Power battery thermal runaway monitoring method based on machine smell

Abstract

Electric development is rapid, and power batteries are more and more applied to the aspects of electric automobiles, vertical take-off and landing aircrafts and the like. And the power battery is thermally runaway and seriously damaged. The thermal runaway is difficult to reverse when entering the middle and later periods, so that a novel monitoring and early warning method is urgently needed to be developed. According to the power battery thermal runaway monitoring method based on machine olfaction, characteristic gas and organic volatile matters released by the battery in a thermal runaway incubation period and an initial period are captured through a plurality of gas collector arrays fused with a plurality of sensing modules, and a multi-modal information fusion recognition technology is combined, so that the thermal runaway of the battery is monitored. According to the method and the system, early warning and evolution stage tracking of thermal runaway are realized, the problems of hysteresis quality, insufficient reliability and susceptibility to interference of an existing power battery thermal runaway monitoring method and system are solved, the safety of a power battery system is favorably improved, and conditions are provided for reliable operation and intelligent operation and maintenance of equipment such as an electric automobile, an airplane and an energy storage system.

Description

Technical Field

[0001] This invention relates to a method for monitoring thermal runaway of power batteries, and more particularly to a method for monitoring thermal runaway of power batteries based on machine olfaction, belonging to the field of power battery safety technology. Background Technology

[0002] With the rapid development of global energy structure transformation and electrification, power batteries are increasingly used in electric vehicles, vertical takeoff and landing aircraft, and other fields. However, the frequent occurrence of thermal runaway accidents in power batteries seriously restricts the safety and reliability of their large-scale application. During the use of power batteries, thermal runaway is prone to occur due to mechanical abuse, electrical abuse, and thermal abuse. For automotive power batteries, once thermal runaway occurs, it can pose a great threat to the personal safety of occupants and those in the surrounding area. Currently, power battery monitoring methods mainly rely on physical parameters such as temperature, voltage, and current, which suffer from problems such as lag, insufficient reliability, and susceptibility to interference. Significant changes in temperature and voltage often occur in the middle and late stages of irreversible thermal runaway, with extremely short warning windows that cannot provide sufficient time for evacuation, demonstrating lag. In the early stages of a battery short circuit, the temperature rise is small and localized, making it difficult for surface temperature sensors to effectively capture it; voltage changes may also be masked by the equalization strategy of the battery management system, resulting in insufficient reliability. The electromagnetic environment of on-board electronic and electrical equipment and energy storage systems is complex, and temperature sensors are easily affected by ambient temperature and heat dissipation conditions, leading to false alarms or missed alarms. Studies have shown that power batteries release characteristic gases before and during thermal runaway. For example, under abuse conditions such as overcharging, internal short circuits, and overheating, the electrolyte decomposes to produce gases such as CO, H2, CH4, and C2H4, as well as electrolyte solvent vapors and volatiles. The types, concentrations, and release sequences of these gases are strongly correlated with the battery failure mode and development stage, representing earlier and more direct characteristic signals than temperature and voltage. Machine olfaction technology mimics the biological olfactory system, using gas sensor arrays to detect gases. Therefore, by using high-sensitivity gas sensors to capture characteristic gases, volatile organic compounds, and electrolyte decomposition products released in the early stages of battery thermal runaway in real time, and combining this with artificial intelligence technology, a machine olfaction-based method for monitoring thermal runaway in power batteries is proposed. This method has significant engineering implications and is beneficial for improving the safety of power batteries. Summary of the Invention

[0003] 1. Purpose of the invention:

[0004] The technical problem to be solved by this invention is the existence of lag, insufficient reliability and susceptibility to interference in existing power battery thermal runaway monitoring methods and systems. This invention provides a power battery thermal runaway monitoring method based on machine olfaction. This method captures the characteristic gases and volatile organic compounds released by the battery during the incubation and early stages of thermal runaway, thereby achieving early warning and tracking of the evolution stage of thermal runaway, significantly improving the timeliness and reliability of the warning.

