Lithium battery thermal runaway early detection device and method based on gas spectrum analysis

Abstract

The invention discloses a lithium battery thermal runaway early detection device and method based on gas spectrum analysis, and relates to the technical field of thermal runaway detection.A miniature gas sampling channel is arranged in a lithium battery pack or near a single battery cell, gas drainage is configured, gas released in the battery operation process is sampled, and gas sample data is obtained; according to the method, a multi-stage gas spectral analysis process is constructed, multi-dimensional spectral characteristics such as spectral peak intensity, spectral position offset and full width at half maximum are extracted from trace release gas collected in the running process of the lithium battery, a thermal runaway risk vector is constructed, and the thermal runaway risk vector is input into a thermal runaway recognition model MOD for real-time risk judgment. The early thermal chemical reaction process in the battery is effectively identified, the pre-judgment of the thermal runaway trend is completed in advance, the early warning of the thermal runaway state of the lithium battery is realized, the preposition of the system response and the safety protection level are improved, and the overall safety performance and reliability of the battery pack are improved.

Description

Technical Field

[0001] This invention relates to the field of thermal runaway detection technology, specifically to an early detection device and method for thermal runaway of lithium batteries based on gas spectral analysis. Background Technology

[0002] Lithium-ion battery thermal runaway detection devices mainly focus on sensing the operating status of the battery pack through temperature sensors, voltage monitors, and gas detection modules to achieve early identification and response control of thermal runaway risks. Most traditional solutions use thermistors or thermocouples distributed on the surface of battery modules or cells to collect temperature rise trends. Some systems also integrate smoke sensors or single-component gas sensors such as CO and HF to detect characteristic gases released during battery thermal decomposition and provide early warning of thermal runaway precursors. Existing lithium battery thermal runaway detection equipment generally lacks the ability to fuse multi-source signals and analyze evolution trends in signal acquisition and judgment logic. Most of them rely on threshold comparison or rule triggering mechanisms and lack data-driven dynamic risk modeling capabilities. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a device and method for early detection of thermal runaway in lithium batteries based on gas spectral analysis.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a lithium battery thermal runaway early detection device and method based on gas spectral analysis, comprising: S1. Set up a micro gas sampling channel inside the lithium battery pack or near the individual cell, and configure gas diversion to sample the gas released during battery operation and obtain gas sample data. S2. Use a Fourier transform infrared spectrometer (FTIR) to absorb gas sample data within a preset wavelength range, and perform background subtraction, spectral normalization, and smoothing on the gas detection data to generate a two-dimensional spectral matrix. S3. Analyze the two-dimensional spectral matrix, extract the peak intensity, peak position and full width at half maximum (FWHM) variation characteristic parameters of the target gas, and construct the gas spectral characteristic parameter set GFP; based on the spectral characteristic parameter set GFP, calculate the peak change rate, occurrence time difference and multi-component co-occurrence index of each gas component, and construct the thermal runaway risk vector GVI. S4. Input the thermal runaway risk vector GVI into the thermal runaway identification model MOD and output the current thermal runaway risk value RSK. S5. When the risk value RSK exceeds the preset threshold Rth, a thermal runaway early warning mechanism is triggered, an early warning signal is output to the battery management system (BMS), and battery isolation operation is performed.

[0005] The gas sampling includes: arranging at least one micro sampling channel inside the lithium battery pack or the casing of a single cell, wherein the micro sampling channel is connected to the gas release area inside the cell, and a controlled negative pressure extraction channel is formed by a micro-drainage pump. The pump extracts the gas released from inside the battery at a constant flow rate, removes moisture from the gas through a dehumidification chamber filled with desiccant, and introduces inert carrier gas to dilute the high-concentration original gas proportionally so that the gas concentration is maintained within the linear detection range of the spectral analysis module.

[0006] The pretreated gas sample is guided into the Fourier transform infrared spectrometer (FTIR) via a constant current and voltage stabilization module, with the working wavelength range set to 2.5 μm to 20 μm. The absorption of light of a gas sample by a specific wavelength is detected by a high-intensity broadband infrared light source, and raw spectral data is collected. Several preprocessing operations are performed on the collected spectral data, including: background subtraction using the static reference spectrum difference method to eliminate environmental background noise; spectral normalization based on the principle of maximum absorption intensity normalization to compress the spectral intensity to the [0,1] interval; and spectral smoothing using the Savitzky-Golay filter to maintain the shape of the spectral peaks and output a two-dimensional spectral matrix.

