Practical training method for digital twinning system of robot

Abstract

The invention provides a digital twin system practical training method for a robot. The method comprises the following steps: respectively executing and acquiring behavior output data when the robot and a digital twin system execute the same operation step, and judging whether responses of the digital twin system and the robot are consistent or not; when the responses are inconsistent, analyzing the system resource use conditions of the digital twin system under different load conditions, judging the abnormal degree of processing the behavior output data by the system resources, and measuring the behavior output data transmission delay from the robot to the digital twin system; and analyzing the stability of behavior output data processing of system resources and the real-time performance of behavior output data transmission so as to evaluate the practical training effect of the digital twin system of the robot. According to the invention, the stability problem of the digital twinning system can be found in time, and the training effect can be adjusted in time, so that the training effect and the stability of the digital twinning system corresponding to the robot are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot simulation training, and in particular to a digital twin system training method for a robot. Background Art

[0002] The digital twin system training of robots refers to the use of digital twin technology to provide a virtual training platform for industrial robot operators or engineers. By building a digital twin system of industrial robots and synchronizing it with the actual robots in real time, trainees can practice operation and troubleshooting in a virtual environment, thereby improving their practical operation skills and problem-solving abilities. This process not only reduces training costs and risks, but also better simulates complex industrial scenarios, ensuring that trainees have sufficient experience and adaptability before entering the actual working environment.

[0003] In the digital twin system training scenario, the digital twin system (i.e., virtual model) can accurately simulate various actions and responses of the robot, including its physical characteristics and behavior patterns. Through real-time interaction with sensor data and actual operation data, trainees can experience actual operation feedback in a virtual environment. This approach not only improves the efficiency and effectiveness of training, but also provides strong support for enterprises in the application and promotion of new technologies and equipment, and promotes the development of intelligent manufacturing. Existing technologies have defects. For example, the complexity of digital twin systems and the characteristics of multi-component integration make them prone to system crashes, software errors or hardware failures. If the digital twin system frequently encounters problems during the training process, it will not only interrupt the training process, but also reduce the trainees' learning experience and effect. More seriously, if this system instability extends to the actual production process, it may cause production line shutdowns, product quality decline, and even safety accidents. Summary of the invention

[0004] The purpose of the present invention is to disclose a digital twin system training method for a robot to solve the aforementioned technical problems, and in particular to realize the technical problems that the complexity of the digital twin system and the characteristics of multi-component integration make it prone to digital twin system crashes, software errors or hardware failures, so as to improve the training effect of the digital twin system and dynamically adjust the training effect.

[0005] To achieve the above object, the present invention provides a robot digital twin system training method, comprising the following steps:

[0006] S1: respectively executing and obtaining the behavior output data of the robot and the digital twin system corresponding to the robot when performing the same operation steps, and using the behavior output data as input, determining whether the responses of the digital twin system and the robot are consistent through the isolation forest algorithm;

[0007] S2: When the responses of the digital twin system and the robot are inconsistent, the system resource usage of the digital twin system under different load conditions is analyzed to determine the degree of abnormality of the system resources in processing the behavior output data and to evaluate the stability of the system resources in processing the behavior output data;

[0008] S3: Use the ping command to measure the behavior output data transmission delay from the robot to the digital twin system, analyze the delay fluctuation of the behavior output data transmission, and evaluate the real-time performance of the behavior output data transmission;

[0009] S4: Analyze the stability of the system resources in processing the behavior output data and the real-time transmission of the behavior output data to evaluate the training effect of the robot's digital twin system.

[0010] As a further improvement of the present invention, the step S1 is specifically as follows:

[0011] When the robot and the digital twin system perform the same operation steps, the behavior output data is preprocessed, and the behavior output data includes one of position data, speed data or posture data or a combination of any several data;

[0012] Taking the pre-processed behavior output data as input, performing anomaly detection on the pre-processed behavior output data by using an isolation forest algorithm to obtain an anomaly detection result;

[0013] Based on the anomaly detection results, outliers in the behavioral output data of the robot and the digital twin system are identified.

[0014] As a further improvement of the present invention, the method of identifying abnormal values ​​in the behavior output data of the robot and the digital twin system according to the abnormality detection result is specifically:

[0015] The difference sd of the number of outliers between the robot and the digital twin system is calculated by the following formula:

[0016] sd=|n actual -n virtua |;

[0017] Where n actual is the number of outliers of the robot, n virtua is the number of outliers in the digital twin system;

[0018] When the difference sd is greater than or equal to the preset standard difference, the digital twin system and the robot respond inconsistently; when the difference sd is less than the preset standard difference, the digital twin system and the robot respond consistently.

