Lithium battery thermal runaway monitoring system and method based on multi-sensor fusion and artificial intelligence analysis

Abstract

The invention discloses a lithium battery thermal runaway monitoring system and method based on multi-sensor fusion and artificial intelligence, which integrates a gas sensitive sensor, a strain gauge, a thermocouple, an acoustic microphone and an electrical parameter module, and realizes early warning through multi-modal data fusion and a deep residual neural network. The system adopts a gas sensitive array to detect gases such as hydrogen and the like, and constructs a space-time incidence matrix in combination with voiceprint, deformation, temperature and electrical parameters; the AI model adopts a ResNet-LSTM architecture, dynamically distributes feature weights through an attention mechanism, outputs risk indexes and triggers three-level responses (ventilation / alarm / power failure, response < = 50ms). And the false alarm revision module can forcibly reset the system, and the AI model is automatically revised. The false alarm rate is reduced by 60%, the hydrogen detection sensitivity reaches 1 ppm, the GB38031-2025 requirement is met, the method is suitable for electric automobiles and energy storage power stations, and safety and timeliness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery safety monitoring, and specifically to a lithium battery thermal runaway monitoring system and method based on multi-sensor fusion and artificial intelligence analysis. The system and method are particularly suitable for real-time safety monitoring and active protection of electric vehicle power battery packs and energy storage power station battery cabinets. Background Art

[0002] Lithium batteries are widely used in the new energy sector due to their high energy density, but the risk of thermal runaway remains a core safety issue. Thermal runaway is typically triggered by internal short circuits, overcharge, overdischarge, or mechanical damage, and is accompanied by the release of gases (such as H2, CO, and alkanes), sudden temperature rise, increased deformation, and abnormal current and voltage.

[0003] The existing technology has the following key defects:

[0004] a. Single parameter reliance: Traditional methods rely primarily on temperature or voltage monitoring. However, temperature sensors have hysteresis (for example, external thermocouples require waiting for heat conduction), and voltage does not change significantly in the early stages of thermal runaway, leading to false alarms or missed alarms.

[0005] b. Limitations of gas detection: Existing gas sensors have problems such as cross-interference and short lifespan. They are not sensitive enough to detect key gases such as hydrogen, making it difficult to meet the new national standard's mandatory requirements for "no fire and no explosion."

[0006] c. Insufficient data fusion: Existing multi-sensor systems mostly use simple threshold judgments or probabilistic models, lacking in-depth exploration of the correlation between temporal features and multimodal features. For example, although some solutions integrate temperature and voltage parameters, they do not incorporate acoustic and deformation data.

[0007] d. Single response mechanism: Most systems only trigger a unified alarm, lack a hierarchical response strategy, and are unable to take differentiated measures (such as ventilation and power outage coordinated control) according to the stage of thermal runaway.

[0008] e. Cost and deployment challenges: Existing internal sensing technologies rely on complex packaging processes, which are costly and difficult to adapt to different battery specifications. Summary of the Invention

[0009] The purpose of the present invention is to provide a lithium battery thermal runaway monitoring system and method with multi-dimensional perception and multi-modal data fusion. By synchronously collecting multi-source information such as gas, deformation, temperature, acoustics, and electrical parameters, combined with a deep residual neural network model with dynamic weight distribution, early warning and three-level response (ventilation, alarm, power outage) can be achieved, significantly improving monitoring accuracy and response timeliness while reducing deployment costs.

[0010] Technical Solution

[0011] 1. System composition

[0012] (1) Gas sensor module (201): It adopts a cross-sensitive gas sensor array, which includes multiple sets of independently calibrated sensors, and sets differentiated detection thresholds for alkanes, hydrogen, carbon monoxide, ammonia and electrolyte decomposition products. The detection range covers 0-5000ppm, and the resolution reaches 1ppm.

[0013] (2) Deformation and temperature monitoring module (202): Full-bridge strain gauges are attached to the surface with the maximum curvature of the battery in an orthogonal symmetrical layout (sampling frequency ≥ 100 Hz).

[0014] (3) Distributed temperature detection module (203): cooperates with distributed thermocouples (three-dimensional grid topology, spacing ≤ 1.2 times the thickness of the battery cell) to capture the deformation rate and temperature gradient in real time.

