1. Introduction
Reliable battery packs are a key component for electric vehicles (EVs). Global competition forces battery producers to increase energy density, to reduce the costs of their products and to maintain an acceptable safety level at the same time. Safety for lithium-ion battery (LIB) applications is challenged by possible defects inside or outside the cell causing exothermic reactions, such as the thermal runaway (TR). During the TR, extensive amounts of heat, toxic and flammable gas and hot particles are generated with critical consequences like fire, explosion and toxic atmosphere [
1,
2,
3,
4]. Thus, TR prevention and early failure detection before serious battery failure incidents need to be addressed.
Developing methods for early failure detection and reducing safety risks from failing high energy LIBs has become a major challenge for industry, research and development [
5]. State-of-the-art battery monitoring equipment applied in the EV battery pack, like cell voltage measurement and temperature sensors, are insufficient to reach an acceptable level of safety and to reliably enable early failure detection. New regulations such as GB 38031-2020 and discussions such as Electrical Vehicle Safety—Global Technical Regulation (EVS-GTR) prescribe a warning for passengers at least five minutes before serious incidents [
6,
7].
In order to enhance battery safety and to fulfill EVS-GTR20 and GB 38031-2020, the use of additional gas sensors combined with state-of-the-art battery failure monitoring such as voltage, temperature, and pressure measurement, is promising and needs to be investigated in more detail.
1.1. Early Failure Detection in Battery Packs
The battery management system (BMS) is the heart of battery operation, performance, monitoring and failure detection inside EVs. The BMS is designed to ensure safe and reliable operation of batteries and should meet functional safety standards ISO 26262. This control unit manages cell balancing, thermal management, charge and discharge control, cell monitoring, state estimation such as state-of-charge (SOC) and state-of-health (SOH) [
8]. A good BMS is designed particularly for the specific cells and modules according to their safety limits. Current functional-safety cell monitoring means monitoring the electric and thermal behavior of the cell in real-time: the cell voltage and current can be measured directly by means of on-board current and voltage sensors and the surface temperature of battery pack components directly with temperature sensors [
9]. While some manufacturers use temperature sensors on each cell surface, others reduce them inside the battery pack to an absolute minimum [
10]. Additional monitoring systems are insulation detection and the high-current fuse. To the authors’ current knowledge, no additional early failure detection methods are used in series products.
As soon as an abnormal behavior of the cell voltage, the current or the cell temperature of one or more batteries is detected, predefined actions are executed by the BMS (e.g., control unit activates cooling or interrupts the power circuit). In case of a detected failure, different shutdown procedures are followed [
6,
11].
However, the standard state-of-the-art battery monitoring equipment in current EV battery packs is inadequate to reach the upcoming level of safety requirements and to reliably enable early failure detection, because battery failure cases such as electrolysis, an open cell housing and electrolyte vaporization or the first venting are very unlikely to be detected by the classical monitoring system. The first venting is when the cell housing (hard case or pouch bag) opens before the TR, since gas is produced inside the cell due to evaporation or decomposition processes [
3]. For cells without a current interrupt device (CID) or overcharge safety device (OSD), this first venting precedes the cell voltage breakdown and is detectable through a small temperature increase directly at the venting position of the failing cell [
3]. It is not possible to measure this venting with thermocouples applied on the cell surface, with cell voltage monitoring or with current monitoring. It is very likely that with these thermocouples, the voltage or current measurement only detect the failing behavior if at least one of the cells is already in the state of a TR. But this opening of the housing leads to emersion of electrolyte (liquid and gaseous), which is known to be flammable, irritant, toxic, and/or corrosive depending on the exact composition of the electrolyte mixture [
12,
13,
14] and would be a valuable warning for an emerging failure such as a TR event.
In this study we show that an additional gas sensor added to the existing monitoring system inside the battery pack can detect battery failures quickly and cost efficiently before serious incidents. A detailed characterization of hazards such as gas emission and vent gas composition before TR is valuable for the community and for finding suitable sensors for early failure detection.
