Digital Twin-based Industry-Education Collaborative Training Method and System

Abstract

The present invention provides a production-education collaborative training method and system based on digital twin. First, a real-time operation data set of industrial equipment and a training case data set of the production-education collaborative knowledge base are obtained. Then, a digital twin of the industrial equipment is constructed based on the real-time operation data. A large language model is used to parse the training case data to generate a multi-modal training interaction feature set. A collaborative feature mapping is performed on the digital twin and the interaction feature set to obtain a virtual-real fusion training scenario feature set. Then, a preset training strategy generation model is used to dynamically optimize the virtual-real fusion training scenario feature set to generate a personalized production-education collaborative training plan. Finally, the personalized production-education collaborative training plan is deployed to the digital twin to trigger real-time interactive training, and the production-education collaborative knowledge base is updated according to the operation feedback data of the trainees, realizing efficient and personalized production-education collaborative training and improving the training effect.

Description

Technical Field

[0001] The present invention relates to the field of digital twin technology, and in particular, to a method and system for industry-education collaborative training based on digital twin. Background Art

[0002] In the current field of industrial education and training, with the rapid development of Industry 4.0 and intelligent manufacturing, the demand for high-skilled talents is increasing day by day. The traditional industry-education collaborative training mode has been difficult to meet the requirements of modern industry for the efficiency, accuracy, and innovation of talent cultivation. Traditional training methods often rely on the direct operation of physical equipment, which is not only limited by factors such as the number of equipment, maintenance costs, and safety, but also difficult to provide personalized and dynamically adjustable teaching plans, resulting in uneven training effects and unable to fully stimulate the learning potential and innovation ability of trainees. Summary of the Invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a method for industry-education collaborative training based on digital twin, the method comprising:

[0004] Obtain a set of real-time operation data of industrial equipment and a set of training case data of an industry-education collaborative knowledge base, the real-time operation data set includes equipment status parameters, operation instruction sequences, and abnormal event logs, and the training case data set includes historical training task descriptions, skill operation standards, and trainee evaluation indicators;

[0005] Construct a digital twin of industrial equipment based on the real-time operation data set, and call a large language model to perform semantic association parsing processing on the training case data set to generate a multi-modal training interaction feature set;

[0006] Perform collaborative feature mapping processing on the digital twin of industrial equipment and the multi-modal training interaction feature set to obtain a virtual-real fusion training scenario feature set, the virtual-real fusion training scenario feature set includes equipment operation mapping relationships, skill execution paths, and abnormal handling strategies;

[0007] Perform dynamic optimization processing on the virtual-real fusion training scenario feature set based on a preset training strategy generation model to generate a personalized industry-education collaborative training plan, the personalized industry-education collaborative training plan includes a training task sequence, skill evaluation nodes, and a dynamic feedback mechanism;

[0008] Deploy the personalized industry-education collaborative training plan to the digital twin of industrial equipment and trigger real-time interactive training, and update the industry-education collaborative knowledge base based on trainee operation feedback data.

[0009] In another aspect, an embodiment of the present invention further provides a production-education collaborative training system based on digital twin, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0010] Based on the above aspects, the embodiment of the present invention realizes the deep integration and efficient utilization of the real-time operation data of industrial equipment and the training case data of the production-education collaborative knowledge base, significantly improving the intelligent and personalized level of training teaching. Specifically, by acquiring and integrating the real-time operation data set of industrial equipment and the training case data set, on this basis, using digital twin technology to build a high-precision virtual model of industrial equipment, and combining the semantic association parsing ability of the large language model, the training case data is transformed into a multi-modal training interaction feature set, making the training content more vivid and intuitive, effectively enhancing the learning experience and participation of trainees. Further, through collaborative feature mapping processing, the digital twin of industrial equipment is organically combined with the multi-modal training interaction feature set to generate a virtual-real fusion training scenario feature set. This virtual-real fusion training scenario feature set not only contains key information such as equipment operation mapping relationships and skill execution paths, but also incorporates abnormal handling strategies, providing comprehensive and systematic training guidance for trainees. More prominently, through the preset training strategy generation model to dynamically optimize the virtual-real fusion training scenario feature set, a personalized production-education collaborative training plan can be generated. This personalized production-education collaborative training plan is customized according to the actual situation and needs of trainees, realizing the serialization of training tasks, the nodalization of skill assessment, and the establishment of a dynamic feedback mechanism, thus greatly improving the flexibility and effectiveness of training teaching. Finally, deploying the personalized production-education collaborative training plan to the digital twin of industrial equipment and triggering real-time interactive training not only realizes the automation and intelligence of the training process, but also can continuously update the production-education collaborative knowledge base based on the operation feedback data of trainees, forming a closed-loop optimization mechanism and continuously promoting the improvement of training teaching quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic execution flow diagram of the production-education collaborative training method based on digital twin provided by the embodiment of the present invention.

[0012] Figure 2 It is a schematic diagram of exemplary hardware and software components of the production-education collaborative training system based on digital twin provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1It is a schematic flowchart of a production-teaching collaborative training method based on digital twin provided by an embodiment of the present invention. The production-teaching collaborative training method based on digital twin will be introduced in detail below.

[0014] Step S110: Obtain the real-time operation data set of industrial equipment and the training case data set of the production-teaching collaborative knowledge base. The real-time operation data set includes equipment status parameters, operation instruction sequences, and abnormal event logs, and the training case data set includes historical training task descriptions, skill operation standards, and trainee evaluation indicators.

[0015] In the actual scenario of industrial production, take the automated welding equipment of a certain automobile manufacturing factory as an example to illustrate this step in detail. This automated welding equipment is a key link in the entire automobile production line and is responsible for the precise welding of various parts of the vehicle body.

[0016] Among them, in the process of obtaining the real-time operation data set, for the equipment status parameters, the status parameters of the automated welding equipment are important bases for reflecting its current operating conditions, and these parameters can be collected through sensors distributed at various key parts of the automated welding equipment. For example, a temperature sensor will continuously monitor the temperature of the welding head to ensure that it is within the appropriate working temperature range. Suppose at a certain moment, the temperature sensor of the welding head shows a temperature of 250 degrees Celsius. A pressure sensor will measure the pressure applied during welding to ensure the stability of welding quality. At this time, the pressure sensor shows a pressure of 500 Newtons. In addition, a position sensor will record the precise position coordinates of the welding head, such as the coordinates in three-dimensional space being (100, 200, 300) millimeters. Combine these different types of parameters in a certain order to form a multi-dimensional equipment status parameter vector. In this example, the equipment status parameter vector can be expressed as (250, 500, 100, 200, 300). As time goes by, continuously collect these parameters to form a sequence of equipment status parameters containing multiple time points.

[0017] For the operation instruction sequence, it records a series of instructions issued by the operator to the automated welding equipment, and these instructions are usually executed in chronological order. For example, the operator first issues the instruction "Move the welding head to the coordinates (150, 250, 350) millimeters", then issues the instruction "Start welding, and the welding time is 10 seconds", and then issues the instruction "Stop welding and return the welding head to the initial position". Arrange these instructions in the order of execution to obtain the operation instruction sequence. These instructions can be stored in the form of text description, such as "Move to (150, 250, 350); Weld for 10 seconds; Return to the initial position".

[0018] For the abnormal event log, it is used to record the abnormal situations that occur during the operation of the device. When an abnormality occurs in the device, the corresponding monitoring system automatically records the time of the abnormality occurrence, the type of the abnormality, and the possible causes. For example, at 10:30 am on a certain day, the motor of the welding device overheats abnormally. Then, this time point is recorded, and at the same time, the type of the abnormality is determined as "motor overheating", and the possible cause is speculated to be that the motor has been working continuously for a long time and has poor heat dissipation. Thus, these abnormal information are sorted into an abnormal event log in chronological order.

[0019] During the acquisition process of the training case data set in the production-education collaborative knowledge base, for the historical training task description, it details the content of the training tasks carried out on this automated welding device in the past. For example, there is a training task that requires trainees to complete the welding work of a specific part of an automobile body within a specified time, and the width of the weld should be between 3 mm and 5 mm, and the welding strength should reach a certain standard. The training task description will clearly indicate information such as the specific welding position, the welding process requirements, and the time limit for completing the task.

[0020] For the skill operation standards, the skill operation standards are the criteria for standardizing the operation behaviors of trainees. For the automated welding device, the skill operation standards include the correct holding method of the welding head, the setting range of welding parameters, the control of welding speed, etc. For example, the holding angle of the welding head should be between 45 degrees and 60 degrees, the welding current should be set between 200 amperes and 300 amperes, and the welding speed should be maintained between 100 mm and 150 mm per minute. These standards are obtained through long-term practice and experience summary, and are the key to ensuring welding quality.

[0021] For the trainee evaluation indicators, they are used to measure the performance of trainees during the training process. These indicators can be evaluated from multiple dimensions, including the accuracy of operation, the time to complete the task, the ability to handle abnormal situations, etc. For example, the accuracy of operation can be evaluated by measuring the deviation between the width of the weld and the standard width, the time to complete the task can be determined by recording the actual time used by the trainee from the start of operation to the completion of the task, and the ability to handle abnormal situations can be evaluated according to whether the measures taken by the trainee when encountering an abnormality are correct and timely. In order to more accurately evaluate the performance of trainees, corresponding weights can be set for each evaluation indicator. For example, the weight of the accuracy of operation is 0.4, the weight of the time to complete the task is 0.3, and the weight of the ability to handle abnormal situations is 0.3.

[0022] Step S120: Construct a digital twin of the industrial device based on the real-time operation data set, and call a large language model to perform semantic association parsing and processing on the training case data set to generate a multi-modal training interaction feature set.

[0023] Step S121: Perform temporal alignment processing on the device status parameters to generate a device operation status temporal vector, and perform instruction decomposition processing on the operation instruction sequence to generate an atomic operation step instruction set.

