Lithium battery safety monitoring method and system based on multi-sensor fusion

Abstract

The present invention discloses a lithium battery safety monitoring method and system based on multi-sensor fusion, belonging to the field of lithium battery safety monitoring technology. The method includes the following steps: continuously monitoring the internal pressure of the battery through an air pressure sensor, combining temperature sensor data for temperature compensation correction, and triggering a sealing or swelling warning when the corrected air pressure value exceeds a preset range; collecting acceleration, angular velocity, vibration frequency, and posture data through an IMU sensor, and determining drop or collision events based on a scene-adaptive threshold; uploading the sensor data to the cloud at fixed intervals or event triggering, and generating a safety risk level through AI algorithm analysis; and receiving and displaying the risk assessment results and warning information on the user terminal. The present invention can realize battery swelling warning, external force impact monitoring, chemical anomaly identification, and dynamic safety assessment, solving the problem of traditional BMS's insufficient monitoring of physical deformation, sudden impact, and chemical anomalies.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery safety monitoring, and in particular to a lithium battery safety monitoring method and system based on multi-sensor fusion. Background Art

[0002] Traditional battery management systems (BMS) estimate battery state of health (SOH) and state of safety (SOS) by monitoring parameters such as cell voltage, current, temperature, and capacity. However, they have the following limitations:

[0003] 1. Unable to provide early warning of physical deformation failures such as battery swelling and internal short circuit;

[0004] 2. Lack of real-time monitoring of battery impacts (such as drops and collisions);

[0005] 3. Relying on empirical models, the prediction accuracy of sudden abnormal conditions is insufficient. Summary of the Invention

[0006] The technical task of this invention is to address these shortcomings by providing a lithium battery safety monitoring method and system based on multi-sensor fusion. This system can provide battery swelling warning, external impact monitoring, chemical anomaly identification, and dynamic safety assessment. This solves the problem of traditional BMS's inadequate monitoring of physical deformation, sudden impact, and chemical anomalies. Through scenario-adaptive thresholds, temperature compensation, and multi-parameter coupling analysis, the system significantly improves warning accuracy and applicability.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] The lithium battery safety monitoring method based on multi-sensor fusion includes the following steps:

[0009] S1: The pressure sensor continuously monitors the internal pressure of the battery and performs temperature compensation correction based on the temperature sensor data. When the corrected pressure value exceeds the preset range, a sealing or bulging warning is triggered.

[0010] S2: The IMU sensor collects acceleration, angular velocity, vibration frequency, and posture data, and determines fall or collision events based on scene-adaptive thresholds.

[0011] S3: Sensor data is uploaded to the cloud at fixed intervals or event triggers, and analyzed by AI algorithms to generate security risk levels;

[0012] S4: The user terminal receives and displays the risk assessment results and warning information.

[0013] This method uses a built-in barometer, inertial navigation sensor, and gas composition sensor, combined with cloud-based AI algorithm edge-side intelligent preprocessing, to achieve battery sealing detection, deformation warning, external force impact monitoring, internal gas anomaly analysis, and dynamic safety assessment.

[0014] Furthermore, the temperature compensation correction in step S1 is implemented based on the ideal gas state equation PV=nRT, where P is the gas pressure, V is the gas volume, n is the amount of gas substance, R is the ideal gas constant, and T is the gas temperature; the correction formula is:

[0015] P0=P 测 ×T0 / T 测 ;

[0016] Among them, P0 is the corrected air pressure reference value, P 测 is the internal pressure value of the battery measured in real time by the pressure sensor, T0 is the absolute temperature corresponding to 25℃, T 测 The real-time absolute temperature.

[0017] Furthermore, if the air pressure value exceeds the preset range (e.g., a drop of more than 5% or an increase of more than 10%), it is determined to be a seal failure or bulging risk, triggering an early warning;

[0018] At the same time, the gas composition sensor is linked. If the H2 concentration is detected to be greater than 50ppm or the CO concentration is greater than 20ppm, it is determined to be a risk of micro-short circuit of the battery cell and the warning level is upgraded.