[0005] 2. Technical Solution:

[0006] A method for monitoring thermal runaway of power batteries based on machine olfaction, characterized by comprising the following steps:

[0007] Step S1, Gas Sampling and Sensing: Multiple gas collectors are placed at key locations above the power battery module, in the exhaust channel and the convergence area inside the battery box to monitor the gas and volatiles generated by the power battery module. The gas to be tested is guided into the gas collector through a micro pump and gas pipeline. The specific gas type and concentration information generated by the power battery module are obtained through the sensing module and the biomimetic snail-shaped spiral gas chamber of the gas collector.

[0008] The gas collector consists of an inlet, a sensing module, a biomimetic snail-shaped spiral gas chamber, a gas collector housing, an outlet, and a signal preprocessing module. The sensing module is a sensor array, which includes sensors sensitive to at least the following gases: a semiconductor gas sensor for monitoring electrolyte solvent vapor, a sensor for monitoring the thermal conductivity or catalytic combustion of hydrogen, an electrochemical or metal oxide semiconductor sensor for monitoring carbon monoxide, and a broad-spectrum semiconductor sensor for monitoring hydrocarbon gases. Each sensor in the array has cross-sensitivity to characteristic gases of the battery, collectively forming a response spectrum for monitoring gases in battery thermal runaway. The biomimetic snail-shaped spiral gas chamber is designed with a spiral ascending gas flow channel, extending the contact time between the gas and the sensor, while using centrifugal force to separate any electrolyte droplets that may be present in the gas, protecting the sensor. The system periodically extracts gas samples at a high flow rate, and after sampling, back-purges the gas path to prevent dust and condensate accumulation. A built-in high-precision temperature and humidity sensor performs real-time software compensation for baseline drift of the gas sensor, ensuring long-term stability.

[0009] Step S2, Data Preprocessing and Feature Extraction: The raw data of gases and volatiles generated by the power battery module are preprocessed through the signal preprocessing module, including filtering and noise reduction, baseline correction, and temperature and humidity compensation. Feature vectors are extracted from the preprocessed response signals. The feature vectors include, but are not limited to, the absolute concentration of each sensor, the rate of change of dynamic response, the integral area of ​​the response curve, the response ratio between sensors, the order of appearance of specific gases, and other time-series and correlation features, which constitute high-dimensional feature vectors.

[0010] Suppose the gas sensor array outputs an n-dimensional eigenvector:

[0011]

[0012] Among them, g i (t) represents the normalized response value of the i-th gas sensor;

[0013] Feature extraction uses a sliding window (window size W):

[0014]

[0015] in: The mean; Standard deviation; It is a first-order difference; Main component characteristics;

[0016] Step S3, Multimodal Information Fusion and Identification: Extract the features of voltage, current, and temperature data provided by the battery management system (BMS), and fuse the identified gas and volatile organic compound feature data with the voltage, current, and temperature data features provided by the battery management system (BMS);

[0017] The battery management system (BMS) provides voltage, current, and temperature data vectors:

[0018]

[0019] Where m is the number of temperature sensors;

[0020] The formulas for feature extraction from voltage, current, and temperature data are as follows:

[0021]

[0022] Due to the different sampling frequencies of the sensors, cubic spline interpolation is used for time alignment:

[0023]

[0024] Among them, t k To merge timestamps;

[0025] A weighted feature splicing and fusion method is adopted:

[0026]

[0027] in, This indicates the Hadamard product; This indicates vector concatenation; The adaptive weight matrix is ​​calculated using an attention mechanism: the weight parameters are updated online to adapt to battery aging.

