Virtual-real linkage control method for dual-mode communication battery simulation teaching

Abstract

The invention relates to the technical field of battery simulation teaching, and discloses a virtual-real linkage control method for simulation teaching of a dual-mode communication battery, so as to solve the problems that virtual and real data of an existing system are asynchronous, fault features are abstract, and a teaching closed loop is imperfect. The method comprises the steps of system initialization, instruction issuing, hardware response, software rendering, data comparison, algorithm verification and fault clearing. A Bluetooth and CAN bus dual-mode communication architecture is adopted, a fault instruction is issued through a mobile fault setter, fault simulation signal output, hardware data acquisition and virtual scene visual linkage are achieved depending on hardware such as a battery simulator, a BMS controller and VR equipment, and virtual and real parameter comparison and algorithm validity verification are completed. The method improves the authenticity and efficiency of teaching practical training, and is suitable for battery professional talent training scenes such as vocational colleges and vehicle enterprise training.

Description

Technical Field

[0001] This invention relates to the field of battery simulation teaching technology, specifically to a virtual-real linkage control method for dual-mode communication battery simulation teaching. Background Technology

[0002] Battery simulation teaching systems, as core equipment for training professionals, can simulate various battery fault scenarios without the actual safety risks of real batteries, helping students master key skills such as battery fault identification, algorithm design, and operation and maintenance debugging. Currently, battery simulation teaching systems on the market are mainly divided into two categories: pure virtual simulation systems and hardware benchtop simulation systems. However, both types of systems have significant technical shortcomings and cannot meet the needs of high-quality teaching. Pure virtual simulation systems: These systems use computer software to build virtual battery models, allowing students to simulate fault scenarios through a software interface. While these systems offer advantages such as low cost and no safety risks, they are completely detached from the physical hardware. The virtual parameters deviate significantly from the characteristics of real batteries, preventing students from perceiving the dynamic changes in hardware parameters. This leads to a disconnect between theory and practice, greatly diminishing the effectiveness of the teaching. Furthermore, these systems only support simple static fault simulations and cannot reproduce the gradual process of battery fault development from its initial stages, making it difficult to cultivate students' ability to identify complex faults.

[0003] Hardware benchtop simulation systems are equipped with real battery simulators, BMS controllers, and other hardware devices, enabling them to output realistic physical signals such as voltage and current. However, these systems typically use a single communication link, which suffers from problems such as link congestion, high signal latency, and weak anti-interference capabilities, easily leading to asynchronous hardware responses and command issuance. Furthermore, the systems lack virtual scenario linkage capabilities; fault status is only displayed through numerical instruments, and fault characteristics are abstract and difficult to understand, making it difficult for trainees to quickly locate fault positions and understand fault mechanisms.

[0004] To address the aforementioned issues, some manufacturers have attempted to develop virtual-real integrated battery simulation teaching systems. However, existing virtual-real integrated solutions still suffer from the following drawbacks: The low synchronization accuracy of virtual and real data, the lack of an effective fusion mechanism between hardware-acquired data and virtual simulation data, the large parameter deviations, and the inconsistency between the fault states displayed in the virtual scene and the actual hardware output can easily mislead trainees into developing incorrect fault perceptions. The fault simulation format is limited, supporting only static fault simulation with fixed parameters, failing to simulate the dynamic gradual process of real battery faults and thus failing to meet trainees' learning needs regarding fault development patterns. The teaching loop is incomplete, lacking a complete teaching chain encompassing fault simulation, algorithm verification, and effect quantification. Trainee-developed fault detection algorithms cannot be compared and verified against real faults simulated by the hardware, resulting in a lack of objective basis for teaching evaluation. Furthermore, insufficient communication reliability, with single communication links susceptible to electromagnetic interference, can lead to command loss and data transmission interruptions, affecting the continuity of the teaching process. Therefore, upgrading the virtual-real linkage control method for battery simulation teaching has become an urgent technical problem to be solved in the industry. Summary of the Invention