[0024] In this embodiment, for the temporal alignment processing of device status parameters, since different sensors may have different sampling frequencies, the device status parameters collected may be inconsistent in time. To accurately reflect the operation status of the device, it is necessary to perform temporal alignment processing on these parameters. Taking an automated welding device as an example, the sampling frequency of the temperature sensor is 1 time per second, while the sampling frequency of the pressure sensor is 2 times per second. During a certain period of time, the temperature data sequence collected by the temperature sensor is (250, 252, 255), and the pressure data sequence collected by the pressure sensor is (500, 502, 505, 508, 510, 512). These data are aligned by the method of timestamp matching. Assuming that the timestamp is in seconds, the timestamps corresponding to the temperature data are (0, 1, 2), and the timestamps corresponding to the pressure data are (0, 0.5, 1, 1.5, 2, 2.5). When performing alignment, for the temperature data, find the pressure data value closest to it at each time point. For example, at time point 0, the temperature is 250, and the corresponding pressure value is 500; at time point 1, the temperature is 252, and the corresponding pressure value is 505. Combine the aligned temperature and pressure data to form a device operation status temporal vector. In this example, the device operation status temporal vector can be expressed as ((250, 500), (252, 505), (255, 510)).

[0025] For the instruction decomposition process of the operation instruction sequence, the operation instruction sequence is usually composed of a series of instructions. To facilitate the simulation and control of the digital twin, these instructions need to be decomposed into an atomic instruction set. Taking the operation instruction sequence of an automated welding device "Move the welding head to the coordinates (150, 250, 350) millimeters, start welding, the welding time is 10 seconds, stop welding, and return the welding head to the initial position" as an example, the instruction decomposition is carried out. First, decompose "Move the welding head to the coordinates (150, 250, 350) millimeters" into three atomic instructions: "Move the X-axis to 150 millimeters", "Move the Y-axis to 250 millimeters", and "Move the Z-axis to 350 millimeters". "Start welding" and "Stop welding" are used as separate atomic instructions respectively. "The welding time is 10 seconds" can be decomposed into two atomic instructions: "Start timing" and "Stop welding when the timing reaches 10 seconds". "Return the welding head to the initial position" can be decomposed into three atomic instructions: "Move the X-axis to the initial X coordinate", "Move the Y-axis to the initial Y coordinate", and "Move the Z-axis to the initial Z coordinate". The final atomic instruction set of the operation steps obtained is "Move the X-axis to 150 millimeters; Move the Y-axis to 250 millimeters; Move the Z-axis to 350 millimeters; Start welding; Start timing; Stop welding when the timing reaches 10 seconds; Move the X-axis to the initial X coordinate; Move the Y-axis to the initial Y coordinate; Move the Z-axis to the initial Z coordinate".

[0026] Step S122: Construct an industrial equipment digital twin based on the device physical model and the device operation status time series vector, and map the atomic instruction set of the operation steps to the operation interface of the industrial equipment digital twin.

[0027] In this embodiment, the device physical model is a mathematical model established according to the actual physical structure and working principle of the automated welding device, which takes into account multiple factors such as the mechanical structure, electrical characteristics, and heat conduction of the device. For example, for the movement of the welding head, the model will consider factors such as the driving force of the motor, the efficiency of the transmission mechanism, and the friction force. By performing numerical simulations on the device physical model, the operating conditions of the device under different states can be predicted.

[0028] Combined with the previously generated device operation status time series vector, construct an industrial equipment digital twin. The digital twin is a virtual mapping of the device and can reflect the actual operating status of the device in real time. For example, when the temperature parameter in the device operation status time series vector increases, the temperature of the welding head in the digital twin will also increase accordingly. Mapping the atomic instruction set of the operation steps to the operation interface of the industrial equipment digital twin means that when an atomic instruction is input to the operation interface of the digital twin, the digital twin can simulate the corresponding operation actions. For example, when the instruction "Move the X-axis to 150 millimeters" is input, the digital twin can calculate the movement trajectory and final position of the welding head under this instruction according to the device physical model.

[0029] Step S123: Invoke a large language model to perform intent recognition processing on the historical training task description, extract task objective features and skill requirement features, and perform semantic enhancement processing on the skill operation standards to generate standard operation path features.

[0030] In this embodiment, taking the historical training task description of an automated welding device "complete the welding work on a specific part of the car body within a specified time, and require the weld width to be between 3 mm and 5 mm, and the welding strength to reach a certain standard" as an example, the large language model extracts task objective features and skill requirement features through semantic analysis of the text. The task objective features include "complete the welding work within the specified time", and the skill requirement features include "the weld width is between 3 mm and 5 mm" and "the welding strength reaches a certain standard".

[0031] For the skill operation standards, which specify the correct holding method of the welding head, the setting range of welding parameters, the control of welding speed, etc. For example, the holding angle of the welding head should be between 45 degrees and 60 degrees, the welding current should be set between 200 amperes and 300 amperes, and the welding speed should be maintained between 100 mm and 150 mm per minute. Through semantic enhancement processing, these standards are further refined and clarified. For example, according to different welding positions and welding materials, specific combinations of welding parameters are determined. Suppose when welding a specific part of the car body, the optimal combination of welding parameters is a welding head holding angle of 50 degrees, a welding current of 250 amperes, and a welding speed of 120 mm per minute. Arrange these specific parameter combinations in the order of welding operations to form standard operation path features. The standard operation path features can be represented as a series of operation steps and corresponding parameter values, such as "Step 1: Adjust the holding angle of the welding head to 50 degrees; Step 2: Set the welding current to 250 amperes; Step 3: Weld at a speed of 120 mm per minute".

[0032] Step S124: Perform multi-dimensional quantization processing on the student evaluation indicators to generate an evaluation weight matrix, and dynamically associate the task objective features, skill requirement features, standard operation path features, and evaluation weight matrix to generate the multi-modal training interaction feature set.

[0033] In this embodiment, the student evaluation indicators include the accuracy of operations, the time to complete tasks, the ability to handle abnormal situations, etc. For the accuracy of operations, it is quantified by measuring the deviation between the weld width of the welding and the standard width. For example, if the standard weld width is 4 mm and the actual weld width of the student's welding is 4.2 mm, then the accuracy deviation of the operation is 0.2 mm. For the time to complete tasks, record the actual time used by the student from the start of the operation to the completion of the task. Assuming that the specified time to complete the task is 30 minutes and the actual time used by the student is 28 minutes, then the time deviation to complete the task is -2 minutes (indicating early completion). For the ability to handle abnormal situations, it is quantified according to whether the measures taken by the student when encountering abnormalities are correct and timely. For example, when the welding equipment has an abnormal situation of overheating of the motor, if the student can take correct cooling measures and resume normal welding within 5 minutes, then the score for the ability to handle abnormal situations is 80 points (out of 100).

[0034] Next, according to these quantified indicators, an evaluation weight matrix is generated. Assuming that the weight of the accuracy of operations is 0.4, the weight of the time to complete tasks is 0.3, and the weight of the ability to handle abnormal situations is 0.3, then the evaluation weight matrix can be expressed as (0.4, 0.3, 0.3).

[0035] Finally, the task objective characteristics, skill requirement characteristics, standard operation path characteristics, and evaluation weight matrix are dynamically associated to generate a multi-modal training interaction feature set. For example, the task objective characteristic of "completing the welding work within the specified time" is associated with the time evaluation indicator for completing tasks. When the time deviation of the student to complete the task exceeds a certain range, it will affect the final evaluation score. The skill requirement characteristic of "the weld width is between 3 mm and 5 mm" is associated with the accuracy evaluation indicator of operations. When the weld width of the student's welding exceeds the standard range, the accuracy score of the operation will decrease. The standard operation path characteristic is associated with the accuracy evaluation indicator of operations. The closer the student's operation steps are to the standard operation path, the higher the accuracy score of the operation. Through this dynamic association, each characteristic and indicator are integrated together to form a multi-modal training interaction feature set.

[0036] Step S130: Perform collaborative feature mapping processing on the industrial equipment digital twin and the multi-modal training interaction feature set to obtain a virtual-real fusion training scenario feature set, and the virtual-real fusion training scenario feature set includes equipment operation mapping relationships, skill execution paths, and abnormal handling strategies.

[0037] Step S131: Perform spatial coordinate matching processing on the equipment operation interface of the industrial equipment digital twin and the standard operation path characteristics to generate an operation action constraint area.

[0038] Continuing with the example of an automated welding device, the device operation interface of the industrial device digital twin defines the operable range and manner of the welding head in three-dimensional space. The standard operation path feature includes the spatial coordinates and parameters corresponding to each operation step during the welding process. When performing spatial coordinate matching processing, it is first necessary to determine whether the initial position of the welding head in the digital twin is consistent with the starting coordinate in the standard operation path feature. Suppose the initial position coordinate of the welding head in the digital twin is (100, 200, 300) millimeters, and the starting coordinate in the standard operation path feature is (100, 200, 300) millimeters, indicating that the two are matched at the starting position.

[0039] Then, based on the subsequent coordinate information in the standard operation path feature, determine the movement trajectory of the welding head during the entire welding process. For example, the standard operation path feature stipulates that the welding head needs to move to the coordinate (150, 250, 350) millimeters for welding. In the digital twin, the welding head is controlled to move to this coordinate through the device operation interface. During the movement, the movement range and speed limit of the welding head need to be considered. Suppose the maximum movement speed of the welding head in the X-axis direction is 10 millimeters per second, the maximum movement speed in the Y-axis direction is 8 millimeters per second, and the maximum movement speed in the Z-axis direction is 6 millimeters per second. According to these limiting conditions, calculate the time and path required for the welding head to move from the initial position to the target position.

[0040] After determining the movement trajectory of the welding head, generate an operation action constraint area. The operation action constraint area is a three-dimensional space range that stipulates the allowable movement range of the welding head during operation. For example, during welding, to ensure welding quality and safety, the welding head cannot exceed a certain spatial range. Suppose the range of the operation action constraint area in the X-axis direction is from 100 millimeters to 200 millimeters, the range in the Y-axis direction is from 200 millimeters to 300 millimeters, and the range in the Z-axis direction is from 300 millimeters to 400 millimeters. Associate this range information with the device operation interface in the digital twin. When the movement of the welding head exceeds this constraint area, the digital twin will issue a warning and stop the operation.

[0041] Step S132: Perform state-goal alignment processing based on the task target feature and the device operation status time series vector to generate a dynamic task trigger condition.

