Power battery production safety monitoring method and system, electronic device, and storage medium

Abstract

The application provides a power battery production safety monitoring method and system, electronic equipment and storage medium, relates to the power battery safety monitoring technical field, and collects multiple source parameters of multiple monitoring points in the power battery production environment; based on the spatial topology relationship of the sensor network, the spatial correlation characteristics between the monitoring points are extracted, and the trend characteristics of the time evolution of the multiple source parameters are extracted; input into the risk identification model, identify the risk mode in the power battery production process; according to the risk mode and the corresponding confidence, corresponding graded early warning information is generated; the isolated single point data is improved to structured information with spatial correlation and time dimension, so that the system can perceive the diffusion path of the risk parameter in space and the evolution law in time, thereby accurately positioning the risk source and predicting the development trend, and accurately identifying in the budding stage of risk occurrence, reducing false positives caused by environmental interference or single sensor fluctuation.

Description

Technical Field

[0001] This invention relates to the field of power battery safety monitoring technology, and in particular to a method, system, electronic device and storage medium for power battery production safety monitoring. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the production scale of power batteries has expanded dramatically. The production lines are highly automated and the processes are complex. They use a large amount of flammable and explosive organic electrolytes and active materials, which means that there are multiple safety risks in the production process, such as thermal runaway, flammable gas explosion, toxic gas leakage and dust explosion. Currently, production lines mainly rely on traditional point-type smoke / temperature detectors, catalytic combustion type or semiconductor type combustible gas detectors for safety monitoring. However, traditional point-type detectors have a delayed response, usually only alarming after an open flame or high temperature has been generated, failing to provide early warning. While conventional combustible gas detectors can monitor total volatile organic compound (TVOCC) concentration, they generally suffer from the fatal flaw of sulfide poisoning. Sulfur in the production environment may originate from raw material impurities, sulfur oxides in the environment, or byproducts of electrolyte decomposition. When the catalytic element of the sensor comes into contact with sulfides, irreversible chemical adsorption or reaction occurs, leading to catalyst deactivation, permanent decrease in sensitivity, or even complete failure, creating a monitoring blind spot. In addition, existing monitoring solutions are mostly discrete single-parameter systems, with each sensor working independently, forming data silos. They lack the ability to comprehensively analyze the coupling and correlation of multiple parameters, resulting in a high false alarm rate. Furthermore, they rely on periodic manual inspections, have poor real-time performance, and cannot meet the safety management needs of large-scale continuous production. Summary of the Invention

[0003] This invention provides a method, system, electronic device, and storage medium for safety monitoring in power battery production, which addresses the shortcomings of existing power battery production line safety monitoring technologies, such as the susceptibility of traditional gas sensors to failure due to sulfur poisoning, the difficulty in achieving early warning and high false alarm rate due to single-parameter monitoring, and the poor real-time performance due to reliance on manual inspection.

[0004] This invention provides a method for monitoring the safety of power battery production, comprising: Multi-source parameters from multiple monitoring points in the power battery production environment are collected, and the multi-source parameters are mapped to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of the sensor network. The physical rule model is used to extract spatial correlation features with physical propagation logic between each monitoring point; Based on the spatial topology of the sensor network, the trend characteristics of the evolution of the multi-source parameters over time are extracted; The spatial correlation features and the trend features are input into a pre-trained risk identification model, which outputs the type of the risk pattern and the corresponding confidence level. Based on the risk pattern and the corresponding confidence level, generate corresponding graded early warning information; The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points. The fused features are then input into the deep neural network classifier to identify risk patterns.

[0005] According to the power battery production safety monitoring method provided by the present invention, the step of extracting spatial correlation features with physical propagation logic between monitoring points using the physical rule model includes: Each sensor is used as a node in a graph neural network, and a dynamic graph structure is constructed based on the physical positional relationship of each sensor in the digital twin and the airflow propagation direction in the production environment. By learning the information transmission patterns between nodes in the dynamic graph structure through graph neural networks, the diffusion paths and propagation patterns of risk parameters in space can be captured.

[0006] According to the power battery production safety monitoring method provided by the present invention, the step of extracting the trend features of the multi-source parameters over time based on the spatial topology relationship of the sensor network includes: The temporal convolutional network module is used to extract local fluctuation features of the data streams from each sensor. The long short-term memory network module is used to extract the long-term dependency features of each sensor data stream; The local fluctuation features are fused with the long-term dependency features to form a complete temporal feature representation.

[0007] According to the power battery production safety monitoring method provided by the present invention, the step of inputting the spatial correlation features and the trend features into a pre-trained risk identification model and outputting the type of the risk pattern and the corresponding confidence level includes: The spatial correlation features and the trend features are mapped into feature vectors of different monitoring parameter types and feature vectors of different monitoring points; Align and concatenate the feature vectors of different monitoring parameter types and feature vectors of different monitoring points; The weights of different parameter types and monitoring points in the current fusion analysis are dynamically allocated through the multi-attention mechanism module. The weighted and fused high-level features are input into the deep neural network classifier, which outputs the probability distribution of risk patterns.

[0008] According to the power battery production safety monitoring method provided by the present invention, the digital twin further includes a physical rule model embedded in the physical environment model, the physical rule model including at least a simplified computational fluid dynamics model of gas diffusion and a heat conduction model; the digital twin is configured to map the equipment layout, sensor positions and real-time monitoring data in the physical world in real time, as well as visualize and dynamically display the risk patterns, risk locations and graded early warning information.