[0005] The present invention aims to provide a dual-mode communication battery simulation teaching virtual-real linkage control method to solve the core problem of asynchronous virtual and real data and large parameter deviation in existing battery simulation teaching equipment, which leads to poor teaching accuracy and practical effect.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A dual-mode communication battery simulation teaching virtual-real linkage control method includes: The system initialization steps include initializing and preparing the hardware devices, pairing them for communication, and configuring the software. The hardware devices include a battery simulator, a BMS controller, a mobile fault setter, an industrial computer, and a VR device. Communication pairing includes Bluetooth and CAN bus communication. Software configuration includes importing the fault parameter library into the algorithm verification module of the industrial computer and setting the CAN communication baud rate. The instruction issuance process involves selecting the target fault from the fault parameter library using the mobile fault setter and transmitting the instruction to the industrial control computer via Bluetooth. In the hardware response steps, after the industrial control computer parses the instruction, it sends a control signal to the battery simulator via the CAN bus. The simulator adjusts the corresponding channel to output a fault simulation signal, and the BMS controller collects hardware data and sends it back to the industrial control computer. In the software rendering step, the industrial control computer synchronizes the hardware data returned by the BMS controller to the 3D simulation module and the VR training module. The 3D simulation module highlights the fault-related components and displays the parameter deviations, while the VR training module outputs fault prompt information. In the data comparison step, the software automatically generates a comparison table between the virtual set value and the actual hardware measured value corresponding to the target fault, and presents the parameter deviation; The algorithm verification process involves injecting fault data through the algorithm verification module, writing a fault detection algorithm, running the fault detection algorithm in the software, comparing the algorithm output with the hardware-simulated fault type, and determining the effectiveness of the algorithm. The fault clearing process involves issuing a fault clearing command via a mobile fault setter, which then controls the industrial control computer to synchronously restore the hardware and software modules to normal operation.

[0007] The principles and advantages of this solution are as follows: In practical applications, dual-mode communication via Bluetooth and CAN bus enables layered transmission of commands and data, solving the problems of high latency and easy data loss in single communication links; software rendering and data comparison enable visualization of fault states and quantitative comparison of parameters, solving the problems of abstract and difficult-to-understand fault characteristics; the algorithm verification process establishes a teaching loop of theoretical design and practical verification, solving the problem of students lacking practical verification of algorithm design capabilities; and the full-process virtual-real linkage ensures a high degree of consistency between the virtual scene and the hardware state, solving the problem of existing equipment being disconnected from the virtual and real worlds and resulting in poor teaching effectiveness.

[0008] Preferably, as an improvement, the algorithm optimization step is also included. After the industrial control computer receives the collected hardware data, it calculates the dynamic variance of the battery data in real time, adjusts the process noise covariance and observation noise covariance, and then fuses the hardware collected data and virtual simulation data to make the parameter deviation meet the threshold. After the fusion is completed, the software rendering step is executed.

[0009] Technical benefits: It facilitates effective filtering of noise interference in hardware-acquired data, supplements the integrity of virtual simulation data, significantly reduces the deviation between virtual and real parameters, ensures the accuracy and reliability of data during the teaching process, and provides high-precision data support for software rendering, data comparison, and algorithm verification.

[0010] Preferably, as an improvement, the algorithm optimization step is implemented based on an LSTM multimodal data fusion model, including: The multimodal raw data acquisition sub-step obtains the hardware measured values ​​collected from the BMS controller and the theoretical calculated values ​​from the virtual simulation module, and performs data preprocessing. The feature fusion sub-step involves extracting modal features from the preprocessed data and then concatenating and fusing them to obtain fused features. The dynamic noise covariance adjustment sub-step calculates the dynamic variance of the fused features within the sliding window in real time, using the following formula:

[0011] Where N is the sliding window, As a feature of fusion, The mean of the fused features within the window; Preset process noise covariance Q and observation noise covariance R, when At that time, according to , Adjustment; when At that time, according to , Adjustment; The fusion calculation sub-step, based on the process noise covariance and observation noise covariance, fuses hardware-acquired data and virtual simulation data, and outputs the fused parameter values; the fusion formula is:

[0012] in, For hardware data acquisition, For virtual simulation data; The model output sub-step inputs the fused parameter values ​​into the classification layer and outputs the fault category.

[0013] Technical effect: By fully capturing the complex characteristics of battery faults through multimodal data fusion, and combining dynamic noise covariance adjustment to achieve adaptive data optimization, the accuracy of fault identification is improved while ensuring the real-time performance of parameter fusion, making it suitable for teaching scenarios with strong temporal characteristics and diverse features of battery faults.

