black (conducting additive) or anode graphite the water-gas shi reaction

Alternative CO 2 producing reactions
CO 2 was the main gaseous product that was identied in the ventgas of discharged cells. Little O 2 is available in cells at 0% SOC and it is questionable if combustion alone can account for all CO 2 . Therefore effort was made to nd further alternative reactions with CO 2 evolution without oxygen involvement. Following reactions were found in the literature: Ring-opening and polymerisation of EC and PC 72-74 e.g.: thermal decomposition of the carbonate esters 50,75,76 e.g.: This values are well below the maximal temperatures reached in our thermal-ramp experiments. Electrolyte decomposition with CO 2 release was also observed in other research. 5,78 The maximum amount of CO 2 generated from purely thermal decomposition of the electrolyte solvents (28)-(30) is only limited by the amount of available electrolyte.
Further CO 2 may be produced from the SEI degradation: The organic SEI produced by EC reduction (2) can decompose in thermally driven reactions, 79 or react with HF analogous to 80 C 2 H 5 OCO 2 Li + HF / C 2 H 5 OH + LiF + CO 2 (33) with the proposed scheme (34) Inorganic SEI can react with HF as well. 48,53 In the presence of impurities such as trace water LiPF 6 may react to POF 3 that in turn reacts with the electrolyte in a decarboxylation reaction with CO 2 release: 50,77,81-83 What is the most signicant CO 2 production mechanisms in oxygen depleted environment? In the case of the NCA cell (Table 1) decomposition of all electrolyte solvent (28)-(30) may translate to 35.0 mmol CO 2 . The amount of SEI is lower than the amount of electrolyte solvents and therefore only 8.7 mmol of CO 2 can be produced with eqn (34) and (35). The reactions involving HF (33)-(36) may be further suppressed by the limited amount of trace ROH and LiPF 6 .

Gas release of a discharged NCA cell
In the Experiment 1 a discharged NCA cell was subject to a thermal-ramp test and absolute amounts of produced gas  (Table 3). Gas analysis with GC gave 23.2 mmol of CO 2 and small amounts of H 2 , CO and hydrocarbons. The overall amount of produced gas n ideal sum inside the heated reactor was 65.4 mmol. This means that the GC system was unable to detect 42.2 mmol of unknown gas components.
What is the source of CO 2 and what is the nature of the not identied gas components? The cathode material of a discharged cell is not fully lithiated and may release a small amount of O 2 (19). The released O 2 can participate in a combustion reaction, but the amount of released O 2 is not sufficient to produce all measured CO 2 ðn a O2 ¼ 6:7 mmol vs: n GC CO2 ¼ 21:9 mmolÞ. We needed to consider alternative reactions in order to account for the measured amounts of gases. Table 4 was calculated with the linear optimisation algorithm. It gives one possible set of reactions to reproduce the measured value of CO 2 and the overall amount of produced gas in the reactor. Because of the elevated temperature all liquid solvents present inside the Li-ion cell either decompose or evaporate ( Table 5). The calculation gives rise to new gaseous components and the amount of those components can be compared to the actual measurements (Table 6).
In this mathematical solution, the missing 42.2 mmol of gas consist mainly of solvent decomposition products (CH 3 OCH 3 , CH 3 OC 2 H 5 ) and remaining solvents as well as water in gaseous state. Such gases can not be found by the GC system due to following reasons: (a) the sampler of the GC runs at room temperature and therefore the solvents condense and are not  Table 8 Initially available material in the cell as well as material that is consumed as a reactant according to the proposed reaction system in Table 7 for the Experiment 13 (over-charged NCA cell)

Gas release of a charged NCA cell
The situation changes when the cells in thermal ramp experiments are charged. High amounts of oxygen and lithium become available and the cells go into distinct thermal runaway. In the Experiment 13 (Table 3) an overcharged NCA cell was tested and the vent-gases were quantied by the GC system using an internal N 2 standard. The cell in Experiment 13 was overcharged to a capacity of 4.03 A h (120% SOC). The lithiation factor of the cathode was x cat Li ¼ 0.08 and the calculated oxygen release (17) was n a O2 ¼ 81:6 mmol. The lithiation factor of the anode was x and Li ¼ 1.00 and the amount of intercalated Li on the anode side equalled n a Li ¼ 151 mmol. In other words, the anode was fully lithiated to the maximum theoretical Li capacity. The cell produced 281.3 mmol of ventgas during thermal runaway and high amounts of CO, H 2 , CO 2 were detected.
To nd one of the possible solutions explaining the measured gas composition the equations in Table 7 were used. In this exemplary mathematical solution all electrolyte solvents, organic SEI, lithium carbonate and the released O 2 were consumed ( Table 8). The major products in the calculation were the gases as measured by the GC as well as the solids LiF and Li 2 O. The overall amount of measured gas n ideal sum and the amounts of the individual gas components n GC i could be reproduced by the calculation ( Table 9). The quantity of LiF and Li 2 O was not measured and therefore not veried by the experiments.
The major reactions which were responsible for the gas and heat production during thermal runaway are summarized in a simplied picture (Fig. 7). In this scheme the released oxygen triggers a chain of exothermic reactions. Because of O 2 insufficiency incomplete combustion of organic material takes place. The resulting H 2 O reacts with the exposed Li with H 2 production. Simultaneously H 2 and CO 2 are produced with the watergas shi reaction. In the end the main gases are CO, CO 2 and H 2 .
Although the calculation shows good agreement of measured and computed amounts of gas it has some aws: (a) the full amount of CO 2 could not be reproduced (b) it is not considered, that the separator material must decompose and add additional gas volume at temperatures >900 C and (c) in reality the reactants are not distributed homogeneously when the reactions take place, instead material is violently expelled from the cell into the reactor during thermal runaway. Further work is needed to take those effects into account.