[0007] Perform multi-peak fitting and peak identification operations on the target gas spectral band in the standardized spectrum; The multi-peak fitting is based on the Gauss-Lorentz mixture model to achieve fine modeling of the characteristic absorption / scattering bands of HF, CO2, CH4, C2H4 and PF5; The peak amplitude of each component at its corresponding main absorption wavelength λ is obtained by using a peak intensity extraction function. Peak shape parameters are extracted using the full width at half maximum (FWHM) calculation formula; By comparing the offset Δλ=λt-λ0 of the same component spectral position λt in the spectra at different times, the evolution trend of its chemical reaction path is determined; a gas spectral feature parameter set GFP is constructed, which includes the peak intensity of the target gas, the full width at half maximum (FWHM), and the spectral position offset Δλ.

[0008] For the gas spectral characteristic parameter set GFP, peak derivative analysis and component correlation analysis are performed in time series order, including: Based on the peak intensity sequence over a continuous time period, the peak value variation rate of each target gas is calculated. Where P is the peak value of a certain gas at the current moment, and Δt is the sampling time interval; Using the event difference model The time difference index ΔTij between typical gases PF5 and HF at the time of their first appearance was obtained. For multiple gaseous components that appear simultaneously in the spectral lines, the co-occurrence ratio of the components is calculated as Cr=Nc / Nt, where Nc is the number of key gas species that appear at the same time and Nt is the total number of target components, which serves as a quantitative indicator of the synergistic effect of multi-component reactions. The Rv, ΔTij, and Cr characteristic data of each component are integrated into a thermal runaway risk vector GVI.

[0009] The thermal runaway identification model MOD is a deep neural network structure consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives the thermal runaway risk vector GVI, which includes n feature dimensions corresponding to the spectral peak change rate Rv, the component occurrence time difference ΔTij, and the co-occurrence ratio Cr. The hidden layer comprises four layers: a first hidden layer (128 nodes), a second hidden layer (256 nodes), a third hidden layer (128 nodes), and a fourth hidden layer (64 nodes), all employing the ReLU activation function to enhance nonlinear expressive power. A BatchNormalization layer is placed between the second and third layers to standardize the feature distribution. Dropout layers are placed between the third and fourth layers, with a dropout rate of 0.3, to suppress overfitting. The output layer consists of a single node that outputs the current thermal runaway risk value RSK, ranging from 0 to 1.

[0010] The training process of the thermal runaway identification model MOD includes: A historical training sample set is constructed, which is formed by multiple sets of lithium battery experimental data with known operating states. Each set of samples includes a gas spectral feature parameter set GFP collected in chronological order, a thermal runaway risk vector GVI calculated from the GFP, and a label of the actual thermal runaway state corresponding to that time period. The training sample set is cleaned and aligned to remove abnormal missing samples, and the GVI is normalized to eliminate the impact of differences in the scale of different features on model training. The processed sample set is divided into a training subset and a validation subset according to a preset ratio. The training subset is input into the thermal runaway identification model MOD. The risk value RSK output by the model is calculated through forward propagation, and the loss function value is calculated with the corresponding real state label. The loss function is used to characterize the deviation between the model output and the real state. The backpropagation algorithm is used to update the model parameters with gradients, and the model weights are gradually optimized through multiple rounds of iterative training, so that the RSK output by the model tends to be consistent with the actual thermal runaway state. After each training cycle, the performance of the current model is evaluated using a validation subset. When the validation error meets the preset convergence condition or reaches the maximum number of training cycles, training is stopped and the model parameters are fixed to form the thermal runaway identification model MOD.

[0011] The thermal runaway early warning mechanism is based on a risk value RSK and a preset early warning threshold Rth for logical judgment. It incorporates a delayed confirmation time ΔT and a buffer coefficient α. The judgment logic includes: When the thermal runaway risk value RSK within a continuous ΔT seconds satisfies RSK≥α·Rth, the thermal runaway risk is determined to be established, and the early warning mechanism is triggered; wherein, the buffer coefficient α ranges from 0.9 to 1.1, and the delayed confirmation time ΔT is set to 30 to 60 seconds; The dynamic risk value RSK′ is calculated using the following formula: Where RSK′ is the average thermal runaway risk value within the time window, and RSK(t′) is the instantaneous risk value at time t′; when RSK′≥Rth, a thermal runaway warning signal is triggered.