[0014] Preferably, as an improvement, the training process of the multimodal data fusion model includes: using the system's built-in fault time series data, with 1000 samples for each type of fault, divided into training set, validation set, and test set in a 7:2:1 ratio; using a joint loss function of cross-entropy loss and mean squared error loss; using the Adam optimizer with a learning rate of 0.001, an epoch number of iterations of 100, a batch size of 32, and training until the loss fluctuation on the validation set is <0.001.

[0015] Technical benefits: It facilitates ensuring sufficient model training and strong generalization ability, avoids overfitting or underfitting, and enables the model to stably output high-precision recognition results and fusion parameters under different fault scenarios, providing a reliable algorithm benchmark for teaching and verification.

[0016] Preferably, as an improvement, the multimodal data fusion model is lightweighted by INT8 quantization using TensorRT and then deployed to an industrial control computer equipped with a GPU, supporting online fine-tuning of model parameters by importing new fault sample data via USB flash drive.

[0017] Technical benefits: It facilitates the reduction of model inference time, adapts to the hardware capabilities of industrial control computers, and supports iterative optimization of models with new fault types, thereby improving the system's adaptability and scalability to diverse teaching needs.

[0018] Preferably, as an improvement, the fault simulation signal supports fault gradual process simulation: based on the LSTM model, the time series prediction branch is extended, and the fault evolution rate is dynamically adjusted according to the parameter change rate fed back by the hardware, and the fault gradual process time is adjusted within the range of 5 to 10 seconds.

[0019] Technical effect: It closely matches the development pattern of real battery failures.

[0020] Preferably, as an improvement, the fault gradient process simulation includes: The model expansion and initialization configuration sub-step adds an independent fault evolution prediction branch to the LSTM-based multimodal data fusion model. This branch shares preprocessed multimodal time series data with the main model, while simultaneously inputting target fault parameters and initial hardware state parameters, and outputting the parameter change per second. In the software configuration stage of system initialization, a 5-10 second adjustable fault transition time interval is preset, which supports manual selection of the transition time through a mobile fault setter or industrial control computer interface, and the calculation of the initial rate. In the dynamic rate adjustment sub-step, the industrial control computer receives hardware parameters collected by the BMS controller via the CAN bus, calculates the actual parameter change rate, inputs the actual parameter change rate, the currently collected parameter value and the remaining gradual change time into the fault evolution prediction branch, outputs the corrected rate, and sends a dynamic adjustment command to the battery simulator according to the corrected rate. The gradual termination and state synchronization sub-steps terminate the gradual process when the deviation between the hardware parameter value and the target fault parameter value is ≤0.01V, or when the actual gradual process takes up the set time. The industrial control computer sends a command to lock the current parameters to the battery simulator, and at the same time synchronizes the final fault parameters and the gradual process curve to the 3D simulation module and the VR training module. The 3D module highlights the faulty cell and marks the gradual completion information, and the VR module outputs the corresponding fault prompt.

[0021] Technical benefits: It facilitates the simulation of the complete evolution process of a fault from its inception to its formation, allowing trainees to intuitively observe the development pattern of faults and deepen their understanding of the dynamic characteristics of battery faults; dynamic rate adjustment ensures that the gradual process is stable and controllable, avoiding teaching misleading effects caused by sudden parameter changes, and enhancing the authenticity and professionalism of the training.

[0022] Preferably, as an improvement, maintenance procedures are also included, such as periodically checking the CAN bus and Bluetooth connection status, calibrating the battery simulator voltage accuracy, and importing update packages via FAT32 formatted USB flash drives for software updates.

[0023] Technical benefits: It facilitates the communication stability and parameter accuracy of hardware devices, ensuring long-term stable operation of the system. Attached Figure Description

[0024] Figure 1This is a flowchart illustrating a virtual-real linkage control method for dual-mode communication battery simulation teaching. Detailed Implementation

[0025] The following detailed description illustrates the specific implementation method: Example 1 The basic implementation examples are as follows: Figure 1 As shown: A dual-mode communication battery simulation teaching virtual-real linkage control method includes system initialization, command issuance, hardware response, software rendering, data comparison, algorithm verification, and fault clearing steps.

[0026] The system initialization steps complete the initialization preparation of hardware devices, communication pairing, and software configuration running on the industrial control computer.