[0009] The present invention also provides a power battery production safety monitoring system, comprising: The acquisition module is used to acquire multi-source parameters from multiple monitoring points in the power battery production environment and map the multi-source parameters to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of the sensor network. The extraction module is used to extract spatial correlation features with physical propagation logic between each monitoring point using the physical rule model, and to extract the trend features of the evolution of the multi-source parameters over time based on the spatial topology of the sensor network. The identification module is used to input the spatial correlation features and the trend features into a pre-trained risk identification model and output the type of the risk pattern and the corresponding confidence level. The generation module is used to generate corresponding graded early warning information based on the risk pattern and the corresponding confidence level. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, and a multi-attention fusion module. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points, and identifies risk patterns based on the fusion results.

[0010] According to the power battery production safety monitoring system provided by the present invention, the acquisition module includes an anti-sulfur poisoning gas sensor array, which includes multiple gas sensor chips. The sensitive material of each gas sensor chip has a core-shell structure. The core of the core-shell structure is a metal oxide semiconductor material, and the outer shell is a porous sulfur-repellent molecular sieve material. The outer shell is used to selectively block or adsorb sulfide molecules.

[0011] According to the power battery production safety monitoring system provided by the present invention, the anti-sulfur poisoning gas sensor array further includes a self-cleaning unit. The self-cleaning unit is used to perform short-term high-temperature pulse heating on the gas sensor chip when a decrease in sensor sensitivity is detected, so as to oxidize and desorb the attached or adsorbed sulfides.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the power battery production safety monitoring method as described in any of the preceding claims.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the power battery production safety monitoring method described in any of the above claims.

[0014] The present invention provides a method, system, electronic device, and storage medium for monitoring the safety of power battery production. It collects multi-source parameters from multiple monitoring points in the power battery production environment and maps these parameters in real time to a digital twin. The digital twin includes a physical environment model and the spatial topology of a sensor network. The physical rule model is used to extract spatial correlation features with physical propagation logic between monitoring points. Based on the spatial topology of the sensor network, the trend features of the multi-source parameters over time are extracted. The spatial correlation features and the trend features are input into a pre-trained risk identification model, which outputs the type of risk pattern and its corresponding confidence level. Based on the risk pattern and its corresponding confidence level, corresponding graded early warning information is generated. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. Through a dual attention mechanism, the weights of different parameter types and different monitoring points are dynamically allocated during fusion to achieve dynamic weighted fusion of multiple parameters and multiple spatial points. The fused features are then input into a deep neural network classifier to identify risk patterns. This invention elevates the originally isolated single-point data into structured information with spatial correlation and temporal dimensions, enabling the system to perceive the spatial diffusion path and temporal evolution of risk parameters, thereby accurately locating the source of risk and predicting its development trend. Furthermore, it can accurately identify risks at the nascent stage, reducing false alarms caused by environmental interference or single sensor fluctuations. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the power battery production safety monitoring method provided in the embodiments of the present invention; Figure 2 This is a functional structure diagram of the power battery production safety monitoring system provided in an embodiment of the present invention; Figure 3 This is a functional structure diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] Figure 1 A flowchart of the power battery production safety monitoring method provided in the embodiments of the present invention is shown below. Figure 1 As shown, the power battery production safety monitoring method provided in this embodiment of the invention includes: Step 101: Collect multi-source parameters from multiple monitoring points in the power battery production environment, and map the multi-source parameters to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology relationship of the sensor network. In this embodiment of the invention, a multi-source parameter coverage "gas-heat-dust-image" multimodal full-domain sensing network is constructed. It not only monitors H2S gas but also simultaneously integrates high-density sensors for temperature, humidity, air pressure, dust concentration, and electrostatic potential, supplemented by infrared thermal imaging video streams from key workstations (such as the liquid injection machine and formation cabinet). This constructs a three-dimensional sensing network, achieving full-domain coverage from "point" to "surface" to "volume".

[0019] In this embodiment of the invention, the multi-source parameters of multiple monitoring points are filtered and denoised; baseline drift correction is performed using a reference sensor installed in a clean air zone; and for the sulfur poisoning gas sensor, its performance degradation coefficient is estimated online and data compensation is performed based on the response change before and after its "self-cleaning pulse".

[0020] Step 102: Extract spatial correlation features with physical propagation logic between each monitoring point using the physical rule model, and extract the trend features of the multi-source parameters over time based on the spatial topology of the sensor network; Step 103: Input the spatial correlation features and the trend features into the pre-trained risk identification model, and output the type of the risk pattern and the corresponding confidence level; Step 104: Generate corresponding graded early warning information based on the risk pattern and the corresponding confidence level; The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points. The fused features are then input into the deep neural network classifier to identify risk patterns.

[0021] In this embodiment of the invention, the deep neural network classifier outputs the risk pattern probability Pr, and at the same time, it combines the absolute level and the rate of increase of the parameters to calculate a comprehensive risk index Rindex (0-100).

[0022] The methods for obtaining the comprehensive risk index Rindex include: (1) Obtain the current time window from the fusion analysis module Read data in real time from various sensors And calculate ; each Cut off at Within the range; each Cut off at Within the range.

[0023] (2) Calculate the absolute level terms separately. sum rate term .

[0024] (3) According to Weighted summation .

[0025] (4) Calculate Rindex=(1-e-λQ)×100, then limit it to between 0 and 100, and round it to the nearest integer.

[0026] When the risk probability is high, the absolute value of the parameter is high, and it rises rapidly. As the value increases, Rindex approaches 100 exponentially, achieving non-linear enhanced early warning; it approaches 0 for low-risk situations. It should be noted that the thresholds and weights involved in the formula can be experimentally calibrated and adaptively adjusted according to the specific production line process.

