Research on accurate fire detection & early warning model for lithium-ion battery packs

To improve the fire detection & early warning accuracy for lithium-ion battery packs and reduce false alarms and missing reports, the fire occurrence mechanism and cause factors of lithium-ion battery packs were analyzed, the calculation method of these characteristic parameters and basic judgment conditions of thermal runaway occurrence were explored, and then a thermal runaway-induced fire spread model of lithium-ion battery packs was established. Next, the scheme and model for accurate detection and early warning of lithium-ion battery pack fire was designed. Furthermore, an accurate fire detection & early warning model of “characteristic parameter sorting and screening + comprehensive conditional judgment + grading implementation” was proposed. Experimental results demonstrate that the accurate fire detection & early warning model for lithium-ion battery packs built in this study would be feasible.


Introduction
Lithium-ion batteries are susceptible to exothermic auxiliary reactions of battery materials under the action of internal and external factors, especially under thermal abuse due to their physicochemical properties of low boiling point, organic characteristics, inflammability, high calorific value, etc.If the reaction heat released is continuously accumulated inside the battery instead of dissipation, the battery temperature will gradually rise, thus resulting in a thermal runaway [1][2][3][4] .For lithium-ion batteries, the thermal runaway is the most serious security incident, which will lead to fire outbreaks and even the explosion of lithium-ion batteries.With the ever-increasing market demands for lithium-ion batteries in recent years, blast accidents occur frequently to batteries, bringing about huge losses to the life and property safety of the masses.Given the safety problems of lithium-ion batteries, mandatory requirements have been proposed in relevant national standards for the alarm and warning against the danger of thermal diffusion triggered by the thermal runaway of lithium-ion batteries [5] .
The thermal runaway detection & early warning techniques of power batteries have been explored by many researchers, achieving rich results regarding mechanical-electric-thermal causes of battery thermal runaways and the prevention & control methods, the occurrence mechanism and inhibition method for battery thermal runaways, battery combustion explosion characteristics and fire safety, thermal runaway spread and heat management of battery systems, early warning characteristic parameters, early warning systems and methods.However, the temperature detection method and gas analysis method are mainly used in the existing fire detection methods for thermal runaway of battery packs, which is easy to cause false alarms, missed alarms, delayed alarms, and inaccurate warnings.As a result, necessary preventive measures cannot be taken in advance at the first time, and the expansion of the fire and the loss are not effectively reduced.
In this study, all characteristic parameters to judge the thermal runaway fire of battery packs and their relationships would be comprehensively analyzed from establishing the causal chain model of thermal runaway fire of battery packs, and then an accurate detection & early warning model of lithiumion battery packs with the core of "characteristic parameter sorting and screening + comprehensive conditional judgment + grading implementation" was proposed, in an attempt to help reduce false alarms and missing reports, and enhance the accuracy of early-warning information.