[0019] Furthermore, the AI ​​algorithm described in step S3 adopts a multimodal fusion network, including a GRU / LSTM layer to process BMS time series data, a CNN branch to analyze the spectral characteristics of the impact event, and a Transformer layer to perform attention weighting on the gas composition time series data. The input data includes air pressure sequence, impact event vector, BMS data matrix and gas composition sequence.

[0020] Furthermore, the method also includes an active intervention mechanism: when the AI ​​algorithm determines that the risk level is high, the edge microprocessor triggers the BMS to perform current limiting, power off or cooling system startup operations, and pushes emergency warning information to the user terminal.

[0021] Furthermore, the AI ​​algorithm is trained through accelerated aging experiment data, extreme environment test data and a variety of typical failure scenario data to reduce false alarm rates and missed alarm rates.

[0022] Furthermore, the step S3 is specifically implemented as follows:

[0023] S3.1: Sensor data is uploaded to the cloud at regular intervals (e.g., every minute) or event-triggered (e.g., when a threshold is exceeded); this includes:

[0024] Air pressure / deformation data: obtained by real-time monitoring of air pressure fluctuations inside the battery cell and shell deformation through embedded sensors;

[0025] Impact event data: records physical impact during transportation or use, and combines accelerometers to capture time, intensity, and direction information;

[0026] BMS timing data: collected core parameters including voltage, current, temperature, SOC / SOH, etc.

[0027] Gas composition data: records the time-series changes in the concentration of characteristic gases such as H2 and CO, and generates the slope of the gas concentration-time curve (for example, an H2 concentration slope > 10 ppm / min is considered abnormal);

[0028] S3.2: Analyze time series data using cloud-based AI algorithms (such as LSTM neural networks), train a safety status model, and dynamically output the risk level (normal / warning / dangerous). The process is as follows:

[0029] (1) Data cleaning and alignment:

[0030] Eliminate outliers (such as those caused by sensor failure) and fill in missing values ​​(such as linear interpolation);

[0031] Unified timestamps ensure time synchronization of multi-source data (e.g. alignment of charge and discharge cycles with shock events);

[0032] Added data consistency check: If the air pressure suddenly rises by more than 5% but the gas composition is normal, it is determined to be sensor drift, and the air pressure sensor is automatically calibrated (calibration coefficient: K = initial air pressure / current air pressure average, effective after 3 consecutive calibrations);

[0033] (2) Multidimensional feature analysis:

[0034] (2.1) Time domain and frequency domain feature extraction:

[0035] Air pressure / deformation characteristics: calculate the air pressure change rate, cumulative deformation, and pressure curve asymmetry;

[0036] Impact characteristics: extract impact peak value, duration, and energy spectrum density;

[0037] BMS dynamic characteristics: Generate derived indicators including voltage / current differential, internal resistance change, temperature gradient, etc.;

[0038] Gas characteristics: calculate the H2 / CO concentration peak, concentration rising rate, and characteristic gas concentration ratio (e.g., if H2 / CO>2.5, it is determined that the electrolyte is decomposed);

[0039] (2.2) Construction of correlation features:

[0040] Cross-analyze the correlation between sudden pressure rise and temperature anomalies (such as precursors to thermal runaway);

[0041] Combined with the voltage drop amplitude and recovery rate after the impact event (to evaluate the impact of mechanical damage on electrochemical performance);

[0042] A three-dimensional correlation model of "pressure rise rate-characteristic gas concentration-temperature gradient" was constructed. When the three are synchronously abnormal (such as pressure rise rate > 0.2kPa / min, H2 rise rate > 15ppm / min, and temperature gradient > 2°C / min), it is determined to be a precursor to thermal runaway.

[0043] (3) Model training and algorithm selection:

[0044] A multimodal fusion network is used: the GRU / LSTM layer processes BMS time series data (capturing long-term dependencies); the CNN branch analyzes the spectral characteristics of the impact event and the pressure curve pattern; the attention mechanism weights key features (such as sudden pressure fluctuations or high-frequency impacts); and the Transformer layer performs weighted attention on the gas composition time series data (focusing on moments of concentration mutation).