[0027] The intermediate variable Q is defined as follows: The probability of the current dominant risk pattern output by the deep neural network classifier, with a value range of... 0 indicates no risk, and 1 indicates that the pattern is certain to occur; For the first Current measured values ​​of several key environmental parameters, such as: TVOC concentration (ppm), temperature (°C), and dust concentration (mg / m³). 3 ), electrostatic potential (kV), etc.; For the first The safety alarm thresholds for each parameter are set according to national standards or process specifications, such as 10% LEL for TVOC, 60℃ for temperature, etc. For the first The normalized risk level of each parameter, when the ratio is ≥1, indicates that the threshold has been exceeded, and the upper limit of the contribution of this item is truncated by 1.5; For the first The first derivative of each key parameter with respect to time, i.e. the current rate of ascent, is used to calculate the linear regression slope over the most recent few seconds (e.g., 10 seconds), with the same unit per parameter per second. For the first The rate alarm threshold for each parameter is set based on historical statistics or safety requirements, such as a temperature rate > 0.5℃ / s being considered dangerous. For the first The weights of the absolute level parameters, =1, determined by the analytic hierarchy process or experiments, such as temperature 0.3, TVOC 0.4, dust 0.2, static electricity 0.1; For the first The weights of each rate parameter, =1, usually different from the horizontal weight, highlights the trend of change; These are the fusion weight coefficients for information sources. Recommended value: (Probability-driven) , λ is the exponential compression coefficient, controlling... The mapping sensitivity to Rindex decreases as λ increases. The faster the growth, the better. Typically, λ... ,recommend ; These are the amplitude limiting functions, ensuring that the final result strictly falls between 0 and 100.

[0028] Based on Rindex and Pr, four levels of early warning are defined: Level 1 (Reminder, Rindex: 20-40): Slight parameter anomaly, possibly due to process fluctuations. The system records this, marks it in yellow on the twin, and sends an SMS to the team leader.

[0029] Level 2 (Warning, Rindex: 40-60): Abnormal pattern confirmed, high probability of risk. Activate enhanced ventilation, sound and light alarms flash, twin marker turns orange, and push notifications to the workshop supervisor and safety officer's app.

[0030] Level 3 (Severe, Rindex: 60-80): Risk is rapidly escalating. Automatically cut off power and liquid supply valves in the affected area, initiate pre-pressurization of the fire suppression system, evacuation orders are pending, twin markers are displayed in red, and the plant-level emergency command center is notified.

[0031] Level 4 (Critical, Rindex: 80-100): Emergency status confirmed. Trigger workshop-wide evacuation broadcast, activate fire suppression system, dispatch firefighting robots, and synchronize all data streams to local fire emergency services.

[0032] Example: Assuming only the electrolyte leakage mode is currently detected, the probability is... The TVOC concentration reached 0.6 times the threshold, the temperature was normal (0.2), and the rate of increase was low (0.1). ,but , This corresponds to the "severe" warning level.

[0033] In this embodiment of the invention, different confidence levels and different risk modes correspond to different warning levels (such as alert, warning, serious, and critical), providing a basis for decision-making for subsequent differentiated emergency response, enabling safety management personnel to take targeted response measures according to the warning level, and transforming from passive response to proactive prevention and control.

[0034] Traditional power battery production lines are highly automated and involve complex processes (including coating, rolling, slitting, assembly, electrolyte injection, and formation), and they use large quantities of flammable and explosive organic electrolytes, metallic lithium / sodium, aluminum foil, and other active materials, resulting in multiple potential safety risks in the production environment. Thermal runaway risk: During processes such as baking, formation, and aging, micro-short circuits may occur inside the battery, generating heat. If heat dissipation is not done properly, it can easily trigger a chain reaction that leads to fire or even explosion.

[0035] Flammable gas explosion risk: When electrolyte solvent vapors accumulate in a confined space and reach the explosion limit (LEL), they will explode upon contact with an open flame or static spark.

[0036] Risk of toxic gas leakage: Electrolyte decomposition or reactions of certain materials may produce highly toxic and corrosive gases such as HF, PF5, and POF3, which may endanger personnel health and equipment safety.

[0037] Dust explosion risk: Dust from positive and negative electrode active materials (such as NCM and graphite) can become explosive when it reaches a certain concentration in the air.

[0038] Current production line safety monitoring primarily relies on traditional point-type smoke / temperature detectors, but their response is delayed, typically only triggering an alarm after an open flame or high temperature has already occurred, failing to anticipate early chemical changes. Catalytic combustion or semiconductor combustible gas detectors are sensitive to total volatile organic compounds (TVOC), but are generally susceptible to sulfide poisoning. In power battery production, sulfur may originate from raw materials (such as impurities in certain binders and conductive agents); trace amounts of SO2 in the environment; residual sulfates decomposing and releasing SO2 / H2S during high-temperature baking of electrodes; and electrolyte salts (LiPF6) decomposing under certain conditions to produce sulfur-containing byproducts. When the sensor's catalytic element comes into contact with sulfides, irreversible chemical adsorption or reaction occurs, leading to catalyst deactivation, permanent decrease in sensitivity, or even complete failure, creating monitoring blind spots and posing significant safety hazards. In discrete, single-parameter monitoring methods, each sensor operates independently, resulting in data silos and a lack of comprehensive analysis of multi-parameter coupling correlations, leading to high false alarm rates and an inability to accurately locate risk sources and assess risk evolution trends. Regular manual inspections and offline testing are inefficient and lack real-time performance, failing to cover the entire process and all time periods.