Analysis of thermal runaway mechanism and cause of lithium-ion battery
When it comes to the occurrence mechanism of the thermal runaway of lithium-ion batteries, the quantity of heat production inside a battery is far higher than the heat dissipation rate, and consequently, a large quantity of heat is accumulated internally.After the occurrence, the thermal runaway will spread downward to trigger chain reactions and finally give rise to a fire outbreak and explosion.Related studies have shown that the thermal runaway process of lithium-ion batteries mainly involves the following group of reactions: the decomposition of SEI membranes, the electrolyte-binder reaction, and the decomposition of electrolytes and positive active materials.The factors inducing the thermal runaway of lithium-ion batteries can be divided into two types: internal factors and external factors, where the former mainly include the internal short circuit caused by the battery defects, and the presence of impurities in battery materials or misuse, which lead to the generation of internal lithium dendrites and further trigger the cathode-anode short circuit.The latter mainly includes the short circuit induced by external factors like extrusion and needling [6] .Hence, the main causes for the thermal runaway of lithium-ion batteries can be concluded as follows: 1) Overheating-induced thermal runaway.The primary causes for battery overheating lie in the unreasonable battery model selection and thermal design, or the temperature rise in batteries due to external short circuits, and the loose cable connection.
2) Overcharging-induced thermal runaway.The overcharged circuit will lose its safety protection function, and the thermal runaway is triggered if charging continues after full charging.
3) Internal short circuit-induced thermal runaway.The internal short circuit can be caused by battery manufacturing impurities, metal particles, charging/discharging expansion/shrinkage, and lithium precipitation.Such an internal short circuit occurs slowly with a very long duration, and it is unknown when a thermal runaway occurs.
4) Machinery-triggered thermal runaway.Collision is a typical mode of machinery-triggered thermal runaway.Needling is usually performed to simulate the collision during a laboratory simulation experiment.
From the aspect of nature, the causes of the thermal runaway can be divided into three types: thermal abuse, mechanical abuse, and electricity abuse.The generation of a thermal runaway is accompanied by an internal short circuit, which is an important mark for the thermal runaway.As for the trigger time of thermal runaways, the generation process of the internal short circuit is divided into the instantaneous trigger and accumulative trigger [7] : 1) As for the former mode, the battery is subjected to a thermal runaway within a very short time, including external short circuit, extrusion, and needling; 2) Accumulative trigger includes overcharging, over-discharging, internal heat source, and external heat source.These forms of abuse gradually pose damage to batteries.The accumulation can last just several minutes, or as long as several months, which meets the principle of quantitative and qualitative changes.
The thermal runaway of lithium-ion batteries is usually the result of joint action of internal and external factors (Figure 1).Triggers not only include the internal factors of the batteries themselves but also external factors, where external factors are the leading factors driving lithium-ion batteries into the thermal runaway state.However, the thermal runaway triggered by a single internal factor is very rarely seen.

Analysis of fire occurrence mechanism and cause of lithium-ion battery packs
The fire outbreak risk, which is the most serious risk for lithium-ion batteries, is mainly caused by the spread of thermal runaway.With the battery system of electric vehicles as an example, the occurrence of thermal runaway-induced fire incidence is largely divided into several stages according to the time sequence, such as the thermal runaway of battery cells, the spread of thermal runaway of battery modules, the heat diffusion of battery modules, battery system fire, and vehicle body fire.With continuous aggregation, the heat cannot be diffused, thus driving the thermal runaway of other nearby battery cells to spread, i.e., heat diffusion.When the thermal runaway cannot be timely inhibited, the battery system is subjected to a serious heat diffusion phenomenon, then a fire incident occurs after the temperature of the battery system reaches the spontaneous ignition point, and finally, the vehicle body will be on fire.[9]   It is proved that internal short circuit is a common link to thermal runaway cause, and the most complicated problem in the study of thermal runaway cause.At present, it is still necessary to study the reasonable experimental method of internal short circuit replacement and further study the internal short circuit model.In addition, the self-triggered internal short circuit has a long-term evolution process before inducing thermal runaway, and early detection based on the characteristics of the self-triggered internal short circuit is one of the problems urgently needed to make a breakthrough in power battery safety research.After thermal runaway is induced, the heat released by the local monomer thermal runaway spreads to the surrounding, which may heat the surrounding batteries and cause thermal runaway of the surrounding batteries, also known as the "expansion" of thermal runaway in the battery packs.The energy released by thermal runaway of a single battery is limited, but if the chain reaction results in the expansion of thermal runaway, the energy of the whole battery pack will be released through thermal runaway, which will cause great harm.It is found in this study that the 5 Ah ternary lithium-ion power battery (with about 1.1 KWH of electric energy) released about 630,000 J, equivalent to 0.15 kg of TNT [10] .

Calculation of heat quantity produced by thermal runaway of lithium-ion batteries
The heat production rate of lithium-ion batteries should be calculated to solve their temperature rise rate and temperature value in case of thermal runaway.Bernardi took the lead in proposing a calculation model for the heat production rate of lithium-ion batteries, as seen in Formula (1) [8] : where Q stands for the overall heat production rate of a battery; Qreaction represents the reactioninduced heat production rate; Qmixing is the heat increasing rate; Qphase-change denotes the heat production rate due to the phase change of the battery; Qheat-capacity is the heat production rate arising from the reaction of substances.Under normal charging/discharging conditions, Qmixing and Qphase-change can be neglected, and the above model can be simplified into Formula (2). - (2) Formula ( 2) can be further extended into the form of Formula (3).