[0012] The thermal runaway early warning mechanism includes: when the thermal runaway risk value RSK exceeds the set threshold Rth, issuing a high-priority early warning command to the battery management system (BMS) and cutting off the high-voltage output circuit through the CAN bus interface.

[0013] A lithium battery thermal runaway early detection device based on gas spectral analysis includes: The gas sampling module is used to arrange miniature gas sampling channels inside the lithium battery pack or near the cells to continuously extract and sample the gas released during the operation of the lithium battery. The spectral analysis module uses a Fourier transform infrared spectrometer to detect the preprocessed gas sample and generate a two-dimensional standard spectral matrix. The feature extraction module extracts spectral feature parameters, including peak intensity, spectral position, and full width at half maximum (FWHM), from the standard spectral matrix. The thermal runaway identification module inputs the risk vector into the risk discrimination model, calculates and outputs the current thermal runaway risk value, and determines whether the battery is in an abnormal state. The early warning decision module uses a confirmation mechanism to determine the risk level based on the current risk value and a preset threshold.

[0014] This invention provides a device and method for early detection of thermal runaway in lithium batteries based on gas spectral analysis. Compared with existing technologies, it has the following advantages: This invention constructs a multi-stage gas spectral analysis process to extract multi-dimensional spectral features such as peak intensity, spectral position shift, and full width at half maximum (FWHM) from trace amounts of gas released during lithium battery operation. This process constructs a thermal runaway risk vector, which is then input into the thermal runaway identification model MOD for real-time risk assessment. This effectively identifies early thermochemical reaction processes inside the battery, allows for early prediction of thermal runaway trends, and provides early warning of lithium battery thermal runaway states. This enhances the proactiveness of system response and safety protection, thereby improving the overall safety performance and reliability of the battery pack. This invention introduces a dynamic judgment mechanism based on the average value of a time window, combined with a risk threshold buffer coefficient and a delayed confirmation time window setting, to improve the robustness of thermal runaway risk judgment and avoid false alarms caused by short-term anomalies; and links the high-voltage isolation control interface to cut off the output path, thereby improving the system's closed-loop control capability for handling anomalies, effectively reducing the risk of fire or explosion caused by thermal runaway, and ensuring the operational safety of electric vehicles, power storage systems and other application scenarios. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the principle framework of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This application provides a device and method for early detection of thermal runaway in lithium batteries based on gas spectral analysis, including: S1. Set up a micro gas sampling channel inside the lithium battery pack or near the individual cell, and configure gas diversion to sample the gas released during battery operation and obtain gas sample data. S2. Use a Fourier transform infrared spectrometer (FTIR) to absorb gas sample data within a preset wavelength range, and perform background subtraction, spectral normalization, and smoothing on the gas detection data to generate a two-dimensional spectral matrix. S3. Analyze the two-dimensional spectral matrix, extract the peak intensity, peak position and full width at half maximum (FWHM) variation characteristic parameters of the target gas, and construct the gas spectral characteristic parameter set GFP; based on the spectral characteristic parameter set GFP, calculate the peak change rate, occurrence time difference and multi-component co-occurrence index of each gas component, and construct the thermal runaway risk vector GVI. S4. Input the thermal runaway risk vector GVI into the thermal runaway identification model MOD and output the current thermal runaway risk value RSK. S5. When the risk value RSK exceeds the preset threshold Rth, a thermal runaway early warning mechanism is triggered, an early warning signal is output to the battery management system (BMS), and battery isolation operation is performed.

[0018] The gas sampling includes: arranging at least one micro sampling channel inside the lithium battery pack or the casing of a single cell, wherein the micro sampling channel is connected to the gas release area inside the cell, and a controlled negative pressure extraction channel is formed by a micro-drainage pump. The pump extracts the gas released from inside the battery at a constant flow rate, removes moisture from the gas through a dehumidification chamber filled with desiccant, and introduces inert carrier gas to dilute the high-concentration original gas proportionally so that the gas concentration is maintained within the linear detection range of the spectral analysis module.