[0039] The power battery production safety monitoring method provided in this invention collects multi-source parameters from multiple monitoring points in the power battery production environment and maps these parameters to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of a sensor network. The method extracts spatial correlation features with physical propagation logic between monitoring points using the physical rule model. It then extracts the trend features of the multi-source parameters over time based on the spatial topology of the sensor network. The spatial correlation features and trend features are input into a pre-trained risk identification model, which outputs the type of risk pattern and its corresponding confidence level. Based on the risk pattern and its corresponding confidence level, corresponding graded early warning information is generated. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used for... Based on the spatial topology of the sensor network, spatial correlation features are extracted. The temporal convolutional network module and long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. Through a dual attention mechanism, the weights of different parameter types and different monitoring points are dynamically allocated during fusion to achieve dynamic weighted fusion of multiple parameters and multiple spatial points. The fused features are then input into a deep neural network classifier to identify risk patterns. This invention elevates the originally isolated single-point data into structured information with spatial correlation and temporal dimensions, enabling the system to perceive the spatial diffusion path and temporal evolution of risk parameters, thereby accurately locating the source of risk and predicting its development trend. Furthermore, it can accurately identify risks at the nascent stage, reducing false alarms caused by environmental interference or single sensor fluctuations.

[0040] Based on any of the above embodiments, the step of extracting spatial correlation features with physical propagation logic between monitoring points using the physical rule model includes: Step 201: Treat each sensor as a node in a graph neural network, and construct a dynamic graph structure based on the physical positional relationship of each sensor in the digital twin and the airflow propagation direction in the production environment. Step 202: Learn the information transmission rules between nodes in the dynamic graph structure through graph neural network, and capture the diffusion path and propagation mode of risk parameters in space.

[0041] This invention utilizes a Graph Neural Network (GNN) to construct a dynamic graph, treating each sensor as a graph node and their spatial adjacency relationships (physical distance, downstream wind direction) as edges. The GNN learns the transmission of information between nodes, effectively capturing spatial propagation patterns such as "air mass spreading from point A to point B".

[0042] Based on any of the above embodiments, the step of extracting the trend features of the multi-source parameters over time based on the spatial topology of the sensor network includes: Step 301: Extract the local fluctuation features of each sensor data stream using the temporal convolutional network module; Step 302: Extract the long-term dependency features of each sensor data stream using the Long Short-Term Memory network module; Step 303: Fuse the local fluctuation features with the long-term dependency features to form a complete temporal feature expression.

[0043] This invention utilizes Temporal Convolutional Network (TCN) and Long Short-Term Memory (LSTM) networks to extract the long-term dependencies and short-term fluctuation features of each sensor data stream. TCN excels at capturing local patterns, while LSTM excels at remembering long-term states.

[0044] Based on any of the above embodiments, the step of inputting the spatial correlation features and the trend features into a pre-trained risk identification model and outputting the type of the risk pattern and the corresponding confidence level includes: Step 401: Map the spatial correlation features and the trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points; Step 402: Align and concatenate the feature vectors of different monitoring parameter types and the feature vectors of different monitoring points; Step 403: Dynamically allocate the weights of different parameter types and monitoring points in the current fusion analysis through the multi-attention mechanism module; Step 404: Input the weighted and fused high-level features into the deep neural network classifier and output the probability distribution of the risk pattern.

[0045] In this embodiment of the invention, feature vectors from different modalities, such as gas concentration, temperature, dust, static electricity, and infrared thermal imaging temperature matrix, are aligned and concatenated, and then input into a multi-attention mechanism fusion module. This module can dynamically assign different importance weights to sensors with different parameters and locations. The fused high-level features are then input into a deep neural network classifier, which is trained to identify multiple risk patterns. Mode A (Minor Electrolyte Leakage): The reading of the specific solvent sensor rises slowly, accompanied by a local increase in total volatile organic compounds (TVOC), while temperature and humidity may remain unchanged.

[0046] Mode B (early internal short circuit): In a certain channel of the formation cabinet, the H2 sensor shows a small but continuous rise before the CO sensor, and the infrared image of the corresponding channel shows a weak but continuous high-temperature hot spot (ΔT 2-5℃).

[0047] Mode C (Dust Concentration Exceeds Standard with Static Electricity Accumulation): When the dust sensor reading exceeds the safety threshold (e.g., 30% of the lower explosive limit) and the electrostatic potential of nearby equipment exceeds the danger threshold (e.g., ±3kV), the system determines it to be a high-risk precursor to a dust explosion.

[0048] Mode D (Sulfide Interference Event): In the sulfur-resistant sensor array, the TVOC reading is stable, but the reading of the unprotected standby conventional sensor drifts or fails. The system automatically records this as a "sulfide event" and triggers the self-cleaning program of the sensors in that area. At the same time, an alarm is triggered to prompt maintenance personnel to check the raw materials or process.

[0049] Mode E (thermal runaway evolution): The concentrations of TVOC, H2, and CO increase exponentially, accompanied by a sharp rise in temperature, and the infrared thermal image shows that the hot spots expand rapidly.

[0050] Based on any of the above embodiments, the digital twin further includes a physical rule model embedded in the physical environment model, the physical rule model including at least a simplified computational fluid dynamics model of gas diffusion and a heat conduction model; the digital twin is configured to map the device layout, sensor location and real-time monitoring data in the physical world in real time, as well as visualize and dynamically display the risk patterns, risk locations and graded early warning information.

