Systems and methods for detecting electrolyte and coolant leaks from lithium-ion battery systems

Abstract

The computer-implemented method includes monitoring a gas analyte level associated with a battery system using a first gas sensor and monitoring at least one variable of the battery system. The method includes determining whether a correlation exists between the monitored gas analyte level and the at least one variable of the battery system. The method includes determining whether an electrolyte leak from the battery system exists based on the correlation determination.

Description

[Technical Field]

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 370,555, filed August 5, 2022, and entitled "SYSTEM AND METHOD FOR DETECTING ELECTROLYTE AND COOLANT LEAKAGE FROM LITHIUM-ION BATTERY SYSTEMS," the contents of which are incorporated herein by reference.

[0002] The present technology includes systems and methods for detecting electrolyte and coolant leaks from battery systems. [Background technology]

[0003] A cell leak occurs when the hermetic seal of a cell in a battery system, such as a battery energy storage system (BESS) or electric vehicle (EV) battery pack, is compromised, resulting in a hole in the cell's internal contents to the atmosphere. This is sometimes referred to as "cold-venting." As a result of the cell hole, high-vapor-pressure carbonate solvents in the battery's electrolyte slowly leak over time, resulting in a dry cell that can affect cell performance. There is also the possibility that the leaked cell contents could create a flammable condition if the leaked solvent is contained in an enclosed space, such as an electric vehicle battery pack. The flash points of these solvents are very low (e.g., 18-25 degrees Celsius), creating the potential for a flammable environment. Furthermore, the presence of a leak in a cell can also be a point of ingress for oxygen and moisture into the cell, which can lead to failures such as thermal runaway. A need exists for systems and methods for detecting electrolyte leaks. Summary of the Invention [Means for solving the problem]

[0004] In one embodiment, a computer-implemented method includes monitoring a gas analyte level associated with a battery system using a first gas sensor and monitoring at least one variable of the battery system. The method includes determining whether a correlation exists between the monitored gas analyte level and the at least one variable of the battery system. The method includes determining whether an electrolyte leak from the battery system exists based on the determination of the correlation.

[0005] In one embodiment, a monitoring system includes at least one gas sensor configured to monitor a gas analyte associated with a battery system and at least one sensor configured to monitor one or more variables of the battery system. The monitoring system includes a controller including a memory for storing machine-readable instructions and a processor for accessing the memory and executing the machine-readable instructions. The machine-readable instructions cause the processor to monitor the gas analyte using the at least one gas sensor, monitor one or more variables of the battery system using the at least one sensor, determine a correlation between the monitored gas analyte level and the one or more variables, and determine whether an electrolyte leak from the battery system exists based on the correlation.

[0006] In one embodiment, a computer-implemented method includes modulating a sensor manipulated variable profile to each gas sensor configured to monitor a gas analyte associated with a battery system. The method includes monitoring the gas analyte using the gas sensor through the modulated sensor manipulated variable profile. The method includes developing a data matrix including sensor signals generated by the gas sensor as a function of the modulated sensor manipulated variable profile. The method includes differentiating gas species of the gas analytes based on a comparison of various features in the data matrix. The method also includes determining a state of the battery system based on the gas species differentiation.

[0007] In another embodiment, a monitoring system includes at least one gas sensor configured to monitor gas analytes associated with a battery system and a controller. The controller includes a memory for storing machine-readable instructions and a processor for accessing the memory and executing the machine-readable instructions. The machine-readable instructions cause the processor to modulate a sensor manipulated variable profile for each of the at least one gas sensor, monitor a gas analyte level using the at least one gas sensor through the modulated sensor manipulated variable profile, develop a data matrix including sensor signals generated by the at least one gas sensor as a function of the modulated sensor manipulated variable profile, differentiate gas species of the gas analytes based on a comparison of various features in the data matrix, and determine a state of the battery system based on the gas species differentiation.

[0008] In the accompanying drawings, chemical formulas, chemical structures, and experimental data are provided that, together with the detailed description provided below, illustrate example embodiments. [Brief explanation of the drawings]

[0009] [Figure 1A] FIG. 1 is a block diagram of an exemplary surveillance system. [Figure 1B] FIG. 2 is a block diagram of another exemplary monitoring system. [Figure 2] 1 is an exemplary method for monitoring gas analytes in a battery system based on data correlation. [Figure 3A] 1 illustrates an exemplary correlation between monitored gas analyte levels and monitored variables of a battery system. [Figure 3B] 10 illustrates another exemplary correlation between monitored gas analyte levels and monitored variables of a battery system. [Figure 4] 1 illustrates an exemplary method for monitoring gas analytes in a battery system based on a modulated temperature profile and monitoring data. [Figure 5A] 5 illustrates an exemplary analysis for differentiating gas species based on a data matrix processed according to the exemplary method of FIG. 4. [Figure 5B] 5 illustrates an exemplary analysis for differentiating gas species based on a data matrix processed according to the exemplary method of FIG. 4. [Figure 5C] 5 illustrates an exemplary analysis for differentiating gas species based on a data matrix processed according to the exemplary method of FIG. 4. [Figure 5D] 5 illustrates an exemplary analysis for differentiating gas species based on a data matrix processed according to the exemplary method of FIG. 4. [Figure 6] 1 illustrates an exemplary machine learning (ML) classification design process. [Figure 7] FIG. 1 is an exemplary flow diagram illustrating the process of pre-training a gas sensor using an ML algorithm based on known gas analytes. [Figure 8] 1 illustrates an exemplary output profile of monitored gas analytes using the systems and methods disclosed herein. [Figure 9] 10 illustrates another exemplary output profile of monitored gas analytes using the systems and methods disclosed herein. DETAILED DESCRIPTION OF THE INVENTION