[0045] Input splicing: merge the pressure sequence, shock event vector, BMS data matrix, and gas composition sequence into a unified input tensor;

[0046] Loss function design: Comprehensive safety risk classification loss (such as thermal runaway probability) and regression loss (such as SOH attenuation prediction) to characterize gas concentration prediction loss;

[0047] (4) Model verification and deployment optimization:

[0048] Verification methods include:

[0049] Verification using accelerated aging test data (e.g., 400-cycle charge-discharge pressure-capacity correlation test);

[0050] Separate independent test sets to evaluate false alarm and false negative rates (e.g., simulate safety valve rupture events);

[0051] Closed-loop optimization of landing applications, including:

[0052] Real-time risk assessment: deploy edge computing modules to dynamically output risk levels (low / medium / high);

[0053] Active intervention mechanism: triggers the BMS to execute current limiting, power off, or start the cooling system (e.g., emergency response when hydrogen concentration exceeds the threshold);

[0054] Continuous iteration: Collect new data through the cloud and regularly update model parameters (such as digital twin-driven SOH calibration).

[0055] The present invention also claims protection for a lithium battery safety monitoring system based on multi-sensor fusion, comprising:

[0056] A battery module, wherein the battery module adopts a sealed housing and has a built-in air pressure sensor, an inertial navigation sensor (IMU), a 4G / 5G communication module, and an edge computing microprocessor;

[0057] Intelligent monitoring cloud platform, used to receive sensor data uploaded by battery modules and perform AI analysis;

[0058] User terminal, used to display monitoring data and warning information;

[0059] The battery module is injected with inert gas and sealed during the production phase to form an initial air pressure reference value. The air pressure sensor and inertial navigation sensor work together to collect battery internal pressure and motion status data, which are uploaded to the cloud platform through the communication module to achieve safety status assessment.

[0060] The system can implement the above method.

[0061] Furthermore, the sealed housing of the battery module meets the IP67 protection standard, and the sealed housing has a built-in gas composition sensor, which is used to detect the concentration of characteristic gases including H2, CO, etc. inside the battery;

[0062] The initial pressure reference value of the battery module includes temperature-related parameters. During the production phase, the purity of the inert gas and the initial gas composition spectrum are recorded simultaneously as a comparison benchmark for subsequent sealing tests.

[0063] The edge computing microprocessor is configured with a pre-processing algorithm for local real-time analysis of sensor data, triggering immediate warnings when super-threshold events are detected, and selectively uploading key data to the cloud.

[0064] The present invention also claims protection for a lithium battery product based on multi-sensor fusion, which includes a battery cell, a BMS module, an air pressure sensor, an IMU sensor, a gas composition sensor, an edge computing chip, and an IP67-rated sealed housing, wherein the housing is made of a flame-retardant ABS+PC composite material;

[0065] This lithium battery product implements lithium battery safety monitoring based on the above method.

[0066] Compared with the prior art, the lithium battery safety monitoring method and system based on multi-sensor fusion of the present invention has the following beneficial effects:

[0067] This system utilizes a built-in barometer, inertial navigation sensor, and gas composition sensor, combined with edge-side preprocessing and cloud-based AI algorithms, to provide battery swelling warning, external impact monitoring, chemical anomaly identification, and dynamic safety assessment. This addresses the inadequacy of traditional BMS monitoring for physical deformation, sudden impact, and chemical anomalies. Scenario-adaptive thresholds, temperature compensation, and multi-parameter coupled analysis significantly improve warning accuracy and applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flowchart of a lithium battery safety monitoring method based on multi-sensor fusion provided by an embodiment of the present invention;

[0069] Figure 2 It is a diagram of the overall process framework provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention will be further described below with reference to specific embodiments.

[0071] The embodiment of the present invention provides a lithium battery safety monitoring method based on multi-sensor fusion, such as Figure 1 As shown, the following steps are included:

[0072] S1: The pressure sensor continuously monitors the internal pressure of the battery and performs temperature compensation correction based on the temperature sensor data. When the corrected pressure value exceeds the preset range, a sealing or bulging warning is triggered.

[0073] S2: The IMU sensor collects acceleration, angular velocity, vibration frequency, and posture data, and determines fall or collision events based on scene-adaptive thresholds.