[0019] The pretreated gas sample is guided into the Fourier transform infrared spectrometer (FTIR) via a constant current and voltage stabilization module, with the working wavelength range set to 2.5 μm to 20 μm. The absorption of light of a gas sample by a specific wavelength is detected by a high-intensity broadband infrared light source, and raw spectral data is collected. In this embodiment, for the gas sampling stage of the early thermal runaway detection device for lithium batteries, at least one miniature sampling port (GSP) is first installed inside the lithium battery pack casing or in the critical gas release area of ​​a single cell casing. This port communicates with the cell cavity and guides the gas released during operation toward the sampling system. A miniature gas extraction pump (GEP) is connected to the sampling port outlet. A constant gas extraction path is formed through a precisely controlled negative pressure extraction mechanism, ensuring that the gas is stably collected from the release source and avoiding leakage or back pressure interference.

[0020] The pump operates in constant flow mode, with the extraction rate typically set to 50–200 mL / min, configured based on the monitored gas diffusion rate and the analysis module response time. The extracted raw gas first passes through a dehumidification chamber (DHC) containing a built-in desiccant material (such as silica gel or molecular sieve) to preliminarily filter moisture and prevent interference with the spectral signal. Next, the gas flows into a dilution unit (DLU), where an inert carrier gas (such as nitrogen (N2) or argon (Ar) of a set concentration and flow rate is injected to dilute the sample gas concentration at a ratio of 1:5 to 1:20. This ensures the mixed gas concentration remains within the linear detection range of the subsequent spectral detection module, preventing saturation absorption or spectral distortion due to high concentrations.

[0021] The diluted gas sample was delivered to a Fourier transform infrared spectrometer (FTIR) at a stable flow rate (e.g., 100 mL / min ± 1%) via a flow control module (FCM) for detection. The FTIR module was set to operate in the wavelength range of 2.5 μm to 20 μm, corresponding to a frequency range of 4000–500 cm⁻¹. -1 It covers the characteristic absorption bands of key organic and inorganic volatile components, including CO2, CH4, C2H4, HF, and PF5.

[0022] Several preprocessing operations are performed on the collected spectral data, including: background subtraction using the static reference spectrum difference method to eliminate environmental background noise; spectral normalization based on the principle of maximum absorption intensity normalization to compress the spectral intensity to the [0,1] interval; and spectral smoothing using the Savitzky-Golay filter to maintain the shape of the spectral peaks and output a two-dimensional spectral matrix.

[0023] Perform multi-peak fitting and peak identification operations on the target gas spectral band in the standardized spectrum; The multi-peak fitting is based on the Gauss-Lorentz mixture model to achieve fine modeling of the characteristic absorption / scattering bands of HF, CO2, CH4, C2H4 and PF5; The peak amplitude of each component at its corresponding main absorption wavelength λ is obtained by using a peak intensity extraction function. Peak shape parameters are extracted using the full width at half maximum (FWHM) calculation formula; By comparing the offset Δλ=λt-λ0 of the same component spectral position λt in the spectra at different times, the evolution trend of its chemical reaction path is determined; a gas spectral feature parameter set GFP is constructed, which includes the peak intensity of the target gas, the full width at half maximum (FWHM), and the spectral position offset Δλ.

[0024] In this embodiment, for the gas spectral data acquired by the Fourier transform infrared spectrometer, spectral preprocessing, peak modeling and feature parameter extraction are performed sequentially to construct a set of gas spectral feature parameters for thermal runaway analysis.

[0025] Specifically, the raw spectral data first undergoes background subtraction processing. In this embodiment, background subtraction employs a static reference spectrum difference method. That is, during the system initialization phase, a background reference spectrum under conditions of no gas input is pre-acquired, and during subsequent real-time detection, the real-time spectrum is subjected to band-by-band difference processing with the reference spectrum. This eliminates the influence of environmental infrared radiation, light source drift, and system baseline changes on the detection results, resulting in net spectral data that characterizes only the gas absorption properties.

[0026] Subsequently, spectral normalization is performed on the net spectrum after background subtraction. In this embodiment, the normalization process is based on the principle of maximum absorption intensity standardization. The absorption intensity corresponding to each wavelength in the spectrum at the same time is proportionally scaled relative to the maximum absorption intensity at that time, so that the amplitude of the processed spectral lines is uniformly distributed within a fixed range. This reduces the amplitude deviation caused by overall changes in gas concentration or differences in detection sensitivity, and improves the comparability of spectral data at different times.

[0027] After normalization, spectral smoothing is performed on the spectral data. In this embodiment, Savitzky Golay filtering is used to smooth the normalized spectrum. High-frequency noise is suppressed through local polynomial fitting while preserving the geometric shape and positional characteristics of the spectral peaks. The final output is a structurally stable two-dimensional spectral matrix for subsequent feature analysis.