[0028]

[0029] Let the probability allocation functions for the gas and volatile matter data generated by the power battery module and the data provided by the BMS system be m respectively. G and m B The output probability after fusion is:

[0030]

[0031] in, For conflict functions;

[0032] Step S4: Thermal Runaway State Judgment: The data preprocessing and feature extraction results are passed to the microprocessor for computation and analysis. A fusion model of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is adopted. CNN is used to process the spatial features extracted from the sensor array response, and LSTM is used to process the time series features of gas release. The input of this model is a multi-dimensional time series signal, and the output is the probability of thermal runaway risk, the probability distribution of fault type, and the development stage. The extracted feature vector is input to a pre-trained thermal runaway recognition model to determine the thermal runaway state. The thermal runaway recognition model is trained using a machine learning algorithm. Its training data comes from the gas response data collected by the gas acquisition array of different battery models under various abuse conditions and different aging states, from the initial abnormality to thermal runaway, and their corresponding known battery state labels, as well as the voltage, current, and temperature state labels of the BMS system. Through the above model, the state of the power battery is determined to be: normal state, early warning state, or thermal runaway warning state. The result is passed to the control module and the graded early warning and linkage control module. The control module is the decision module, and the graded early warning and linkage control module is the execution module.

[0033] Step S5, Graded Early Warning and Linkage Control: The graded early warning and linkage control module executes the following based on the power battery status:

[0034] If the model output is normal, continue monitoring;

[0035] If the model output is an early warning state, corresponding to a slight abnormality in the battery, such as trace gas evolution, then a level one warning will be activated, data will be recorded, a check will be prompted, an audible and visual alarm will be triggered, and the cloud platform will be notified.

[0036] If the model output indicates a thermal runaway warning state, corresponding to the start of a thermal runaway chain reaction and a rapid change in the concentration of characteristic gases, a level-two warning will be immediately activated. The highest-level alarm will be sent to the BMS system via the communication interface, triggering a series of coordinated measures including forced power outage, activation of the directional fire suppression system, audible and visual alarms, and prompting personnel to evacuate. The cloud platform will also be notified.

[0037] 3. The advantages of the present invention, "A method for monitoring thermal runaway of power batteries based on machine olfaction," are as follows:

[0038] (1) The early warning time is advanced. The characteristic gas released by the thermal runaway chemical reaction of the power battery is directly detected, which can provide an early warning several minutes to tens of minutes earlier than the traditional electrical signal or temperature monitoring, thus winning valuable time for safe disposal.

[0039] (2) High reliability. The gas signal is the direct product of the internal chemical reaction of the battery and is not affected by the external thermal management strategy. The false alarm rate is low. The cross-response mode and pattern recognition algorithm of the sensor array can effectively distinguish between battery fault gas and background ambient gas. The false alarm rate is reduced by more than one order of magnitude compared with the traditional monitoring scheme.

[0040] (3) It is easy to set up, highly adaptable, and non-invasive. The gas sampler probe does not need to be in direct contact with the battery cell. The sensor probe only needs to be placed in the air gap or ventilation channel inside the battery pack. There is no need to make structural modifications to the battery cell. It is suitable for battery packs with different packaging forms. The system can be trained by data from different battery systems and battery products from different manufacturers to adapt to multiple battery types. It has good versatility and expandability, and it is easy to modify or add to existing systems.

[0041] In summary, this invention can realize the monitoring and early warning of thermal runaway of power battery systems, which is conducive to improving the safety of power battery systems and provides conditions for the reliable operation and intelligent maintenance of equipment such as electric vehicles, aircraft, and energy storage systems. Attached Figure Description

[0042] The invention can be better understood by referring to the accompanying drawings and the following description of non-limiting preferred embodiments;

[0043] Figure 1 This is a schematic diagram of the system composition for implementing the present invention;

[0044] Figure 2 This is a schematic diagram of the gas sensor composition of the present invention;

[0045] Figure 3 This is a flowchart of the steps of the present invention;

[0046] Figure 1 and Figure 2 The symbols in the diagram are explained as follows: 1-Power battery module, 2-Battery box, 3-Gas collector, 4-Microprocessor, 5-Control module, 6-Graded early warning and linkage control module, 301-Air inlet, 302-Sensing module, 303-Bionic snail-shaped spiral air chamber, 304-Gas collector housing, 305-Air outlet, 306-Signal preprocessing module. Detailed Implementation

[0047] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings:

[0048] As attached Figure 1 and Figure 2 As shown, the components and systems involved in the power battery thermal runaway monitoring method based on machine olfaction of the present invention include: power battery module 1, battery box 2, gas collector 3, microprocessor 4, control module 5, graded early warning and linkage control module 6, air inlet 301, sensing module 302, biomimetic snail-shaped spiral air chamber 303, gas collector shell 304, air outlet 305, and signal preprocessing module 306.