(
) (3) where I is the total current of the lithium-ion battery, A; U denotes the open-circuit voltage, V; E is the electromotive force, V; T stands for the average battery temperature, K.
To accurately describe the heat production rate of lithium-ion batteries, the heat production model was improved, as seen in Formula (4).

( ) ( ) (
) ) where Qr(t) denotes the heat production rate due to the chemical reactions inside the battery, and Qe(t) stands for the electric power released by the short circuit inside the battery.Qr(t) is calculated through the following formula: ) where QSEI stands for the heat quantity produced by the decomposition reaction of SEI membranes; Qanode is the heat quantity produced by the cathode-electrolyte reaction; Qseparator is the heat quantity absorbed by the membrane decomposition; Qcathode represents the heat quantity produced by the anode decomposition; Qelectrolyte is the heat quantity produced by the electrolyte decomposition; QPVDF denotes the heat quantity produced by the decomposition reaction of binders.
QSEI is calculated as below [9] : where HSEI is the total energy released by the decomposition reaction of SEI membranes, J; CSEI (t) denotes the normalized concentration of SEI membranes; ASEI denotes the frequency factor of SEI membrane reaction, s-1; Ea, SEI represent the activation energy of chemical reaction, J/mol; R is the ideal gas constant; Ti(t) is the temperature of the battery cell at time t.The calculation formulas for Qanode, Qseparator, Qcathode, Qelectrolyte, and QPVDF are obtained by altering the subscripts in Formula (6) into the anode, separator, cathode, electrolyte, or PVDF, respectively.Qe(t) is calculated as follows: where Ashort is the velocity factor of the weak shot circuit; b denotes the index term of the short circuit; ∆H represents the total energy released by the short circuit; ∆t is the average reaction time; stands for the energy of the weak short circuit that already happens; Qshort is the reaction-produced thermal power in case of a weak short circuit inside the battery; Tonset is the melting temperature of membranes inside the battery; Ti(t) represents the internal temperature of the battery at time t; Tshort is the battery temperature in the event of weak short circuit.

Thermal runaway characteristic parameters of lithium-ion batteries
The thermal runaway characteristic parameters of lithium-ion batteries mainly include temperature parameters, voltage parameters, gaseous-phase parameters, smoke parameters, electrolyte leakage, and flame parameters, among which flame parameters can be detected only in case of combustion after the spread of thermal runaway.
1) Temperature parameters.The battery temperature continuously rises as the heat accumulation is aggravated with time.After a slow and stable temperature rise for a certain stage, the battery is subjected to a violent exothermic reaction at the critical point of thermal runaway, and the temperature suddenly rises rapidly and reaches the peak value.The surface temperature rise rate fluctuated within a very small range before the thermal runaway.When the battery reaches the critical point of thermal runaway, the temperature rise rate reaches the peak value with the violent exothermic reaction and then it is weakened [10]   .The relations of the short circuit duration with the temperature rise, thermal state, and various internal reactions of the battery are displayed in Figure 2. 2) Voltage parameters.Tests have shown that the battery voltage is relatively stable during the first half stage of the heating process.As the temperature rises, the battery experiences a short circuit, with the voltage sharply dropping to zero.
3) Gaseous-phase parameters.Y. Fernandes, from the National University of Orleans, France, proposed a new method to analyze the gases generated in the overcharging process of lithium-ion batteries and explored the varieties and quantity of gases produced in this process.With the cylindrical 25550LFP battery taken as the study object in this experiment, the varieties and volume fractions of gases released from the battery during overcharging were mainly: CO2 (47%), H2 (23%), C2H4 (10%), CO (4.9%), and C2H5F (4.6%).
4) Smoke parameters: When it comes to thermal runaway, a battery first experiences the combustion process after smoking.From the early-warning angle, the smoking process should be the first concern.The temperature [11]  of gases, the components, and the concentrations could be detected at 100 mm above the safety valve in the battery.5) Electrolyte leakage.Electrolyte leakage is also one of the important symptoms that may trigger the thermal runaway.Hence, technical means should be adopted to detect electrolyte leakage to realize early discovery and treatment.
The battery surface temperature, temperature rise rate, and voltage value are the primary characteristic parameters for the early detection and early warning of a thermal runaway.Gaseous-phase parameters and electrolyte leakage are comprehensive parameters used to judge the occurrence and spread of the thermal runaway.