[0010] The present disclosure generally relates to systems and methods for detecting electrolyte and / or coolant leaks in batteries. Over time, batteries gradually degrade, resulting in reduced capacity, cycle life, and safety issues. A degrading battery may release gas. The gas may result from an electrolyte leak, a coolant leak, or a battery off-gassing / venting event. An electrolyte leak can cause problems with capacity loss, quality issues, potential compliance concerns, and potential safety issues. Regular inspection of batteries for any signs of degradation will be noted, and repairs will be required, particularly in the case of electrolyte leaks and insulation failures (e.g., IEC 62485-5).

[0011] In an electrolyte leak event, the gas released is electrolyte solvent vapor, which is the primary gas in a battery off-gassing / venting event. In a battery off-gassing / cell venting event, trace amounts of hydrogen (H), carbon monoxide (CO), and carbon dioxide (CO) may be released, allowing them to be detected by H and / or CO detectors in close proximity to the venting cell. However, in an electrolyte leak event, only solvent vapor would theoretically be present, which would not be detectable using H and / or CO sensors. Specifically, electrolyte leakage can occur very slowly, while off-gassing / cell venting occurs quickly. Unlike off-gassing / cell venting, electrolyte leak detection is difficult because it may not occur at a distinct time and there may be no ability to establish a baseline or reference (e.g., if the cell is leaking prior to monitoring, a change is likely to go undetected).

[0012] The systems and methods described herein can detect electrolyte and / or coolant leaks. Additionally, the systems and methods described herein can be configured to monitor electrolyte and / or coolant leaks in any type of battery, including lithium-ion batteries and lead-acid batteries.

[0013] As used herein, the term "gas analyte" refers to gases released by a battery. Gas analytes can include electrolyte gases, such as volatile electrolyte solvents, volatile components of a battery's electrolyte mixture, or the like, and off-gases (i.e., "emitted gases" and "gas analytes"), including coolants (e.g., ethylene glycol / water mixtures). Volatile electrolyte or off-gas analyte species can include one or more of the following flammable or toxic gases: lithium-ion battery off-gas, dimethyl carbonate, diethyl carbonate, methyl ethyl carbonate, ethylene carbonate, propylene carbonate, vinylene carbonate, carbon dioxide, carbon monoxide, hydrocarbons, methane, ethane, ethylene, propylene, propane, benzene, toluene, hydrogen, oxygen, nitrogen oxides, volatile organic compounds, toxic gases, hydrogen chloride, hydrogen fluoride, hydrogen sulfide, sulfur oxides, ammonia, and chlorine, or the like. As used herein, gas sensors can detect gas analytes when present in the atmosphere.

[0014] Additionally, the systems and methods described herein may be configured with multiple battery storage devices. Accordingly, the systems and methods described herein may be used to monitor gas analytes emitted by one or more batteries located within the battery storage device. As used herein, the term "battery storage device" refers to any enclosure capable of at least partially enclosing one or more batteries. In one example, the battery storage device may include a ventilated battery storage device and a non-ventilated battery storage device. A ventilated battery storage device may include a ventilation system, which may include an intake and an exhaust device. In another example, the battery storage device may include a battery cabinet. In another example, the battery storage device may include a battery enclosure for a vehicle battery system. In a further example, the battery storage device may include a battery shipping container.

[0015] Additionally, as used herein, the term "processor" may refer to any device capable of executing machine-readable instructions, such as a computer, a controller, an integrated circuit (IC), a microchip, or any other device capable of implementing logic. As used herein, the term "memory" may refer to a non-transitory computer storage medium, such as volatile memory (e.g., random access memory), non-volatile memory (e.g., a hard disk drive, solid-state drive, flash memory, or the like), or a combination thereof.

[0016] 1A illustrates an example of a system 100 for monitoring electrolyte and / or coolant leaks. The system 100 includes at least one gas sensor 102 configured to monitor a gas analyte 104 released from a battery 106. The at least one gas sensor 102 may include any type of gas sensor, such as a chemiresistive sensor, an electrochemical sensor, a semiconductive metal oxide sensor, a catalytic sensor, a thermal conductivity sensor, a metal oxide semiconductor, a potentiometric sensor, an optical sensor, an infrared (IR) sensor, an amperometric sensor, a micro-hotplate sensor, or the like. The at least one gas sensor 102 may be positioned in proximity to the battery 106. The at least one gas sensor 102 is configured to generate a (real-time) sensor signal 110 to communicate the monitored gas analyte 104 of the battery 106.

