Laboratory real-time state digital management equipment

Abstract

The invention discloses laboratory real-time state digital management equipment. The equipment comprises a data acquisition module which is responsible for acquiring internal data of a battery and a laboratory; and the data processing and analysis module is responsible for processing and analyzing the collected data, predicting the health state and the service life of the battery, constructing a heterogeneous data relation graph based on a graph neural network technology, mining the incidence relation between cross-modal data by using a graph neural network, constructing a prediction model, and predicting the health state and the service life of the battery. The prediction model excavates a change rule of battery parameter data, and provides a virtual parameter curve and anomaly detection information for the digital twin module; the digital twinborn module is responsible for integrating an electrochemical model and actually measured data, simulating charging and discharging behaviors of the battery in real time, identifying model deviation by comparing a virtual parameter curve output by a twinborn body with the actually measured data, and optimizing a test strategy; the personnel behavior monitoring module is used for detecting dangerous operation by adopting a SlowFast network; and the display and control module provides a user interface.

Description

Technical Field

[0001] The present invention relates to the technical field of laboratory management, and in particular to real-time digital management equipment for laboratories. Background Art

[0002] In modern society, batteries have become an indispensable energy storage solution for everyday life and industry. With the continuous advancement of technology, a variety of new battery technologies have emerged, improving energy density, extending service life, and continuously achieving breakthroughs in safety performance. From electric vehicles to portable electronic devices, from energy storage stations to aerospace applications, battery applications are becoming increasingly diverse, and the performance requirements are becoming increasingly demanding. To support the research and development and production of these high-performance battery products, rigorous battery performance testing has become indispensable. The testing process not only provides data on battery performance under ideal conditions, but also enables manufacturers to understand how batteries behave under extreme conditions, enabling them to improve battery design and ensure the end product has excellent performance and sufficient safety. Accelerated battery life testing is particularly critical, allowing researchers to estimate the product's service life, predict the rate of performance degradation, and identify potential failure modes. Establishing a comprehensive safety monitoring system for battery performance testing laboratories is crucial.

[0003] In existing technologies, insufficient historical data makes it difficult for models to fully understand the characteristics of battery charge and discharge performance. Consequently, when faced with new charge and discharge conditions, the models often fail to generate accurate prediction curves, resulting in significant discrepancies between predicted results and actual performance. This not only affects research accuracy but also increases experimental risk and cost. Therefore, a device for digitally managing real-time laboratory status has been proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a laboratory real-time status digital management device.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] Laboratory real-time status digital management equipment, including:

[0007] Data acquisition module: responsible for collecting battery and laboratory internal data, including battery parameters, environmental parameters and image features in the laboratory;

[0008] Data processing and analysis module: This module processes and analyzes the collected data to predict the health status and lifespan of the battery. Based on graph neural network technology, it constructs a heterogeneous data relationship graph of battery parameters, environmental parameters, and image features. Graph neural networks are used to mine the correlation between cross-modal data and build a prediction model. The prediction model mines the variation patterns of battery parameter data and provides virtual parameter curves and anomaly detection information for the digital twin module.

[0009] Digital twin module: This module integrates electrochemical models with measured data, simulates battery charge and discharge behavior in real time, and identifies model deviations and optimizes test strategies by comparing virtual parameter curves output by the twin with measured data.

[0010] Personnel behavior monitoring module: uses the SlowFast network to detect dangerous operations and integrates environmental sensor data to identify violations;

[0011] Display and control module: provides a user interface to display laboratory information and allows administrators to remotely control and adjust laboratory equipment.

[0012] The above technical solution further includes:

[0013] Furthermore, the data acquisition module includes a battery parameter acquisition unit, an environmental parameter acquisition unit and an image acquisition unit. The battery parameter acquisition unit monitors key parameters of the battery, such as the number of cycles (each charge and discharge is a cycle), temperature (real-time temperature of the battery surface), open circuit voltage (the voltage of the battery in the open circuit state), etc. The environmental parameter acquisition unit is responsible for monitoring environmental parameters such as temperature and humidity. The image acquisition unit captures image features in the laboratory, including changes in battery appearance, personnel operating behavior, etc.