[0051] This invention utilizes 3D modeling software (such as Unity3D, a digital twin platform) to reproduce the physical environment of a power battery production line at a 1:1 high precision, including the plant structure, equipment layout, pipeline routing, and sensor locations. The twin is not only a visual shell but also embeds physical rules (such as a simplified computational fluid dynamics model for gas diffusion and a heat conduction model).

[0052] This invention establishes a digital twin model of key equipment and areas on the production line, mapping sensor data from the physical world in real time. Graph Neural Networks (GNNs) are used to analyze the spatial topological relationships of the sensor network, and Temporal Convolutional Networks (TCNs) and Long Short-Term Memory Networks (LSTMs) are combined to uncover deep coupling relationships in multi-parameter time series (e.g., "slow rise in TVOC accompanied by immediate local temperature rise" is a strong characteristic of electrolyte leakage, and "sudden increase in CO accompanied by infrared hotspots at specific locations" is a sign of early internal short circuits). This achieves a leap from "threshold alarm" to "pattern warning." Based on the warning level (alert, warning, severe, critical), a graded response mechanism is automatically triggered, such as activating enhanced ventilation, closing specific valves, cutting off power to the area, initiating pre-filling of fire extinguishing media, dispatching inspection robots to confirm the situation, and pushing emergency response plans to the industrial control system and the responsible personnel's terminals, forming a closed loop of "perception-analysis-decision-control."

[0053] Based on any of the above embodiments, embodiments of the present invention further include using historical accident data, simulated experimental data (simulating leakage, heating, etc. in a safety experimental chamber), and a large amount of normal production data to conduct supervised training on the above deep learning model.

[0054] And online self-learning of the model: After the system is put into operation, data is continuously collected. For false alarms or missed alarms confirmed by humans, their features and labels are added to the training set, and the model is incrementally trained and optimized in the background on a regular basis (such as weekly), so that the system can continuously adapt to changes in production line processes and become more and more "intelligent" with use.

[0055] The methods provided in the embodiments of the present invention will now be described in detail using Cases 1 and 2: Case 1: Deployment and testing in a 2GWh power battery formation workshop. A pre-programmed micro-short-circuited battery cell was placed in the formation cabinet. The system accurately warned of "early short circuit risk (Level 2) in formation cabinet A-12 channel" when the cell temperature rise was only 3.2℃ and the H2 concentration reached 120ppm (far below the levels of open flame or large amounts of smoke generation), demonstrating accurate location.

[0056] Case 2: After three months of continuous monitoring at the outlet of the coating oven, the sensitivity of traditional sensors decreased by 60% due to sulfide poisoning, while the performance fluctuation of the sulfur-resistant sensor in this system was less than 5%, and it maintained stability through a self-cleaning process.

[0057] False alarm rate test: During six consecutive months of operation, the false alarms caused by process fluctuations (such as the formation and gas generation of a large number of new batteries) were reduced by about 85% compared with traditional threshold alarm systems after multi-parameter fusion analysis of the system.

[0058] The power battery production safety monitoring method provided in this invention eliminates the biggest weakness of traditional combustible gas monitoring by its anti-sulfur poisoning design; the multi-parameter fusion early warning advances the time of safety hazard detection from "when the accident occurs" to "the budding stage", winning valuable time for emergency response (which may range from a few minutes to tens of minutes).

[0059] Through multi-sensor cross-validation and AI pattern recognition, false alarms caused by single sensor failures or environmental interference are significantly reduced, minimizing production interruption losses. The system can self-diagnose sensor performance degradation and predictively prompt maintenance. The digital twin platform provides intuitive, holistic security situation awareness, improving management efficiency. As a crucial component of the smart factory, the system generates massive amounts of safety data that can be used for process optimization (e.g., analyzing the relationship between leaks and process parameters), quality traceability (e.g., linking safety incidents to battery batches), insurance rate assessment, and other derivative value creation. Preventing a major fire or explosion can recover losses far exceeding the system investment. Simultaneously, preventative maintenance reduces equipment downtime and improves overall production efficiency.

[0060] The power battery production safety monitoring system provided by the present invention is described below. The power battery production safety monitoring system described below can be referred to in correspondence with the power battery production safety monitoring method described above.

[0061] Figure 2 The functional structure diagram of the power battery production safety monitoring system provided in the embodiments of the present invention is as follows: Figure 2 As shown, the power battery production safety monitoring system provided in this embodiment of the invention includes: The acquisition module 201 is used to acquire multi-source parameters from multiple monitoring points in the power battery production environment and map the multi-source parameters to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of the sensor network. Extraction module 202 is used to extract spatial correlation features with physical propagation logic between each monitoring point using the physical rule model, and to extract the trend features of the evolution of the multi-source parameters over time based on the spatial topology of the sensor network. The identification module 203 is used to input the spatial correlation features and the trend features into a pre-trained risk identification model, and output the type of the risk pattern and the corresponding confidence level. The generation module 204 is used to generate corresponding graded early warning information based on the risk pattern and the corresponding confidence level. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, and a multi-attention fusion module. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points, and identifies risk patterns based on the fusion results.

[0062] In this embodiment of the invention, the acquisition module includes an anti-sulfur poisoning gas sensor array, which includes multiple gas sensor chips. The sensitive material of each gas sensor chip has a core-shell structure. The core of the core-shell structure is a metal oxide semiconductor material, and the outer shell is a porous sulfur-repellent molecular sieve material. The outer shell is used to selectively block or adsorb sulfide molecules.