1.2. Literature on Battery Failure Detection with Gas Sensors
The basic idea of using gas sensors for detecting battery vent gases produced during battery failures, additional to the existing BMS monitoring sensors, is not new. Researchers have published suggestions for using gas sensors for detecting battery failures [
5,
11,
15,
16,
17,
18] and have reported field experiments on gas sensors in battery TRs [
10,
11,
19]. Patent applications have also been published proposing the implementation of gas sensors for TR detection [
5,
20]. The following literature review is divided into two groups: literature focusing on gases produced during the TR and literature focusing on gases produced before the TR reaction.
For detecting the TR itself and gases developed during the TR reaction, gases such as CO, CO
2, H
2, CH
4, C
2H
4, C
2H
6, C
3H
8 are addressed [
2], and sensors to detect at least one of these gases are chosen.
Cai et al. used early failure detection by means of detection of CO
2 gas produced at the TR of the first 18650 NMC cell [
15]. They stated that the measurement of CO
2 allows a significantly faster TR detection than conventional surface temperature sensing. Koch et al. tested a sensor set consisting of voltage sensor, temperature sensor, pressure sensor, gas sensor, smoke sensor, and creep distance sensor in TR experiments of NMC pouch cells [
11]. They found that the gas sensor can detect the TR event earlier than the other sensors during nail-penetration tests. Koch et al. used a SnO
2 gas sensor (metal oxide semiconductor (MOx, also named MOS)) sensitive to CH
4, C
3H
8 and CO. Each tested sensor detected the TR independently of battery size and energy density. The authors stated that the combination of several sensors might lead to an improvement of the system. Mateev et al. proposed gas detection with a MQ-7 (TR gases CO and H
2 are the target gases) MOx gas sensor showing results of failing 18650 LIBs [
16]. One sensitive MOx layer was applied to the relatively large analogue sensor. Liao et al. published a survey of methods for monitoring and detection of TR of LIBs [
5]. They stated that the combination of voltage, cell surface temperature, inner cell temperature and gas monitoring in battery applications is the most efficient method to promote safety of LIBs. Concerning vent gas detection, they described target gases produced during the TR, such as CO, CO
2, C
2H
4, C
2H
6, H
2O, C
3H
6 and O
2, but they stated that gas sensors can detect battery failure related signals seven to eight minutes before the TR, which would mean earlier detection of the gas signal compared to the voltage drop or a temperature signal.
Other researchers have focused on volatile organic compounds (VOCs) such as electrolyte components, which can be measured at battery failure stages before the TR:
Cummings et al. proposed in the US patent application monitoring of electrolyte vapor such as diethyl carbonate (DEC) and dimethyl carbonate (DMC) [
20]. Hill et al. introduced the principle of off gas sensing prior to TR in overcharge experiments using cylindrical, pouch and prismatic cells [
19]. They tested a chemi-resistive sensor, sensitive to CO, CH
4, C
2H
6, VOCs, C
2H
4, C
3H
8, HF, but did not disclose details on the sensing material or operating conditions. The lifetime of the tested sensor is lower than five years. They observed in overcharge tests that the sensor reacted to battery off-gassing 10 min before the TR itself. Swartz et al. promoted the same chemi-resistive sensor element for H
2 measurements and announced refined formulations for detecting CO, CH
x and VOCs [
21]. Wenger et al. presented insight into the gas sensor response of MOx sensors during electrolyte leakage and battery overcharge experiments of pouch cells [
10]. They stated that a first venting eventually occurs and that the time between the first venting and the TR depends on the applied current in the overcharge test. The manufacturer of the sensor is not disclosed in their paper. Herold et al. promoted sensors fabricated by AMS for detecting critical battery states (overcharge, nail penetration) [
18]. They tested their MOx gas sensor in abuse tests such as nail-penetration, overcharging, short circuit and leakage on pouch cells, as well as during charging and temperature cycles, and demonstrated the reaction of the MOx sensor to electrolyte vapor. They suggested a focus on the resistance change relative to the background rather than to the absolute resistance value.