[0015] (4) Acoustic feature acquisition module (204): Configure a bandpass filter (2-8kHz) and wavelet packet decomposition algorithm to identify the transient soundprint characteristics of gas release (such as the boiling sound frequency of electrolyte), and combine it with a high-sensitivity microphone to achieve non-contact monitoring.

[0016] (5) Electrical parameter monitoring module (205): configure the current and voltage monitoring module to dynamically monitor the input and output electrical parameters of the lithium battery.

[0017] (6) Data processing and analysis module (101): It uses a micro embedded computer to communicate with the data acquisition and control module (102) through serial communication. It has an artificial intelligence monitoring model embedded inside and communicates with the lower computer data acquisition and control module in real time to perform monitoring and early warning.

[0018] (7) Data acquisition and control module (102): A single chip microcomputer is used to perform closed-loop control on the above-mentioned gas sensing, deformation and temperature sensing, acoustic feature acquisition, current and voltage monitoring, and active ventilation modules.

[0019] (8). Early warning execution module (3): Three early warning modes are used in different early warning stages, namely low-speed ventilation, sound and light alarm, prompting manual intervention and emergency power off. Active ventilation control uses a two-speed centrifugal fan driven by a relay, and dynamically adjusts the air volume according to the risk index to dilute the concentration of combustible gas.

[0020] (9). False alarm revision module (4): Using keyboard input interactive model, after checking and confirming that there are no errors, the monitoring system can be forced to enter the initial mode and the artificial intelligence model algorithm can be revised.

[0021] 2. Artificial Intelligence Model

[0022] (1) Using a deep residual network (ResNet) architecture, the input layer integrates a temporal coding unit, the hidden layer cascades the LSTM module (to extract temporal dependencies) and the attention mechanism layer (to dynamically focus on key features), and the output layer calculates the risk index (0-1 range) through the Softmax function.

[0023] (2) The training dataset covers multimodal data from laboratory accelerated aging experiments (overcharging, puncture, and extrusion), and the weights of each sensor are optimized through feature importance evaluation. For example, in mechanical abuse scenarios, voiceprint and deformation features are prioritized.

[0024] 3. Graded response mechanism

[0025] (1) Primary warning (risk index 0.4-0.6): Start low-speed ventilation to reduce the concentration of combustible gas;

[0026] (2) Intermediate warning (0.6-0.8): triggers an audible and visual alarm (sound pressure level ≥ 90dB) and prompts manual intervention;

[0027] (3) Advanced warning (≥0.8): Execute emergency power off (response time ≤50ms). BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 : System architecture diagram (including sensor composition and data flow);

[0029] Figure 2 :System workflow diagram;

[0030] Figure 3 : Flowchart of the three-level response mechanism. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] Example 1: Electric vehicle battery pack monitoring

[0033] Deployment steps:

[0034] 1. Sensor installation:

[0035] Gas sensor array: Installed at the entrance of the battery module exhaust channel, with a sensor spacing of 10mm and a 45° angle to the airflow direction, it detects the concentration of combustible gases that may be released from the electrolyte, such as H2, CO and alkanes.

[0036] Acoustic module: An explosion-proof microphone is installed on the side wall of the module, with its axis aligned with the cell spacing area to capture the electrolyte boiling soundprint.

[0037] Strain gauges and thermocouples: Strain gauges are affixed to the easily deformed areas of the casing in a "cross" layout; thermocouples are attached to the outer wall of the battery pack to monitor heat diffusion trends.

[0038] Current and voltage monitoring module: connected in series with the input and output terminals of the battery pack to detect the current and voltage status.

[0039] 2. Data processing and early warning:

[0040] Data fusion: The single-chip microcomputer (STM32F407) synchronously collects gas concentration, soundprint energy, deformation rate, temperature gradient and electrical parameters to construct a spatiotemporal correlation matrix.

[0041] Model inference: An embedded computer (NVIDIA Jetson Nano) runs the ResNet-LSTM model. When the H2 concentration exceeds 100 ppm and the soundprint energy suddenly increases by 3 times, it is determined to be in the early stages of thermal runaway and triggers an intermediate warning.