[0019] As a further improvement of the present invention, the determination of the abnormality of the system resources in processing the behavior output data in step S2 is specifically as follows:

[0020] Analyze the deviation of the average response time of the output data of the system resource processing behavior of the digital twin system to obtain the response time deviation index WD;

[0021] The specific method for obtaining the response time deviation index WD is:

[0022] Collect response time series of output data of system resource processing behavior;

[0023] Select a prior distribution p(θ), where the parameter (θ) represents the distribution parameter of the response time;

[0024] Construct the likelihood function p({t e}|θ) as a given parameter (θ), and the system resource processing behavior output data response time t is calculated by the following formula e The probability of occurrence p({t e}|μ,σ 2 ):

[0025]

[0026] Wherein, μ is the average value of the response time series of the output data of the system resource processing behavior, σ is the standard deviation of the response time series of the output data of the system resource processing behavior, and the response time series is {t1, t2, t3, ..., t e ,...,t q}, where e=1, 2, 3, ..., q, q is a positive integer greater than 0;

[0027] Use Bayesian theorem to update the prior distribution p(θ) to obtain the output data response time t of system resource processing behavior e The parameter distribution of

[0028] The new response time distribution p(t new ∣{t e}),

[0029] p(t new ∣{t e})=∫p(t new |p(θ){t e})dθ;

[0030] According to the new response time distribution p(t new ∣{t e}) and calculate the response time deviation index WD by the following formula:

[0031]

[0032] In the formula, σ new is the standard deviation of the predicted new response time series.

[0033] As a further improvement of the present invention, it also includes:

[0034] The obtained response time deviation index WD is compared with the standard response time deviation index threshold pre-set in the historical data. If the response time deviation index WD is greater than or equal to the standard response time deviation index threshold, a system resource data processing abnormal signal is generated; if the response time deviation index WD is less than the standard response time deviation index threshold, a system resource data processing normal signal is generated.

[0035] As a further improvement of the present invention, the step S3 is specifically as follows:

[0036] Execute the ping command to obtain the RTT value sequence of the data packets formed by each output data transmission, and select a fixed-size window N, which represents the moving average window size;

[0037] Starting from the Nth data packet, calculate the average value of the first N data packets as the Nth moving average value, and move the window backward in sequence. Each time a new data packet is added, calculate the average value of all data packets in the current window to obtain the moving average value DG of the RTT of the next data packet.

[0038] The moving average

[0039] Where x is the total number of packets, RTT j is the RTT value of the jth data packet;

[0040] Perform a difference operation on each moving average value DG and the RTT value of the data packet to obtain the fluctuation value sequence of each data packet RTT, and calculate the standard deviation GK of the fluctuation value sequence.

[0041] As a further improvement of the present invention, step S3 further includes:

[0042] The RTT fluctuation value of the data packet is compared with the RTT fluctuation value of the data packet under the standard state in the historical data. If the RTT fluctuation value of the data packet is greater than or equal to the RTT fluctuation value of the data packet under the standard state in the historical data, the output behavior outputs a data transmission abnormality signal; if the RTT fluctuation value of the data packet is less than the RTT fluctuation value of the data packet under the standard state in the historical data, the output behavior outputs a data transmission normal signal.

[0043] As a further improvement of the present invention, the step S4 further includes:

[0044] The response time deviation index WD and the data packet RTT fluctuation value are normalized, and the training effect evaluation coefficient GH is calculated by the normalized response time deviation index WD and the data packet RTT fluctuation value;

[0045] The training effect evaluation coefficient

[0046] Where a1 and a2 are the proportional coefficients of the response time deviation index WD and the standard deviation GK of the fluctuation value sequence, respectively, and a2>a1>0.

[0047] As a further improvement of the present invention, step S5 is also included: according to the evaluation result of the training effect, the training effect is divided into high training effect, general training effect and low training effect; when the training effect is a general training effect, a fault risk prediction evaluation is performed on the digital twin system through fuzzy rules, and the training effect is dynamically adjusted according to the result of the fault risk prediction evaluation.