[0028] For the gas spectral characteristic parameter set GFP, peak derivative analysis and component correlation analysis are performed in time series order, including: Based on the peak intensity sequence over a continuous time period, the peak value variation rate of each target gas is calculated. Where P is the peak value of a certain gas at the current moment, and Δt is the sampling time interval; Using the event difference model The time difference index ΔTij between typical gases PF5 and HF at the time of their first appearance was obtained. For multiple gaseous components that appear simultaneously in the spectral lines, the co-occurrence ratio of the components is calculated as Cr=Nc / Nt, where Nc is the number of key gas species that appear at the same time and Nt is the total number of target components, which serves as a quantitative indicator of the synergistic effect of multi-component reactions. The Rv, ΔTij, and Cr characteristic data of each component are integrated into a thermal runaway risk vector GVI.

[0029] The thermal runaway identification model MOD is a deep neural network structure consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives the thermal runaway risk vector GVI, which includes n feature dimensions corresponding to the spectral peak change rate Rv, the component occurrence time difference ΔTij, and the co-occurrence ratio Cr. The hidden layer comprises four layers: a first hidden layer (128 nodes), a second hidden layer (256 nodes), a third hidden layer (128 nodes), and a fourth hidden layer (64 nodes), all employing the ReLU activation function to enhance nonlinear expressive power. A BatchNormalization layer is placed between the second and third layers to standardize the feature distribution. Dropout layers are placed between the third and fourth layers, with a dropout rate of 0.3, to suppress overfitting. The output layer consists of a single node that outputs the current thermal runaway risk value RSK, ranging from 0 to 1.

[0030] In this embodiment, the thermal runaway identification model MOD is designed as a typical feedforward deep neural network structure, consisting of an input layer, four hidden layers, a normalization and regularization layer, and an output layer. It receives the thermal runaway risk vector GVI output by the spectral feature extraction module and completes the inference and output of the risk value RSK. This structure possesses good nonlinear fitting ability and anti-overfitting ability, making it suitable for complex lithium battery thermal runaway feature identification tasks.

[0031] Specifically, the input layer is designed to accept an n-dimensional thermal runaway risk vector GVI, where n is the number of feature dimensions. In this embodiment, GVI includes three main feature dimensions: peak change rate Rv, component appearance time difference ΔTij, and multi-component co-occurrence ratio Cr, which are used to reflect the intensity dynamics, first response time differences, and multi-gas co-release behavior of specific gas components during the reaction process, respectively, serving as the core input basis for the model to determine the thermal runaway trend.

[0032] The hidden layer consists of four fully connected layers, with the following structure: the first hidden layer has 128 neural nodes, used for preliminary feature mapping and transformation of low-dimensional input features; the second hidden layer has 256 neural nodes, expanding the feature dimension to enhance the model's expressive power; the third hidden layer returns to 128 nodes, further compressing features to improve generalization; and the fourth hidden layer has 64 nodes, serving as the final layer for deep semantic feature extraction.

[0033] To enhance the stability of the network during training and accelerate convergence, a BatchNormalization layer is embedded between the second and third hidden layers to perform mean-variance normalization on each batch of input features, preventing gradient explosion or vanishing. At the same time, Dropout layers are set in the third and fourth layers respectively, with a Dropout rate of 0.3, which means that some neuron outputs in this layer are randomly blocked with a 30% probability, thereby breaking the complex dependencies between nodes and effectively preventing the model from overfitting on the training set.

[0034] The training process of the thermal runaway identification model MOD includes: A historical training sample set is constructed, which is formed by multiple sets of lithium battery experimental data with known operating states. Each set of samples includes a gas spectral feature parameter set GFP collected in chronological order, a thermal runaway risk vector GVI calculated from the GFP, and a label of the actual thermal runaway state corresponding to that time period. The training sample set is cleaned and aligned to remove abnormal missing samples, and the GVI is normalized to eliminate the impact of differences in the scale of different features on model training. The processed sample set is divided into a training subset and a validation subset according to a preset ratio. The training subset is input into the thermal runaway identification model MOD. The risk value RSK output by the model is calculated through forward propagation, and the loss function value is calculated with the corresponding real state label. The loss function is used to characterize the deviation between the model output and the real state. The backpropagation algorithm is used to update the model parameters with gradients, and the model weights are gradually optimized through multiple rounds of iterative training, so that the RSK output by the model tends to be consistent with the actual thermal runaway state. After each training cycle, the performance of the current model is evaluated using a validation subset. When the validation error meets the preset convergence condition or reaches the maximum number of training cycles, training is stopped and the model parameters are fixed to form the thermal runaway identification model MOD.