[0063] Based on a novel metal-oxide-semiconductor (MOS) gas sensor, its sensitive material adopts a core-shell structure of "noble metal / metal oxide@porous sulfur-repellent molecular sieve". The outer molecular sieve selectively filters / adsorbs sulfide molecules, preventing them from contacting the core sensitive material; at the same time, it integrates a periodic "electrothermal pulse self-cleaning" unit, which uses instantaneous high temperature (e.g., 500-600℃) to oxidize and desorb sulfides that occasionally penetrate or are adsorbed, restoring the sensor's activity and achieving long-term stable operation.

[0064] With sulfur poisoning-resistant chemical sensors as the core breakthrough, this system combines multi-parameter fusion sensing, digital twins, and artificial intelligence early warning algorithms to construct a fully closed-loop safety defense line of "perception-cognition-early warning-control". It is expected to fundamentally improve the safety level of the power battery manufacturing process and provide a solid guarantee for the high-quality and sustainable development of the industry.

[0065] In this embodiment of the invention, the sulfur poisoning gas sensor array further includes a self-cleaning unit, which is used to perform short-term high-temperature pulse heating on the gas sensor chip when a decrease in sensor sensitivity is detected, so as to oxidize and desorb the attached or adsorbed sulfides.

[0066] The sulfur poisoning-resistant multi-parameter intelligent sensing module adopts a modular design, with a main body size of approximately 150mm x 100mm x 50mm and an IP65 protection rating. The sulfur poisoning-resistant gas sensor unit utilizes classic MOS materials (such as SnO2, ZnO, with doping with noble metals like Pt and Pd to enhance selectivity) that exhibit high sensitivity and fast response to target gases (such as TVOC and H2). A hydrothermal synthesis method is used to grow a multi-level porous zeolite imidazolate framework (ZIF) or modified silica-alumina molecular sieve with a controllable thickness (approximately 50-100 nanometers) in situ on the surface of the core material. This shell layer possesses the following characteristics: Pore ​​size sieving: The pore size is designed to be 0.3-0.5nm, allowing small molecules of electrolyte solvent (DMC molecule size is about 0.45nm), H2, CO, etc. to pass through, but effectively blocking larger molecules such as H2S (~0.36nm but with a large kinetic diameter) and SO2 (~0.36nm), or by surface modification to make it have a strong chemical adsorption selectivity for sulfides, thus "intercepting" them.

[0067] Sulfide-repellent functional groups: Introducing amino groups, metal sites, etc. into the molecular sieve framework to form strong bonds with sulfides and prevent their penetration.

[0068] Sensor Structure: A micro-hotplate structure based on a Micro-Electro-Mechanical System (MEMS) is employed, with the aforementioned core-shell materials coated onto the hotplate. The hotplate integrates: a working heater to maintain the sensor at its optimal operating temperature (e.g., 300-400℃); a self-cleaning pulse heater with independent circuit control. Every 24 hours or when a decrease in sensitivity is detected, the system automatically triggers a short, high-temperature pulse (550-650℃) lasting several seconds, oxidizing and desorbing sulfides adhering to the outer shell or those that have penetrated slightly, without damaging the core sensitive material structure; a resistance temperature detector (RTD) for precise temperature control; and a gas sensing array: a module integrating four sensor chips with different core-shell material combinations, exhibiting cross-selective responses to total TVOC, hydrogen (H2, early short-circuit gas production), carbon monoxide (CO, a high-temperature pyrolysis indicator), and characteristic solvents (e.g., EC). Enhanced ability to identify complex gas mixtures through array signal pattern recognition; optional electrochemical sensor slot: reserved standard interface, pluggable installation for monitoring special gases such as HF (using a special solid polymer electrolyte anti-interference electrochemical sensor).

[0069] In the embodiments of the present invention, a high-precision thermocouple / digital temperature sensor is used to monitor the ambient temperature. A laser scattering dust concentration sensor is used to monitor the concentration of active substance dust in the air in real time (mg / m 3 ). A non-contact electrostatic potentiometer is used to monitor the static electricity accumulation voltage on the surface of equipment and materials. A digital humidity / pressure sensor is used to collect humidity and pressure. The local microprocessor is responsible for sensor driving, AD conversion, data preprocessing (filtering, preliminary feature extraction), self-check (including diagnosis of sensor performance degradation), and communication.

[0070] They are arranged in a grid (such as every 10m x 10m) under the workshop ceiling to monitor the overall environment. They are arranged near the injection needle and inside the recovery hood to focus on monitoring the solvent vapor leakage. They are arranged in the exhaust passage or at key internal points of each cabinet to monitor the early gas production of single cells. Oven / Oven outlet: Monitor the possible organic matter volatilization and sulfide release during the pole piece baking process. The powder feeding station and mixing tank are focused on strengthening the monitoring of dust concentration and static electricity. They are arranged at a high density in the electrolyte warehouse and the hazardous waste temporary storage area.

[0071] Panoramic infrared thermal imagers are deployed in areas such as the formation workshop and aging room to online scan the temperature field distribution of battery modules and charging and discharging cabinets. They are linked with visible light cameras to achieve dual-light verification. The data is transmitted back through gigabit Ethernet.