[0049] As attached Figure 3 As shown, the present invention provides a method for monitoring thermal runaway of power batteries based on machine olfaction, comprising the following steps:

[0050] Step S1, Gas Sampling and Sensing: Multiple gas collectors 3 are arranged at key locations above the power battery module 1, in the exhaust channel and the confluence area inside the battery box 2 to monitor the gas and volatiles generated by the power battery module 1. The gas to be tested is guided into the gas collector 3 through a micro pump and gas pipeline. The specific gas type and concentration information generated by the power battery module 1 are obtained through the sensing module 302 and the biomimetic snail-shaped spiral gas chamber 303 of the gas collector 3.

[0051] The gas collector 3 consists of an inlet 301, a sensing module 302, a biomimetic snail-shaped spiral gas chamber 303, a gas collector housing 304, an outlet 305, and a signal preprocessing module 306. The sensing module 302 is a sensor array, which includes at least sensors sensitive to the following gases: a semiconductor gas sensor for monitoring electrolyte solvent vapor, a thermal conductivity or catalytic combustion sensor for monitoring hydrogen, an electrochemical or metal oxide semiconductor sensor for monitoring carbon monoxide, and a broad-spectrum semiconductor sensor for monitoring hydrocarbon gases. The sensors in the array exhibit cross-sensitivity to characteristic gases of the battery, collectively forming a response spectrum for monitoring gases during battery thermal runaway. The biomimetic snail-shaped spiral gas chamber 303 design features a spiral ascending gas flow channel, extending the contact time between the gas and the sensor. Simultaneously, centrifugal force is used to separate any electrolyte droplets that may be present in the gas, protecting the sensor. The system periodically extracts gas samples at high flow rates, and after sampling, the gas path is back-purged to clean it, preventing the accumulation of dust and condensate. A built-in high-precision temperature and humidity sensor performs real-time software compensation for baseline drift of the gas sensor, ensuring long-term stability.

[0052] Step S2, Data Preprocessing and Feature Extraction: The signal preprocessing module 306 preprocesses the raw data of gas and volatiles generated by the power battery module 1, including filtering and noise reduction, baseline correction, and temperature and humidity compensation. Feature vectors are extracted from the preprocessed response signals. The feature vectors include, but are not limited to, the absolute concentration of each sensor, the rate of change of dynamic response, the integral area of ​​the response curve, the response ratio between sensors, the order of appearance of specific gases, and other time-series and correlation features, which constitute high-dimensional feature vectors.

[0053] Suppose the gas sensor array outputs an n-dimensional eigenvector:

[0054]

[0055] Among them, g i (t) represents the normalized response value of the i-th gas sensor;

[0056] Feature extraction uses a sliding window (window size W):

[0057]

[0058] in: The mean; Standard deviation; It is a first-order difference; Main component characteristics;

[0059] Step S3, Multimodal Information Fusion and Identification: Extract the features of voltage, current, and temperature data provided by the battery management system (BMS), and fuse the identified gas and volatile organic compound feature data with the voltage, current, and temperature data features provided by the battery management system (BMS);

[0060] The battery management system (BMS) provides voltage, current, and temperature data vectors:

[0061]

[0062] Where m is the number of temperature sensors;

[0063] The formulas for feature extraction from voltage, current, and temperature data are as follows:

[0064]

[0065] Due to the different sampling frequencies of the sensors, cubic spline interpolation is used for time alignment:

[0066]

[0067] Among them, t k To merge timestamps;

[0068] A weighted feature splicing and fusion method is adopted:

[0069]

[0070] in, This indicates the Hadamard product; This indicates vector concatenation; The adaptive weight matrix is ​​calculated using an attention mechanism: the weight parameters are updated online to adapt to battery aging.