[0035] The thermal runaway early warning mechanism is based on a risk value RSK and a preset early warning threshold Rth for logical judgment. It incorporates a delayed confirmation time ΔT and a buffer coefficient α. The judgment logic includes: When the thermal runaway risk value RSK within a continuous ΔT seconds satisfies RSK≥α·Rth, the thermal runaway risk is determined to be established, and the early warning mechanism is triggered; wherein, the buffer coefficient α ranges from 0.9 to 1.1, and the delayed confirmation time ΔT is set to 30 to 60 seconds; The dynamic risk value RSK′ is calculated using the following formula: Where RSK′ is the average thermal runaway risk value within the time window, and RSK(t′) is the instantaneous risk value at time t′; when RSK′≥Rth, a thermal runaway warning signal is triggered.

[0036] This invention introduces a dynamic judgment mechanism based on the average value of a time window, combined with a risk threshold buffer coefficient and a delayed confirmation time window setting, to improve the robustness of thermal runaway risk judgment and avoid false alarms caused by short-term anomalies; and links the high-voltage isolation control interface to cut off the output path, thereby improving the system's closed-loop control capability for handling anomalies, effectively reducing the risk of fire or explosion caused by thermal runaway, and ensuring the operational safety of electric vehicles, power storage systems and other application scenarios.

[0037] The thermal runaway early warning mechanism includes: when the thermal runaway risk value RSK exceeds the set threshold Rth, issuing a high-priority early warning command to the battery management system (BMS) and cutting off the high-voltage output circuit through the CAN bus interface.

[0038] In this embodiment, the thermal runaway early warning mechanism is built on the output of the thermal runaway identification model MOD. Based on the dynamic monitoring results of the risk value RSK, it executes the linkage response logic between the battery management system (BMS) to ensure that once a potential thermal runaway trend is detected, a high-priority emergency response command can be triggered, thereby ensuring the safety and stability of the battery system operation.

[0039] Specifically, when the risk value RSK output by the thermal runaway identification model MOD exceeds the system's preset thermal runaway threshold Rth, the early warning logic module will determine that a "thermal runaway event has occurred" and immediately initiate the early warning response process. First, the early warning module pushes a high-priority early warning command to the battery management system (BMS). This command is transmitted in real time through a preset CAN bus interface, and the data frame contains the current RSK value, trigger timestamp, and device identification code to ensure that the lower-level system can accurately locate the source of the anomaly.

[0040] Subsequently, after receiving the high-priority warning command, the BMS immediately stopped the normal data processing flow according to the priority scheduling strategy, executed the emergency control logic, and scheduled the power control module to cut off the high-voltage output circuit and disconnect the high-voltage connection between the powertrain and the battery, thereby effectively isolating the possible thermal runaway propagation path and avoiding further thermal spread or chain reaction caused by battery failure.

[0041] Beneficial effects: This invention constructs a multi-stage gas spectral analysis process to extract multi-dimensional spectral features such as peak intensity, spectral position shift, and full width at half maximum (FWHM) from trace amounts of gas released during lithium battery operation. This constructs a thermal runaway risk vector, which is then input into the thermal runaway identification model MOD for real-time risk assessment. This effectively identifies early thermochemical reaction processes inside the battery, predicts thermal runaway trends in advance, and provides early warning of lithium battery thermal runaway states. This enhances the proactiveness of system response and safety protection, and improves the overall safety performance and reliability of the battery pack.

[0042] A lithium battery thermal runaway early detection device based on gas spectral analysis includes: The gas sampling module is used to arrange miniature gas sampling channels inside the lithium battery pack or near the cells to continuously extract and sample the gas released during the operation of the lithium battery. The spectral analysis module uses a Fourier transform infrared spectrometer to detect the preprocessed gas sample and generate a two-dimensional standard spectral matrix. The feature extraction module extracts spectral feature parameters, including peak intensity, spectral position, and full width at half maximum (FWHM), from the standard spectral matrix. The thermal runaway identification module inputs the risk vector into the risk discrimination model, calculates and outputs the current thermal runaway risk value, and determines whether the battery is in an abnormal state. The early warning decision module uses a confirmation mechanism to determine the risk level based on the current risk value and a preset threshold.