[0042] 3. Response Verification:

[0043] Overcharge test: When the voltage rises to 4.5V, the CO concentration rises to 300ppm within 120 seconds, the temperature gradient reaches 8°C / min, the model outputs a risk index of 0.7 within 30 seconds, starts low-speed ventilation, and triggers an audible and visual alarm.

[0044] Example 2: Monitoring battery cabinets in energy storage power stations

[0045] Deployment steps:

[0046] 1. Sensor configuration:

[0047] Gas sensor array: Multiple groups of gas sensor arrays with the same function are set up and installed on the top, side and between battery packs of the battery cabinet to detect the concentration of combustible gases that may be released from the electrolyte, such as H2, CO and alkanes.

[0048] Acoustic module: Explosion-proof microphones are installed on the side walls of the energy storage power station to capture the soundprint of electrolyte boiling.

[0049] Strain gauges and thermocouples: Strain gauges are affixed to the easily deformed areas of the shell in a "cross" layout; thermocouples are arranged on the surface of the lithium battery to monitor the heat diffusion trend.

[0050] Current and voltage monitoring module: connected in series with the input and output terminals of the battery pack to monitor the current and voltage status of the energy storage power station in real time.

[0051] 2. Feature extraction and response:

[0052] Data fusion: The single-chip microcomputer (STM32F407) synchronously collects gas concentration, soundprint energy, deformation rate, temperature gradient and electrical parameters to construct a spatiotemporal correlation matrix.

[0053] Acoustic signal processing: Acoustic signals are collected in real time. After bandpass filtering, the energy in the 3.5-4.5kHz frequency band (corresponding to the electrolyte boiling soundprint) is extracted through Morlet wavelet transform. When the energy exceeds the baseline by three times for five consecutive cycles, it is marked as an abnormal event.

[0054] Electrical parameter analysis: Calculates the voltage change rate (dV / dt) and triggers the fast diagnosis mode when ΔV ≥ 10% / s.

[0055] Model inference: An embedded computer (NVIDIA Jetson Nano) runs the ResNet-LSTM model. When the H2 concentration exceeds 100 ppm and the soundprint energy suddenly increases by 3 times, it is determined to be in the early stages of thermal runaway and triggers an intermediate warning.

[0056] Linkage control: When the H2 and CO concentrations are >200ppm and the voltage sag is ≥10% / s, the model outputs a risk index of 0.92 within 2 seconds, triggering an advanced warning and the relay cutting off the main power supply (response time ≤30ms).

[0057] Beneficial effects

[0058] (1) Multi-parameter collaborative monitoring: integrating gas, acoustic, deformation, temperature, and electrical parameters, the false alarm rate is reduced by 60% compared with a single sensor system.

[0059] (2) Intelligent dynamic decision-making: The attention mechanism dynamically assigns feature weights, and the model accuracy reaches 98.5%, supporting adaptation to multiple scenarios such as overcharging and mechanical damage.

[0060] (3) Cost and real-time advantages: Modular design adapts to different battery specifications, and edge computing enables localized processing (response time ≤ 50ms), avoiding the complex process of implantable sensors.

[0061] (4) Comply with the new national standard: The hydrogen detection sensitivity reaches 1ppm, and the active ventilation system reduces the H2 concentration to below the lower explosion limit within 10 seconds, meeting the mandatory standard of GB38031-2025.