[0017] The battery 106 may be any type of battery or battery system that includes an electrolyte. The battery 106 may further include a cooling system having a coolant for cooling the battery 106. The battery 106 may be a lithium-ion battery, a lead-acid battery, or any other type of rechargeable or non-rechargeable battery. The battery 106 is fully or partially enclosed by the enclosure 108.

[0018] Unlike most other systems, the deployed gas sensor 102 eliminates the requirement for using a separate reference sensor in the system 100 to calculate a running average from the (real-time) sensor signal 110 for detecting the gas analyte 104 released by the battery 106.

[0019] The system 100 also includes at least one sensor 112 for monitoring at least one variable of the battery 106. The at least one variable of the battery 106 includes one or more of the temperature of the battery 106 (e.g., cell temperature or overall temperature of the battery system), the current of the battery 106 (e.g., charge / discharge current), the relative humidity in the ambient environment in which the battery 106 is located, the airflow (e.g., flow rate) of the ambient air around the battery 106, or a combination thereof. The at least one sensor 112 includes a temperature sensor, a humidity sensor, an airflow sensor, and a current or resistivity sensor. The at least one sensor 112 may be located on or proximate to the battery 106, inside and / or outside the storage device 108. In some embodiments, the at least one gas sensor 102 and / or the at least one sensor 112 may be or include a sensor in a battery management system (BMS) of the battery 106. At least one sensor 112 is configured to generate a (real-time) sensor signal 114 communicating at least one monitored variable of the battery 106 .

[0020] The system 100 includes a controller 116 configured to operate and coordinate the operation of various components within the system 100. The controller 116 may include any suitable processor 118 (e.g., a microprocessor, a MOSFET, an IGBT, etc.) and a memory 120. The controller 116 may be programmed to implement certain procedures or predetermined procedures. In one example, an analytical procedure or algorithm 122 is stored in the memory 120 to process / analyze the sensor signals 110 and 114 to determine in real time any released gas analytes 104 as events, including one or more of the following: an electrolyte leak event, a coolant leak event, a water ingress event, a poisoned metal oxide sensor event, an off-gas event (OGE), a thermal runaway event (TRE), and an interfering outgassing event (i.e., non-OGE). A sensor poisoning event, as discussed herein, refers to poisoning of a sensor by a chemical that prevents the sensor from properly performing its function; the sensor poisoning event is not limited to a poisoned metal oxide sensor.

[0021] In another example, the analytical procedure or algorithm 122 may include a machine learning (ML) or deep learning (DL) algorithm (program code) to be executed by the processor 118 to enable the at least one gas sensor 102 to detect and classify, in real time, any emitted gas analyte 104 as being an event including one or more of the following: electrolyte leak, coolant leak, water ingress, poisoned metal oxide sensor, off-gassing event (OGE), thermal runaway event (TRE), and interfering outgassing event (i.e., non-OGE).

[0022] The controller 116 may include any suitable wired or wireless communication device / mechanism for outputting a signal or alarm 124 based on the analysis. The output signal or alarm 124 may be a warning alarm or a logic signal sent for on-screen warning or display to take preventative action indicating a condition of the battery 106 (e.g., electrolyte leak, coolant leak, water ingress, poisoned metal oxide sensor, OGE, TRE, non-OGE, etc.).

[0023] The at least one gas sensor 102 may be pre-trained and the ML or DL algorithm 122 may be stored as a candidate model in the memory 120 to distinguish between sensor signals 110 detected by the at least one gas sensor 102 without the need for a reference gas sensor or the further need to retrain the at least one gas sensor 102 once deployed in the field.

[0024] The machine learning and training of the algorithm 122 steps may be performed a priori at the factory during the manufacturing process or offline at any time prior to physical commissioning or installation of the at least one gas sensor 102 in the system 100. Once the at least one gas sensor 102 is operational in the system 100, real-time adaptation may not be necessary. Alternately, in another option, the ML or DL algorithm 122 may be retrained or updated by the at least one gas sensor 102 learning new encounters with other gas analytes not previously retrained or listed in the database. The purpose of this pre-training using the ML or DL algorithm is not only to detect OGE and refrigerant, but also to be able to identify other gas sources detected by the at least one gas sensor 102, thereby eliminating the need for a reference sensor.

[0025] 1A, the algorithm 122 may be implemented in an embedded microcontroller of the at least one gas sensor 102. The at least one gas sensor 102 and the controller 116 may be an integrated chip 126, such as an ASIC semiconductor chip. Alternatively, the at least one gas sensor 102 and the controller 116 may each be separate components electrically connected via a wiring harness or mounted on a printed circuit board (PCB).

[0026] In another embodiment, as shown in FIG. 1B, the controller 116 may be a separate computer (e.g., a computer in the BMS of the battery 106), and the sensor signal 110 may be sent to the controller 116 via wired or wireless communication.