[0072] An industrial-grade edge computing gateway is deployed in each workshop area, responsible for aggregating the data of about 50-100 intelligent sensing modules in this area. The gateway is built-in with a lightweight AI model, which can perform local real-time anomaly judgment (such as the data of a single sensor exceeding the threshold), achieve a millisecond-level fast response, and reduce the pressure on the central server. The "star + ring" redundant industrial Ethernet is used as the backbone. The intelligent sensing module and the edge gateway use long-distance radio or wireless addressable remote sensor high-speed channels for wireless communication, reducing the wiring complexity and facilitating renovation and flexible adjustment. The key control instructions are issued through the wired network to ensure reliability. Central server: Deploy a high-performance server cluster to run the digital twin platform, core early warning algorithms, and databases. The actuator is integrated with the existing production line control system (programmable logic controller / distributed control system) and can be联动 controlled, such as: Variable frequency explosion-proof fan: Adjust the exhaust air volume as needed; Combustible gas automatic cut-off valve: Installed on the electrolyte supply pipeline; Inert gas (such as N2) fire extinguishing system: Prefilled pipeline valves; Sound and light alarm, emergency broadcast; Automated guided vehicle / patrol robot: Receive instructions to go to the warning point for image verification and secondary gas detection.

[0073] This embodiment of the invention takes a suspected leak in the injection room as an example. At 10:05:00, the "characteristic solvent" sensor of intelligent sensing module 3, located in area B of the injection machine, showed a step increase of 0.5% in its reading of the Lower Explosive Limit (LEL). Subsequently, the TVOC sensor reading slowly increased, while the temperature sensor reading remained unchanged. The reading of the adjacent module 4 sensor did not change. Data from module 3 was transmitted to the edge gateway via LoRa. The gateway's local model determined it to be a "suspected trace leak," triggering a Level 1 alert and marking a timestamp. The digital twin platform showed node 3 turning yellow. Based on the spatial topology and real-time workshop ventilation direction data, the GNN model predicted that the air mass might spread southeast. The TCN-LSTM model analyzed the time-series data of node 3, confirming the continued upward trend. One minute later (10:06:00), the solvent reading of node 3 rose to 1.2% LEL, and node 4 also began to respond. The multi-attention fusion model, integrating information such as gas concentration, diffusion pattern, and historical failure rate of the injection machine, raised the risk level to Level 2 (Warning). The system automatically activated the twin, causing areas 3 and 4 and the predicted diffusion path to flash orange; triggered the injection room ventilation system to increase to maximum airflow; and sent an alarm to the injection machine operator's handheld terminal and the workshop supervisor's APP: "Suspected solvent leak in injection machine area B. Please immediately confirm equipment sealing. Enhanced ventilation has been activated." Upon receiving the alarm, the operator immediately inspected the area and found a slight leak at a hose joint, which was immediately tightened. The leak stopped, and the sensor readings gradually decreased. The operator clicked "Confirm, handled" on the APP. The system recorded the entire event (including alarm, response, and handling results) as case data for model optimization. The warning was lifted.

[0074] The power battery production safety monitoring system provided in this invention fundamentally solves the industry pain point of sensor sulfur poisoning failure through innovation in materials science and MEMS technology, ensuring long-term monitoring reliability. Through multi-parameter fusion and artificial intelligence analysis, it achieves a leap from "delayed alarm" to "early warning," and from "single-point threshold" to "pattern recognition," significantly improving the accuracy and foresight of early warnings. Through digital twins and closed-loop control, it achieves panoramic visualization of the safety situation and automated, precise emergency response, greatly improving safety management efficiency. The system possesses self-learning and evolutionary capabilities, continuously adapting to new production environments and process changes. This system not only provides robust safety assurance for power battery manufacturing, but its generated data assets can also be used for process optimization and quality traceability, yielding significant economic and social benefits.

[0075] It should be noted that a long-term stability test of at least 12 months is required in a real production line environment to assess the structural stability and sulfur resistance life of the molecular sieve shell under high temperature, high humidity, and complex gas environments. Initially, the cost of the sulfur poisoning-resistant sensor will be higher than that of traditional sensors. However, with large-scale production and application, the cost will decrease significantly. The total cost of ownership (TCO) of the system should be comprehensively evaluated based on its ability to avoid accident losses, reduce production downtime, and lower maintenance costs, with an expected return on investment within 2-3 years. The system is designed as a relatively independent safety network, using a unified communication architecture based on a standard open platform and industrial protocols such as the Modicon bus transmission control protocol to interact with the factory's existing PLC / DCS and manufacturing execution systems for data exchange and command issuance, making integration difficult to manage. The design of this system complies with the explosion-proof requirements of GB 3836 Explosive Atmospheres and GB 50493 Design Standard for Detection and Alarm of Combustible and Toxic Gases in Petrochemical Industry, and aims to become a new standard for safety monitoring in the power battery industry.

[0076] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The memory 330 includes a computer program, an operating system, and acquired data. The processor 310 can call the logical instructions in the memory 330 to execute a power battery production safety monitoring method. This method includes: collecting multi-source parameters from multiple monitoring points in the power battery production environment; extracting spatial correlation features between monitoring points based on the spatial topology of the sensor network, and extracting trend features of the multi-source parameters over time; inputting the spatial correlation features and the trend features into a pre-trained risk identification model to identify risk patterns in the power battery production process; and generating corresponding graded early warning information based on the risk patterns and corresponding confidence levels. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network; the temporal convolutional network module and the long short-term memory network module are used to extract trend features; and the multi-attention fusion module is used to dynamically weight and fuse the spatial correlation features and trend features, and input the fused features into the deep neural network classifier to identify risk patterns.