Claims

1. A lithium battery thermal runaway monitoring system and method based on multi-sensor fusion and artificial intelligence analysis, characterized in that: include: a data analysis control module (1), comprising a data processing and analysis module (101) and a data acquisition control module (102), wherein the data analysis module (101) is an embedded computer, and runs an artificial intelligence monitoring model, which is a pre-trained neural network model for multi-modal data fusion analysis; the data acquisition control module (102) uses a single-chip microcomputer to synchronously collect data from various sensors through a multi-channel interface, executes the functional actions of the early warning execution module (3), and receives information input from the error correction module (4); the data analysis module (101) and the data acquisition control module (102) realize information exchange through serial communication; b. A gas sensor module (201) for detecting the concentration of alkanes, combustible gases, hydrogen, ammonia and electrolyte decomposition products released by lithium batteries; c. a resistance strain gauge module (202), attached to the surface of the lithium battery, for real-time monitoring of the deformation of the lithium battery; d. a distributed temperature monitoring module (203), comprising a plurality of thermocouple temperature sensors attached to the surfaces of the lithium battery; e. an acoustic feature acquisition module (204), comprising a microphone sensor and a frequency analysis unit for capturing gas release acoustic wave signals and analyzing their spectral characteristics; f. Electrical parameter monitoring module (205) for real-time acquisition of lithium battery input and output current and voltage data; g. Early warning execution module (3), including a buzzer alarm, a power supply system circuit breaker and a relay-driven forced exhaust device for guiding the release of gas into the detection area; h. False alarm correction module (4), including external command input function, guiding the monitoring system to enter the initial mode and artificial intelligence model algorithm correction.

2. The system according to claim 1, wherein The gas sensor module (201) adopts a cross-sensitive gas sensing array, comprising at least five groups of independently calibrated gas sensors, and setting differential detection thresholds for C1-C4 alkanes, CO / H2 mixed gas, NH3 and characteristic decomposition products of electrolyte respectively.

3. The system according to claim 1, wherein: The resistive strain gauge module (202) adopts a full-bridge strain measurement circuit, the strain gauges are arranged in an orthogonal symmetric manner on the surface of the lithium battery with the maximum curvature, and the sampling frequency is not less than 100 Hz.

4. The system according to claim 1, wherein: The distributed temperature monitoring module (203) comprises no less than 6 K-type thin film thermocouples arranged on the surface of the lithium battery in a three-dimensional grid topology structure, with the distance between adjacent measuring points not exceeding 1.2 times the thickness of the battery cell.

5. The system according to claim 1, wherein: The acoustic feature acquisition module (204) is configured with a bandpass filter bank, the passband range of which is set to 2kHz-8kHz, and includes a transient voiceprint feature extraction algorithm based on wavelet packet decomposition.

6. The system according to claim 1, wherein: The electrical parameter monitoring module (205) has the capability of monitoring the current and voltage at the input and output terminals of the lithium battery. When the voltage / current fluctuates, the artificial intelligence monitoring model makes a comprehensive judgment and provides a corresponding processing strategy.

7. The system according to claim 1, wherein: The early warning execution module (3) has a step-by-step response early warning mechanism, wherein the three early warning stages respectively adopt three early warning modes: low-speed ventilation, sound and light alarm, prompting manual intervention, and emergency power off; and an active ventilation dual-mode driving strategy: when the detection value of any sensor exceeds a first threshold, low-speed ventilation is started, and when it exceeds a second threshold, the high-speed ventilation mode is switched. The second threshold is at least 30% higher than the first threshold.

8. The system according to claim 1, wherein: The false alarm revision module (4) uses an external keyboard to revise the monitoring results, and can also guide the monitoring system to enter the initial mode according to the instructions, and revise and improve the artificial intelligence model algorithm.

9. The system according to claim 1, wherein: The artificial intelligence monitoring model adopts a deep residual neural network architecture, the input layer includes a temporal feature encoding unit, and the network hidden layer includes a cascade structure of a long short-term memory module and an attention mechanism layer.

10. A method for monitoring thermal runaway of a lithium battery based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Establish a multi-physics field coupling feature database and simultaneously collect sensor data under different failure modes through accelerated aging experiments; S2. Construct a spatiotemporal correlation feature matrix and normalize and fuse the gas concentration, deformation, temperature gradient, voiceprint spectrum, and electrical parameter change rate; S3. Train a neural network model with dynamic weight allocation and optimize the contribution weight of each sensor data through feature importance evaluation; S4. Deploy an online monitoring process. When the risk index output by the model exceeds the preset safety boundary, execute three levels of response in sequence: the primary warning initiates active ventilation, the intermediate warning triggers the sound and light alarm, and the high-level warning executes the emergency shutdown of the power supply system.

11. A lithium battery thermal runaway monitoring system and method based on multi-sensor fusion and artificial intelligence analysis, comprising the lithium battery thermal runaway monitoring system and method according to any one of claims 1 to 10.

Images