Determination conditions for the thermal runaway of lithium-ion batteries
Krishna et al. proposed a new model to analyze the heat conduction characteristics inside a battery, describe the heat production and dissipation process, state that the thermal runaway was related to the heat conductivity of the battery, and thus introduce the heat conductivity parameter to generate a dimensionless parameter, thermal runaway number (TRN), to judge the thermal safety state of the battery, as seen in Formula (8) [12] .Then, the heat dissipation rate μ1 of the battery surface with the external environment, the heat production rate β, and the battery radius R are calculated through Formulas ( 1)-( 7).
The heat production rate β, heat dissipation coefficient, and heat conductivity coefficient of the lithium-ion battery are key parameters controlling its thermal runaway.A greater β value will contribute to a greater TRN value.When TRN>1, the battery will be subjected to a thermal runaway, and it will not if TRN<1.It is noteworthy that β is not a fixed value, instead, it continuously increases with the temperature rise, and so will the TRN value.
The determination conditions triggering the thermal runaway as recommended by relevant standards in China are as follows [5] : a) The trigger object generates a voltage drop, with the drop value exceeding 25% of the initial voltage; b) The temperature at the monitoring point reaches the maximum working temperature stipulated by the manufacturer; c) The temperature rise rate of the monitoring point satisfies dT/dt ≥ 1℃/s and lasts over 3s.When a) and c) or b) and c) occur, the thermal runaway is identified.

Fire spread modeling of thermal runaway
To perform accurate detection and early warning of thermal runaway-induced fire in lithium-ion batteries, the thermal runaway-induced fire spread laws must be abided by to establish a thermal simulation model, a thermal runaway model, and a diffusion model of the battery system.On this basis, the heat propagation model of the main battery subjected to the thermal runaway towards adjacent batteries under the multi-factor influences and comprehensive heat transfer conditions (heat conduction, heat radiation, and heat convection) should be established, to construct a Domino effect model (Figure 3) for the thermal runaway-induced fire spread of the battery system.

Accurate fire detection & early warning solution
In this study, all-round detection of characteristic parameters and multi-sensor coupling was performed for sorting and screening using the above-established Domino effect model according to three different stages: occurrence of thermal runaway, diffusion and spread, and fire outbreak.Then, the comprehensive parameter conditions were judged for grading warning as per the determination conditions for the occurrence of thermal runaway, spread, and fire outbreaks.First, the battery temperature, temperature rise rate, and voltage value were detected for judging comprehensive conditions, with the results as the early warning basis for the occurrence of thermal runaway.Second, smoke, gases, and electrolyte leakage were detected to judge comprehensive conditions, with the results as the early warning basis for the spread of thermal runaway.Third, gas, smoke, and flame parameters were detected to judge comprehensive conditions with the results as the early-warning basis for the thermal runaway-induced fire incident.This thermal runaway-induced fire early-warning solution consisting of "characteristic parameter sorting and screening + comprehensive conditional judgment + grading implementation" can not only realize accurate detection & early warning of thermal runaway-induced fire and effectively reduce missing reports and false alarms but also provide grading warning signals for taking pertinent preventive and control measures (Figure 4).