[0027] 2 illustrates an example computer-implemented method 200 performed by system 100 to determine whether there is an electrolyte leak from the battery system. Method 200 includes monitoring a gas analyte level using a first gas sensor (step 202). At least one gas sensor 102 (e.g., any suitable gas sensor) is configured to detect a gas analyte 104. The at least one gas sensor 102 is positioned on or in close proximity to the battery 106 such that the at least one gas sensor 102 continuously monitors / measures the gas present and detects the gas analyte 104 released by the battery 106. The at least one gas sensor 102 generates a sensor signal 110 corresponding to the amount of gas detected.

[0028] The method 200 includes monitoring at least one variable of the battery system (step 204). At least one sensor 112 (e.g., any suitable sensor) is configured to monitor at least one variable of the battery 106, such as temperature, relative humidity, charge / discharge current, airflow around the battery 106, etc.

[0029] The method 200 includes determining whether a correlation exists between a monitored gas analyte level of the battery system 106 and at least one monitored variable (step 206), and determining a state of the battery system and / or a state of the first gas sensor based on the correlation (step 208).

[0030] In step 208, determining the status of the battery system may include determining whether there is an electrolyte leak from the battery system based on determining the correlation. The sensor signals 110 and 114, data, or information, are processed or analyzed by the controller 116 to determine whether there is a correlation between the two. Figure 3A shows an example plot 300 having a reference data series trend 302 and a test data series trend 304 plotted on an X-axis corresponding to at least one monitored variable 306 (e.g., a variable of the battery 106 or a variable corresponding to an environmental condition of the battery 106) and a Y-axis corresponding to a monitored gas analyte level 308.

[0031] The reference data series trend 302 indicates an expected correlation: the monitored gas analyte level remains relatively constant / unchanging while at least one monitored variable changes (e.g., increases or decreases). In this case, the monitored gas analyte level does not correlate with at least one variable of the battery 106 (e.g., temperature, relative humidity, charge / discharge current, airflow around the battery 106, etc.), indicating the absence of electrolyte leakage.

[0032] In the test data series trend 304, the monitored gas analyte level increases in a relatively linear manner as at least one monitored variable increases, indicating a relatively linear correlation between the two variables. In this case, the correlation between at least one variable of the battery 106 (e.g., temperature, relative humidity, charge / discharge current, airflow around the battery 106, etc.) and the monitored gas analyte level deviates from that of the reference data series trend 302. This deviation in correlation indicates an electrolyte leak.

[0033] The linear correlation in plot 300 is shown by way of non-limiting example only. Non-linear correlations between the monitored gas analyte levels of the battery 106 and the at least one monitored variable may exist. Any correlation (e.g., linear, non-linear, exponential, logarithmic, cubic, etc.) between the monitored gas analyte levels of the battery 106 and the at least one monitored variable that deviates from an expected correlation (e.g., reference data series trend 302) is indicative of electrolyte leakage.

[0034] The controller 116 or the algorithm 122 may be configured to determine the presence of an electrolyte leak based on the correlation coefficient. For example, the controller 116 or the algorithm 122 may be configured to determine that an electrolyte leak exists if the absolute value of the correlation coefficient of the test data series trend 304 is greater than a predetermined value. As another example, the controller 116 or the algorithm 122 may be configured to determine that an electrolyte leak exists if the difference between the correlation coefficient of the reference data series trend 302 and the correlation coefficient of the test data series trend 304 is greater than a predetermined threshold.

[0035] In step 208, determining the status of the first gas sensor may include determining whether the first gas sensor is poisoned based on the correlation determination. The sensor signals 110 and 114, data, or information, are processed or analyzed by the controller 116 to determine whether there is a correlation between the two and to determine the degree of data variance. Figure 3B shows an example plot 310 having a reference data series trend 312 and a test data series trend 314 plotted on an X-axis corresponding to at least one monitored variable 306 (e.g., a variable of the battery 106 or a variable corresponding to an environmental condition of the battery 106) and a Y-axis corresponding to a monitored gas analyte level 308.

[0036] The reference data series trend 312 shows the expected correlation with the expected data variance if no correlation exists between the two variables. The data variance shows the natural variance as expected because the at least one gas sensor 102 is expected to respond to natural variations in the background gas (e.g., ambient gas) within the battery housing 108. This results in a substantial variance in the data around the expected correlation 312 over time. The reference data series trend 312 indicates that the gas sensor (e.g., the at least one gas sensor 102) is not poisoned.

[0037] In the test data series trend 314, the monitored gas analyte level remains relatively constant / unchanging while at least one monitored variable changes. The degree of data variance / scatter is lower than that of the reference data series trend 312, indicating that a gas sensor (e.g., at least one gas sensor 102) is poisoned. If a gas sensor is poisoned, its response to background gas fluctuations will be lower, resulting in less variance in the data from the gas sensor.