[0027] The hardware equipment includes a battery simulator, BMS controller, mobile fault setter, industrial computer and VR equipment; the purpose of the initialization preparation of the hardware equipment is to complete the power supply, startup and physical status check of the hardware equipment, and lay the foundation for communication pairing and functional operation. Specifically, it includes: (1) Power supply configuration of the training platform: connect the training platform to 220V AC mains power, confirm that the power supply wiring is firm, pull the emergency stop switch on the platform, and put it in a non-triggering state to ensure operation safety; (2) Startup of the industrial computer: turn on the power supply of the industrial computer host and the matching monitor. After the operating system (such as Windows 10) has started, double-click to run the "battery pack system.exe" program and the "start-web" program on the desktop to start the system server and ensure that the core control program runs in the background; (3) VR equipment preparation: press and hold the VR headset power button until the indicator light is constantly white to confirm that the VR headset has started successfully; install two 1.5V batteries in the VR controller. AA battery, turn off the power of the handle and wait for subsequent pairing; (4) Mobile fault setter preparation: turn on the mobile fault setter (such as a tablet device), ensure that the device has sufficient power, enter the system settings interface to search for the corresponding wireless signal (WiFi) of the industrial control computer, complete the connection, and ensure that it is on the same local area network as the industrial control computer; (5) Hardware status check: observe whether the power indicator lights of the battery simulator and BMS controller are lit normally, and confirm that the physical connection of each hardware module is not loose (such as CAN bus wiring, sensor connection line).

[0028] Communication pairing is used to realize the communication connection between each hardware device and the industrial control computer, and to ensure smooth transmission of instructions and data. Specifically, it includes: (1) Bluetooth pairing between VR device and industrial control computer: turn on the Bluetooth function of industrial control computer and search for surrounding Bluetooth devices; press and hold the power button of VR controller to enter Bluetooth pairing mode; find the device name corresponding to VR controller in the Bluetooth device list of industrial control computer, click to pair, until pairing is successful; (2) Communication confirmation between mobile fault setter and industrial control computer: after the mobile fault setter connects to the wireless signal of industrial control computer, open the "communication test" function on the device interface and send test instructions (such as "connection request"); after the industrial control computer server receives the instruction, it will send back the "connection successful" signal, and the mobile fault setter interface will show that the communication status is normal, and the pairing is completed; (3) CAN bus communication pre-verification: confirm that the physical connection of CAN bus between battery simulator, BMS controller and industrial control computer is normal; after the industrial control computer starts, it will automatically detect the CAN bus interface status, and the software interface will show "CAN bus connected", and the CAN communication link preparation is completed.

[0029] The purpose of software configuration is to ensure that the software modules are compatible with the hardware device protocols and that fault data is available. Specifically, this includes: (1) Importing fault parameter library: In the algorithm verification module interface of the industrial control computer, find the parameter library management function entry; select to import the 27 types of teaching fault parameter library preset by the system, which includes parameter data (such as fault threshold and characteristic parameters) for fault types such as low SOH, abnormal temperature, poor consistency of single cell voltage, and overvoltage / undervoltage of single cell battery; (2) Configuring communication protocol parameters: In the communication configuration interface of the industrial control computer, set the CAN communication baud rate to 500kbps, which is consistent with the CAN communication parameters of the hardware device; confirm the CAN 2.0 optimization protocol has been enabled to ensure that the communication protocol of the software is compatible with that of the battery simulator and BMS controller, and to avoid data transmission format mismatch; (3) Software function self-check: Start the 3D simulation module (Unity engine) to confirm that the battery pack 3D model can be loaded normally and supports component rotation, scaling, and exploded view switching; Start the VR training module to confirm that the VR headset can receive the virtual scene data transmitted by the industrial control computer normally, and the screen is smooth and without delay; Test the Python editing function and data injection function in the algorithm verification module to ensure that the code editing area can be input normally and the dataset can be imported; (4) Initialization status confirmation: The software interface displays status prompts such as normal hardware connection, complete parameter library import, and communication protocol matching; The 3D simulation module displays the normal battery status, and the VR scene displays that the training system is ready and the software configuration is complete.

[0030] The instruction issuance process involves selecting and confirming the target fault using a mobile fault setter. The target fault is chosen from a pre-imported fault parameter library. The standardized fault instruction is transmitted to the industrial control computer via a paired Bluetooth link. CRC verification is used during transmission to ensure data integrity and prevent instruction loss. The industrial control computer's Bluetooth module monitors the received signal in real time. Upon successful instruction reception, the built-in parsing program decodes the instruction data and extracts key information such as the fault type and target parameters to prepare for subsequent hardware response steps.