[0042] In this embodiment, the task target feature includes the specific requirements of the training task, such as completing the welding work within a specified time and ensuring welding quality. The device operation status time series vector records the operation status parameters of the device at different time points, such as temperature, pressure, position, etc. When performing state-goal alignment processing, it is necessary to analyze the impact of the device operation status on the task target.

[0043] Continuing with the example of an automated welding equipment, the task objective features require that the welding work on a certain component of the car body be completed within 30 minutes, and the welding quality should meet certain standards. The time series vector of the equipment operation status shows that the temperature of the welding head is continuously rising. When the temperature exceeds a certain threshold, it may affect the welding quality. Through the analysis of the time series vector of the equipment operation status, it is found that when the temperature of the welding head exceeds 300 degrees Celsius, the welding quality will significantly decline. To ensure the achievement of the task objective, measures need to be taken before the temperature of the welding head reaches 300 degrees Celsius, such as reducing the welding current or increasing heat dissipation.

[0044] Based on the above analysis, dynamic task trigger conditions are generated. The dynamic task trigger conditions are determined according to the relationship between the equipment operation status and the task objective. In this example, the dynamic task trigger condition can be set to "when the temperature of the welding head reaches 280 degrees Celsius, automatically reduce the welding current by 10%". When the temperature of the welding head in the time series vector of the equipment operation status reaches 280 degrees Celsius, the digital twin will automatically trigger this task and adjust the welding parameters to ensure the welding quality and the timely completion of the task.

[0045] Step S133: Perform pattern recognition processing on the abnormal event log, extract the abnormal event features, and perform policy matching processing on the abnormal event features and the skill requirement features to generate an abnormal handling policy.

[0046] In this embodiment, when performing pattern recognition processing on the abnormal event log, first, the information in the abnormal event log needs to be sorted and classified. Taking the automated welding equipment as an example, the abnormal event log records abnormal situations such as motor overheating, sensor failure, and unstable welding quality. Classify these abnormal situations. For example, classify motor overheating and sensor failure as equipment hardware abnormalities, and classify unstable welding quality as process abnormalities.

[0047] Then, extract the abnormal event features. The abnormal event features include the time of occurrence of the abnormality, the type of abnormality, the severity of the abnormality, etc. For example, for the motor overheating abnormality, the time of occurrence of the abnormality is 10:30 am, the type of abnormality is motor overheating, and the severity of the abnormality can be judged according to the increase in the motor temperature. Assume that the normal operating temperature of the motor is 80 degrees Celsius. When the temperature rises to 120 degrees Celsius, the severity of the abnormality is judged to be moderate.

[0048] Next, perform a strategy matching process on the abnormal event features and the skill requirement features. The skill requirement features include requirements for what measures trainees should take when encountering abnormal situations. For example, the skill requirement features stipulate that when encountering an abnormal situation of overheating of the motor, trainees should take cooling measures within 5 minutes. Based on the abnormal event features and the skill requirement features, generate an abnormal handling strategy. For the abnormal situation of overheating of the motor, the abnormal handling strategy can be: when it is monitored that the motor temperature reaches 100 degrees Celsius (this temperature setting is based on the analysis of the motor performance and past abnormal data, and it is generally considered that when the temperature exceeds this value, the motor may have a risk of performance degradation or even damage), the digital twin immediately issues a first-level warning, prompting the trainee to closely monitor the motor status. The warning information includes key data such as the current motor temperature and the temperature rising rate. The temperature rising rate can be obtained by dividing the difference between two adjacent temperature monitoring values by the time interval. For example, if the previous monitored temperature was 90 degrees Celsius and it was monitored as 100 degrees Celsius 5 minutes later, then the temperature rising rate is (100 - 90) ÷ 5 = 2 degrees Celsius per minute.

[0049] When the motor temperature continues to rise to 120 degrees Celsius, the digital twin issues a second-level warning and automatically triggers preliminary cooling measures. The preliminary cooling measures include reducing the working load of the motor. For example, reduce the current power output of the motor from 100% to 80%. The amplitude of power reduction is determined based on the heat dissipation capacity of the motor and historical abnormal handling experience, which can reduce heat generation to a certain extent while ensuring that the equipment does not stop working completely and maintaining a certain production progress. At the same time, turn on the auxiliary heat dissipation device of the motor, such as a cooling fan, and adjust the fan speed to the high gear to accelerate air circulation and take away the heat generated by the motor.

[0050] If the motor temperature continues to rise to 150 degrees Celsius, at this time, the abnormal severity is determined to be highly dangerous. The digital twin immediately issues an emergency alarm and automatically stops the operation of the motor to prevent further damage to the motor and even cause safety accidents. At the same time, the system will provide detailed fault troubleshooting guidelines to guide trainees to check whether the heat dissipation channel of the motor is blocked, whether the power supply is stable, and whether there is a short circuit in the motor winding. For example, prompt trainees to check the heat dissipation channel to see if there is debris accumulation, which can be judged by observing whether there is an obvious ventilation blockage at the heat dissipation outlet; check the power supply, use a professional voltage detection device to measure whether the input voltage of the motor is within the normal range, and the normal range is generally ±5% of the rated voltage; check the motor winding, use an insulation resistance tester to measure the insulation resistance value of the winding. If the insulation resistance value is lower than the specified standard, there may be a short circuit problem.

[0051] For sensor failure anomalies, when the abnormal event characteristics show abnormal fluctuations in the data output by the sensor, such as the temperature value displayed by the temperature sensor changing rapidly within a short period of time, or the reading of the pressure sensor exceeding the normal operating range, first conduct a preliminary self-check on the sensor. The self-check process includes checking whether the connection wires of the sensor are loose or damaged. The appearance of the wires can be observed with the naked eye, and the wires can be gently shaken to see if it affects the sensor output data. Check whether the power supply of the sensor is normal. Use a multimeter to measure the power supply voltage of the sensor to ensure that it is within the specified power supply range.

[0052] If no problems are found in the preliminary self-check, further calibrate the sensor. The calibration process requires the use of standard reference equipment. For example, for a temperature sensor, use a high-precision thermometer as a reference. Place the sensor and the reference thermometer in a stable temperature environment at the same time, and record the reading difference between the two. Correct the output data of the sensor according to the reading difference. The correction formula is: Corrected reading = Original reading of the sensor + (Reading of the reference thermometer - Original reading of the sensor) × Calibration coefficient. The calibration coefficient is determined according to the accuracy and characteristics of the sensor, and is generally between 0.9 and 1.1.

[0053] If the sensor still cannot work properly after calibration, it is determined that the sensor is damaged and needs to be replaced in time. When replacing the sensor, ensure that the model and specifications of the new sensor are the same as those of the original sensor to ensure the normal operation of the equipment. At the same time, re-calibrate and test the sensor after replacement to ensure that its output data is accurate and reliable.

[0054] For welding quality instability anomalies, when the abnormal event characteristics are manifested as uneven weld width, insufficient welding strength, etc., first check the welding parameters. Check whether the welding current is stable. The current display value of the welding power supply can be observed, and multiple current data can be recorded within a period of time to calculate its fluctuation range. If the fluctuation range exceeds ±3%, it is considered that the welding current is unstable, and the parameter settings of the welding power supply need to be adjusted to make it output a stable current.

[0055] Check whether the welding speed is appropriate. The welding speed can be determined by measuring the distance that the welding head moves within a unit time. For example, if the welding head moves 120 millimeters in 10 seconds, the welding speed is 120÷10 = 12 millimeters per second. If the welding speed is too fast, it may result in a too narrow weld width and insufficient welding strength; if the welding speed is too slow, it may result in a too wide weld and the appearance of welding beads and other problems. Adjust the welding speed to the appropriate range according to the welding process requirements and the actual welding situation.

[0056] Check whether the flow rate and purity of the welding gas meet the requirements. The welding gas flow rate can be measured by a gas flow meter. Generally, a suitable gas flow rate range is determined according to the welding materials and process requirements. For example, for stainless steel welding, the argon gas flow rate is generally between 10 - 15 liters per minute. The gas purity can be checked using a gas analyzer. If the gas purity is lower than the specified standard, the gas cylinder needs to be replaced or the gas needs to be purified.

[0057] If no problems are found in the welding parameter inspection, further check the surface quality of the welding materials and the workpiece. Check whether the model and specifications of the welding materials are correct and whether there are problems such as moisture absorption and oxidation. For example, obvious rust or moisture absorption on the surface of the welding materials may affect the welding quality, and qualified welding materials need to be replaced. Check whether the surface of the workpiece is clean and whether there are impurities such as oil stains and rust. The surface of the workpiece can be cleaned using sandpaper, cleaning agents, etc. to ensure that the welding surface is clean and flat.

[0058] Step S134: Perform spatio-temporal fusion processing on the operation action constraint region, dynamic task trigger conditions, and abnormal handling strategies to obtain the virtual-real fusion training scenario feature set.

[0059] In this embodiment, after obtaining the operation action constraint region, dynamic task trigger conditions, and abnormal handling strategies, it is necessary to perform spatio-temporal fusion processing on them to form a complete virtual-real fusion training scenario feature set. Spatio-temporal fusion processing integrates these features in terms of time and space, enabling them to be interrelated and work collaboratively.

[0060] Continuing with the example of an automated welding device, the operation action constraint region defines the operable range of the welding head in three-dimensional space. The dynamic task trigger conditions determine when to trigger specific tasks based on the device operation status and task objectives. The abnormal handling strategies provide corresponding handling measures for different abnormal situations.

[0061] In the time dimension, arrange the dynamic task trigger conditions and abnormal handling strategies in chronological order. For example, the dynamic task trigger condition stipulates that when the temperature of the welding head reaches 280 degrees Celsius, the welding current is automatically reduced by 10%. The abnormal handling strategy stipulates that when the temperature of the motor reaches 100 degrees Celsius, a first-level warning is issued. Arrange these conditions and strategies in ascending order of temperature to form a time series. During the operation of the device, according to the real-time monitored temperature data, trigger the corresponding tasks and warnings in sequence according to this time series.