1.3. What Is Missing in the Current Literature and What Is Needed
A detailed analysis of gases produced during battery failure cases before TR is poorly reported in current literature. Most literature is based on gases produced at the TR itself [
2,
14,
22,
23,
24,
25]. In this scientific field it is relevant to gain insight into the gas production even before the TR in order to identify suitable gas sensors. The possible battery failure case of unwanted electrolysis between two voltage carrying parts and resulting H
2 production is rarely addressed in the literature but is a possible and serious battery failure case because of the high flammability of H
2. While detecting the TR allows the setting up of actions to prevent TR propagation to the neighboring cells, focusing on the failure stages before the TR is indispensable to prevent the TR itself. Thus, a detailed analysis of evolving gases at battery failure stages before the TR would be valuable.
Even though TR monitoring with gas sensors has been claimed to be more efficient than voltage and temperature monitoring for failure detection [
5], validation and comparison of several possible gas sensors for early detection of battery failures is currently insufficiently addressed in the literature but would be valuable for research and industry concerned with battery safety.
Because the first venting does not occur in all failure cases [
3] and the exact time when the cell housing opens and releases gases depends on more factors than just the overcharge current, it is relevant to investigate which battery failure cases can be detected and if it is possible to detect battery failures with gas sensors at an early stage.
Furthermore, an algorithm for event detection is needed and the measurement with the gas sensor approach needs to be stable against false positives. Considerations to enable distinguishing between battery failure cases and preventing false positives is necessary, but is not mentioned in the reviewed literature and currently unsolved.
In this study we investigate battery failure cases BEFORE the TR in detail, including electrolysis, and we identify and quantify the gas components and several selected commercially available sensors which detect compounds emitted during the gas producing events before the TR. Different sensor principles and sensors from different manufacturers were benchmarked in special test setups. The most promising gas sensors were tested inside a TR test bed using more than 30 different state-of-the-art automotive LIBs in three different TR triggers: overtemperature, overcharge and nail-penetration. Thus, the influence of the failure case on battery failing behavior is explored and algorithms for event detection are suggested. Additionally, sensors were targeted, which enables distinction between different failure cases, and which allows algorithms to prevent false positives produced by surrounding gases.
4. Summary and Conclusions
This study discusses early detection of battery failures with gas sensors. The use of gas sensors was tested for four battery failure cases, including three failure cases before the TR: unwanted electrolysis of voltage carrying parts, electrolyte vapor, first venting of the cell due to increasing pressure inside the cell, and the TR. This contribution shows that it is possible to detect battery failures involving gas emissions at an early stage quickly and cost efficiently with gas sensors.
First, the produced gases in the mentioned failure cases were studied in detail and it turned out that H2 and especially VOC gases (e.g., electrolyte vapor) were produced at battery failures before the TR. Since the electrolyte of currently used LIBs consists of a varying mixture of DMC, DEC, EMC and EC or PC, which are released after the opening of the cell housing, the electrolyte vapor is an attractive tracer gas for failure detection before the TR. Second, based on the results and literature research, a total of 16 sensors for potential application as battery failure detectors in series products was chosen, implemented on a sensor platform, and benchmarked. Different sensor principles (MOx, NDIR, electrochemical sensors, thermometer and hygrometer) and sensors from different manufacturers were compared. Currently MOx sensors seem to be the most promising technology for early failure detection of battery failures as they show a significant response to all the mentioned battery failures within seconds, have a high sensitivity, are of low cost, and some investigated sensors can be easily connected with I²C to the BMS. Although the electrochemical and the NDIR (CO2) sensors are more selective than the MOx sensors, they are currently not recommended, because they are not able to detect all the introduced battery failures. Especially the MOx sensors Sensirion SGP30, the newer version SGP4x_eng, and the MiCS-6814 fulfil defined requirements because of their high sensitivity to H2 and electrolyte vapor and they use a multi-pixel sensor array for improving selectivity and enabling the prevention of false positives.