[0048] As a further improvement of the present invention, the same operating steps are standard operating steps predefined by the robot and the digital twin system corresponding to the robot.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention comprehensively analyzes the stability of data processing and the real-time performance of data transmission of the digital twin system's system resources, comprehensively evaluates the training effect of the robot's digital twin system, and dynamically adjusts it according to the evaluation results. The comprehensive evaluation mechanism disclosed in the present invention can not only timely discover the stability problems of the digital twin system corresponding to the robot, but also timely adjust the training effect, thereby improving the training effect of the digital twin system and the stability of the digital twin system corresponding to the robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is an overall flow chart of the digital twin training method of the robot of the present invention;

[0052] Figure 2 Another overall flow chart of the digital twin training method for the robot of the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below in conjunction with the various embodiments shown in the accompanying drawings, but it should be noted that these embodiments are not limitations of the present invention, and any equivalent transformations or substitutions in functions, methods, or structures made by ordinary technicians in the field based on these embodiments are all within the scope of protection of the present invention.

[0054] Please refer to Figure 1 As shown, this embodiment discloses a specific implementation method of a digital twin system training method of a robot of the present invention, which includes the following steps S1 to S4.

[0055] First, step S1: respectively execute and obtain the behavior output data of the robot and the digital twin system corresponding to the robot when performing the same operation steps, and use the behavior output data as input to determine whether the responses of the digital twin system and the robot are consistent through the isolation forest algorithm. In step S1, the so-called same operation steps are standard operation steps predefined by the robot and the digital twin system corresponding to the robot. The purpose of obtaining the behavior output data of the actual robot and its corresponding digital twin system when performing the same operation steps is to compare their actions, states and other information to evaluate the accuracy of the digital twin system simulation.

[0056] Specifically, step S1 is as follows: when the robot and the digital twin system perform the same operation steps, the behavior output data is collected, and the behavior output data is preprocessed. The behavior output data includes one of the position data, speed data or posture data, or a combination of any several data. The behavior output data is acquired using sensors and data acquisition equipment to ensure accurate recording of the process and results of the robot's operation. Taking the preprocessed behavior output data as input, the preprocessed behavior output data is subjected to anomaly detection through the isolation forest algorithm to obtain anomaly detection results. According to the anomaly detection results, the outliers in the behavior output data of the robot and the digital twin system are identified. The preprocessing in this application includes removing noise, processing missing values, normalizing data, etc. Ensure the quality and reliability of the data.

[0057] The Isolation Forest Algorithm is a tree-based unsupervised anomaly detection algorithm that identifies outliers by building a tree structure. Based on the results of the Isolation Forest Algorithm, outliers in the output data of the actual robot and the digital twin system are identified. Outliers are data points that are isolated in the tree structure and are generally considered to be abnormal. The Isolation Forest Algorithm assigns an anomaly score to each data point and identifies outliers based on the anomaly score; analyzes outliers in the output data of the actual robot and the digital twin system, and compares whether the number and distribution of outliers between the two are consistent.

[0058] In this specific implementation, the behavior output data of the actual robot and the digital twin system when performing the same operation steps are obtained by the isolation forest algorithm for analysis, which can help evaluate the accuracy and reliability of the digital twin system in simulating the actual robot. In this specific implementation, by comparing the outliers in the behavior output data of the actual robot and the digital twin system, the difference in the behavior output data of the robot and the digital twin system can be deeply understood, so as to judge whether the response between the digital twin system and the actual robot is consistent. If the outlier distribution and the number of outliers of the robot and the digital twin system are basically consistent, it can be shown that the digital twin system has a good simulation effect on the actual physical robot and can accurately reflect the behavior characteristics of the actual robot; otherwise, it shows that the digital twin system has a poor simulation effect on the actual physical robot. The behavior output data of the actual robot and the digital twin system when performing the same operation steps are obtained by the isolation forest algorithm for analysis, and the consistency is judged, which is of great significance for the design, optimization and application of the digital twin system.

[0059] Depending on the actual situation, the robot may need to be calibrated and configured to ensure that it accurately performs the operation steps. According to the pre-defined standard operation steps, the operator performs the corresponding command operations on the actual robot. The actual robot performs the operation and completes the corresponding task or action. In the digital twin system, by inputting the corresponding instructions or control signals, the actual robot is simulated to perform the same operation steps. The digital twin system simulates the movements and behaviors of the robot to complete the corresponding tasks or actions. During the operation of the digital twin system, the output data of the system is recorded, such as the position, speed, acceleration and other status information of the simulated robot. The digital twin system usually automatically records various parameters and status information during the simulation process. The pre-defined standard operation steps contain the initial state and environmental parameters of the digital twin system.