[0014] Furthermore, the data processing and analysis module processes and analyzes the collected data, including the following steps:

[0015] Data preprocessing: preprocessing the collected battery parameters (such as cycle number, temperature, open circuit voltage, etc.), environmental parameters (such as temperature, humidity, etc.) and image features;

[0016] Constructing a heterogeneous data relationship graph: Based on the graph neural network (GNN) technology, the battery parameters, environmental parameters and image features are constructed into a heterogeneous data relationship graph. In the heterogeneous data relationship graph, the nodes represent different data points (such as a measurement value of a battery parameter), and the edges represent the relationship between data points (such as temporal continuity, spatial proximity, etc.). For node i, its updated representation Computed by aggregating the representations of its neighbor nodes: Among them, N(i) is the set of neighbor nodes of node i, W( l) and b(l) are the weights and biases of the lth layer, σ is the activation function, and when constructing the heterogeneous data relationship graph, the Bayesian network is used to determine the causal relationship between nodes;

[0017] Cross-modal data fusion and prediction model construction: Graph neural networks are used to mine the correlations between cross-modal data. Bayesian networks and transfer learning are combined to build a prediction model. The prediction model mines the changing patterns of battery parameter data and predicts the health status and lifespan of the battery.

[0018] Online incremental learning and model updating: Using online incremental learning and reinforcement learning techniques, we update the prediction model and parameter fluctuation thresholds in real time, adaptively responding to sudden anomalies and nonlinear changes. We use gradient descent to update the model weights each time new data is received. Among them, L(W ( l ) ) is the loss function (which measures the difference between the model's predicted value and the actual value), η is the learning rate, is the gradient of the loss function with respect to the weights;

[0019] Anomaly detection and early warning: Anomaly detection is performed based on the updated prediction model and the threshold of the parameter fluctuation range. If the predicted value of a data point exceeds the preset threshold range, an early warning signal is triggered.

[0020] Furthermore, the specific steps of cross-modal data fusion and prediction model construction;

[0021] Constructing a Bayesian network model: The Bayesian network uses nodes to represent variables and directed edges to represent the causal relationship between variables. In cross-modal data fusion, battery parameters, environmental parameters, and image features are used as nodes of the Bayesian network, and the causal relationship between nodes is constructed based on historical data and expert knowledge. In the Bayesian network, the conditional probability distribution of a node is expressed as: P(Xi|Pa(Xi)), where Xi represents node i and Pa(Xi) represents the set of parent nodes of node i. For given observation data, Bayes' theorem is used to calculate the posterior probability: P(H|D)=P(D|H)*P(H) / P(D), where H represents the hypothesis and D represents the observation data.

[0022] Transfer learning: Use transfer learning techniques to transfer knowledge from other related tasks. Initially, there is a pre-trained model with weights W_pre. Its weights are used as the initial weights of the new model and fine-tuned: W_new = W_pre + ΔW, where ΔW is the weight update obtained by training with the new task data.

[0023] Build a prediction model: Combining Bayesian networks and transfer learning, we build a prediction model to mine the changing patterns of battery parameter data and predict the health status and life of the battery. We use Bayesian networks to infer the causal relationship between different modal data, and use transfer learning to fuse these data to build a prediction model.

[0024] Furthermore, the purpose of establishing the electrochemical model is to simulate the charge and discharge behavior of the battery. The electrochemical model is expressed as Wherein, C is the concentration of lithium ions, t is time, D is the diffusion coefficient, j is the current density, and F is the Faraday constant. The electrochemical model describes the diffusion of lithium ions inside the battery and the concentration change caused by the current.

[0025] Furthermore, the digital twin module constructs the digital twin in the following specific steps:

[0026] Real-time data acquisition and preprocessing: The data acquisition module collects battery voltage, current, temperature and other parameters in real time;

[0027] Digital twin construction and simulation: The preprocessed data is input into the electrochemical model to construct a digital twin, which simulates the battery's charge and discharge behavior in real time and outputs a virtual parameter curve.

[0028] Model deviation identification: Compare the virtual parameter curve output by the digital twin with the measured data. By calculating the difference between the two, the model deviation is identified.