[0031] The hardware response steps involve the industrial control computer parsing the instruction. The instruction contains information such as the fault type code, target parameter threshold, and channel identifier. For example, "poor single-cell voltage consistency" corresponds to code 003, target voltage 3.0V, and channel 10 (the 10th cell). The industrial control computer's built-in instruction parsing program decodes the instruction. Specifically: it identifies the fault type, matches the preset fault parameters in the fault simulation module according to the code (e.g., code 003 corresponds to poor single-cell voltage consistency), and calls the hardware adjustment rules for that fault. Key parameters are extracted, determining the target adjustment channel (e.g., the 10th channel in a 96-cell battery simulator), the target parameter value (e.g., voltage 3.0V), and the adjustment accuracy requirement (e.g., deviation ≤ ±1mV). A control signal is generated, converting the parsed parameters into a digital control signal recognizable by the battery simulator (e.g., using the frame format specified by the CAN 2.0 optimized protocol, including address, data, and check segments). The data segment clearly states "channel 10, voltage 3.0V, adjustment mode: gradual / instantaneous."

[0032] Control signals are sent to the battery simulator via the CAN bus. The simulator adjusts the corresponding channel to output fault simulation signals. The BMS controller collects hardware data and sends it back to the industrial control computer. The collected hardware data includes: (1) Voltage acquisition: The real-time voltage value of each channel is collected by connecting the 96 channel output terminals of the battery simulator through the sampling line, including the target fault channel and other normal channels; (2) Temperature acquisition: The data of the 24-point cell temperature sensor built into the battery simulator and the ambient temperature data are collected. If it is an "over-temperature" fault, the temperature signal ≥60℃ output by the simulator is collected; (3) Current acquisition: The real-time current value of the battery simulator under the simulated charging and discharging state is collected, such as 0.5A; (4) Data preprocessing: The BMS controller filters the collected raw data, removes high-frequency noise, and converts it into standardized digital data, such as voltage unit V and temperature unit ℃.

[0033] The BMS controller transmits the collected data back to the industrial computer via the CAN bus (which shares the same bus with the control signal transmission and is time-multiplexed): (1) Data packaging: The collected voltage, temperature and current data are packaged into data frames in the format of "channel number-parameter value-collection timestamp". Each frame contains parameter data of 16 channels and is transmitted in multiple frames (96 voltage data are transmitted in 6 frames); (2) Real-time transmission: The transmission is started immediately after the collection is completed. The transmission time for a single frame is ≤3ms, and the transmission time for the full transmission of 96 voltage + temperature + current data is ≤8ms; (3) After the industrial computer receives the data, it aligns the timestamp with the control signal issuance time to ensure the correlation between the data and the fault command, and provides accurate hardware measurement data for subsequent software rendering and data comparison.

[0034] In the software rendering process, the industrial control computer synchronizes the hardware data returned by the BMS controller to the 3D simulation module and the VR training module. The 3D simulation module highlights the fault-related components and displays the parameter deviations, while the VR training module outputs fault prompt information.

[0035] The data comparison step automatically generates a comparison table between the virtual setpoints and the actual hardware measurements corresponding to the target fault, and presents the parameter deviations. The virtual setpoints are the input targets for fault simulation, stored in the fault simulation module's 27-category fault parameter library. For example, when the fault type is poor single-cell voltage consistency, virtual setpoint examples include target fault channel voltage (e.g., 3.0V for the 10th cell), normal channel voltage (e.g., 3.3V for the other 95 cells), and voltage deviation threshold (e.g., >5%). When the fault type is battery overheating, virtual setpoint examples include target cell temperature (e.g., 60℃), ambient temperature (e.g., 35℃), and temperature rise rate (e.g., 5℃ / min). When the fault type is low SOH, virtual setpoints are used for other fault types. Examples of proposed setpoints include battery health (e.g., 75%), rated capacity (e.g., 100Ah), and actual simulated capacity (e.g., 75Ah); for fault types such as single-cell overvoltage / undervoltage, examples of virtual setpoints include overvoltage setpoint (e.g., 4.3V), undervoltage setpoint (e.g., 2.4V), and duration (e.g., 30s); for fault types such as abnormal capacity consistency, examples of virtual setpoints include capacity values ​​for each channel (e.g., 90Ah for section 5, 100Ah for other channels), and capacity deviation threshold (e.g., >10%); for fault types such as abnormal current, examples of virtual setpoints include charge / discharge current setpoint (e.g., discharge current 2A, normal 1A), and current fluctuation range (e.g., ±0.3A).