Accurate detection & early warning model for the thermal runaway-induced fire of lithium-ion batteries
The accurate detection & early warning model of "characteristic parameter sorting and screening + comprehensivet conditional judgment + grading implementation" for the thermal runaway-induced fire of lithium-ion batteries should generally include three relatively independent but mutually supporting sub-modules: internal short-circuit fault early diagnosis sub-module, omnidirectional characteristic parameter detection and multi-sensor coupling sub-module, and grading warning and result output submodule.1) Internal short-circuit fault early diagnosis sub-module.The internal short circuit marks the beginning of the thermal runaway in the battery.Hence, the internal short-circuit fault of battery cells should be diagnosed first to realize early discovery and warning of the thermal runaway-induced fire incident.The basic internal short-circuit fault early diagnosis model is as follows: First, the battery cells probably subjected to internal short-circuit fault are judged based on the mean-difference model of the battery pack.Then, the internal short circuit characteristic parameters are judged based on the heatelectricity coupling model of the internal short circuit, thus determining the internal short-circuit fault [13, 14]   .In the internal short-circuit early diagnosis sub-module, the temperature, temperature rise rate, and voltage value of battery cells are taken as the main characteristic parameters to be detected for the sake of comprehensive conditional judgment.In case of a thermal runaway, an initial early warning signal will be sent.
2) Omni-directional characteristic parameter detection and multi-sensor coupling sub-module.Based on the internal short-circuit fault diagnosis sub-module, the gas detection and electrolyte leakage detection should be set from typical characteristic parameters in the development and spread stages of thermal runaway to determine the starting point and early state of thermal runaway, and then send an initial early-warning signal indicating the spread of thermal runaway.In addition, smoke sensing and video detection should be set to send an alerting signal of thermal runaway-induced fire incidence in the event of the spread of thermal runaway and the appearance of visible smoke.Moreover, photosensitive detection was set to send a signal of taking manual fire extinction measures in the face of flame and combustion.Furthermore, temperature sensing should be set to send a signal to start the automatic fire extinguishing system in the event of high heat quantity and fire incidents.On this basis, an omnidirectional characteristic parameter detection and multi-sensor coupling sub-module is constructed, as shown in Figure 5 a and b.
In the omnidirectional characteristic parameter detection and multi-sensor coupling sub-module, all the detected characteristic parameters were sorted and screened according to three successive stages: occurrence of thermal runaway, diffusion and spread, and fire outbreak, followed by the comprehensive conditional judgment (Figure 5 c).Next, a grading warning was performed according to the judgment results.3) Grading warning and result output sub-module.After the omnidirectional characteristic parameter detection and multi-sensor coupling sub-module output the results, the grading early warning and result output sub-module output the corresponding early-warning signal according to the set early-warning level, giving a warning of the current thermal safety state.In general, the early warning of thermal runaway-induced fire incidents is divided into three levels: Level I is the most serious early-warning level (Figure 6) (gas parameters for example).The early warning of different levels has the following meanings: Level I early warning indicates that the battery inside the battery box is already subjected to or about to experience an obvious thermal runaway.Level II early warning means that the battery inside the battery box may be about to experience a heat self-production phenomenon.Level III early warning If any characteristic parameter or its change rate exceeds the first threshold, a Level I early-warning signal will be sent, during which the battery pack already has the risk of fire outbreak and combustion, the automatic fire extinguishing apparatus is immediately started, and the sound-light alarm is started to remind the field personnel of evacuation.If any characteristic parameter or its change rate exceeds the second threshold, a Level II early-warning signal is sent.When the ultrasonic testing data or its change rate exceeds the third threshold, and the temperature detection data or its change rate exceeds the third threshold simultaneously, a Level II early-warning signal will also be sent [15] .In this case, the thermal management system is immediately started, and the thermal runaway is inhibited or slowed down by forced cooling to ensure the system′s safety.If the ultrasonic testing data or its change rate exceeds the third threshold while the temperature detection data or its change rate does not exceed the third threshold, no exothermic chain side reactions will be triggered by the heat self-production inside the battery, and then a Level III early-warning signal will be sent.If the ultrasonic testing data or its change rate does not exceed the third threshold, the temperature detection data or its change rate exceeds the third threshold, and the change rate of voltage exceeds the third threshold, a Level III early-warning signal will be sent [15] .In this case, the temperature of each battery does not reach the scope of triggering exothermic side reactions, and the thermal management system is immediately started to strengthen the ventilation cooling.At the same time, from the test data of fire danger signal (Table 1.), the average missing alarm rate of fire warning is only 0.588%, and the average accuracy rate output of warning signal is 98.24%.