[0029] Optimize testing strategies: Based on identified model deviations, optimize the testing strategy. For example, if the model has large deviations under certain conditions, increase the number of experiments under those conditions to improve model accuracy. Alternatively, if a model parameter is inaccurate, calibrate the parameter using experimental data.

[0030] Furthermore, the personnel behavior monitoring module uses the SlowFast network to detect dangerous operations, including the following steps:

[0031] SlowFast network model input: The preprocessed image frame sequence is input into the SlowFast network in chronological order;

[0032] Feature extraction: Static features: The Slow path (low frame rate path) is responsible for capturing static scene information in the video, such as the layout of experimental equipment and the posture of the experimenter; Dynamic features: The Fast path (high frame rate path) is responsible for capturing fast motion information in the video, such as the details of the experimenter's movements and the rapid movement of objects; The SlowFast network, through its unique fast-slow path structure, fuses static features and dynamic features to form a feature representation;

[0033] Dangerous operation identification: Based on the extracted features, the SlowFast network classifies the operations in the image. During the training phase, the network adjusts weights through a backpropagation algorithm to minimize the error between the predicted probability and the true label. During the inference phase, the network outputs the most likely dangerous operation category based on the input feature vector.

[0034] Furthermore, the personnel behavior monitoring module integrates environmental sensor data to identify specific steps of illegal behavior:

[0035] Data fusion: Fusing the alcohol concentration sensor data with the output of the SlowFast network;

[0036] Threshold judgment: Sensor data is judged based on the preset threshold to determine whether a violation has occurred and set a safety threshold for alcohol concentration. If the sensor data exceeds the threshold, it is considered a violation (such as operating under the influence of alcohol);

[0037] Comprehensive judgment: The system combines the output of the SlowFast network with environmental sensor data to make a comprehensive judgment. If dangerous operations and illegal behaviors (such as dangerous operations under the influence of alcohol) are detected at the same time, an early warning will be triggered.

[0038] The present invention has the following beneficial effects:

[0039] In this invention, cross-modal data fusion technology based on graph neural networks allows the device to more deeply explore the correlations between battery parameters, environmental parameters, and image features, improving the accuracy and efficiency of data fusion. By building a digital twin of battery testing, the device can simulate the battery's charging and discharging behavior in real time, comparing virtual parameter curves with measured data to identify model deviations. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a system block diagram of the laboratory real-time status digital management equipment proposed by the present invention. DETAILED DESCRIPTION

[0041] 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] See also Figure 1 As shown, the present invention is a laboratory real-time status digital management device, comprising:

[0043] Data acquisition module: responsible for collecting battery and laboratory internal data, including battery parameters, environmental parameters and image features in the laboratory. The battery parameter acquisition unit and environmental parameter acquisition unit in the data acquisition module can be connected to the data processing and analysis module through wired mode;

[0044] Data processing and analysis module: Responsible for processing and analyzing the collected data and predicting the health status and life of the battery. Based on graph neural network (GNN) technology, it constructs a heterogeneous data relationship graph of battery parameters, environmental parameters, and image features. The graph neural network is used to mine the correlation between cross-modal data and build a prediction model. The prediction model mines the variation pattern of battery parameter data and provides virtual parameter curves and anomaly detection information for the digital twin module. The data processing and analysis module is placed in the server room or data center to ensure the efficiency and security of data processing. For data that needs to be processed in real time, the data acquisition module and the data processing and analysis module can be connected through a high-speed network;

[0045] Digital twin module: This module integrates electrochemical models with measured data, simulates battery charge and discharge behavior in real time, identifies model deviations, and optimizes test strategies by comparing virtual parameter curves output by the twin with measured data. The digital twin module shares server resources with the data processing and analysis module, or can be deployed independently on devices with high-performance computing capabilities.

[0046] Personnel behavior monitoring module: uses the SlowFast network to detect dangerous operations and integrates environmental sensor data such as alcohol concentration to identify violations;

[0047] The Display and Control Module provides a user interface that displays real-time laboratory status, abnormality warnings, and other information, allowing administrators to remotely control and adjust laboratory equipment. The Personnel Behavior Monitoring Module transmits information about dangerous operations and violations to the Display and Control Module for real-time prompts and warnings. The Display and Control Module also allows administrators to understand the results of personnel behavior monitoring through the real-time status display unit and take appropriate management measures.