Finally, MOx gas sensors were tested inside a TR test bed using more than 30 state-of-the-art automotive LIBs in three different TR triggers. The results illustrate that the detection of the first venting event before the TR is possible with gas sensors in overtemperature and overcharge experiments. In nail-penetration experiments, the first venting took place at the same time as the TR. Consequently, in the nail-penetration failure case, only the TR can be detected. Two event detectors were suggested in this study, which could detect each first venting event in 21 overtemperature and overcharge experiments, electrolysis and evolving electrolyte vapor of different state-of-the-art automotive LIBs with a SNR >>5.
To fulfil the new regulations GB 38031-2020 and EVS-GTR and warn passengers at least five minutes before serious incidents, gas sensors may significantly contribute to failure detection and improvement of battery safety. To sum up, the combination of different monitoring techniques such as voltage and current control, temperature sensing and including a multi-pixel gas sensor is recommended, also to detect electrolysis, electrolyte vapor and the first venting. The sensor combination allows for error allocation and potentially prevents false positives. With this combination of gas sensors, battery failures are detectable earlier than with the current state-of-the-art monitoring systems only. The high sensitivity of the proposed gas sensor enables a warning at certain gas concentrations inside the battery pack, before exposure, flammability or explosion limits are reached. Under the assumption that the price range of MOx sensors at volumes reaching a few million pieces per year will be between 1 and 2 euros, one can estimate an additional cost of less than €10–20 for placing up to 10 sensors inside one battery pack. Among the investigated sensors in this study, the MOx-technology is the most cost efficient.
Still challenging and currently unsolved are the following points:
Automotive certificate of the sensors. This typically requires a sensor lifetime of 15 years and a demanding robustness against mechanical, electrical, and environmental stress. The market of automotive grade MOx gas sensors is limited to complete modules including read-out electronics, data interface and housing. To our knowledge there are currently no isolated automotive grade MOx gas sensor elements with 15 years lifetime on the market.
Secure prevention of false positives. This investigation shows the high sensitivity of MOx gas sensors to gases produced at battery failures but also to gases which might be transported into the battery pack such as gasoline vapor, solvents used for cleaning the car, etc.
Detection of the target gas in relation to the background. The weakest point of the MOx sensor technology is a rather poor selectivity with respect to differentiating classes of oxidizing or reducing compounds. Therefore, a sufficiently large gas volume in comparison to background gases is an essential requirement for the reliable detection of early failure cases. Additionally, the gas dilution during the diffusion from its source to the gas sensor could have a limiting impact on its detectability.
For challenge number 2, a solution is the designing of the MOx sensor with multi-pixels with slightly different sensitivities to different gases, and the finding of fingerprints for relevant tracer gases. This would enable a clear differentiation of gases produced at battery failing states from interfering gases. Concerning challenge number 3, a solution could be the use of multiple gas sensors evenly distributed over the battery pack. Given that the MOx technology is very cost efficient, this approach seems realistic and might even be preferred by manufactures due to the safety increase by redundancy. Clearly, the optimum number of gas sensors and their locations would be the subject of further application focused research.
Beside the detection of LIB failures, methods for controlling the mitigation of TR of neighboring cells should be the focus of future research. Materials stopping abrasive gas and particle flow and thermal propagation are relevant in this context.
In our current work we aim to develop functional polymers, which release tracer gases even before the first venting, which can be measured with the presented gas sensors to further improve battery safety. The use of gas sensors for the detection of gases produced at battery failures is recommended for battery pack applications, but also for battery storage and transport.