[0071]

[0072] Let the probability allocation functions for the gas and volatile matter data generated by power battery module 1 and the data provided by the BMS system be m respectively. G and m B The output probability after fusion is:

[0073]

[0074] in, For conflict functions;

[0075] Step S4: Thermal runaway state judgment: The data preprocessing and feature extraction results are passed to the microprocessor 4 for computation and analysis. A fusion model of convolutional neural network (CNN) and long short-term memory network (LSTM) is adopted. CNN is used to process the spatial features extracted from the sensor array response, and LSTM is used to process the time series features of gas release. The input of this model is a multi-dimensional time series signal, and the output is the probability of thermal runaway risk, the probability distribution of fault type, and the development stage. The extracted feature vector is input to the pre-trained thermal runaway recognition model to judge the thermal runaway state. The thermal runaway recognition model is trained by machine learning algorithm. Its training data comes from the gas response data collected by the gas collector 3 array and its corresponding known battery status labels during the entire process from initial abnormality to thermal runaway of different battery models under various abuse conditions and different aging states, as well as the voltage, current, and temperature status labels of the BMS system. Through the above model, the state of the power battery is determined to be: normal state, early warning state, or thermal runaway warning state. The result is passed to the control module 5 and the graded warning and linkage control module 6. The control module 5 is the decision module, and the graded warning and linkage control module 6 is the execution module.

[0076] Step S5, Graded Early Warning and Linkage Control: The graded early warning and linkage control module 6 executes the following based on the power battery status:

[0077] If the model output is normal, continue monitoring;

[0078] If the model output is an early warning state, corresponding to a slight abnormality in the battery, such as trace gas evolution, then a level one warning will be activated, data will be recorded, a check will be prompted, an audible and visual alarm will be triggered, and the cloud platform will be notified.

[0079] If the model output is a thermal runaway warning state, the corresponding thermal runaway chain reaction has been initiated and the concentration of characteristic gas has changed drastically. Then, a level two warning will be immediately activated, and the highest level alarm will be sent to the BMS system through the communication interface. This will trigger a series of linkage measures, including forced power outage, activation of the directional fire protection system, audible and visual alarms, prompting personnel to evacuate, and notification to the cloud platform.

[0080] The present invention has been described according to specific embodiments, but is not limited to the above examples. Any technical solution that conforms to the ideas of the present invention and is obtained by using similar structures and material substitution methods is within the protection scope of the present invention.