[0048] In one embodiment, the data acquisition module includes a battery parameter acquisition unit, an environmental parameter acquisition unit, and an image acquisition unit. The battery parameter acquisition unit monitors key battery parameters, such as cycle number (each charge and discharge is one cycle), temperature (real-time temperature of the battery surface), open circuit voltage (the voltage of the battery in the open circuit state), etc. The environmental parameter acquisition unit is responsible for monitoring environmental parameters such as temperature and humidity. The image acquisition unit captures image features in the laboratory, including changes in battery appearance and personnel operating behavior.

[0049] The battery parameter acquisition unit is installed near the battery test area to facilitate direct connection to the battery and real-time data acquisition. The battery parameter acquisition unit is connected to the battery testing equipment or sensor to ensure the accuracy and real-time nature of data acquisition.

[0050] Environmental parameter acquisition units are evenly distributed in different locations in the laboratory to ensure that environmental changes in the laboratory can be fully monitored and monitoring parameters such as temperature and humidity can be set. These parameters are crucial to ensuring that the laboratory environment meets experimental requirements.

[0051] Image acquisition units are installed in key areas, such as battery testing areas and personnel operation areas, to capture important image features. Multiple high-resolution cameras are installed in the laboratory. These cameras can capture changes in the battery's appearance during the charging and discharging process, such as expansion and deformation, as well as whether personnel operating behavior complies with regulations.

[0052] In one embodiment, the data processing and analysis module processes and analyzes the collected data, including the following steps:

[0053] Data preprocessing: preprocessing the collected battery parameters (such as cycle number, temperature, open circuit voltage, etc.), environmental parameters (such as temperature, humidity, etc.) and image features;

[0054] Constructing a heterogeneous data relationship graph: Based on the graph neural network (GNN) technology, the battery parameters, environmental parameters and image features are constructed into a heterogeneous data relationship graph. In the heterogeneous data relationship graph, the nodes represent different data points (such as a measurement value of a battery parameter), and the edges represent the relationship between data points (such as temporal continuity, spatial proximity, etc.). For node i, its updated representation Computed by aggregating the representations of its neighbor nodes: Among them, N(i) is the set of neighbor nodes of node i, W ( l) and b) ( l) is the weight and bias of the lth layer, σ is the activation function, and when constructing a heterogeneous data relationship graph, a Bayesian network is used to determine the causal relationship between nodes;

[0055] Cross-modal data fusion and prediction model construction: Graph neural networks are used to mine the correlations between cross-modal data. Bayesian networks and transfer learning are combined to build a prediction model. The prediction model mines the changing patterns of battery parameter data and predicts the health status and lifespan of the battery.

[0056] Online incremental learning and model updating: Using online incremental learning and reinforcement learning techniques, we update the prediction model and parameter fluctuation thresholds in real time, adaptively responding to sudden anomalies and nonlinear changes. We use gradient descent to update the model weights each time new data is received. Among them, L(W ( l ) ) is the loss function (which measures the difference between the model's predicted value and the actual value), η is the learning rate, is the gradient of the loss function with respect to the weights;

[0057] Anomaly detection and early warning: Anomaly detection is performed based on the updated prediction model and the threshold of the parameter fluctuation range. If the predicted value of a data point exceeds the preset threshold range, an early warning signal is triggered.

[0058] In one embodiment, the specific steps of cross-modal data fusion and prediction model construction are:

[0059] Constructing a Bayesian network model: The Bayesian network uses nodes to represent variables and directed edges to represent the causal relationship between variables. In cross-modal data fusion, battery parameters, environmental parameters, and image features are used as nodes of the Bayesian network, and the causal relationship between nodes is constructed based on historical data and expert knowledge. In the Bayesian network, the conditional probability distribution of a node is expressed as: P(Xi|Pa(Xi)), where Xi represents node i and Pa(Xi) represents the set of parent nodes of node i. For given observation data, Bayes' theorem is used to calculate the posterior probability: P(H|D)=P(D|H)*P(H) / P(D), where H represents the hypothesis and D represents the observation data.