[0062] In the spatial dimension, the operation action constraint area is associated with the dynamic task trigger condition and the abnormal handling strategy. For example, when the welding head operates normally within the operation action constraint area, the operation is carried out according to the normal task process; when the welding head approaches the boundary of the operation action constraint area, the corresponding warning task is triggered to prompt the trainee to pay attention to adjusting the position of the welding head. At the same time, when an abnormal situation occurs, the implementation of the abnormal handling strategy also needs to consider the limitations of the operation action constraint area. For example, during the abnormal handling of overheating of the motor, if it is necessary to reduce the working load of the motor, it is necessary to ensure that the movement of the welding head after reducing the load is still within the operation action constraint area.

[0063] Through spatio-temporal fusion processing, the operation action constraint area, the dynamic task trigger condition and the abnormal handling strategy are integrated together to form a virtual-real fusion training scenario feature set. This virtual-real fusion training scenario feature set contains information such as equipment operation mapping relationships, skill execution paths and abnormal handling strategies, providing a basis for the subsequent generation of training programs. For example, the equipment operation mapping relationship clarifies the actual corresponding equipment actions and effects when the trainee operates in the digital twin; the skill execution path stipulates the operation steps and sequences that the trainee should follow when completing the training task; the abnormal handling strategy provides the trainee with coping methods and guidance when encountering abnormal situations.

[0064] Step S140: Dynamically optimize the virtual-real fusion training scenario feature set based on a preset training strategy generation model to generate a personalized production-education collaborative training program, and the personalized production-education collaborative training program includes a training task sequence, skill assessment nodes and a dynamic feedback mechanism.

[0065] Step S141: Extract the skill weakness characteristics based on the historical operation data of the trainee, and perform priority sorting on the skill weakness characteristics and the virtual-real fusion training scenario feature set to generate training task weight parameters.

[0066] In this embodiment, the historical operation data of the trainee records the operation behaviors and performances of the trainee during previous training processes. By analyzing the historical operation data of the trainee, the skill weakness characteristics of the trainee can be extracted. Taking the training of an automated welding equipment as an example, the historical operation data of the trainee includes information such as the weld width, welding strength, and operation time of the welding.

[0067] Analyze the weld width data and calculate the deviation between the weld width of each welding by the trainee and the standard weld width. For example, the standard weld width is 4 mm, and the weld widths of the trainee's multiple weldings are 3.5 mm, 4.2 mm, 3.8 mm, etc. By calculating the average value and standard deviation of these deviations, evaluate the trainee's skill level in weld width control. If the average deviation is large and the standard deviation is also large, it indicates that there are significant problems with the trainee in weld width control, and this is regarded as one of the characteristics of weak skills.

[0068] Analyze the welding strength data. By conducting tensile tests on the welded joints and other methods, obtain the welding strength data. Compare the trainee's welding strength data with the standard welding strength. If the trainee's welding strength is generally lower than the standard value, it indicates that there is a weak skill point for the trainee in ensuring welding strength.

[0069] Analyze the operation time data. Record the actual time taken by the trainee to complete each training task and compare it with the specified standard time. If the trainee exceeds the standard time to complete the task multiple times, it indicates that there are problems with the trainee's operation efficiency, which is also one of the weak skill points.

[0070] Perform a priority sorting process on the extracted weak skill point characteristics and the set of virtual-real fusion training scenario characteristics. The set of virtual-real fusion training scenario characteristics includes different training tasks and skill requirements. According to the severity of the weak skill point characteristics and their impact on the training objectives, assign a priority to each weak skill point. For example, weld width control and welding strength assurance are crucial for welding quality, and their priorities are set to high; the operation efficiency issue affects the task completion time, but relatively has a smaller direct impact on welding quality, and its priority is set to medium.

[0071] According to the priority sorting result, generate the training task weight parameters. The training task weight parameters are used to determine the importance of each training task in the personalized production-education collaborative training plan. For training tasks related to high-priority weak skill points, assign a higher weight; for training tasks related to low-priority weak skill points, assign a lower weight. For example, for training tasks related to weld width control and welding strength assurance, the weight is set to 0.4; for training tasks related to operation efficiency improvement, the weight is set to 0.2.

[0072] Step S142: Perform a path adjustment process on the standard operation path characteristics according to the training task weight parameters to generate an adaptive operation training path.

[0073] In this embodiment, the standard operation path characteristics specify the standard operation steps and parameters for completing the training tasks. Perform a path adjustment process on the standard operation path characteristics according to the training task weight parameters to generate an adaptive operation training path that is more suitable for the trainee's skill level.

[0074] Continuing with the example of an automated welding device, the standard operation path features specify information such as the movement trajectory of the welding head and the welding parameter settings. For training tasks with higher weights, such as tasks related to weld width control and welding strength assurance, the standard operation path is refined and optimized. For example, in the standard operation path, the movement speed of the welding head is a fixed value. To better control the weld width, according to the weak points of the trainees in weld width control, the movement speed of the welding head is adjusted in segments. At the beginning stage of the weld, the welding speed is appropriately reduced to ensure that the width at the start of the weld meets the requirements; in the middle stage of the weld, according to the welding material and process requirements, the welding speed is adjusted to an appropriate range; at the end stage of the weld, the welding speed is appropriately reduced again to avoid the problem of too narrow a width at the end of the weld.

[0075] For training tasks with lower weights, such as tasks related to improving operation efficiency, on the premise of ensuring welding quality, the standard operation path is appropriately simplified and optimized. For example, some unnecessary operation steps are reduced, or the operation sequence is adjusted to improve operation efficiency. However, it should be noted that this kind of simplification and optimization cannot affect the welding quality and must be carried out within the scope of skill requirements.

[0076] By performing path adjustment processing on the standard operation path features, an adaptive operation training path is generated. The adaptive operation training path can be adjusted personalized according to the weak points of the trainees, enabling the trainees to specifically improve their skill levels during the training process.

[0077] Step S143: Perform dynamic feedback association processing on the adaptive operation training path and the evaluation weight matrix to generate a real-time evaluation trigger node.

[0078] In this embodiment, the evaluation weight matrix is generated according to the trainee evaluation indicators and is used to measure the performance of trainees in different aspects. The adaptive operation training path and the evaluation weight matrix are subjected to dynamic feedback association processing to determine when to conduct an evaluation during the training process, that is, to generate a real-time evaluation trigger node.

[0079] Taking the training of an automated welding device as an example, the adaptive operation training path specifies operation steps such as the movement trajectory of the welding head and the welding parameter settings. The evaluation weight matrix includes the weights of evaluation indicators such as operation accuracy, task completion time, and the ability to handle abnormal situations.

[0080] In the adaptive operation training path, the real-time evaluation trigger nodes are determined according to the criticality of the operation steps and their impact on the evaluation metrics. For example, during the welding process, the starting, middle, and ending stages of the weld are key links affecting the weld width and welding strength, and these stages are set as the real-time evaluation trigger nodes. When the welding head reaches these nodes, the system automatically triggers the evaluation task and evaluates the trainee's operation performance according to the evaluation weight matrix.

[0081] For the evaluation metric of operation accuracy, at the real-time evaluation trigger nodes, parameters such as the weld width and welding strength are measured and compared with the standard values, and the operation accuracy score is calculated according to the weights in the evaluation weight matrix. For the evaluation metric of task completion time, the time from the start of the training to reaching the real-time evaluation trigger node is recorded and compared with the specified standard time to calculate the task completion time score. For the evaluation metric of the ability to handle abnormal situations, if an abnormal situation occurs before reaching the real-time evaluation trigger node, the corresponding score is calculated based on the trainee's handling measures and effects, combined with the weights in the evaluation weight matrix.

[0082] By dynamically feedback-correlating the adaptive operation training path with the evaluation weight matrix to generate the real-time evaluation trigger nodes, trainees can receive timely feedback during the training process, understand their operation performance, and make targeted improvements.

[0083] Step S144: Construct a feedback closed-loop mechanism based on the real-time evaluation trigger nodes and the abnormal situation handling strategy, and integrate the adaptive operation training path, real-time evaluation trigger nodes, and feedback closed-loop mechanism into the personalized production-education collaborative training plan.

[0084] In this embodiment, the feedback closed-loop mechanism refers to a process of circular feedback in which, during the training process, the training tasks and guidance methods are adjusted in a timely manner according to the trainee's operation performance and real-time evaluation results. Exemplarily, a feedback closed-loop mechanism can be constructed based on the real-time evaluation trigger nodes and the abnormal situation handling strategy.

[0085] When reaching the real-time evaluation trigger nodes, the trainee's operation performance can be evaluated and corresponding measures can be taken according to the evaluation results. If the trainee's operation performance is good and meets the skill requirements, the subsequent training tasks are continued according to the adaptive operation training path. If there are problems with the trainee's operation performance, such as the weld width not meeting the requirements or the welding strength being insufficient, the system provides corresponding guidance and suggestions according to the abnormal situation handling strategy. For example, if the weld width is too narrow, the system prompts the trainee to appropriately reduce the welding speed or increase the welding current; if the welding strength is insufficient, the system prompts the trainee to check the welding materials and welding process parameters.

[0086] Meanwhile, optimize and adjust the abnormal handling strategy according to the trainees' ability to handle abnormal situations. If a trainee can take timely and correct handling measures when encountering an abnormal situation, it indicates that the trainee has a good grasp of the abnormal handling strategy, and the difficulty and complexity of the abnormal situation can be appropriately increased to further improve the trainee's response ability. If problems occur when a trainee handles an abnormal situation, the system analyzes the reasons and refines and improves the abnormal handling strategy, providing more detailed guidance and training.

[0087] Finally, integrate the adaptive operation training path, real-time evaluation trigger nodes, and feedback closed-loop mechanism into a personalized production-education collaborative training plan. The adaptive operation training path provides specific operation guidance for trainees, the real-time evaluation trigger nodes enable trainees to receive timely feedback, and the feedback closed-loop mechanism ensures the dynamic adjustment and optimization of the training plan. Through this integration, the personalized production-education collaborative training plan can be customized according to the trainees' skill levels and operation performances, improving the training effect and quality.

[0088] Step S150: Deploy the personalized production-education collaborative training plan to the industrial equipment digital twin and trigger real-time interactive training, and update the production-education collaborative knowledge base based on the trainees' operation feedback data.

[0089] Step S151: Load the adaptive operation training path to the operation interface of the industrial equipment digital twin and generate operation trajectory data based on the trainees' real-time operation actions.