Claims

1. A method for monitoring thermal runaway of power batteries based on machine olfaction, characterized in that, Includes the following steps: Step S1, Gas Sampling and Sensing: Multiple gas collectors are placed at key locations above the power battery module, in the exhaust channel and the convergence area inside the battery box to monitor the gas and volatiles generated by the power battery module. The gas to be tested is guided into the gas collector through a micro pump and gas pipeline. The specific gas type and concentration information generated by the power battery module are obtained through the sensing module and the biomimetic snail-shaped spiral gas chamber of the gas collector. The gas collector consists of an inlet, a sensing module, a biomimetic snail-shaped spiral gas chamber, a gas collector housing, an outlet, and a signal preprocessing module. The sensing module is a sensor array, which includes sensors sensitive to at least the following gases: a semiconductor gas sensor for monitoring electrolyte solvent vapor, a sensor for monitoring the thermal conductivity or catalytic combustion of hydrogen, an electrochemical or metal oxide semiconductor sensor for monitoring carbon monoxide, and a broad-spectrum semiconductor sensor for monitoring hydrocarbon gases. Each sensor in the array has cross-sensitivity to characteristic gases of the battery, collectively forming a response spectrum for monitoring gases in battery thermal runaway. The biomimetic snail-shaped spiral gas chamber is designed with a spiral ascending gas flow channel, extending the contact time between the gas and the sensor, while using centrifugal force to separate any electrolyte droplets that may be present in the gas, protecting the sensor. The system periodically extracts gas samples at a high flow rate, and after sampling, back-purges the gas path to prevent dust and condensate accumulation. A built-in high-precision temperature and humidity sensor performs real-time software compensation for baseline drift of the gas sensor, ensuring long-term stability. Step S2, Data Preprocessing and Feature Extraction: The raw data of gases and volatiles generated by the power battery module are preprocessed through the signal preprocessing module, including filtering and noise reduction, baseline correction, and temperature and humidity compensation. Feature vectors are extracted from the preprocessed response signals. The feature vectors include, but are not limited to, the absolute concentration of each sensor, the rate of change of dynamic response, the integral area of ​​the response curve, the response ratio between sensors, the order of appearance of specific gases, and other time-series and correlation features, which constitute high-dimensional feature vectors. Suppose the gas sensor array outputs an n-dimensional eigenvector: Among them, g i (t) represents the normalized response value of the i-th gas sensor; Feature extraction uses a sliding window (window size W): in: The mean; Standard deviation; It is a first-order difference; Main component characteristics; Step S3, Multimodal Information Fusion and Identification: Extract the features of voltage, current, and temperature data provided by the battery management system (BMS), and fuse the identified gas and volatile organic compound feature data with the voltage, current, and temperature data features provided by the battery management system (BMS); The battery management system (BMS) provides voltage, current, and temperature data vectors: Where m is the number of temperature sensors; The formulas for feature extraction from voltage, current, and temperature data are as follows: Due to the different sampling frequencies of the sensors, cubic spline interpolation is used for time alignment: Among them, t k To merge timestamps; A weighted feature splicing and fusion method is adopted: in, This indicates the Hadamard product; This indicates vector concatenation; The adaptive weight matrix is ​​calculated using an attention mechanism: the weight parameters are updated online to adapt to battery aging. Let the probability allocation functions for the gas and volatile matter data generated by the power battery module and the data provided by the BMS system be m respectively. G and m B The output probability after fusion is: in, For conflict functions; Step S4: Thermal Runaway State Judgment: The data preprocessing and feature extraction results are passed to the microprocessor for computation and analysis. A fusion model of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is adopted. CNN is used to process the spatial features extracted from the sensor array response, and LSTM is used to process the time series features of gas release. The input of this model is a multi-dimensional time series signal, and the output is the probability of thermal runaway risk, the probability distribution of fault type, and the development stage. The extracted feature vector is input to a pre-trained thermal runaway recognition model to determine the thermal runaway state. The thermal runaway recognition model is trained using a machine learning algorithm. Its training data comes from the gas response data collected by the gas acquisition array of different battery models under various abuse conditions and different aging states, from the initial abnormality to thermal runaway, and their corresponding known battery state labels, as well as the voltage, current, and temperature state labels of the BMS system. Through the above model, the state of the power battery is determined to be: normal state, early warning state, or thermal runaway warning state. The result is passed to the control module and the graded early warning and linkage control module. The control module is the decision module, and the graded early warning and linkage control module is the execution module. Step S5, Graded Early Warning and Linkage Control: The graded early warning and linkage control module executes the following based on the power battery status: If the model output is normal, continue monitoring; If the model output is an early warning state, corresponding to a slight abnormality in the battery, such as trace gas evolution, then a level one warning will be activated, data will be recorded, a check will be prompted, an audible and visual alarm will be triggered, and the cloud platform will be notified. If the model output indicates a thermal runaway warning state, corresponding to the start of a thermal runaway chain reaction and a rapid change in the concentration of characteristic gases, a level-two warning will be immediately activated. The highest-level alarm will be sent to the BMS system via the communication interface, triggering a series of coordinated measures including forced power outage, activation of the directional fire suppression system, audible and visual alarms, and prompting personnel to evacuate. The cloud platform will also be notified.

Images