[0060] Transfer learning: Use transfer learning techniques to transfer knowledge from other related tasks. Initially, there is a pre-trained model with weights W_pre. Its weights are used as the initial weights of the new model and fine-tuned: W_new = W_pre + ΔW, where ΔW is the weight update obtained by training with the new task data.

[0061] Build a prediction model: Combining Bayesian networks and transfer learning, we build a prediction model to mine the changing patterns of battery parameter data and predict the health status and life of the battery. We use Bayesian networks to infer the causal relationship between different modal data, and use transfer learning to fuse these data to build a prediction model.

[0062] Consider a new energy vehicle R&D laboratory conducting charge and discharge tests on a newly developed battery. Using real-time digital laboratory status management equipment, we collect battery parameters such as cycle count, temperature, and open-circuit voltage, as well as environmental parameters such as laboratory temperature and humidity. High-resolution cameras are also used to capture image features of the battery's appearance and the personnel operating the battery.

[0063] In the process of cross-modal data fusion and predictive model construction, a Bayesian network model was first constructed, using battery parameters, environmental parameters, and image features as nodes. Causal relationships between nodes were established based on historical data and expert knowledge. Then, transfer learning techniques were used to transfer knowledge from other battery testing tasks to improve the model's generalization capabilities. Finally, a predictive model based on Bayesian networks and transfer learning was constructed to predict the battery's state of health and lifespan.

[0064] This approach can more effectively mine correlations between cross-modal data and build more accurate predictive models. This helps laboratory managers take timely measures to ensure safe laboratory operations and improve research efficiency and quality. Furthermore, the causal reasoning capabilities of Bayesian networks can be used to perform root cause analysis and uncertainty quantification, providing enhanced decision support for laboratory operations and management.

[0065] In one embodiment, the purpose of establishing the electrochemical model is to simulate the charge and discharge behavior of the battery. The electrochemical model is expressed as Wherein, C is the concentration of lithium ions, t is time, D is the diffusion coefficient, j is the current density, and F is the Faraday constant. The electrochemical model describes the diffusion of lithium ions inside the battery and the concentration change caused by the current.

[0066] In one embodiment, the digital twin module constructs the digital twin in the following specific steps:

[0067] Real-time data acquisition and preprocessing: The data acquisition module collects battery voltage, current, temperature and other parameters in real time;

[0068] Digital twin construction and simulation: The preprocessed data is input into the electrochemical model to construct a digital twin, which simulates the battery's charge and discharge behavior in real time and outputs a virtual parameter curve.

[0069] Model deviation identification: Compare the virtual parameter curve output by the digital twin with the measured data. By calculating the difference between the two, the model deviation is identified.

[0070] Optimize testing strategies: Based on identified model deviations, optimize the testing strategy. For example, if the model has large deviations under certain conditions, increase the number of experiments under those conditions to improve model accuracy. Alternatively, if a model parameter is inaccurate, calibrate the parameter using experimental data.

[0071] In one embodiment, the human behavior monitoring module uses a SlowFast network to detect dangerous operations, including the following steps:

[0072] SlowFast network model input: The preprocessed image frame sequence is input into the SlowFast network in chronological order;

[0073] Feature extraction: Static features: The Slow path (low frame rate path) is responsible for capturing static scene information in the video, such as the layout of experimental equipment and the posture of the experimenter; Dynamic features: The Fast path (high frame rate path) is responsible for capturing fast motion information in the video, such as the details of the experimenter's movements and the rapid movement of objects; The SlowFast network, through its unique fast-slow path structure, fuses static features and dynamic features to form a feature representation;

[0074] Dangerous operation identification: Based on the extracted features, the SlowFast network classifies the operations in the image. During the training phase, the network adjusts weights through a backpropagation algorithm to minimize the error between the predicted probability and the true label. During the inference phase, the network outputs the most likely dangerous operation category based on the input feature vector.