[0074] S3: Sensor data is uploaded to the cloud at fixed intervals or event triggers, and analyzed by AI algorithms to generate security risk levels;

[0075] S4: The user terminal receives and displays the risk assessment results and warning information.

[0076] The temperature compensation correction in step S1 is implemented based on the ideal gas state equation PV=nRT, where P is the gas pressure in kPa, V is the gas volume, L and n are the amount of gas substance in mol, R is the ideal gas constant, which is 8.314 J / (mol·K), and T is the gas temperature in K. The correction formula is:

[0077] P0=P 测 ×T0 / T 测 ;

[0078] Among them, P0 is the corrected air pressure reference value, unit is kPa, P 测The pressure inside the battery is measured in real time by the pressure sensor in kPa. T0 is the absolute temperature corresponding to 25°C in K. T 测 It is the real-time absolute temperature in K.

[0079] Step S1 also includes gas composition correlation analysis: when the air pressure value is abnormal, if the gas composition sensor detects that the H2 concentration is greater than 50ppm or the CO concentration is greater than 20ppm, the warning level is automatically upgraded.

[0080] The AI ​​algorithm described in step S3 adopts a multimodal fusion network, including a GRU / LSTM layer to process BMS time series data, a CNN branch to analyze the spectral characteristics of the impact event, and a Transformer layer to perform attention weighting on the gas composition time series data. The input data includes air pressure sequence, impact event vector, BMS data matrix and gas composition sequence.

[0081] The method also includes an active intervention mechanism: when the AI ​​algorithm determines that the risk level is high, the edge microprocessor triggers the BMS to perform current limiting, power off or cooling system startup operations, and pushes emergency warning information to the user terminal.

[0082] The AI ​​algorithm is trained using accelerated aging experiment data, extreme environment test data, and data from a variety of typical failure scenarios to reduce false alarm rates and missed alarm rates.

[0083] The specific implementation process of this method is as follows:

[0084] S1: Battery sealing and swelling monitoring.

[0085] During the production phase, a fixed amount of gas is injected into the battery and sealed, and the internal pressure is continuously monitored using a manometer.

[0086] Added temperature compensation mechanism: Combined with the ambient temperature collected by the built-in temperature sensor, based on the ideal gas state equation (PV=nRT) (P is the gas pressure in kPa, V is the gas volume in L, n is the amount of gas in mol, R is the ideal gas constant of 8.314J / (mol·K), and T is the gas temperature in K), the pressure reference value is corrected in real time (correction formula: P0=P 测 ×T0 / T 测 , where P0 is the corrected air pressure reference value in kPa, P 测 The unit is kPa, which is the internal pressure value of the battery measured by the pressure sensor in real time. T0 is the absolute temperature unit K corresponding to 25℃. 测 is the real-time absolute temperature unit K).

[0087] If the air pressure exceeds the preset range (e.g., it drops by more than 5% or rises by more than 10%), it is judged as a seal failure or bulging risk, triggering an early warning.

[0088] At the same time, the gas composition sensor is linked. If the H2 concentration is detected to be greater than 50ppm or the CO concentration is greater than 20ppm, it is determined to be a risk of micro-short circuit of the battery cell and the warning level is upgraded.

[0089] S2: External force impact monitoring.

[0090] The IMU sensor collects battery acceleration, vibration frequency and posture data in real time.

[0091] If the instantaneous acceleration exceeds the threshold (such as >5g) or the abnormal vibration continues, it is determined to be a drop or collision event, and the impact intensity, timestamp, and attitude angle change are recorded.

[0092] S3: Data upload and AI analysis.

[0093] Sensor data is uploaded to the cloud at fixed intervals (e.g., every minute) or when triggered by an event (e.g., exceeding a threshold). This includes:

[0094] Air pressure / deformation data: Real-time monitoring of internal pressure fluctuations and shell deformation of the battery cell through embedded sensors;

[0095] Impact event data: records physical impact during transportation or use, and combines accelerometers to capture time, intensity, and direction information;

[0096] BMS timing data: collects core parameters such as voltage, current, temperature, SOC / SOH, etc.

[0097] Gas composition data: records the time-series changes in the concentration of characteristic gases such as H2 and CO, and generates the slope of the gas concentration-time curve (for example, an H2 concentration slope > 10ppm / min is considered abnormal).