[0077] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0078] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program performs the power battery production safety monitoring method provided by the above methods. The method includes: collecting multi-source parameters from multiple monitoring points in the power battery production environment; extracting spatial correlation features between monitoring points based on the spatial topology of a sensor network, and extracting trend features of the multi-source parameters over time; inputting the spatial correlation features and the trend features into a pre-trained risk identification model to identify risk patterns in the power battery production process; and generating corresponding graded early warning information based on the risk patterns and corresponding confidence levels. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network; the temporal convolutional network module and the long short-term memory network module are used to extract trend features; the multi-attention fusion module is used to dynamically weight and fuse the spatial correlation features and trend features, and input the fused features into the deep neural network classifier to identify risk patterns.

[0079] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0080] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the safety of power battery production, characterized in that, include: Multi-source parameters from multiple monitoring points in the power battery production environment are collected, and the multi-source parameters are mapped to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of the sensor network. The physical rule model is used to extract spatial correlation features with physical propagation logic between each monitoring point; Based on the spatial topology of the sensor network, the trend characteristics of the evolution of the multi-source parameters over time are extracted; The spatial correlation features and the trend features are input into a pre-trained risk identification model, which outputs the type of the risk pattern and the corresponding confidence level. Based on the risk pattern and the corresponding confidence level, generate corresponding graded early warning information; The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, a multi-attention fusion module, and a deep neural network classifier. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points. The fused features are then input into the deep neural network classifier to identify risk patterns.

2. The method for monitoring the safety of power battery production according to claim 1, characterized in that, The extraction of spatial correlation features with physical propagation logic between monitoring points using the physical rule model includes: Each sensor is used as a node in a graph neural network, and a dynamic graph structure is constructed based on the physical positional relationship of each sensor in the digital twin and the airflow propagation direction in the production environment. By learning the information transmission patterns between nodes in the dynamic graph structure through graph neural networks, the diffusion paths and propagation patterns of risk parameters in space can be captured.

3. The method for monitoring the safety of power battery production according to claim 1, characterized in that, The extraction of the trend features of the multi-source parameters over time based on the spatial topology of the sensor network includes: The temporal convolutional network module is used to extract local fluctuation features of the data streams from each sensor. The long short-term memory network module is used to extract the long-term dependency features of each sensor data stream; The local fluctuation features are fused with the long-term dependency features to form a complete temporal feature representation.

4. The method for monitoring the safety of power battery production according to claim 1, characterized in that, The step of inputting the spatial correlation features and the trend features into a pre-trained risk identification model and outputting the type of the risk pattern and the corresponding confidence level includes: The spatial correlation features and the trend features are mapped into feature vectors of different monitoring parameter types and feature vectors of different monitoring points; Align and concatenate the feature vectors of different monitoring parameter types and feature vectors of different monitoring points; The weights of different parameter types and monitoring points in the current fusion analysis are dynamically allocated through the multi-attention mechanism module. The weighted and fused high-level features are input into the deep neural network classifier, which outputs the probability distribution of risk patterns.

5. The method for monitoring the safety of power battery production according to claim 1, characterized in that, The digital twin also includes a physical rule model embedded in the physical environment model, which includes at least a simplified computational fluid dynamics model for gas diffusion and a heat conduction model. The digital twin is configured to map the device layout, sensor locations, and real-time monitoring data in the physical world in real time, as well as to visualize and dynamically display the risk patterns, risk locations, and graded early warning information.

6. A power battery production safety monitoring system, characterized in that, include: The acquisition module is used to acquire multi-source parameters from multiple monitoring points in the power battery production environment and map the multi-source parameters to a digital twin in real time. The digital twin includes a physical environment model and the spatial topology of the sensor network. The extraction module is used to extract spatial correlation features with physical propagation logic between each monitoring point using the physical rule model, and to extract the trend features of the evolution of the multi-source parameters over time based on the spatial topology of the sensor network. The identification module is used to input the spatial correlation features and the trend features into a pre-trained risk identification model and output the type of the risk pattern and the corresponding confidence level. The generation module is used to generate corresponding graded early warning information based on the risk pattern and the corresponding confidence level. The risk identification model includes a graph neural network module, a temporal convolutional network module, a long short-term memory network module, and a multi-attention fusion module. The graph neural network module is used to extract spatial correlation features based on the spatial topology of the sensor network. The temporal convolutional network module and the long short-term memory network module are used to extract trend features. The multi-attention fusion module is used to map the spatial correlation features and trend features into feature vectors of different monitoring parameter types and feature vectors of different monitoring points, respectively. It dynamically allocates the weights of different parameter types and different monitoring points during fusion through a dual attention mechanism to achieve dynamic weighted fusion of multiple parameters and multiple spatial points, and identifies risk patterns based on the fusion results.

7. The power battery production safety monitoring system according to claim 6, characterized in that, The acquisition module includes an anti-sulfur poisoning gas sensor array, which includes multiple gas sensor chips. Each gas sensor chip has a core-shell structure as its sensitive material. The core of the core-shell structure is a metal oxide semiconductor material, and the outer shell is a porous sulfur-repellent molecular sieve material. The outer shell is used to selectively block or adsorb sulfide molecules.

8. The power battery production safety monitoring system according to claim 7, characterized in that, The sulfur poisoning gas sensor array also includes a self-cleaning unit, which is used to perform short-term high-temperature pulse heating on the gas sensor chip when a decrease in sensor sensitivity is detected, so as to oxidize and desorb the attached or adsorbed sulfides.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the power battery production safety monitoring method as described in any one of claims 1 to 5.

10. A non-transitory readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the power battery production safety monitoring method as described in any one of claims 1 to 5.

Images