Conclusions and outlook
We measured the thermal runaway characteristics of commercial Li-ion cells in destructive thermal ramp experiments in inert atmosphere. Our samples were 23 NCA and LFP based Li-ion batteries with the geometrical format 18650 charged to different SOC. The main ndings of this work are: (1) The cell material and cell design (e.g. high energy density vs. high power density) have a high inuence on the maximum cell temperature and on the released gases in thermal runaway conditions (Table 10). Charged NCA cells showed a drastic thermal runaway behaviour. NCA cells could reach maximum temperatures of 1075 C and they released up to 317 mmol of gas (equal to 7.1 L at standard conditions). Charged LFP cells exhibited a less pronounced thermal runaway: maximum cell temperatures as high as 448 C were observed and the LFP cells released up to 61 mmol of gas.
(2) Discharged cells showed no thermal runaway upon heating up to $250 C. Both cell types needed to be at least partially charged in order to go into thermal runaway.
(3) The severity of the thermal runaway increased with increasing SOC.
(4) The thermal runaway reactions produced high amounts of CO, H 2 and CO 2 thus making the gas ammable and potentially toxic. The gas composition depended on the cell type and SOC. NCA cells produced more CO and H 2 than LFP The SOC and the cell type had less effect on the onset temperature, as long as no Li plating occurred. Overcharge may cause metallic Li deposition on the anode which compromises the thermal stability. The onset temperature of overcharged cells decreased dramatically from 140 C to values as low as 65 C! (6) In three experiments, the absolute amounts of gases from NCA cells were quantied. It is shown, that it is theoretically possible to explain the absolute amounts of the measured gases with a set of chemical degradation reactions and with known amounts of initial material in the cell. (7) We think that the main reactions in charged cells are combustion of carbonous material and Li oxidation. Both are strong exothermic reactions which contribute to the energy release during the thermal runaway of a Li-ion battery. The amounts of O 2 and Li available to degradation reactions depend on the SOC as well as on the amount and type of active cathode and anode material. Higher SOC increases the O 2 release of the cathode and the amount of intercalated Li in the anode. In over-charged cells these amounts increase further and deposition of highly reactive metallic Li may occur on the anode.
(8) It is proposed that both, the cathode and anode side participate in the reaction system. Therefore experiments with only one electrode may not cover the full picture.
Many open questions concerning the safety of Li-ion batteries remain. The industry needs scaling rules to evaluate the safety of large battery systems with hundreds of cells based on results of misuse experiments with individual cells. Many test results exist for small 18650 cells but we think that more effort must be made to understand the thermal runaway behaviour of large cells with capacities as high as 60 A h. It is yet to prove, if specic amount of gas and heat are the same for small and large cells. The risks of re and toxicity (including HF) of vent gas must be quantied for real life applications including misuse of battery packs for electric vehicles, airliners and for home storage of solar energy.
Our future work in the next three years will include (a) additional testing of 18650 cells in an improved test rig, (b) experiments with large automotive Li-ion cells in a new large test rig, (c) bottom up thermodynamic calculations of the chemical reaction systems and (d) top down FEM simulation of failure propagation and the reaction kinetics in large battery packs.

Nomenclature
Dm Mass loss of the cell, caused by temperature ramp experiment (g) q Cell temperature ( C) q m Maximum cell temperature during the temperature ramp experiment ( C) q o Onset temperature of the thermal runaway ( C) q gas Gas temperature inside the reactor (K) C irr Charge associated with n irr Li (A h) C nom Typical cell capacity as specied in the datasheet (A h) C res Charge associated with n res Li (A h) C u and Theoretically usable capacity of the anode (A h) C u cat Theoretically usable capacity of the cathode (A h) F Faraday constant (F ¼ 96 485 A s mol À1 ) n a i Amount of substance i in a pristine cell, at the start of a thermal ramp experiment (mol) n p i Theoretically calculated amount of product i, which is produced by chemical reactions during the thermal runaway (mol) n a j Amount of material j in the cell, that is available for chemical reactions during the thermal runaway (mol) n a cat Amount of either LFP or NCA units in the cathode n GC i Absolute amount of gas component i in the reactor, calculated from GC results (mol) n GC sum Amount of gas produced by a cell during a temperature ramp experiment, calculated from GC results (mol) n ideal sum Amount of gas produced by a cell during a temperature ramp experiment, calculated with the ideal gas law (mol) n irr Li Amount of irreversibly trapped Li in the anode caused by initial cell formation (mol) n res Li Amount of residual Li in the anode of a cell which is discharged to V min (mol) n r j Theoretically calculated amount of reagent j in the cell, which is consumed by chemical reactions during the thermal runaway (mol) n 0 Initial amount of gas in the reactor at the start of the experiment (mol) n N 2 Actual amount of N 2 in the reactor (mol) P Gas pressure in the reactor (Pa) R Gas constant (R ¼ 8.314 J mol À1 K À1 ) r Lithiation factor of the anode or cathode