[0090] In this embodiment, the adaptive operation training path is loaded to the operation interface of the industrial equipment digital twin, enabling the digital twin to simulate the operation of the equipment according to the adaptive operation training path. Taking an automated welding equipment as an example, the adaptive operation training path stipulates operation steps such as the movement trajectory of the welding head and the setting of welding parameters. Then, the above step information is input into the operation interface of the digital twin, and the digital twin controls the movement of the welding head and the setting of welding parameters according to the above information.

[0091] During the trainees' real-time interactive training, the digital twin monitors the trainees' operation actions in real time. The trainees control the movement of the welding head and the adjustment of welding parameters by operating the operation interface of the digital twin. The digital twin records each operation action of the trainees, including the operation time and the operation content (such as moving the welding head, adjusting the welding current, etc.), and generates operation trajectory data based on these operation actions. The operation trajectory data is a sequence containing multiple time points and operation information, reflecting the operation process and behavior pattern of the trainees during the training process. For example, the operation trajectory data may record that the trainees move the welding head to the coordinates (110, 210, 310) millimeters at the 10th second and adjust the welding current from 250 amperes to 260 amperes at the 20th second, etc.

[0092] Step S152: Perform deviation analysis on the operation trajectory data and the standard operation path features to generate skill execution error features.

[0093] Step S1521: Extract the action timestamp sequence and the spatial coordinate sequence from the operation trajectory data, and perform spatio-temporal discretization on the standard operation path features to generate a standard spatio-temporal grid.

[0094] In this embodiment, the operation trajectory data contains the timestamps and spatial coordinate information of the trainee's operation actions. By extracting this information, an action timestamp sequence and a spatial coordinate sequence are formed. Taking an automated welding device as an example, the action timestamp sequence records the time points of each operation of the trainee, such as (10, 20, 30,...) seconds; the spatial coordinate sequence records the spatial coordinates of the welding head at each operation, such as ((110, 210, 310), (120, 220, 320),...) millimeters.

[0095] Next, perform spatio-temporal discretization on the standard operation path features to generate a standard spatio-temporal grid. The standard operation path features specify the ideal positions and states of the welding head at different time points. Discretize the standard operation path according to a certain time interval and spatial resolution, and divide it into small spatio-temporal grids. For example, set the time interval to 1 second and the spatial resolution to 1 millimeter. Then, within each 1-second time interval, divide the space into grids with a unit of 1 millimeter. Each grid represents an ideal state of the welding head at a specific time and space.

[0096] Step S1522: Map the action timestamp sequence and the spatial coordinate sequence to the standard spatio-temporal grid, and calculate the spatio-temporal offset of each grid node.

[0097] In this embodiment, for each operation action, find its corresponding standard spatio-temporal grid node. For example, when the trainee moves the welding head to the coordinate (110, 210, 310) millimeters at the 10th second, find the time layer corresponding to the 10th second in the standard spatio-temporal grid, and find the grid node corresponding to the coordinate (110, 210, 310) millimeters in this time layer.

[0098] The spatio-temporal offset includes a time offset and a space offset. The time offset refers to the difference between the actual time of the trainee's operation action and the standard operation time. For example, if the standard operation requires the welding head to reach a certain position at the 10th second, and the trainee reaches it at the 12th second, then the time offset is 2 seconds. The space offset refers to the difference between the actual spatial coordinates of the welding head during the trainee's operation and the standard spatial coordinates. Suppose the standard spatial coordinates are (100, 200, 300) millimeters, and the coordinates of the trainee's actual operation are (110, 210, 310) millimeters. Then the space offset in the X-axis direction is 110 - 100 = 10 millimeters, in the Y-axis direction is 210 - 200 = 10 millimeters, and in the Z-axis direction is 310 - 300 = 10 millimeters.

[0099] For each grid node corresponding to an operation action, the spatio-temporal offset is calculated according to the above method. The spatio-temporal offsets of all grid nodes are recorded to form a spatio-temporal offset sequence. For example, within a certain period of time, there are 5 operation actions, and the corresponding spatio-temporal offset sequence may be ((2, 10, 10, 10), (-1, 5, -3, 2), (3, 8, 6, 4), (0, -2, -1, 0), (1, 3, 2, 1)), where the first value of each element represents the time offset, and the following three values represent the space offsets in the X, Y, and Z axis directions respectively.

[0100] Step S1523: Construct an error distribution heat map based on the spatio-temporal offset, and perform pattern clustering processing on the error distribution heat map to generate key error regions.

[0101] In this embodiment, the error distribution heat map is a visual representation method used to intuitively display the distribution of the trainee's operation errors in time and space.

[0102] First, determine the time and space ranges of the heat map. The time range can be determined according to the total duration of the training task, and the space range can be determined according to the operation space of the welding head. For example, the total duration of the training task is 60 seconds, and the operation space of the welding head is 0 - 200 millimeters in the X-axis direction, 0 - 300 millimeters in the Y-axis direction, and 0 - 400 millimeters in the Z-axis direction.

[0103] Then, map the spatio-temporal offset to the corresponding position in the heat map. For the spatio-temporal offset of each grid node, find the corresponding position in the heat map according to its time and space coordinates, and assign a corresponding color value according to the magnitude of the offset. The larger the offset, the darker the color, indicating a greater error; the smaller the offset, the lighter the color, indicating a smaller error. For example, red can be used to represent large errors, yellow for medium errors, and green for small errors.

[0104] After constructing the error distribution heat map, perform pattern clustering on it. The purpose of pattern clustering is to group regions with similar error distributions together in order to identify key error regions. Clustering algorithms such as the K-means clustering algorithm can be used. First, determine the number of clusters. For example, divide the error distribution heat map into 3 categories. Then, randomly select 3 initial cluster centers. Based on the similarity between each grid node and the cluster centers (which can be measured by calculating the distance of spatio-temporal offsets), assign the grid nodes to the nearest cluster center. Next, update the positions of the cluster centers and repeat the above assignment and update process until the cluster centers no longer change significantly.

[0105] After clustering is completed, analyze the characteristics of each cluster. The key error region usually refers to the region where the errors are large and concentrated. For example, if the spatio-temporal offsets of the grid nodes in a certain cluster are generally large and form a continuous region in the heat map, then this region can be determined as the key error region. By analyzing the key error regions, the key areas with problems in the trainees' operations can be found, providing targeted guidance for subsequent skill improvement.

[0106] Step S1524: Perform feature quantization processing on the key error region, extract operation delay features, spatial deviation features, and path redundancy features, and integrate them into the skill execution error features.

[0107] In this embodiment, perform feature quantization processing on the key error region to extract more targeted skill execution error features.

[0108] Among them, the operation delay feature is mainly determined by analyzing the time offsets within the key error region. Calculate the average value and standard deviation of the time offsets of all grid nodes within the key error region. The average value represents the overall level of operation delay, and the standard deviation represents the degree of fluctuation of the operation delay. For example, there are 10 grid nodes within the key error region, and the time offsets are (2, 3, 1, 4, 2, 3, 5, 2, 3, 1) seconds respectively. Calculate the average value as (2 + 3 + 1 + 4 + 2 + 3 + 5 + 2 + 3 + 1) ÷ 10 = 2.6 seconds, and the standard deviation is calculated through the corresponding statistical method. The operation delay feature can be represented by the average value and the standard deviation, such as (2.6, 0.8), where 2.6 represents the average operation delay time and 0.8 represents the degree of fluctuation of the operation delay.

[0109] The spatial deviation feature is determined by analyzing the spatial offsets within the key error region. The average value and standard deviation of the spatial offsets of all grid nodes within the key error region in the X, Y, and Z axis directions are calculated respectively. For example, in the X axis direction, the spatial offsets of the grid nodes are (10, 12, 8, 15, 11, 9, 13, 10, 12, 8) millimeters. The average value is calculated as (10 + 12 + 8 + 15 + 11 + 9 + 13 + 10 + 12 + 8) ÷ 10 = 10.8 millimeters, and the standard deviation is calculated through the corresponding statistical method. Similarly, the average values and standard deviations in the Y and Z axis directions are calculated. The spatial deviation feature can be represented by a vector composed of the average values and standard deviations in the three directions, such as ((10.8, 1.2), (8.5, 1.0), (9.2, 0.9)), which respectively represent the spatial deviation conditions in the X, Y, and Z axis directions.

[0110] The path redundancy feature is determined by analyzing the difference between the trainee's operation path and the standard operation path within the key error region. The ratio of the length of the trainee's operation path to the length of the standard operation path within the key error region can be calculated. If the ratio is greater than 1, it indicates that there is redundancy in the trainee's operation path. For example, if the length of the standard operation path is 100 millimeters and the length of the trainee's operation path within the key error region is 120 millimeters, then the path redundancy feature is 120 ÷ 100 = 1.2.

[0111] Integrate the operation delay feature, spatial deviation feature, and path redundancy feature into the skill execution error feature. These features can be concatenated together in a certain order to form a multi-dimensional vector. For example, the skill execution error feature can be represented as ((2.6, 0.8), ((10.8, 1.2), (8.5, 1.0), (9.2, 0.9)), 1.2), which completely describes the errors existing in the trainee during the skill execution process.

[0112] Step S153: According to the real-time evaluation trigger node, call the evaluation weight matrix to perform a quantitative evaluation process on the skill execution error feature, and generate a trainee ability evaluation report.

[0113] Step S1531: Based on the real-time evaluation trigger node, determine the evaluation time window, and extract the time series change curve of the skill execution error feature within the evaluation time window.

[0114] In this embodiment, the real-time evaluation trigger node is a time point preset during the training process for evaluation. For example, during the training of an automated welding device, the starting, middle, and ending stages of the welding process are set as the real-time evaluation trigger nodes respectively. Taking the evaluation trigger node in the starting stage as an example, assuming the time corresponding to this node is the 10th second, in order to comprehensively evaluate the operation performance of the trainee at this stage, an evaluation time window is determined. For example, centered on the 10th second, it extends 5 seconds forward and backward respectively, so the evaluation time window is from the 5th second to the 15th second.

[0115] The skill execution error features include operation delay features, spatial deviation features, path redundancy features, etc. For the operation delay features, the operation delay values at each time point within the evaluation time window are recorded to form an operation delay time series change curve. For example, the operation delay is 1 second at the 5th second, 1.2 seconds at the 6th second, 0.8 seconds at the 7th second, etc. Connecting these values in chronological order gives the operation delay time series change curve. Similarly, the time series change curves of the spatial deviation features and path redundancy features within the evaluation time window are extracted.