[0036] The algorithm verification process involves injecting fault data through the algorithm verification module, writing a fault detection algorithm, running the fault detection algorithm in the software, comparing the algorithm output with the hardware-simulated fault type, and determining the effectiveness of the algorithm.

[0037] The fault clearing process involves issuing a fault clearing command via a mobile fault setter, which then controls the industrial control computer to synchronously restore the hardware and software modules to normal operation.

[0038] It also includes maintenance steps, regularly checking the CAN bus and Bluetooth connection status, calibrating the battery simulator voltage accuracy, and importing update packages via FAT32 formatted USB flash drive for software updates.

[0039] Example 2 The algorithm also includes an optimization step. After the industrial control computer receives the collected hardware data, it calculates the dynamic variance of the battery data in real time, adjusts the process noise covariance and observation noise covariance, and then fuses the hardware-acquired data and virtual simulation data to ensure that the parameter deviation is ≤0.5%. After fusion, the software rendering step is executed. The algorithm optimization step is implemented based on the LSTM multimodal data fusion model and includes: The multimodal raw data acquisition sub-step obtains the hardware measured values ​​from the BMS controller and the theoretical calculation values ​​from the virtual simulation module, and performs data preprocessing. The acquired values ​​include four types of core input modal data, of which voltage data (96 individual cell voltages + total voltage), current data (charging / discharging current), and temperature data (24-point cell temperature + ambient temperature) are obtained from the hardware measured values ​​from the BMS controller, and SOC change data (real-time SOC value + rate of change) are obtained from the theoretical calculation values ​​from the virtual simulation module. The sampling frequencies for each modal data are 10Hz for voltage, 10Hz for current, 5Hz for temperature, and 1Hz for SOC change.

[0040] The feature fusion sub-step involves extracting modal features from the preprocessed data and then concatenating and fusing them to obtain fused features. Specifically, two LSTM layers are set up as voltage / current LSTM subnetworks, with 64 neurons in the first layer and 32 neurons in the second layer, using the tanh activation function to output a 32-dimensional electrical time-series feature vector; two LSTM layers are set up as temperature LSTM subnetworks, with 32 neurons in the first layer and 16 neurons in the second layer, outputting a 16-dimensional temperature time-series feature vector; two fully connected layers are set up as SOC change fully connected layers, with 16 neurons in the first layer and 8 neurons in the second layer, using the ReLU activation function to output an 8-dimensional SOC feature vector; the above three types of feature vectors are concatenated into a 56-dimensional unified feature vector, which is then input into a fully connected fusion layer with 64 neurons, using the ReLU activation function. The fusion formula is as follows:

[0041] Where W is a 56*64 weight matrix and b is a bias term.

[0042] The dynamic noise covariance adjustment sub-step calculates the dynamic variance of the fused features within a sliding window (10 time steps) in real time, using the following formula:

[0043] Where N is the sliding window, As a feature of fusion, The mean of the fused features within the window; The process noise covariance Q (value 0.01) and observation noise covariance R (value 0.1) are preset. At that time, according to , Adjustment; when At that time, according to , Adjustment.

[0044] The fusion calculation sub-step, based on the process noise covariance and observation noise covariance, fuses hardware-acquired data and virtual simulation data, and outputs the fused parameter values; the fusion formula is:

[0045] in, For hardware data acquisition, For virtual simulation data; The model output sub-step inputs the 64-dimensional fused feature vector into the classification layer containing the Softmax activation function, outputs the probability distribution of 27 types of faults, and takes the category corresponding to the maximum probability as the recognition result; outputs the fused accurate parameter values ​​to ensure that the parameter deviation is ≤0.5%.

[0046] The training process of the multimodal data fusion model includes: using 27 types of fault time-series data built into the system, with 1000 samples for each type of fault, divided into training set, validation set, and test set in a 7:2:1 ratio; the loss function adopts a joint loss function of cross-entropy loss and mean squared error loss, and the function model is as follows:

[0047] in, For cross-entropy loss, The mean squared error loss is used. The Adam optimizer is employed with a learning rate of 0.001, an epoch count of 100, a batch size of 32, and the loss fluctuation on the training to validation set is <0.001.