[0098] Cloud-based AI algorithms (such as LSTM neural networks) analyze time series data, train security status models, and dynamically output risk levels (normal / warning / dangerous).

[0099] S3.1 Data cleaning and alignment:

[0100] (1) Eliminate abnormal values ​​(such as outliers caused by sensor failure) and fill in missing values ​​(such as linear interpolation).

[0101] (2) Unify timestamps to ensure time synchronization of multi-source data (such as alignment of charge and discharge cycles with shock events).

[0102] (3) Add data consistency check: If the air pressure rises suddenly by more than 5% but the gas composition is normal, it is judged as sensor drift and the air pressure sensor is automatically calibrated (calibration coefficient: K = initial air pressure / current air pressure average, effective after 3 consecutive calibrations).

[0103] S3.2 Multidimensional feature analysis:

[0104] Time domain and frequency domain feature extraction:

[0105] (1) Air pressure / deformation characteristics: Calculate the air pressure change rate, cumulative deformation, and pressure curve asymmetry.

[0106] (2) Impact characteristics: Extract impact peak value, duration, and energy spectrum density.

[0107] (3) BMS dynamic characteristics: Generate derived indicators such as voltage / current differential, internal resistance change, and temperature gradient.

[0108] (4) Gas characteristics: Calculate the H2 / CO concentration peak, concentration rise rate, and characteristic gas concentration ratio (e.g., if H2 / CO>2.5, it is determined to be electrolyte decomposition).

[0109] Associative feature construction:

[0110] (1) Cross-analysis of the correlation between sudden pressure rise and temperature anomalies (such as precursors to thermal runaway).

[0111] (2) Combine the voltage drop amplitude and recovery rate after the impact event (to evaluate the impact of mechanical damage on electrochemical performance).

[0112] (3) A three-dimensional correlation model of “pressure rise rate-characteristic gas concentration-temperature gradient” is constructed. When the three are synchronously abnormal (such as pressure rise rate > 0.2 kPa / min, H2 rise rate > 15 ppm / min, temperature gradient > 2 °C / min), it is determined to be a precursor to thermal runaway.

[0113] S3.3 Model training and algorithm selection:

[0114] (1) A multimodal fusion network is used: the GRU / LSTM layer processes the BMS time series data (capturing long-term dependencies). The CNN branch analyzes the spectral characteristics and pressure curve patterns of the impact event. The attention mechanism weights key features (such as sudden pressure fluctuations or high-frequency impacts). The Transformer layer performs attention weighting on the gas composition time series data (focusing on the moment of concentration mutation).

[0115] (2) Input splicing: Combine the pressure sequence, shock event vector, BMS data matrix, and gas composition sequence into a unified input tensor.

[0116] (3) Loss function design: Comprehensive safety risk classification loss (such as thermal runaway probability) and regression loss (such as SOH attenuation prediction) to characterize gas concentration prediction loss.

[0117] S3.4 Model Verification and Deployment Optimization:

[0118] Verification method:

[0119] (1) Verification using accelerated aging test data (e.g., 400-cycle charge-discharge pressure-capacity correlation test).

[0120] (2) Divide independent test sets to evaluate false alarm rates and missed alarm rates (e.g., simulate safety valve rupture events).

[0121] Closed-loop optimization of landing applications:

[0122] (1) Real-time risk assessment: deploy edge computing modules to dynamically output risk levels (low / medium / high risk).

[0123] (2) Active intervention mechanism: triggering the BMS to execute current limiting, power off or start the cooling system (such as emergency response when the hydrogen concentration exceeds the threshold).

[0124] (3) Continuous iteration: Collect new data through the cloud and regularly update model parameters (such as digital twin-driven SOH calibration).

[0125] S4: User terminal interaction.

[0126] Display air pressure, motion status, and AI risk assessment results through mobile apps or mini-programs;

[0127] Supports historical data backtracking, early warning push, and maintenance suggestion generation (such as "battery replacement recommendation").

[0128] This method uses a barometer, inertial navigation sensor, and gas composition sensor to coordinate and monitor all aspects of physical deformation, external force impact, and chemical anomalies (cell micro-short circuits, electrolyte decomposition), achieving multi-modal sensing fusion.