[0116] Step S1532: Perform convolution processing on the time series change curve and the evaluation weight matrix to generate a dynamic evaluation score.

[0117] In this embodiment, the evaluation weight matrix is generated according to the trainee evaluation indicators and is used to measure the importance of different evaluation indicators. The time series change curve of the skill execution error features is subjected to convolution processing with the evaluation weight matrix to generate a dynamic evaluation score.

[0118] Taking the time series change curve of the operation delay features as an example, assuming the weight of the operation delay features in the evaluation weight matrix is 0.3. Perform convolution processing on the operation delay time series change curve with the weight 0.3. The process of convolution processing is to multiply each value in the time series change curve by the weight 0.3 to obtain a new curve. For example, the value of the operation delay time series change curve at a certain time point is 2 seconds, and after multiplying by the weight 0.3, it becomes 0.6. Perform such processing on all the values in the operation delay time series change curve to obtain the dynamic evaluation curve of the operation delay features.

[0119] Similarly, perform convolution processing on the time series change curves of the spatial deviation features and path redundancy features with their corresponding weights respectively to obtain the dynamic evaluation curves of the spatial deviation features and path redundancy features.

[0120] Integrate the dynamic evaluation curves of the operation delay feature, spatial deviation feature, and path redundancy feature to generate a dynamic evaluation score. The values of these three dynamic evaluation curves at each time point can be added together to obtain the dynamic evaluation score at that time point. For example, at a certain time point, the dynamic evaluation value of the operation delay feature is 0.6, the dynamic evaluation value of the spatial deviation feature is 0.8, and the dynamic evaluation value of the path redundancy feature is 0.4. Then the dynamic evaluation score at this time point is 0.6 + 0.8 + 0.4 = 1.8. As time goes by, a series of dynamic evaluation scores are obtained, forming a dynamic evaluation score sequence.

[0121] Step S1533: Perform a severity level classification process on the key error areas of the skill execution error feature to generate an error level label, and perform an association mapping process between the error level label and the dynamic evaluation score.

[0122] In this embodiment, the key error areas can be divided into different levels according to the severity of the operation delay feature, spatial deviation feature, and path redundancy feature within the key error areas. For example, key error areas with overly long operation delay time, overly large spatial deviation degree, and serious path redundancy can be classified as high level; key error areas with moderate operation delay time, small spatial deviation degree, and small path redundancy can be classified as medium level; key error areas with short operation delay time, very small spatial deviation degree, and almost no path redundancy can be classified as low level.

[0123] Assign an error level label to each level. For example, the high level corresponds to "severe error", the medium level corresponds to "medium error", and the low level corresponds to "minor error".

[0124] Perform an association mapping process between the error level label and the dynamic evaluation score. According to the magnitude of the dynamic evaluation score, map it to the corresponding error level label. For example, it is set that the dynamic evaluation score between 0 - 1 corresponds to "minor error", between 1 - 2 corresponds to "medium error", and above 2 corresponds to "severe error". When the dynamic evaluation score at a certain time point is 1.8, map it to the error level label of "medium error".

[0125] Step S1534: Generate a visual evaluation chart based on the dynamic evaluation score and the error level label, and perform a comparative analysis process on the visual evaluation chart and the historical training data to generate the student ability evaluation report.

[0126] In this embodiment, a line chart can be used to show the change of the dynamic evaluation score over time, and the corresponding error level labels are marked on the line chart. For example, different colored dots are used to represent different error levels, with red dots representing "severe error", yellow dots representing "medium error", and green dots representing "slight error". In this way, the operation performance and error situation of the trainee during the training process can be intuitively seen.

[0127] The visual evaluation chart is compared and analyzed with the historical training data. The historical training data records the operation performance and evaluation results of other trainees in the same or similar training tasks. The comparative analysis can be carried out from multiple aspects, such as the average value and fluctuation of the dynamic evaluation score, the distribution of the error level labels, etc.

[0128] If the average value of the dynamic evaluation score of the current trainee is lower than the average value of the historical training data, it indicates that the overall operation performance of this trainee is relatively poor; if the fluctuation of the dynamic evaluation score is large, it indicates that the operation stability of this trainee is poor. By comparing the distribution of the error level labels, it can be found in which aspects the current trainee has more severe errors and what differences there are compared with the historical trainees.

[0129] According to the results of the comparative analysis, a trainee ability evaluation report is generated. The trainee ability evaluation report includes the summary of the trainee's operation performance, the comparative analysis results with historical trainees, existing problems and improvement suggestions, etc. For example, the report may point out that there are major problems with the trainee's operation delay and it is necessary to strengthen the control of the operation time; in terms of spatial deviation, the errors in certain areas are relatively severe and targeted training is required. Through the trainee ability evaluation report, the trainee can clearly understand his / her skill level and existing deficiencies, providing guidance for subsequent learning and training.

[0130] Step S154: Perform an association analysis process on the trainee ability evaluation report and the abnormal handling strategy to generate a knowledge base update instruction, and optimize the skill operation standard in the production-education collaborative knowledge base based on the knowledge base update instruction.

[0131] Step S1541: Extract the attention error pattern in the trainee ability evaluation report, and perform a matching degree calculation process on the attention error pattern and the abnormal handling strategy to generate a strategy failure index.

[0132] In this embodiment, the trainee ability evaluation report may contain various error patterns of the trainee during the training process, and the attention error pattern refers to the operation error caused by the trainee's inattention. For example, during the training of an automated welding equipment, the trainee may miss key operation steps due to distraction, or may not respond to abnormal situations in a timely manner during the operation.

[0133] Extract attention error patterns from trainee competency assessment reports. Attention error patterns can be determined by analyzing information such as operational error records, the time and scenario of error occurrence in the assessment report. For example, if a trainee is found to have operational delays at the same time point multiple times during welding, and there are other interference factors near that time point, such as a sudden increase in equipment noise, then it can be inferred that this is an error pattern caused by the trainee's attention being disturbed.

[0134] The attention error pattern and the abnormal handling strategy are matched and calculated. The abnormal handling strategy is a response measure formulated for various abnormal situations. When calculating the matching degree, the similarity between the attention error pattern and the abnormal situations and response measures involved in the abnormal handling strategy is analyzed. For example, the abnormal handling strategy mentions that when equipment noise interference occurs, the trainee should immediately stop the operation and check the equipment. If the trainee's attention error pattern matches this situation, it means that the matching degree is high.

[0135] Through a detailed analysis of the attention error pattern and the abnormal handling strategy, a matching value is calculated for each combination of the attention error pattern and the abnormal handling strategy. For example, the matching value is set between 0 and 1, where 0 indicates a complete mismatch and 1 indicates a complete match. All matching values are aggregated to generate a strategy failure index. The strategy failure index can be represented by the average of the matching values. For example, the five matching values calculated are 0.2, 0.3, 0.5, 0.6, and 0.4, respectively. Then the strategy failure index is (0.2+0.3+0.5+0.6+0.4)÷5=0.4.

[0136] Step S1542: When the strategy failure index exceeds a preset threshold, reverse deduction is performed on the abnormal handling strategy to locate the strategy defect characteristics.

[0137] In this embodiment, the preset threshold is a standard value set based on experience and actual conditions, and is used to judge the severity of strategy failure. For example, the preset threshold is set to 0.5. When the strategy failure index exceeds the preset threshold, it indicates that the abnormal handling strategy has certain problems in dealing with the trainee's attention error pattern, and the abnormal handling strategy needs to be reversely deduced.

[0138] Reverse deduction processing starts from the result of strategy failure, reversely analyzes the formulation and execution process of the abnormal handling strategy, and finds out the reasons that may cause the strategy failure. For example, through analysis, it was found that although the abnormal handling strategy mentioned measures to deal with equipment noise interference, it did not specify how to judge whether the noise would affect the students' attention, nor did it provide specific methods to improve the students' attention in a noisy environment. These are the characteristics of strategy defects.

[0139] The process of locating the defect features of the positioning strategy requires a detailed review of all aspects of the exception handling strategy, including the identification criteria for abnormal situations, the specific content of countermeasures, the execution process, etc. By comparing with the attention error pattern in the trainee ability assessment report, the loopholes and deficiencies in the strategy are identified. For example, it is found that there is no corresponding countermeasure in the exception handling strategy for the situation where a trainee is disturbed by the voices of other trainees during the operation process, and this is a defect feature of the strategy.

[0140] Step S1543: Construct strategy optimization suggestions based on the strategy defect features, and perform a similarity matching process between the strategy optimization suggestions and the historical handling cases in the production-education collaborative knowledge base. When the similarity matching result is lower than the preset standard, generate a knowledge base update instruction including an iterative handling strategy, and add the iterative handling strategy to the production-education collaborative knowledge base.

[0141] Step S1543-1: Invoke a large language model to perform multi-dimensional analysis processing on the strategy defect features to generate defect description text.

[0142] In this embodiment, the large language model has powerful semantic understanding and text generation capabilities. The strategy defect features can be input into the large language model, and it is required to perform analysis from multiple dimensions. For example, for the strategy defect feature that "the exception handling strategy does not specifically state how to determine whether noise will affect the attention of trainees", the large language model can analyze from dimensions such as the type, intensity, and duration of the noise.

[0143] The large language model generates defect description text, which details the specific situation of the strategy defect and the possible impacts. For example, the defect description text may be "The exception handling strategy is insufficient in dealing with noise interference and does not clarify the judgment criteria for noise affecting trainees' attention. Different types of noise (such as equipment noise, environmental noise) and different intensities and durations of noise have different degrees of impact on trainees' attention. Due to the lack of judgment criteria, it is difficult for trainees to accurately determine when to take countermeasures during actual operation, which may lead to operation errors and untimely handling of abnormal situations."

[0144] Step S1543-2: Perform knowledge retrieval processing in the production-education collaborative knowledge base based on the defect description text to obtain relevant technical documents and expert experience data.

[0145] In this embodiment, the production-education collaborative knowledge base contains a large amount of technical documents, historical handling cases, and expert experience data. Use methods such as keyword matching to search for content related to the defect description text in the knowledge base.