Conclusions
In this study, the causal chain model on thermal runaway fire of lithium-ion battery packs is established, all characteristic parameters to judge thermal runaway fire and their mutual relationships are comprehensively analyzed, the calculation method of these characteristic parameters and basic judgment conditions of thermal runaway occurrence are explored, and the scheme and model on accurate detection and early warning of lithium-ion battery packs fire is designed.Experimental results demonstrate that by using the multi-sensor coupling detection and judgment model and hierarchical early warning model established in this study, the internal short circuit that may cause serious thermal runaway accidents can be detected at least 15 minutes in advance, the average missing alarm rate of fire warning can be reduced to 0.588% and the average accuracy rate can be increased to 98.24%.It is concluded that the accurate fire detection & early warning model for lithium-ion battery packs built in this study would be feasible.

Figure 1 .
Figure 1.Thermal-runaway fire accident-inducing factors and their interactions in lithium-ion batteries.

Figure 2 .
Figure 2. Reaction-Temperature Relations in Different Thermal Runaway Stages of Lithium-Ion Batteries.

Figure 3 .
Figure out the leading spread path of thermal runaway-induced fire incident

Figure 4 .
Figure 4. Accurate warning solution for fire of lithium-ion batteries.

Figure 6 .
Figure 6.Grading Early Warning Model.Concerning the grading early warning and result output model of thermal runaway in lithium-ion batteries, a new grading early warning and result output sub-module was established, as shown in Figure7.In this new grading early warning and result output sub-module, the following groups of symbols indicate three-level critical thresholds for related characteristic parameters from high to low: {T1, T2, T3} denote the first, second and third thresholds for the battery surface temperature; {U1, U2, U3} stand for the first, second and third thresholds for the voltage; {G1, G2, G3} represent the first, second and third thresholds for the internal gas distribution; {S1, S2, S3} are the first, second and third thresholds for smoke particle concentration; {DG1, DG2, DG3} are the first, second and third thresholds for the change rate of internal gas distribution; {DT1, DT2, DT3} represent the first, second and third thresholds for the change rate of temperature; {DU1, DU2, DU3} denote the first, second and third thresholds for the change rate of voltage; {DS1, DS2, DS3} stand for the first, second and third thresholds for the change rate of smoke particle concentration.If any characteristic parameter or its change rate exceeds the first threshold, a Level I early-warning signal will be sent, during which the battery pack already has the risk of fire outbreak and combustion, the automatic fire extinguishing apparatus is immediately started, and the sound-light alarm is started to remind the field personnel of evacuation.If any characteristic parameter or its change rate exceeds the second threshold, a Level II early-warning signal is sent.When the ultrasonic testing data or its change rate exceeds the third threshold, and the temperature detection data or its change rate exceeds the third threshold simultaneously, a Level II early-warning signal will also be sent[15] .In this case, the thermal management system is immediately started, and the thermal runaway is inhibited or slowed down by forced cooling to ensure the system′s safety.If the ultrasonic testing data or its change rate exceeds the third threshold while the temperature detection data or its change rate does not exceed the third threshold, no exothermic chain side reactions will be triggered by the heat self-production inside the battery, and then a Level III early-warning signal will be sent.If the ultrasonic testing data or its change rate does not exceed the third threshold, the temperature detection data or its change rate exceeds the third threshold, and the change rate of voltage exceeds the third threshold, a Level III early-warning signal will be sent[15] .In this case, the temperature of each battery does not reach the scope of triggering exothermic side reactions, and the thermal management system is immediately started to strengthen the ventilation cooling.

r D G > D G 3 (Figure 7 .
Figure 7. Grading Early-Warning Result Output Model 4.4.Experimental study on the model For the above model, taking electric vehicle battery packs as an example (Figure 8 a), a real-time monitoring platform for the real thermal runaway of battery system (Figure 8 b) and various chemical parameters in the process of fire occurrence are established.Four groups of internal short circuit resistances were selected to test by using a multi-sensor coupling detection platform device.The data show that the multi-sensor coupling detection model can identify the internal short circuit which may cause serious thermal runaway accidents at least 15 minutes in advance.At the same time, from the test data of fire danger signal (Table1.), the average missing alarm rate of fire warning is only 0.588%, and the average accuracy rate output of warning signal is 98.24%.

Table 1 .
Fire danger alarm signal test data.