[0129] Utilize cloud-based incremental learning optimization algorithms to achieve dynamic AI modeling and improve adaptability to different battery aging stages;

[0130] This method adopts an event-driven mechanism: data upload is triggered by thresholds, combined with edge-side preprocessing to reduce communication energy consumption and extend battery monitoring life.

[0131] Scenario adaptation and fault location: Accurate early warning and fault cell location are achieved through scenario-based threshold adjustment and multi-cell parameter difference analysis;

[0132] Temperature compensation and gas correlation analysis resolve the interference of temperature on air pressure monitoring and improve the accuracy of fault identification through the coupling relationship between gas composition and air pressure.

[0133] A specific application example of this method is as follows:

[0134] 1. Sensor selection: MEMS barometer (accuracy ±0.1kPa), 6-axis IMU (including accelerometer + gyroscope), gas sensor (detection of H2: 0-1000ppm, CO: 0-500ppm, accuracy ±5%FS), temperature sensor (accuracy ±0.5℃).

[0135] 2. Data fusion: Integrate sensor data acquisition modules into the BMS and eliminate noise through extended Kalman filtering.

[0136] 3. AI algorithm example:

[0137] Input: air pressure time series data (including temperature correction), gas composition time series data, IMU impact events, BMS traditional parameters;

[0138] Output: SOS score (0-100%) based on the random forest-neural network fusion model (0-100%, score <30 is dangerous, 30-70 is warning, >70 is normal).

[0139] 4. Low power consumption strategy: The sensor sleep and wake-up mechanism activates high-frequency sampling only when an anomaly is detected, and the edge-side data compression algorithm (such as wavelet transform compression ratio of 5:1) reduces upload traffic.

[0140] The present invention also provides a lithium battery safety monitoring system based on multi-sensor fusion, which can implement the lithium battery safety monitoring method based on multi-sensor fusion described in the above embodiment. It includes:

[0141] A battery module, wherein the battery module adopts a sealed housing and has a built-in air pressure sensor, an inertial navigation sensor (IMU), a 4G / 5G communication module, and an edge computing microprocessor;

[0142] Intelligent monitoring cloud platform, used to receive sensor data uploaded by battery modules and perform AI analysis;

[0143] User terminal, used to display monitoring data and warning information;

[0144] Among them, the battery module is injected with inert gas and sealed during the production stage to form an initial air pressure reference value. The air pressure sensor and inertial navigation sensor work together to collect the internal pressure and motion status data of the battery, and upload it to the cloud platform through the communication module to realize safety status assessment.

[0145] The sealed housing of the battery module meets the IP67 protection standard, and the sealed housing has a built-in gas composition sensor, which is used to detect the concentration of characteristic gases including H2, CO, etc. inside the battery.

[0146] The initial air pressure reference value of the battery module includes temperature-related parameters. The purity of the inert gas and the initial gas composition spectrum are recorded synchronously during the production stage as a comparison benchmark for subsequent sealing testing.

[0147] The edge computing microprocessor is equipped with a pre-processing algorithm that can perform local real-time analysis of sensor data, trigger immediate warnings when super-threshold events are detected, and selectively upload key data to the cloud.

[0148] The specific implementation of this system includes:

[0149] Edge layer: The barometer, IMU sensor, gas composition sensor and low-power communication module are embedded in the battery body, and the chip with integrated edge computing function realizes local data preprocessing and rapid response.

[0150] Cloud layer: data storage module, AI analysis engine (training historical data + real-time inference), digital twin model library (including virtual mapping of different battery models).

[0151] Application layer: user terminal apps / applets, operation and maintenance management platforms, and third-party system interfaces (such as logistics management systems and equipment monitoring platforms).

[0152] An embodiment of the present invention also provides a lithium battery product based on multi-sensor fusion. The product includes a battery cell, a BMS module, an air pressure sensor, an IMU sensor, a gas composition sensor, an edge computing chip, and an IP67-rated sealed housing made of a flame-retardant ABS+PC composite material. This lithium battery product implements lithium battery safety monitoring based on the multi-sensor fusion lithium battery safety monitoring method described in the above embodiment.