[0146] For example, according to the above defect description text, search in the knowledge base for research reports on the impact of noise on trainees' attention, successful cases of dealing with noise interference, and relevant suggestions given by experts, etc. Some technical documents may be retrieved, which mention that by measuring the decibel value of noise, it can be judged whether the noise will affect people's attention. Generally, when the noise decibel value exceeds 70 decibels, it may interfere with people's attention. At the same time, expert experience data may also be retrieved, which mentions that by observing changes in the operation behavior of trainees, such as a sudden slowdown in operation speed or an increase in operation errors, etc., it can be used to assist in judging whether the noise affects the attention of trainees.

[0147] Step S1543-3: Perform semantic fusion processing on the relevant technical documents and expert experience data to generate a set of candidate optimization strategies.

[0148] In this embodiment, semantic fusion processing is to integrate information from different sources and different formats, extract the key information therein, and transform it into consistent and logical strategy suggestions.

[0149] First, perform text analysis on the relevant technical documents and expert experience data. Identify the key content regarding dealing with noise interference, improving trainees' attention, and improving the abnormal handling strategy. For example, extract from the technical document that "70 decibels is the judgment threshold for noise affecting attention", and extract from the expert experience data key information such as "observing changes in trainees' operation behavior for auxiliary judgment".

[0150] Then, perform semantic association and integration on the above key information. Associate the information related to judging the impact of noise on attention to form a complete judgment method. For example, combining the technical document and expert experience data, a judgment method can be obtained: when the noise decibel value exceeds 70 decibels and it is observed that the operation speed of the trainee suddenly slows down or the operation errors increase, it is determined that the noise affects the attention of the trainee.

[0151] Based on the integrated information, generate a set of candidate optimization strategies. The set of candidate optimization strategies includes various possible optimization strategies to address the defects in the abnormal handling strategy. For example, in addition to the above judgment method, some coping measure strategies can also be generated, such as when it is determined that the noise affects the attention of trainees, immediately pause the operation, turn on the sound insulation equipment to reduce the noise; or provide sound insulation tools such as earplugs for trainees, and at the same time remind trainees to concentrate their attention, etc. These strategies together constitute the set of candidate optimization strategies.

[0152] Step S1543-4: Perform simulation verification processing on the set of candidate optimization strategies through the industrial equipment digital twin body, and screen out the optimization strategies that meet the preset safety standards as the strategy optimization suggestions.

[0153] Step S1543-4-1: Load the candidate optimization strategy set into the abnormal event injection interface of the industrial equipment digital twin to trigger a simulated abnormal scenario.

[0154] In this embodiment, the industrial equipment digital twin is a virtual mapping of the actual industrial equipment and has functions and behaviors similar to those of the actual equipment. The candidate optimization strategy set is loaded into the abnormal event injection interface of the industrial equipment digital twin to simulate various abnormal scenarios in the digital twin.

[0155] For the digital twin of an automated welding equipment, assume that there is a strategy for dealing with noise interference in the candidate optimization strategy set, such as the judgment method and countermeasures mentioned above. In the digital twin, different noise environments with different decibel values are simulated through the abnormal event injection interface. For example, scenarios with noise decibel values of 60 dB, 70 dB, 80 dB, etc. are set, and at the same time, the operation behaviors of the trainees in these scenarios are simulated to observe the operation speed and error situations of the trainees.

[0156] Step S1543-4-2: Collect the equipment response data and operation feedback data of the industrial equipment digital twin in the simulated abnormal scenario.

[0157] In this embodiment, the equipment response data includes the operating state of the equipment in a noise environment, such as the motion accuracy of the welding head and the stability of welding parameters. For example, in the scenario with a noise decibel value of 80 dB, record the position deviation of the welding head, the fluctuations of welding current and voltage, etc.

[0158] The operation feedback data is the feedback information about the operation behavior of the trainee, such as the operation speed and the number of operation errors. Through the monitoring system of the digital twin, the operation situations of the trainee in different noise scenarios are recorded in real time. For example, after the noise starts to increase, record the time for the trainee to complete each operation step and count the number of operation errors within a certain period of time.

[0159] Step S1543-4-3: Perform stability evaluation processing on the equipment response data to generate equipment safety indicators, and perform complexity evaluation processing on the operation feedback data to generate operation feasibility indicators.

[0160] In this embodiment, the device safety index reflects the operating stability and safety of the device in abnormal scenarios. For example, for the position deviation of the welding head, calculate its average value and standard deviation over a period of time. If the average value of the position deviation is large and the standard deviation is also large, it indicates that the movement of the welding head is unstable, which may affect the welding quality and device safety. Comprehensive analysis is performed on data such as the average value and standard deviation of the position deviation to set a calculation method for the device safety index. For example, device safety index = (average value of position deviation × weight 1 + standard deviation of position deviation × weight 2), where weight 1 and weight 2 are set according to the actual situation. For example, weight 1 is 0.6 and weight 2 is 0.4.

[0161] Perform complexity evaluation processing on the operation feedback data to generate an operation feasibility index. The operation feasibility index measures the difficulty for trainees to execute the optimization strategy in abnormal scenarios. For example, for the part of the strategy to cope with noise interference that requires trainees to observe changes in operation behaviors to judge the impact of noise, evaluate the complexity of trainees' execution of this operation. It can be evaluated by counting data such as the proportion of trainees correctly judging the impact of noise and the time required for judgment. If the proportion of trainees' correct judgment is low and the time required for judgment is long, it indicates that the operation complexity is high and the operation feasibility is poor. Based on these data, set a calculation method for the operation feasibility index, such as operation feasibility index = (proportion of correct judgment × weight 3 + reciprocal of the time required for judgment × weight 4), and weight 3 and weight 4 are also set according to the actual situation. For example, weight 3 is 0.7 and weight 4 is 0.3.

[0162] Step S1543-4-4: When the device safety index and the operation feasibility index both meet the preset threshold, mark the corresponding candidate optimization strategy as an effective strategy; otherwise, perform iterative optimization processing until the conditions are met.

[0163] In this embodiment, the preset threshold is set according to actual requirements and safety standards. For example, the preset threshold for the device safety index is set to 0.8, and the preset threshold for the operation feasibility index is set to 0.7. When the device safety index and the operation feasibility index generated by a certain candidate optimization strategy in the simulated abnormal scenario both reach or exceed the preset threshold, mark this candidate optimization strategy as an effective strategy.

[0164] If the device safety index or operation feasibility index of a certain candidate optimization strategy does not meet the preset threshold, iterative optimization processing needs to be performed on this strategy. For example, in the strategy for dealing with noise interference, since trainees are required to observe operation behaviors and noise decibel values simultaneously, the operation feasibility index is relatively low. The strategy can be adjusted, such as simplifying the judgment method and only requiring trainees to judge the noise impact based on the number of operation errors. Then, load the adjusted strategy into the digital twin of the industrial device for simulation verification again, and repeat the above processes of collecting data and evaluating indicators until both the device safety index and the operation feasibility index meet the preset threshold.

[0165] Take the selected effective strategies as suggestions for strategy optimization. These suggestions for strategy optimization are specific improvement plans proposed for the defects existing in the abnormal handling strategies, and have high feasibility and safety.

[0166] Step S1543-5: Perform a similarity matching process between the suggestions for strategy optimization and the historical handling cases in the production-education collaborative knowledge base. When the similarity matching result is lower than the preset standard, generate a knowledge base update instruction including the iterative handling strategy, and add the iterative handling strategy to the production-education collaborative knowledge base.

[0167] In this embodiment, the similarity matching process can be carried out by methods such as text matching and feature matching. For example, compare the text content of the suggestions for strategy optimization with the text of the historical handling cases to calculate the similarity score between them. Or extract the key features in the suggestions for strategy optimization and the historical handling cases, such as the types of abnormalities to be dealt with, the measures taken, etc., and calculate the similarity by comparing these features.

[0168] The preset standard is a similarity threshold set according to the actual situation, for example, set to 0.3. When the similarity matching result is lower than the preset standard, it indicates that the suggestions for strategy optimization are a new strategy with a large difference from the historical handling cases. At this time, generate a knowledge base update instruction including the iterative handling strategy. The iterative handling strategy is a strategy improved by combining the suggestions for strategy optimization on the basis of the original abnormal handling strategy.

[0169] For example, in the original abnormal handling strategy, there is no clear judgment method and corresponding measures for noise interference, while the suggestions for strategy optimization propose a judgment method based on noise decibel values and trainees' operation behaviors and the corresponding measures. Integrate these optimization suggestions into the original abnormal handling strategy to form an iterative handling strategy. Then send the knowledge base update instruction including the iterative handling strategy to the production-education collaborative knowledge base, and add the iterative handling strategy to the knowledge base to complete the optimization of the skill operation standards in the production-education collaborative knowledge base. In this way, the skill operation standards in the production-education collaborative knowledge base are updated and improved, and can better guide trainees in practical training and dealing with abnormal situations in actual production.

[0170] In summary, the embodiments of the present invention achieve the deep integration and efficient utilization of the real-time operation data of industrial equipment and the training case data of the production-education collaborative knowledge base, significantly improving the intelligent and personalized level of training teaching. Specifically, by acquiring and integrating the real-time operation data set of industrial equipment and the training case data set, on this basis, using digital twin technology to construct a high-precision virtual model of industrial equipment, and combining the semantic association parsing ability of the large language model, the training case data is transformed into a multi-modal training interaction feature set, making the training content more vivid and intuitive, and effectively enhancing the learning experience and participation of trainees. Further, through collaborative feature mapping processing, the digital twin of industrial equipment and the multi-modal training interaction feature set are organically combined to generate a virtual-real fusion training scenario feature set. This virtual-real fusion training scenario feature set not only includes key information such as equipment operation mapping relationships and skill execution paths, but also incorporates abnormal handling strategies, providing comprehensive and systematic training guidance for trainees. More prominently, through the preset training strategy generation model to dynamically optimize the virtual-real fusion training scenario feature set, a personalized production-education collaborative training plan can be generated. This personalized production-education collaborative training plan is customized according to the actual situation and needs of trainees, realizing the serialization of training tasks, the nodalization of skill assessment, and the establishment of a dynamic feedback mechanism, thus greatly improving the flexibility and effectiveness of training teaching. Finally, deploying the personalized production-education collaborative training plan to the digital twin of industrial equipment and triggering real-time interactive training not only realizes the automation and intelligence of the training process, but also can continuously update the production-education collaborative knowledge base based on the operation feedback data of trainees, forming a closed-loop optimization mechanism and continuously promoting the improvement of training teaching quality.