[0038] The controller 116 or algorithm 122 may be configured to determine whether a gas sensor (e.g., at least one gas sensor 102) is poisoned based on the degree of data variance or variability index. In one example, the controller 116 or algorithm 122 may be configured to determine that a gas sensor is poisoned if the data variance or variability of the test data series trend 314 is less than a predetermined value or threshold. As another example, the controller 116 or algorithm 122 may be configured to determine that a gas sensor is poisoned if the difference between the data variance or variability of the test data series trend 314 and that of the reference data series trend 312 is greater than a predetermined value or threshold.

[0039] In one non-limiting example, the at least one variable monitored in FIGS. 3A and 3B is the temperature of the battery 106.

[0040] In addition to detecting electrolyte leaks based on correlations between monitored gas analyte levels and monitored variables in the battery (e.g., correlation methods for determining electrolyte leaks), system 100 is configured to address electrolyte solvent vapor leaks in the presence of other gases. At least one gas sensor 102 can be sensitive to other gases that may originate from other volatile organic compounds (VOCs), such as adhesives or off-gassing gasket materials within the battery module. System 100 can reliably differentiate when a sensor response is caused by small amounts of electrolyte vapor (true positive) or other VOCs (false positive).

[0041] In embodiments in which the system 100 is configured to differentiate between gas species, the at least one gas sensor 102 may include a gas sensor with an induced or modulated gas sensor manipulated variable. For example, the at least one gas sensor 102 may include multiple micro-hotplate sensors (e.g., a second gas sensor), and the operating temperature of the second gas sensor is modulated to resolve the difference between true-positive and false-positive scenarios. The change in temperature causes different gas species to react differently with the sensor electrode. This creates different signatures on the raw gas sensor signal that can be resolved to differentiate gases originating from positive sources (e.g., battery solvent vapors) or false-positive sources (e.g., off-gassing adhesives). Furthermore, based on this approach, the system 100 can detect and classify battery coolant leaks. Coolants used in liquid-cooled modules of batteries typically include a glycol-water mixture (e.g., a 50:50 ethylene glycol / water mixture in internal combustion engine automotive radiators for engine coolant). The system 100 is further configured to detect battery coolant leaks and therefore can monitor for unique failure modes in batteries.

[0042] As an example, system 100 is configured to sort the response of a second gas sensor (e.g., a micro-hotplate sensor) into carbonate and non-carbonate. The solvent in lithium-ion batteries is a carbonate-based solvent, and therefore the presence of carbonate-based solvent may indicate a leaking battery cell (e.g., electrolyte leakage), while the presence of hydrogen may indicate electrolysis occurring within the battery module due to coolant leakage or water ingress.

[0043] 4 illustrates an example computer-implemented method 400 performed by the system 100 to differentiate gas species and determine the state of the battery 106 based on the gas species differentiation. The method 400 includes modulating a gas sensor manipulated variable profile for each secondary gas sensor (step 402). A gas sensor manipulated variable, as discussed herein, refers to any variable for controlling / operating a gas sensor, including, but not limited to, temperature and a bias applied to a sensor element, such as power, voltage, current, or polarity bias. For purposes of discussion, temperature modulation is described below as an example, but the method 400 may be performed based on modulation of any one or more of the gas sensor manipulated variables.

[0044] Step 402 includes modulating the electrode temperature of each second gas sensor (e.g., a micro-hotplate sensor) in any suitable waveform (e.g., any periodic temperature change as a function of time). Gas sensor temperature profile modulation can be achieved through any suitable modulation of a gas sensor manipulated variable (e.g., hotplate temperature, bias applied to the sensor element, e.g., power, voltage, current, polarity, etc.).

[0045] In one example, the electrode temperature of each second gas sensor is modulated between an initial temperature (e.g., a temperature at 0% capacity or a minimum temperature) and a final temperature (e.g., a temperature at 100% capacity or a maximum temperature) at a predetermined change (e.g., increase or decrease) level and a predetermined time interval. Each predetermined change may be a predetermined temperature change of 5 degrees Celsius (°C), 10°C, 15°C, 20°C, etc., or a predetermined gas sensor manipulated variable change (e.g., a power or voltage change) of 5%, 10%, 15%, etc. In one example, step 402 includes modulating the plurality of electrodes of the second gas sensor 102 at a rate of 10°C per second from a minimum temperature of 100°C to a maximum temperature of 400°C. Step 402 includes holding the plurality of electrodes of the second gas sensor 102 at each temperature for a valid period. In step 402, the electrodes of the second gas sensor 102 may have the same or a different modulated temperature profile.

[0046] The method 400 includes monitoring a gas analyte through a modulated gas sensor manipulated variable profile (e.g., a temperature profile) (step 404). The method 400 also includes developing a data matrix containing monitored sensor data as a function of the modulated gas sensor manipulated variable profile (e.g., a temperature profile) and differentiating gas species of the monitored gas analyte based on a comparison of various features (step 406). The processor 118 receives the sensor signal 110 (e.g., impedance) and temperature data from the second gas sensor 102. The collected sensor signal 110 (e.g., impedance data) at every sensor temperature variable step (e.g., impedance at each temperature on all electrodes) is a "feature" that can be used for pattern recognition / classification. Any suitable algorithm or analysis technique 122 stored in the memory 120 can be used to perform step 406.