[0153] The specific structure of the lithium battery product includes:

[0154] Battery module: including battery cells, BMS modules and corresponding accessories, using a sealed housing that meets IP67 standards. The housing is made of flame-retardant ABS+PC composite material (with 15% glass fiber added to enhance deformation resistance);

[0155] Built-in air pressure sensor to detect the gas pressure inside the battery;

[0156] Built-in inertial navigation sensor (IMU) for collecting acceleration, angular velocity and motion trajectory;

[0157] Built-in gas composition sensor for internal H2, CO and other characteristic gas concentrations;

[0158] When the battery is packaged, an inert gas (such as nitrogen) is injected and sealed to form an initial pressure reference value;

[0159] Built-in 4G module, can communicate with the intelligent lithium battery safety monitoring cloud platform;

[0160] Built-in microprocessor supports local real-time data preprocessing and rapid early warning response.

[0161] The above specific embodiments will allow those skilled in the art to easily implement the present invention. However, it should be understood that the present invention is not limited to the above specific embodiments. Based on the disclosed embodiments, those skilled in the art can arbitrarily combine different technical features to implement different technical solutions.

[0162] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A lithium battery safety monitoring method based on multi-sensor fusion, characterized in that: The following steps are involved: S1: The pressure sensor continuously monitors the internal pressure of the battery and performs temperature compensation correction based on the temperature sensor data. When the corrected pressure value exceeds the preset range, a sealing or bulging warning is triggered. S2: The IMU sensor collects acceleration, angular velocity, vibration frequency, and posture data, and determines fall or collision events based on scene-adaptive thresholds. S3: Sensor data is uploaded to the cloud at fixed intervals or event triggers, and analyzed by AI algorithms to generate security risk levels; S4: The user terminal receives and displays the risk assessment results and warning information; The AI ​​algorithm described in step S3 uses a multimodal fusion network, including a GRU / LSTM layer to process BMS time series data, a CNN branch to analyze the spectral characteristics of the impact event, and a Transformer layer to perform attention weighting on the gas composition time series data. The input data includes the pressure sequence, the impact event vector, the BMS data matrix, and the gas composition sequence. The AI ​​algorithm is trained using accelerated aging experiment data, extreme environment test data, and data from a variety of typical failure scenarios to reduce false alarm rates and missed alarm rates.

2. The lithium battery safety monitoring method based on multi-sensor fusion according to claim 1, characterized in that: The temperature compensation correction in step S1 is implemented based on the ideal gas state equation PV=nRT, where P is the gas pressure, V is the gas volume, n is the amount of gas substance, R is the ideal gas constant, and T is the gas temperature; the correction formula is: P0=P 测 ×T0 / T 测 ; Among them, P0 is the corrected air pressure reference value, P 测 is the internal pressure value of the battery measured in real time by the pressure sensor, T0 is the absolute temperature corresponding to 25℃, T 测 The real-time absolute temperature.

3. The lithium battery safety monitoring method based on multi-sensor fusion according to claim 1 or 2, characterized in that: If the air pressure value exceeds the preset range, it is judged as a seal failure or bulging risk, triggering an early warning; At the same time, the gas composition sensor is linked. If the H2 concentration is detected to be greater than 50ppm or the CO concentration is greater than 20ppm, it is determined to be a risk of micro-short circuit of the battery cell and the warning level is upgraded.

4. The lithium battery safety monitoring method based on multi-sensor fusion according to claim 1, characterized in that: The method also includes an active intervention mechanism: when the AI ​​algorithm determines that the risk level is high, the edge microprocessor triggers the BMS to perform current limiting, power off or cooling system startup operations, and pushes emergency warning information to the user terminal.