[0171] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of a digital-twin-based production-education collaborative training system 100 that can implement the idea of the present invention provided by some embodiments of the present invention. For example, the processor 120 can be used on the digital-twin-based production-education collaborative training system 100 and is used to execute the functions in the present invention.

[0172] The digital-twin-based production-education collaborative training system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the digital-twin-based production-education collaborative training method of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0173] For example, the digital-twin-based production-education collaborative training system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROMs, or RAMs, or any combination thereof. Exemplarily, the digital-twin-based production-education collaborative training system 100 may further include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention may be implemented according to these program instructions. The digital-twin-based production-education collaborative training system 100 further includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0174] For ease of illustration, only one processor is described in the digital-twin-based production-education collaborative training system 100. However, it should be noted that the digital-twin-based production-education collaborative training system 100 in the present invention may further include multiple processors. Therefore, the steps executed by one processor described in the present invention may also be jointly executed or separately executed by multiple processors. For example, if the processor of the digital-twin-based production-education collaborative training system 100 executes steps A and B, it should be understood that steps A and B may also be jointly executed by two different processors or separately executed in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0175] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above digital-twin-based production-education collaborative training method is implemented.

[0176] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A digital-twin-based production-education collaborative training method, characterized in that The method includes: Obtaining a real-time operation data set of industrial equipment and a training case data set of an industry-education collaborative knowledge base, where the real-time operation data set includes equipment status parameters, operation instruction sequences, and abnormal event logs, and the training case data set includes historical training task descriptions, skill operation standards, and trainee evaluation indicators; Constructing a digital twin of industrial equipment based on the real-time operation data set, and invoking a large language model to perform semantic association parsing processing on the training case data set to generate a multi-modal training interaction feature set; Performing collaborative feature mapping processing on the digital twin of industrial equipment and the multi-modal training interaction feature set to obtain a virtual-real fusion training scenario feature set, where the virtual-real fusion training scenario feature set includes equipment operation mapping relationships, skill execution paths, and abnormal handling strategies; Performing dynamic optimization processing on the virtual-real fusion training scenario feature set based on a preset training strategy generation model to generate a personalized industry-education collaborative training plan, where the personalized industry-education collaborative training plan includes a training task sequence, skill evaluation nodes, and a dynamic feedback mechanism; Deploying the personalized industry-education collaborative training plan to the digital twin of industrial equipment and triggering real-time interactive training, and updating the industry-education collaborative knowledge base based on trainee operation feedback data; The constructing a digital twin of industrial equipment based on the real-time operation data set, and invoking a large language model to perform semantic association parsing processing on the training case data set to generate a multi-modal training interaction feature set includes: Performing time series alignment processing on the equipment status parameters to generate an equipment operation status time series vector, and performing instruction decomposition processing on the operation instruction sequence to generate an atomic operation step instruction set; Constructing a digital twin of industrial equipment based on the equipment physical model and the equipment operation status time series vector, and mapping the atomic operation step instruction set to the operation interface of the digital twin of industrial equipment; Invoking a large language model to perform intention recognition processing on the historical training task description, extracting task target features and skill requirement features, and performing semantic enhancement processing on the skill operation standards to generate standard operation path features; Performing multi-dimensional quantization processing on the trainee evaluation indicators to generate an evaluation weight matrix, and dynamically associating the task target features, skill requirement features, standard operation path features, and evaluation weight matrix to generate the multi-modal training interaction feature set.

2. The collaborative training method for production and education based on digital twin according to claim 1, wherein The performing collaborative feature mapping processing on the digital twin of industrial equipment and the multi-modal training interaction feature set to obtain a virtual-real fusion training scenario feature set includes: Performing spatial coordinate matching processing on the equipment operation interface of the digital twin of industrial equipment and the standard operation path features to generate an operation action constraint area; Performing state-goal alignment processing based on the task target features and the equipment operation status time series vector to generate a dynamic task trigger condition; Performing pattern recognition processing on the abnormal event logs, extracting abnormal event features, and performing strategy matching processing on the abnormal event features and the skill requirement features to generate an abnormal handling strategy; Perform spatio-temporal fusion processing on the operation action constraint area, dynamic task trigger conditions, and exception handling strategies to obtain the virtual-real fusion training scenario feature set.

3. The collaborative training method for production and education based on digital twin according to claim 1, wherein Perform dynamic optimization processing on the virtual-real fusion training scenario feature set based on a preset training strategy generation model to generate a personalized production-education collaborative training plan, including: Extract skill weak point features based on the historical operation data of the trainees, and perform priority sorting processing on the skill weak point features and the virtual-real fusion training scenario feature set to generate training task weight parameters; Perform path adjustment processing on the standard operation path features according to the training task weight parameters to generate an adaptive operation training path; Perform dynamic feedback association processing on the adaptive operation training path and the evaluation weight matrix to generate a real-time evaluation trigger node; Construct a feedback closed-loop mechanism based on the real-time evaluation trigger node and the exception handling strategy, and integrate the adaptive operation training path, real-time evaluation trigger node, and feedback closed-loop mechanism into the personalized production-education collaborative training plan.

4. The digital-twin-based production-education collaborative training method according to claim 3, wherein, Deploy the personalized production-education collaborative training plan to the digital twin of industrial equipment and trigger real-time interactive training, and update the production-education collaborative knowledge base based on the operation feedback data of the trainees, including: Load the adaptive operation training path to the operation interface of the digital twin of industrial equipment, and generate operation trajectory data based on the real-time operation actions of the trainees; Perform deviation analysis processing on the operation trajectory data and the standard operation path features to generate skill execution error features; Call the evaluation weight matrix according to the real-time evaluation trigger node to perform quantitative evaluation processing on the skill execution error features to generate a trainee ability evaluation report; Perform association analysis processing on the trainee ability evaluation report and the exception handling strategy to generate a knowledge base update instruction, and optimize the skill operation standard in the production-education collaborative knowledge base based on the knowledge base update instruction.

5. The method for production-education collaborative training based on digital twin according to claim 4, characterized in that, The performing deviation analysis processing on the operation trajectory data and the standard operation path features to generate skill execution error features includes: Extract the action timestamp sequence and spatial coordinate sequence in the operation trajectory data, and perform spatio-temporal discretization processing on the standard operation path features to generate a standard spatio-temporal grid; Map the action timestamp sequence and spatial coordinate sequence to the standard spatio-temporal grid, and calculate the spatio-temporal offset of each grid node; Construct an error distribution heat map based on the spatio-temporal offset, and perform pattern clustering processing on the error distribution heat map to generate a key error area; Perform feature quantization processing on the key error area, extract operation delay features, spatial deviation features, and path redundancy features, and integrate them into the skill execution error features.

6. The method for industry-education collaborative training based on digital twin according to claim 4, wherein The performing quantitative evaluation processing on the skill execution error features according to the real-time evaluation trigger node to call the evaluation weight matrix to generate a trainee ability evaluation report includes: Determine an evaluation time window based on the real-time evaluation trigger node, and extract the time series change curve of the skill execution error features within the evaluation time window; Perform a convolution operation on the time series change curve and the evaluation weight matrix to generate a dynamic evaluation score; Perform a severity level classification process on the key error regions of the skill execution error characteristics to generate error level labels, and perform an association mapping process on the error level labels and the dynamic evaluation score; Generate a visual evaluation chart based on the dynamic evaluation score and the error level label, and perform a comparative analysis process on the visual evaluation chart and historical training data to generate the trainee ability evaluation report.

7. The method for collaborative training of production and education based on digital twin according to claim 4, wherein, The associative analysis process of associating the trainee ability evaluation report with the abnormal handling strategy to generate a knowledge base update instruction includes: Extract the attention error pattern in the trainee ability evaluation report, and perform a matching degree calculation process on the attention error pattern and the abnormal handling strategy to generate a strategy failure index; When the strategy failure index exceeds the preset threshold, perform a reverse derivation process on the abnormal handling strategy to locate the strategy defect characteristics; Construct a strategy optimization suggestion based on the strategy defect characteristics, and perform a similarity matching process on the strategy optimization suggestion and the historical handling cases in the production-education collaborative knowledge base. When the similarity matching result is lower than the preset standard, generate a knowledge base update instruction including an iterative handling strategy, and add the iterative handling strategy to the production-education collaborative knowledge base.

8. The method for production-education collaborative training based on digital twin according to claim 7, wherein, The constructing a strategy optimization suggestion based on the strategy defect characteristics includes: Call a large language model to perform a multi-dimensional analysis process on the strategy defect characteristics to generate a defect description text; Perform a knowledge retrieval process in the production-education collaborative knowledge base based on the defect description text to obtain relevant technical documents and expert experience data; Perform a semantic fusion process on the relevant technical documents and expert experience data to generate a candidate optimization strategy set; Perform a simulation verification process on the candidate optimization strategy set through the industrial equipment digital twin to screen out the optimization strategies that meet the preset safety standards as the strategy optimization suggestions; Among them, the performing a simulation verification process on the candidate optimization strategy set through the industrial equipment digital twin includes: Load the candidate optimization strategy set into the abnormal event injection interface of the industrial equipment digital twin to trigger a simulated abnormal scenario; Collect the equipment response data and operation feedback data of the industrial equipment digital twin in the simulated abnormal scenario; Perform a stability evaluation process on the equipment response data to generate an equipment safety index, and perform a complexity evaluation process on the operation feedback data to generate an operation feasibility index; When the equipment safety index and the operation feasibility index both meet the preset threshold, mark the corresponding candidate optimization strategy as an effective strategy, otherwise perform an iterative optimization process until the conditions are met.

9. A production-education collaborative training system based on digital twin, characterized in that, It includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions or codes. The processor is used to execute the programs, instructions or codes in the memory to implement the digital twin-based production-education collaborative training method according to any one of claims 1-8 above.

Images