[0047] In one example, machine learning or deep learning techniques can be applied to perform step 406. Specifically, based on machine learning, step 406 includes developing features (step 408) based on the monitored gas analyte data and the modulated gas sensor manipulated variable profile (e.g., temperature profile). The number of features depends on the collected data set. If the monitored gas analyte data is collected from three electrodes in a temperature profile ramped from 100°C to 400°C in 5°C intervals, the total number of features is 183, the average impedance at each temperature over 1 minute of the temperature ramp from 100 to 400°C (61 temperature intervals). Therefore, method 400 can optionally include reducing the number of features (step 410). Principal Component Analysis (PCA) can be applied to reduce the number of features. This step is not necessarily required to fit and evaluate the model, but is an option to reduce dimensionality. Additionally, additional environmental variables of the battery 106, such as information measured by at least one sensor 112 (e.g., temperature, relative humidity, battery current, airflow, etc.), may be monitored. Alternatively, a unique signature may be generated through any suitable modulation of the gas sensor operational variables (e.g., hotplate temperature, bias applied to the sensor element, power, voltage, current, polarity, etc.) in step 406, and the signature may be developed based on the monitored gas analyte data and the modulated bias power, voltage, current, and / or polarity.

[0048] Method 400 includes feeding the features into a model (e.g., a machine learning or ML model) and outputting a classification to determine the gas species of the gas analyte (step 412). The candidate ML model is stored in memory 120 (e.g., algorithm 122) for analyzing the features. The candidate ML model can be any supervised machine learning model, such as a k-Nearest Neighbors (kNN) technique.

[0049] The method 400 includes determining 414 the state of the battery based on the determined species of the gas analytes 104. Specifically, the system 100 and method 400 can sort the gas analytes (e.g., into carbonates and non-carbonates) and / or differentiate or classify the gas analytes 104 into particular species to determine the state of the battery 106. For example, detection of hydrogen indicates electrolysis occurring within the battery 106 due to a coolant leak or water ingress, and detection of one or more electrolyte vapors indicates an electrolyte leak.

[0050] At least one gas sensor 102 and candidate model are pre-trained prior to deployment, and steps 402 through 410 can be performed offline to tune / adapt the model, while steps 402 through 414 are performed in real time for classification and decision making.

[0051] FIG. 5A shows an example impedance versus temperature plot for system 100 with three electrodes (e.g., three second gas sensors) when the temperature profile is modulated to ramp from 100°C to 400°C. FIG. 5B shows measurable features of the example monitoring data in FIG. 5A for each of the three electrodes. Based on PCA, the number of features is reduced as shown in FIG. 5C. When the features are fed into the ML model, the monitored gas analytes are distinguishable after one full sensor modulation period between a typical VOC interferent (a false positive detection) and an actual electrolyte leak (a true positive detection), as shown in FIG. 5D. Classification, such as the one shown in FIG. 5D, can be performed after every modulation period, after every sampling period, or both.

[0052] 6 and 7 show an example of how an ML classifier design process or ML algorithm can be developed. A pre-training (supervised learning) approach can combine multiple signal features (from at least one gas sensor 102 and at least one sensor 112) using both discovery-based and physics-based impedance information in the data pre-processing step of the algorithm development phase. It can also include environmental measurements such as temperature, relative humidity, airflow, and charge / discharge currents included in the sensor set. For example, signal features can include a moving average, Bollinger bands, minimum electrode impedance, maximum rate of impedance change, maximum rate of recovery of impedance for each electrode, principal component analysis (PCA), and linear discriminant analysis. In addition, pre-training (supervised learning) of gas sensors to distinguish between non-OGE and OGE or TRE using other techniques may also include, for example, but not limited to, the following classification techniques: support vector machines, discriminant analysis or nearest neighbor methods, basic statistics of time series value dispersion (e.g., location, spread, Gaussianity, outlier properties), linear correlation (e.g., autocorrelation, power spectrum features), stationarity (e.g., sliding window measures, forecast error), information theoretic and entropy / complexity, etc.

[0053] 7, an exemplary flow diagram illustrates a process 700 for pre-training at least one gas sensor 102 based on multiple known gas analytes using an ML algorithm. For example, in step 702, raw sensor signals 110, such as resistance and capacitance, for each known gas analyte may be generated from multiple electrodes of the at least one gas sensor 102 and sent to the processor 118 for feature extraction (e.g., change in impedance or transfer function over time) in step 704. The feature extraction step 704 may include a time-frequency transformation (e.g., a discrete cosine transform (DCT) or a discrete Fourier transform (DFT)) to convert the time-domain analog signal into a frequency-domain signal. In step 706, the extracted features may be organized accordingly. In step 708, an ML algorithm is applied to the organized data to establish a candidate model 710 (e.g., a multidimensional decision boundary construction). By repeating steps 702 through 710 for the remainder of the plurality of known gas analytes, the candidate model 710 may be updated to establish a database or to construct a synthetic decision boundary plot to complete training of the ML algorithm 122 stored in memory 120 to be executed by the processor 118 (see step 712). More specifically, step 708 may be accomplished by repeating training steps. Each training step may include sequentially performing the following operations: convolution, rectified linear unit (ReLU), and pooling operations. The deployed model 714 will become a field-ready ML algorithm when working in conjunction with a multi-electrode gas sensor to perform gas analyte classification.