[0075] In one embodiment, the personnel behavior monitoring module integrates environmental sensor data to identify illegal behaviors in the following specific steps:

[0076] Data fusion: Fusing the alcohol concentration sensor data with the output of the SlowFast network;

[0077] Threshold judgment: Sensor data is judged based on the preset threshold to determine whether a violation has occurred and set a safety threshold for alcohol concentration. If the sensor data exceeds the threshold, it is considered a violation (such as operating under the influence of alcohol);

[0078] Comprehensive judgment: The system combines the output of the SlowFast network with environmental sensor data to make a comprehensive judgment. If dangerous operations and illegal behaviors (such as dangerous operations under the influence of alcohol) are detected at the same time, an early warning will be triggered.

[0079] Assume that an alcohol concentration sensor is deployed in a laboratory to monitor the alcohol concentration in real time. Simultaneously, a SlowFast network is used to detect dangerous operating behaviors by the experimenter. As the experimenter performs an operation, the alcohol concentration sensor collects alcohol concentration data in real time, and the SlowFast network outputs the probability of a dangerous operation based on the experimenter's movements and posture.

[0080] The alcohol concentration data is used as an additional feature and concatenated with the output of the SlowFast network. In this way, a composite feature vector containing alcohol concentration and the probability of dangerous operation is obtained.

[0081] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. Laboratory real-time status digital management equipment, characterized by: include: Data acquisition module: responsible for collecting battery and laboratory internal data, including battery parameters, environmental parameters and image features in the laboratory; Data processing and analysis module: This module processes and analyzes the collected data to predict the health status and lifespan of the battery. Based on graph neural network technology, it constructs a heterogeneous data relationship graph of battery parameters, environmental parameters, and image features. Graph neural networks are used to mine the correlation between cross-modal data and build a prediction model. The prediction model mines the variation patterns of battery parameter data and provides virtual parameter curves and anomaly detection information for the digital twin module. Digital twin module: This module integrates electrochemical models with measured data, simulates battery charge and discharge behavior in real time, and identifies model deviations and optimizes test strategies by comparing virtual parameter curves output by the twin with measured data. Personnel behavior monitoring module: uses the SlowFast network to detect dangerous operations and integrates environmental sensor data to identify violations; Display and control module: provides a user interface to display laboratory information and allows administrators to remotely control and adjust laboratory equipment.

2. The laboratory real-time status digital management equipment according to claim 1, characterized in that: The data acquisition module includes a battery parameter acquisition unit, an environmental parameter acquisition unit and an image acquisition unit. The battery parameter acquisition unit monitors key parameters of the battery, the environmental parameter acquisition unit is responsible for monitoring environmental parameters, and the image acquisition unit captures image features in the laboratory.

3. The laboratory real-time status digital management equipment according to claim 1, characterized in that: The data processing and analysis module processes and analyzes the collected data, including the following steps: Data preprocessing: preprocess the collected battery parameters, environmental parameters and image features; Constructing heterogeneous data relationship graph: Based on graph neural network technology, battery parameters, environmental parameters and image features are constructed into a heterogeneous data relationship graph. In the heterogeneous data relationship graph, nodes represent different data points, and edges represent the relationship between data points. For node i, its updated representation Computed by aggregating the representations of its neighbor nodes: Among them, N(i) is the set of neighbor nodes of node i, W ( l) and b(l) are the weights and biases of layer l, and σ is the activation function; Cross-modal data fusion and prediction model construction: Graph neural networks are used to mine the correlations between cross-modal data. Bayesian networks and transfer learning are combined to build a prediction model. The prediction model mines the changing patterns of battery parameter data and predicts the health status and lifespan of the battery. Online incremental learning and model updating: Using online incremental learning and reinforcement learning techniques, we update the prediction model and parameter fluctuation thresholds in real time, adaptively responding to sudden anomalies and nonlinear changes. We use gradient descent to update the model weights each time new data is received. Among them, L(W ( l ) ) is the loss function, η is the learning rate, is the gradient of the loss function with respect to the weights; Anomaly detection and early warning: Anomaly detection is performed based on the updated prediction model and the threshold of the parameter fluctuation range. If the predicted value of a data point exceeds the preset threshold range, an early warning signal is triggered.