[0048] The multimodal data fusion model is then processed by TensorRT with INT8 quantization for lightweighting and deployed to an industrial control computer equipped with a GPU. It supports online fine-tuning of model parameters by importing new fault sample data via USB flash drive.

[0049] The fault simulation signal supports fault gradient process simulation: based on an LSTM model with an extended time series prediction branch, the fault evolution rate is dynamically adjusted according to the parameter change rate fed back by the hardware, and the fault gradient time is adjusted within the range of 5-10 seconds. The fault gradient process simulation includes: The model expansion and initialization configuration sub-step adds an independent fault evolution prediction branch to the LSTM-based multimodal data fusion model. This branch includes two LSTM layers (48 neurons in the first layer and 24 neurons in the second layer) and one fully connected output layer (16 neurons, with ReLU activation function). It shares preprocessed multimodal time-series data with the main model, and simultaneously inputs target fault parameters and initial hardware state parameters, outputting the parameter change per second. In the system initialization software configuration stage, a 5-10 second adjustable fault gradient duration interval is preset. Manual selection of the gradient duration is supported via a mobile fault setter or industrial control computer interface, and the initial rate is calculated using the following formula:

[0050] in, For the target fault parameter value, These are the initial hardware parameter values. Select the gradient duration.

[0051] In the dynamic rate adjustment sub-step, the industrial control computer receives hardware parameters collected by the BMS controller every 100ms via the CAN bus and calculates the actual parameter change rate. The calculation formula is as follows:

[0052] in, The currently collected parameter value. The parameter value collected last time. The time interval for data collection is, for example, 0.1 seconds.

[0053] Rate of change of actual parameters The currently collected parameter values and remaining gradient time (for (Subtract the time already elapsed) Input the fault evolution prediction branch, output the corrected rate. : like ,but ; like ,but ; like V / second, then maintain ; The industrial control computer is based on the corrected speed. A dynamic adjustment command is sent to the battery simulator every 100ms.

[0054] The gradual termination and state synchronization sub-steps occur when the hardware parameter values... With the target fault parameter value The deviation is ≤0.01V, or the actual gradual change time reaches the set time. At that time, the gradual change process terminates; the industrial control computer sends a command to lock the current parameters to the battery simulator, and at the same time synchronizes the final fault parameters and the gradual change process curve to the 3D simulation module and the VR training module. The 3D module highlights the faulty cell and marks the gradual change completion information, and the VR module outputs the corresponding fault prompt.

[0055] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A virtual-real linkage control method for dual-mode communication battery simulation teaching, characterized in that, include: The system initialization steps include initializing and preparing the hardware devices, pairing them for communication, and configuring the software. The hardware devices include a battery simulator, a BMS controller, a mobile fault setter, an industrial computer, and a VR device. Communication pairing includes Bluetooth and CAN bus communication. Software configuration includes importing the fault parameter library into the algorithm verification module of the industrial computer and setting the CAN communication baud rate. The instruction issuance process involves selecting the target fault from the fault parameter library using the mobile fault setter and transmitting the instruction to the industrial control computer via Bluetooth. In the hardware response steps, after the industrial control computer parses the instruction, it sends a control signal to the battery simulator via the CAN bus. The simulator adjusts the corresponding channel to output a fault simulation signal, and the BMS controller collects hardware data and sends it back to the industrial control computer. In the software rendering step, the industrial control computer synchronizes the hardware data returned by the BMS controller to the 3D simulation module and the VR training module. The 3D simulation module highlights the fault-related components and displays the parameter deviations, while the VR training module outputs fault prompt information. In the data comparison step, the software automatically generates a comparison table between the virtual set value and the actual hardware measured value corresponding to the target fault, and presents the parameter deviation; The algorithm verification process involves injecting fault data through the algorithm verification module, writing a fault detection algorithm, running the fault detection algorithm in the software, comparing the algorithm output with the hardware-simulated fault type, and judging the effectiveness of the algorithm. The fault clearing process involves issuing a fault clearing command via a mobile fault setter, which then controls the industrial control computer to synchronously restore the hardware and software modules to normal operation.