[0054] Alternatively, these desired classifications can be achieved with deep learning algorithms using pre-trained convolutional neural networks (e.g., convolutional neural networks CNN and long short-term memory (LSTM)) and automatic signal feature extraction.

[0055] The methods and analyses disclosed herein enable differentiation of gas analytes 104 based on ML techniques. Figures 8 and 9 show two example output summaries based on the systems and methods disclosed herein. Figure 8 shows an example where the methods and systems disclosed herein can accurately differentiate gas analytes 104 into carbonates and non-carbonates without false positives or false negatives. Figure 9 is an example positive prediction rate matrix for various gas species.

[0056] Further, method 400 can be modified to determine the state of a gas sensor, e.g., a poisoned gas sensor. For example, the method can include modulating a gas sensor temperature profile for each gas sensor, monitoring a gas analyte through the modulated gas sensor temperature profile, developing a data matrix including gas sensor data monitored as a function of the modulated gas sensor temperature profile, and differentiating the gas sensor response based on a comparison of various features. The features are fed to a model, which outputs a classification to determine a unique gas sensor response characteristic. The method then determines whether the gas sensor is poisoned based on the determined gas sensor response. The method can include training the model with a data set collected from a wide range of gas sensor environments using poisoned gas sensors to develop the poisoned gas sensor data. The gas sensor features uniquely identify when a gas sensor is poisoned, and thus the model can be trained to differentiate poisoned gas sensors.

[0057] Methods 200 and 400 disclosed herein may be used alone or in combination. For example, method 400 may be used to confirm the results of method 200 and further confirm the detected gas species. In some embodiments, method 200 and / or method 400 may include steps for reporting or alerting the determined battery condition (e.g., electrolyte leak, coolant leak, etc.) and / or the detected gas species. For example, upon determination, method 200 or 400 may include steps for sending an early warning including a logic signal output, an audible alarm, a visual alarm, fire suppression, and / or communication with other systems and users.

[0058] The terms "comprise" or "comprising," to the extent that they are used in this specification or the claims, are intended to be inclusive in a manner similar to the term "comprises," as that term is interpreted when used as a transitional word in a claim. As used in this specification and the claims, the singular forms "a," "an," and "the" include the plural. Furthermore, to the extent that the term "or" (e.g., A or B) is used, it is intended to mean "A or B or both." Finally, when the term "about" or "approximately," when used in conjunction with a number, it is intended to include within ±5%, ±4%, ±3%, ±2%, ±1%, or ±0.5% of that number.

[0059] As noted above, the present application has been illustrated by description of embodiments, and while the embodiments have been described in considerable detail, it is not intended that such details be constrained or in any way limiting to the scope of the appended claims. Additional advantages and modifications will be readily apparent to those skilled in the art with the benefit of this application. Therefore, in its broader aspects, the present application is not limited to the specific details and illustrative examples shown. Departures may be made from such details and examples without departing from the spirit or scope of the general inventive concept.

Claims

1. monitoring a gas analyte level associated with the battery system using a first gas sensor; monitoring at least one variable of the battery system; determining whether a correlation exists between the monitored gas analyte level and the monitored at least one variable of the battery system; determining whether an electrolyte leak from the battery system exists based on the correlation determination; and A computer-implemented method comprising:

2. 2. The computer-implemented method of claim 1, wherein the correlation comprises an increasing trend of the monitored gas analyte level as the at least one monitored variable of the battery system increases.

3. The computer-implemented method of claim 1 , wherein the battery system is a lithium-ion battery system.

4. The computer-implemented method of claim 1 , wherein the monitored gas analytes include electrolyte gases and non-off-gas event (non-OGE) interfering gases.

5. The computer-implemented method of claim 4 , wherein the non-OGE interfering gas comprises hydrogen and / or a coolant.

6. The computer-implemented method of claim 1 , wherein the at least one monitored variable of the battery system includes a temperature of the battery system.

7. The computer-implemented method of claim 1 , wherein the at least one monitored variable of the battery system includes a current of the battery system.

8. The computer-implemented method of claim 1 , wherein the at least one monitored variable of the battery system includes an ambient relative humidity of the battery system.

9. The computer-implemented method of claim 1 , wherein the at least one monitored variable of the battery system includes airflow around the battery system.

10. modulating a sensor manipulated variable profile for each second gas sensor configured to monitor the gas analyte associated with the battery system; monitoring the gas analyte using the second gas sensor through the modulated sensor manipulated variable profile; developing a data matrix including a sensor signal generated by the second gas sensor as a function of the modulated sensor manipulated variable profile; differentiating gas species of the gas analytes based on a comparison of various features in the data matrix; determining a state of the battery system based on the differentiation of gas species; The computer-implemented method of claim 1 , comprising:

11. 11. The computer-implemented method of claim 10, comprising pre-training the second gas sensor based on a machine learning (ML) algorithm prior to initial field deployment of the second gas sensor.