5. The lithium battery safety monitoring method based on multi-sensor fusion according to claim 1, characterized in that: The step S3 is specifically implemented as follows: S3.1: Sensor data is uploaded to the cloud at regular intervals or event triggers; the data includes: Air pressure / deformation data: obtained by real-time monitoring of air pressure fluctuations inside the battery cell and shell deformation through embedded sensors; Impact event data: records physical impact during transportation or use, and combines accelerometers to capture time, intensity, and direction information; BMS timing data: collected core parameters including voltage, current, temperature, SOC / SOH; Gas composition data: records the time series changes in the concentration of characteristic gases including H2 and CO, and generates the slope of the gas concentration-time curve; S3.2: Analyze time series data using cloud-based AI algorithms, train security status models, and dynamically output risk levels. The process is as follows: (1) Data cleaning and alignment: Eliminate outliers and fill in missing values; Unified timestamps to ensure time synchronization of multi-source data; Added data consistency check: If the air pressure suddenly rises by more than 5% but the gas composition is normal, it is determined to be sensor drift and the air pressure sensor is automatically calibrated; (2) Multidimensional feature analysis: (2.1) Time domain and frequency domain feature extraction: Air pressure / deformation characteristics: calculate the air pressure change rate, cumulative deformation, and pressure curve asymmetry; Impact characteristics: extract impact peak value, duration, and energy spectrum density; BMS dynamic characteristics: Generate derived indicators including voltage / current differential, internal resistance change, and temperature gradient; Gas characteristics: calculate H2 / CO concentration peak, concentration rising rate, and characteristic gas concentration ratio; (2.2) Construction of correlation features: Cross-analyze the correlation between sudden pressure rise and temperature anomaly; Combined with the voltage drop amplitude and recovery rate after the impact event; A three-dimensional correlation model of "pressure rise rate-characteristic gas concentration-temperature gradient" is constructed. When the three are synchronously abnormal, it is determined to be a precursor to thermal runaway. (3) Model training and algorithm selection: A multimodal fusion network is used: the GRU / LSTM layer processes BMS time series data; the CNN branch analyzes the spectral characteristics and pressure curve patterns of the impact event; the attention mechanism weights key features; and the Transformer layer performs attention weighting on the gas composition time series data. Input splicing: merge the pressure sequence, shock event vector, BMS data matrix, and gas composition sequence into a unified input tensor; Loss function design: Comprehensive safety risk classification loss and regression loss function to predict gas concentration loss; (4) Model verification and deployment optimization: Verification methods include: Verified using accelerated aging test data; Divide independent test sets to evaluate false alarm rate and missed alarm rate; Closed-loop optimization of landing applications, including: Real-time risk assessment: deploy edge computing modules to dynamically output risk levels; Active intervention mechanism: triggers the BMS to execute current limiting, power off or start the cooling system; Continuous iteration: Collect new data through the cloud and update model parameters regularly.

6. The lithium battery safety monitoring system based on multi-sensor fusion is characterized by: include: A battery module, which uses a sealed housing and has a built-in air pressure sensor, inertial navigation sensor, 4G / 5G communication module, and edge computing microprocessor; Intelligent monitoring cloud platform, used to receive sensor data uploaded by battery modules and perform AI analysis; User terminal, used to display monitoring data and warning information; The battery module is injected with inert gas and sealed during the production phase to form an initial air pressure reference value. The air pressure sensor and inertial navigation sensor work together to collect battery internal pressure and motion status data, which are uploaded to the cloud platform through the communication module to achieve safety status assessment. The system can implement the method described in any one of claims 1 to 5.

7. The lithium battery safety monitoring system based on multi-sensor fusion according to claim 6, characterized in that: The sealed housing of the battery module meets the IP67 protection standard and has a built-in gas composition sensor for detecting the characteristic gas concentrations inside the battery, including H2 and CO. The initial pressure reference value of the battery module includes temperature-related parameters. During the production phase, the purity of the inert gas and the initial gas composition spectrum are recorded simultaneously as a comparison benchmark for subsequent sealing tests. The edge computing microprocessor is configured with a pre-processing algorithm for local real-time analysis of sensor data, triggering immediate warnings when super-threshold events are detected, and selectively uploading key data to the cloud.

8. Lithium battery products based on multi-sensor fusion, characterized by: The product includes battery cells, BMS modules, air pressure sensors, IMU sensors, gas composition sensors, edge computing chips and IP67-rated sealed housings made of flame-retardant ABS+PC composite materials. The lithium battery product realizes lithium battery safety monitoring based on the method described in any one of claims 1 to 5.

Images