4. The laboratory real-time status digital management equipment according to claim 3, characterized in that: The specific steps of cross-modal data fusion and prediction model construction; Constructing a Bayesian network model: The Bayesian network uses nodes to represent variables and directed edges to represent the causal relationship between variables. In cross-modal data fusion, battery parameters, environmental parameters, and image features are used as nodes of the Bayesian network, and the causal relationship between nodes is constructed based on historical data and expert knowledge. In the Bayesian network, the conditional probability distribution of a node is expressed as: P(Xi|Pa(Xi)), where Xi represents node i and Pa(Xi) represents the set of parent nodes of node i. For given observation data, Bayes' theorem is used to calculate the posterior probability: P(H|D)=P(D|H)*P(H) / P(D), where H represents the hypothesis and D represents the observation data. Transfer learning: Use transfer learning techniques to transfer knowledge from other related tasks. Initially, there is a pre-trained model with weights W_pre. Its weights are used as the initial weights of the new model and fine-tuned: W_new = W_pre + ΔW, where ΔW is the weight update obtained by training with the new task data. Build a prediction model: Combining Bayesian networks and transfer learning, we build a prediction model to mine the changing patterns of battery parameter data and predict the health status and life of the battery. We use Bayesian networks to infer the causal relationship between different modal data, and use transfer learning to fuse these data to build a prediction model.

5. The laboratory real-time status digital management equipment according to claim 1, characterized in that: The purpose of establishing the electrochemical model is to simulate the charge and discharge behavior of the battery. The electrochemical model is expressed as Wherein, C is the concentration of lithium ions, t is time, D is the diffusion coefficient, j is the current density, and F is the Faraday constant. The electrochemical model describes the diffusion of lithium ions inside the battery and the concentration change caused by the current.

6. The laboratory real-time status digital management equipment according to claim 1, characterized in that: The specific steps of the digital twin module to build a digital twin are: Real-time data acquisition and preprocessing: Real-time data acquisition of battery parameters through the data acquisition module; Digital twin construction and simulation: The preprocessed data is input into the electrochemical model to construct a digital twin, which simulates the battery's charge and discharge behavior in real time and outputs a virtual parameter curve. Model deviation identification: Compare the virtual parameter curve output by the digital twin with the measured data, and identify the model deviation by calculating the difference between the two; Optimize test strategy: Based on the identified model deviations, optimize the test strategy.

7. The laboratory real-time status digital management equipment according to claim 1, characterized in that: The personnel behavior monitoring module uses the SlowFast network to detect dangerous operations, including the following steps: SlowFast network model input: The preprocessed image frame sequence is input into the SlowFast network in chronological order; Feature extraction: Static features: The Slow path is responsible for capturing static scene information in the video; Dynamic features: The Fast path is responsible for capturing fast motion information in the video; The SlowFast network fuses static features and dynamic features to form a feature representation; Dangerous operation identification: Based on the extracted features, the SlowFast network classifies the operations in the image.

8. The laboratory real-time status digital management equipment according to claim 7, characterized in that: The specific steps of the personnel behavior monitoring module to integrate environmental sensor data and identify violations are as follows: Data fusion: Fusing the alcohol concentration sensor data with the output of the SlowFast network; Threshold judgment: Sensor data is judged based on preset thresholds to determine whether violations have occurred; Comprehensive judgment: Combines the output of the SlowFast network and environmental sensor data to make a comprehensive judgment; if dangerous operations and violations are detected at the same time, an early warning is triggered; Assume that an alcohol concentration sensor is deployed in the laboratory to monitor the alcohol concentration in real time. Simultaneously, a SlowFast network is used to detect dangerous operating behaviors by the experimenter. As the experimenter performs an operation, the alcohol concentration sensor collects alcohol concentration data in real time, and the SlowFast network outputs the probability of a dangerous operation based on the experimenter's movements and posture. The alcohol concentration data is taken as an additional feature and concatenated with the output of the SlowFast network; in this way, a composite feature vector containing alcohol concentration and the probability of dangerous operation is obtained.

Images