12. 11. The computer-implemented method of claim 10, wherein the condition of the battery system includes one or more of an electrolyte leak, a coolant leak, a cell ejection, a thermal runaway, water ingress, and off-gassing.

13. The computer-implemented method of claim 10 , further comprising identifying a poisoned one of the second gas sensors.

14. The computer-implemented method of claim 10 , wherein the sensor manipulated variables include temperature, power, voltage, polarity, and / or current.

15. at least one gas sensor configured to monitor a gas analyte associated with the battery system; at least one sensor configured to monitor one or more variables of the battery system; a memory for storing machine-readable instructions; and a processor for accessing the memory and executing the machine-readable instructions; a controller comprising:

1. A monitoring system comprising: monitoring the gas analyte using the at least one gas sensor; monitoring the one or more variables of the battery system using the at least one sensor; determining a correlation between the monitored gas analyte level and the one or more variables; and determining whether an electrolyte leak from the battery system exists based on the correlation; a monitoring system that causes the processor to perform the following:

16. 16. The monitoring system of claim 15, wherein the battery system is a lithium ion battery system.

17. The monitoring system of claim 15 , wherein the at least one sensor includes one or more of a temperature sensor, a relative humidity sensor, a current sensor, and an airflow sensor.

18. The machine-readable instructions: modulating a sensor manipulated variable profile for the at least one gas sensor; monitoring a gas analyte level using the at least one gas sensor through the modulated sensor manipulated variable profile; developing a data matrix including sensor signals generated by the at least one gas sensor as a function of the modulated sensor manipulated variable profile; differentiating gas species of the gas analytes based on a comparison of various features in the data matrix; determining a state of the battery system based on the differentiation of gas species; The monitoring system of claim 15 , further comprising:

19. 20. The monitoring system of claim 18, wherein the machine-readable instructions cause the processor to determine whether there is a coolant leak in the battery system based on the differentiation of gas species.

20. 20. The monitoring system of claim 18, wherein the machine-readable instructions cause the processor to determine whether there is water ingress within the battery system based on the differentiation of gas species.

21. 20. The monitoring system of claim 18, wherein the machine-readable instructions cause the processor to determine whether the at least one gas sensor includes a poisoned sensor.

22. The monitoring system of claim 18 , wherein the sensor manipulated variables include temperature, power, voltage, polarity, and / or current.

23. modulating a sensor manipulated variable profile for each gas sensor configured to monitor a gas analyte associated with the battery system; monitoring the gas analyte using the gas sensor through the modulated sensor manipulated variable profile; developing a data matrix including sensor signals generated by the gas sensor as a function of the modulated sensor manipulated variable profile; differentiating gas species of the gas analytes based on a comparison of various features in the data matrix; determining a state of the battery system based on the differentiation of gas species; A computer-implemented method comprising:

24. 24. The computer-implemented method of claim 23, comprising pre-training a second gas sensor based on a machine learning (ML) algorithm prior to initial field deployment of the gas sensor.

25. 24. The computer-implemented method of claim 23, wherein the condition of the battery system includes one or more of an electrolyte leak, a coolant leak, a cell ejection, a thermal runaway, water ingress, and off-gassing.

26. differentiating gas sensor responses based on a comparison of various features in the data matrix; identifying a poisoned one of the second gas sensors based on said differentiation of gas sensor responses; 24. The computer-implemented method of claim 23, comprising:

27. 24. The computer-implemented method of claim 23, wherein the sensor manipulated variables include temperature, power, voltage, polarity, and / or current.

28. at least one gas sensor configured to monitor a gas analyte associated with the battery system; a memory for storing machine-readable instructions; and a processor for accessing the memory and executing the machine-readable instructions; a controller comprising:

1. A monitoring system comprising: modulating a sensor manipulated variable profile for each of the at least one gas sensor; monitoring a gas analyte level using the at least one gas sensor through the modulated sensor manipulated variable profile; developing a data matrix including sensor signals generated by the at least one gas sensor as a function of the modulated sensor manipulated variable profile; differentiating gas species of the gas analytes based on a comparison of various features in the data matrix; and determining a state of the battery system based on the differentiation of gas species; a monitoring system that causes the processor to perform the following:

29. 30. The monitoring system of claim 28, wherein the machine-readable instructions cause the processor to determine whether there is a coolant leak in the battery system based on the differentiation of gas species.

30. 30. The monitoring system of claim 28, wherein the machine-readable instructions cause the processor to determine whether there is water ingress within the battery system based on the differentiation of gas species.

31. 29. The monitoring system of claim 28, wherein the machine-readable instructions cause the processor to differentiate gas sensor responses based on a comparison of various features in the data matrix and identify a poisoned one of the at least one gas sensor based on the differentiation of gas sensor responses.

32. 30. The monitoring system of claim 28, wherein the sensor manipulated variables include temperature, power, voltage, polarity, and / or current.

Images