2. The virtual-real linkage control method for dual-mode communication battery simulation teaching according to claim 1, characterized in that: It also includes an algorithm optimization step. After the industrial control computer receives the collected hardware data, it calculates the dynamic variance of the battery data in real time, adjusts the process noise covariance and observation noise covariance, and then fuses the hardware-collected data and virtual simulation data to make the parameter deviation meet the threshold. After the fusion is completed, the software rendering step is executed.

3. The dual-mode communication battery simulation teaching virtual-real linkage control method according to claim 2, characterized in that, The algorithm optimization steps are implemented based on the LSTM multimodal data fusion model, including: The multimodal raw data acquisition sub-step obtains the hardware measured values ​​collected from the BMS controller and the theoretical calculated values ​​from the virtual simulation module, and performs data preprocessing. The feature fusion sub-step involves extracting modal features from the preprocessed data and then concatenating and fusing them to obtain fused features. The dynamic noise covariance adjustment sub-step calculates the dynamic variance of the fused features within the sliding window in real time, using the following formula: Where N is the sliding window, As a feature of fusion, The mean of the fused features within the window; Preset process noise covariance Q and observation noise covariance R, when At that time, according to , Adjustment; when At that time, according to , Adjustment; The fusion calculation sub-step, based on the process noise covariance and observation noise covariance, fuses hardware-acquired data and virtual simulation data, outputting fused parameter values; the fusion formula is: in, For hardware data acquisition, For virtual simulation data; The model output sub-step inputs the fused parameter values ​​into the classification layer and outputs the fault category.

4. The dual-mode communication battery simulation teaching virtual-real linkage control method according to claim 3, characterized in that, The training process of the multimodal data fusion model includes: using the system's built-in fault time series data, with 1000 samples for each type of fault, divided into training set, validation set, and test set in a 7:2:1 ratio; using a joint loss function of cross-entropy loss and mean squared error loss; employing the Adam optimizer with a learning rate of 0.001, 100 epochs, and a batch size of 32, training until the loss fluctuation on the validation set is <0.

001.

5. The virtual-real linkage control method for dual-mode communication battery simulation teaching according to claim 3, characterized in that: The multimodal data fusion model is then processed by TensorRT with INT8 quantization for lightweighting and deployed to an industrial control computer equipped with a GPU. It supports online fine-tuning of model parameters by importing new fault sample data via USB flash drive.

6. The virtual-real linkage control method for dual-mode communication battery simulation teaching according to claim 3, characterized in that: The fault simulation signal supports fault gradual process simulation: based on the LSTM model, the time series prediction branch is extended, and the fault evolution rate is dynamically adjusted according to the parameter change rate fed back by the hardware. The fault gradual process time is adjusted within the range of 5 to 10 seconds.

7. The dual-mode communication battery simulation teaching virtual-real linkage control method according to claim 6, characterized in that, The fault gradual change process simulation includes: The model expansion and initialization configuration sub-step adds an independent fault evolution prediction branch to the LSTM-based multimodal data fusion model. This branch shares preprocessed multimodal time series data with the main model, while simultaneously inputting target fault parameters and initial hardware state parameters, and outputting the parameter change per second. In the software configuration stage of system initialization, a 5-10 second adjustable fault transition time interval is preset, which supports manual selection of the transition time through a mobile fault setter or industrial control computer interface, and the calculation of the initial rate. In the dynamic rate adjustment sub-step, the industrial control computer receives hardware parameters collected by the BMS controller via the CAN bus, calculates the actual parameter change rate, inputs the actual parameter change rate, the currently collected parameter value and the remaining gradual change time into the fault evolution prediction branch, outputs the corrected rate, and sends a dynamic adjustment command to the battery simulator according to the corrected rate. The gradual termination and state synchronization sub-steps terminate the gradual process when the deviation between the hardware parameter value and the target fault parameter value is ≤0.01V, or when the actual gradual process takes up the set time. The industrial control computer sends a command to lock the current parameters to the battery simulator, and at the same time synchronizes the final fault parameters and the gradual process curve to the 3D simulation module and the VR training module. The 3D module highlights the faulty cell and marks the gradual completion information, and the VR module outputs the corresponding fault prompt.

8. The virtual-real linkage control method for dual-mode communication battery simulation teaching according to claim 1, characterized in that: It also includes maintenance steps, regularly checking the CAN bus and Bluetooth connection status, calibrating the battery simulator voltage accuracy, and importing update packages via FAT32 formatted USB flash drive for software updates.

Images