[0060] In this embodiment, the behavior output data also includes: the position data of the robot, the position coordinates of the robot, for example, including the position information of the robot on the x, y, and z axes. The speed data is the movement speed of the robot in all directions. The posture data describes the posture of the robot, and the posture data specifically includes: data such as orientation and rotation angle. The sensor data is the environmental data collected by the sensor connected to the robot. If the robot is equipped with sensors, such as cameras, laser radars, etc., the environmental information collected by the sensors can also be obtained. Execution action data, the specific actions performed by the robot, for example, specific actions such as moving, grabbing, and rotating. At the same time, the behavior output data of the robot and the digital twin system are visualized, and the motion trajectory, speed change, etc. of the robot and the digital twin system corresponding to the robot are compared. The data can be presented in the form of charts, animations, etc., and the difference between the two can be observed intuitively. The actual robot is a physical robot and has the same concept as the robot in this application, while the digital twin system is a virtual model corresponding to the actual robot (i.e., robot), and a digital mapping system connected to the robot can be established through the digital twin system simulation.

[0061] Then, execute step S2: when the response of the digital twin system and the robot is inconsistent, analyze the system resource usage of the digital twin system under different load conditions, determine the degree of abnormality of the system resources in processing the behavior output data, and evaluate the stability of the system resources in processing the behavior output data. In this application, according to the abnormality detection results, the identification of abnormal values ​​in the behavior output data of the robot and the digital twin system is specifically: the difference sd of the number of abnormal values ​​of the robot and the digital twin system is calculated by the following formula. sd=|n actual -n virtua |; In the formula, n actual is the number of outliers of the robot, n virtua is the number of outliers in the digital twin system. When the difference sd is greater than or equal to the preset standard difference, it means that the digital twin system and the robot respond inconsistently; when the difference sd is less than the preset standard difference, it means that the digital twin system and the robot respond consistently.

[0062] The more specific meaning of analyzing the system resource usage of the digital twin system under different load conditions (i.e., workload) in step S2 is to identify potential performance bottlenecks and unstable factors by monitoring and comparing indicators such as CPU usage, memory usage, network bandwidth, disk I / O, and response time, thereby providing a basis for optimizing the digital twin system and ensuring that it can efficiently and reliably simulate the actual robot in practical applications. The system resources of the digital twin system include: CPU, memory, disk I / O, and network bandwidth, among which the CPU is responsible for performing calculations and task processing, the memory is used to store and quickly access data, the disk I / O processes the persistent storage and reading of data, and the network bandwidth ensures data transmission and communication inside and outside the system, thereby ensuring the overall performance and stability of the digital twin system.

[0063] Determining the abnormal degree of the system resource processing of the behavior output data in step S2 is specifically: analyzing the deviation of the average response time of the system resource processing behavior output data of the digital twin system to obtain the response time deviation index WD.

[0064] Specifically, the response time deviation index WD is obtained by: collecting the response time series of the output data of the system resource processing behavior. Selecting the prior distribution p(θ), where the parameter (θ) represents the distribution parameter of the response time. Constructing the likelihood function p({t e}|θ) as a given parameter (θ), and the system resource processing behavior output data response time t is calculated by the following formula e The probability of occurrence p({t e}|μ,σ 2 ): Where μ is the average value of the response time series of the output data of the system resource processing behavior, σ is the standard deviation of the response time series of the output data of the system resource processing behavior, and the response time series of the system resource processing data is collected, and the response time series is {t1, t2, t3, ..., t e ,...,t q}, where e = 1, 2, 3, ..., q, and q is a positive integer greater than 0. Use Bayes' theorem to update the prior distribution p(θ) and obtain the output data response time t of the system resource processing behavior. e The parameter distribution of The new response time distribution p(t new ∣{t e}), p(t new ∣{t e})=∫p(t new |p(θ){t e According to the new response time distribution p(t new ∣{te}) and calculate the response time deviation index WD by the following formula: In the formula, σ new is the standard deviation of the predicted new response time series.

[0065] More specifically, step S2 in the present application also includes: comparing the acquired response time deviation index WD with the standard response time deviation index threshold pre-set in the historical data; if the response time deviation index WD is greater than or equal to the standard response time deviation index threshold, generating a system resource data processing abnormal signal; if the response time deviation index WD is less than the standard response time deviation index threshold, generating a system resource data processing normal signal.

[0066] Then, execute step S3: use the ping command to measure the behavior output data transmission delay from the robot to the digital twin system, analyze the delay fluctuation of the behavior output data transmission, and evaluate the real-time performance of the behavior output data transmission.

[0067] In this embodiment, step S3 is specifically as follows: execute the ping command to obtain the RTT value sequence of the data packets formed by each output data transmission, select a fixed-size window N, which represents the window size of the moving average; starting from the Nth data packet, calculate the average value of the first N data packets as the Nth moving average value, and move the window backward in sequence. Each time a new data packet is added, calculate the average value of all data packets in the current window to obtain the moving average value DG of the RTT of the next data packet.