[0043] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0044] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0045] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for early detection of thermal runaway of lithium batteries based on gas spectroscopy analysis, characterized in that, include: S1. Set up a micro gas sampling channel inside the lithium battery pack or near the individual cell, and configure gas diversion to sample the gas released during battery operation and obtain gas sample data. S2. Use a Fourier transform infrared spectrometer (FTIR) to absorb gas sample data within a preset wavelength range, and perform background subtraction, spectral normalization, and smoothing on the gas detection data to generate a two-dimensional spectral matrix. S3. Analyze the two-dimensional spectral matrix, extract the peak intensity, peak position and full width at half maximum (FWHM) variation characteristic parameters of the target gas, and construct the gas spectral characteristic parameter set GFP; Based on the spectral characteristic parameter set GFP, the peak change rate, occurrence time difference and multi-component co-occurrence index of each gas component are calculated, and the thermal runaway risk vector GVI is constructed. S4. Input the thermal runaway risk vector GVI into the thermal runaway identification model MOD, and output the current thermal runaway risk value RSK. S5. When the risk value RSK exceeds the preset threshold Rth, the thermal runaway early warning mechanism is triggered, and an early warning signal is output to the battery management system (BMS) to perform battery isolation operation.

2. The method for early detection of thermal runaway of lithium battery based on gas spectrum analysis according to claim 1, characterized in that, The gas sampling includes: arranging at least one micro sampling channel inside the lithium battery pack or the casing of a single cell, wherein the micro sampling channel is connected to the gas release area inside the cell, and a controlled negative pressure extraction channel is formed by a micro-drainage pump. The pump extracts the gas released from inside the battery at a constant flow rate, removes moisture from the gas through a dehumidification chamber filled with desiccant, and introduces inert carrier gas to dilute the high-concentration original gas proportionally so that the gas concentration is maintained within the linear detection range of the spectral analysis module.

3. The method for early detection of thermal runaway of lithium battery based on gas spectrum analysis according to claim 1, characterized in that, The pretreated gas sample is guided into the Fourier transform infrared spectrometer (FTIR) via a constant current and voltage stabilization module, with the working wavelength range set to 2.5 μm to 20 μm. The absorption of light of a gas sample by a specific wavelength is detected by a high-intensity broadband infrared light source, and raw spectral data is collected. Several preprocessing operations are performed on the acquired spectral data, including: background subtraction using the static reference spectrum difference method to eliminate environmental background noise; Spectral line normalization is based on the principle of maximum absorption intensity normalization, compressing the spectral intensity to the [0,1] interval; spectral line smoothing uses a Savitzky-Golay filter to maintain the shape of the spectral peaks and outputs a two-dimensional spectral matrix. 4.The early detection of thermal runaway of lithium battery based on gas spectrum analysis method according to claim 1, characterized in that, Perform multi-peak fitting and peak identification operations on the target gas spectral band in the standardized spectrum; The multi-peak fitting is based on the Gauss-Lorentz mixture model to achieve fine modeling of the characteristic absorption / scattering bands of HF, CO2, CH4, C2H4 and PF5; The peak amplitude of each component at its corresponding main absorption wavelength λ is obtained by using a peak intensity extraction function. Peak shape parameters are extracted using the full width at half maximum (FWHM) calculation formula; By comparing the offset Δλ=λt-λ0 of the same component spectral position λt in the spectra at different times, the evolution trend of its chemical reaction path is determined; a gas spectral feature parameter set GFP is constructed, which includes the peak intensity of the target gas, the full width at half maximum (FWHM), and the spectral position offset Δλ.

5. The method for early detection of thermal runaway of lithium battery based on gas spectroscopy analysis according to claim 1, characterized in that, For the gas spectral characteristic parameter set GFP, peak derivative analysis and component correlation analysis are performed in time series order, including: Based on the peak intensity sequence over a continuous time period, the peak value variation rate of each target gas is calculated. Where P is the peak value of a certain gas at the current moment, and Δt is the sampling time interval; Using the event difference model The time difference index ΔTij between typical gases PF5 and HF at the time of their first appearance was obtained. For multiple gaseous components that appear simultaneously in the spectral lines, the co-occurrence ratio of the components is calculated as Cr=Nc / Nt, where Nc is the number of key gas species that appear at the same time and Nt is the total number of target components, which serves as a quantitative indicator of the synergistic effect of multi-component reactions. The Rv, ΔTij, and Cr characteristic data of each component are integrated into a thermal runaway risk vector GVI.

6. The method for early detection of thermal runaway in lithium batteries based on gas spectral analysis according to claim 1, characterized in that, The thermal runaway identification model MOD is a deep neural network structure consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives the thermal runaway risk vector GVI, which includes n feature dimensions corresponding to the spectral peak change rate Rv, the component occurrence time difference ΔTij, and the co-occurrence ratio Cr. The hidden layer comprises four layers: a first hidden layer (128 nodes), a second hidden layer (256 nodes), a third hidden layer (128 nodes), and a fourth hidden layer (64 nodes), all employing the ReLU activation function to enhance nonlinear expressive power. A BatchNormalization layer is placed between the second and third layers to standardize the feature distribution. Dropout layers are placed between the third and fourth layers, with a dropout rate of 0.3, to suppress overfitting. The output layer consists of a single node that outputs the current thermal runaway risk value RSK, ranging from 0 to 1.

7. The method for early detection of thermal runaway in lithium batteries based on gas spectral analysis according to claim 1, characterized in that, The training process of the thermal runaway identification model MOD includes: A historical training sample set is constructed, which is formed by multiple sets of lithium battery experimental data with known operating states. Each set of samples includes a gas spectral feature parameter set GFP collected in chronological order, a thermal runaway risk vector GVI calculated from the GFP, and a label of the actual thermal runaway state corresponding to that time period. The training sample set is cleaned and aligned to remove abnormal missing samples, and the GVI is normalized to eliminate the impact of differences in the scale of different features on model training. The processed sample set is divided into a training subset and a validation subset according to a preset ratio. The training subset is input into the thermal runaway identification model MOD. The risk value RSK output by the model is calculated through forward propagation, and the loss function value is calculated with the corresponding real state label. The loss function is used to characterize the deviation between the model output and the real state. The backpropagation algorithm is used to update the model parameters with gradients, and the model weights are gradually optimized through multiple rounds of iterative training, so that the RSK output by the model tends to be consistent with the actual thermal runaway state. After each training cycle, the performance of the current model is evaluated using a validation subset. When the validation error meets the preset convergence condition or reaches the maximum number of training cycles, training is stopped and the model parameters are fixed to form the thermal runaway identification model MOD.

8. The method for early detection of thermal runaway in lithium batteries based on gas spectral analysis according to claim 1, characterized in that, The thermal runaway early warning mechanism is based on a risk value RSK and a preset early warning threshold Rth for logical judgment. It incorporates a delayed confirmation time ΔT and a buffer coefficient α. The judgment logic includes: When the thermal runaway risk value RSK within a continuous ΔT seconds satisfies RSK≥α·Rth, the thermal runaway risk is determined to be established, and the early warning mechanism is triggered; wherein, the buffer coefficient α ranges from 0.9 to 1.1, and the delayed confirmation time ΔT is set to 30 to 60 seconds; The dynamic risk value RSK′ is calculated using the following formula: Where RSK′ is the average thermal runaway risk value within the time window, and RSK(t′) is the instantaneous risk value at time t′; when RSK′≥Rth, a thermal runaway warning signal is triggered.

9. The method for early detection of thermal runaway in lithium batteries based on gas spectral analysis according to claim 1, characterized in that, The thermal runaway early warning mechanism includes: when the thermal runaway risk value RSK exceeds the set threshold Rth, issuing a high-priority early warning command to the battery management system (BMS) and cutting off the high-voltage output circuit through the CAN bus interface.

10. A lithium battery thermal runaway early detection device based on gas spectral analysis, based on the lithium battery thermal runaway early detection method based on gas spectral analysis according to any one of claims 1-9, characterized in that, include: The gas sampling module is used to arrange miniature gas sampling channels inside the lithium battery pack or near the cells to continuously extract and sample the gas released during the operation of the lithium battery. The spectral analysis module uses a Fourier transform infrared spectrometer to detect the preprocessed gas sample and generate a two-dimensional standard spectral matrix. The feature extraction module extracts spectral feature parameters, including peak intensity, spectral position, and full width at half maximum (FWHM), from the standard spectral matrix. The thermal runaway identification module inputs the risk vector into the risk discrimination model, calculates and outputs the current thermal runaway risk value, and determines whether the battery is in an abnormal state. The early warning decision module uses a confirmation mechanism to determine the risk level based on the current risk value and a preset threshold.

Images