[0068] Moving Average

[0069] Where x is the total number of packets, RTT j is the RTT value of the jth data packet.

[0070] Perform a difference operation on each moving average value DG and the RTT value of the data packet to obtain the fluctuation value sequence of each data packet RTT, and calculate the standard deviation GK of the fluctuation value sequence.

[0071] In this specific embodiment, it is necessary to ensure that the robot and its corresponding digital twin system are connected to the same network or can access each other through the network (for example, ISCSI or internal LAN). Get the IP address or host name of the digital twin system. Open the control terminal on the robot, and open the command line interface on the control terminal of the robot through SSH connection, physical connection or other remote access methods. Enter the ping command on the visual user graphical interface (GUI) displayed by the robot's terminal, and specify the IP address or host name of the digital twin system. Thus, by typing the ping command, the round-trip time of each data packet is output, and a certain number of data packet samples are collected. The ping command is usually executed for a few seconds or a certain number of packets is specified; the ping command will give statistical results after completion, including: the RTT value of the minimum deviation, the RTT value of the maximum deviation, the RTT value of the average deviation, and the RTT value of the standard deviation. Record the RTT value of each ping and analyze the fluctuation of data transmission delay.

[0072] Step S3 also includes: comparing the RTT fluctuation value of the data packet with the RTT fluctuation value of the data packet under the standard state in the historical data. If the RTT fluctuation value of the data packet is greater than or equal to the RTT fluctuation value of the data packet under the standard state in the historical data, it indicates that the real-time performance of the data transmission is worse, and thus the output behavior outputs a data transmission abnormality signal; if the RTT fluctuation value of the data packet is less than the RTT fluctuation value of the data packet under the standard state in the historical data, it indicates that the real-time performance of the data transmission is better, and thus the output behavior outputs a data transmission normal signal.

[0073] Finally, step S4 is executed: the stability of the system resources in processing the behavior output data and the real-time transmission of the behavior output data are analyzed to evaluate the training effect of the digital twin system of the robot. Preferably, in this specific embodiment, step S4 also includes: normalizing the response time deviation index WD and the data packet RTT fluctuation value, and calculating the training effect evaluation coefficient GH through the normalized response time deviation index WD and the data packet RTT fluctuation value.

[0074] Training effect evaluation coefficient

[0075] Where a1 and a2 are the proportional coefficients of the response time deviation index WD and the standard deviation GK of the fluctuation value sequence, respectively, and a2>a1>0.

[0076] It should be noted that instability means that the increase in the RTT fluctuation value of the data packet means that the delay of data transmission is unstable, that is, the round-trip time of the data packet varies greatly. If there is instability, it may cause the delay fluctuation of the data packet during transmission, making the data transmission time unpredictable, thereby affecting the real-time performance of data transmission. Delay variation means that a large fluctuation value means that there is a large difference between the RTT values ​​of the data packets, and sometimes there may be a sudden increase or decrease in delay. Delay variation will increase the uncertainty of data transmission and reduce the degree to which the digital twin system meets the real-time requirements. Packet loss risk means that the increase in the RTT fluctuation value of the data packet may increase the risk of data packet loss. During network transmission, if the RTT value of the data packet fluctuates greatly, it may cause the data packet to time out or retransmit during transmission, thereby increasing the possibility of data loss and reducing the real-time performance and reliability of data transmission. The response time is unstable. For application scenarios that require instant response, a large RTT fluctuation value will cause the response time of the digital twin system to be unstable and the user experience to be poor. For example, in scenarios such as real-time video conferencing or online games, if the real-time performance of data transmission is poor, it may cause problems such as screen freezes and sound delays, affecting the user experience.

[0077] Large fluctuations in response time mean that when the digital twin system processes the same or similar tasks, the response time varies greatly and cannot guarantee consistent processing speed. Unpredictable performance. Due to the large deviation between the actual response time and the predicted time, the system performance becomes unpredictable and may show obvious performance degradation in critical tasks or high-load conditions. Poor user experience. For applications with high real-time requirements, such as robot control systems, large response time deviations will cause the system to respond untimely, thus affecting user experience and task execution efficiency. Potential resource bottlenecks. A large deviation index may indicate bottlenecks or uneven use of system resources (such as CPU, memory, disk I / O, or network bandwidth), resulting in abnormally high or low utilization of system resources under certain load conditions.

[0078] Combination Figure 1 and Figure 2 As shown, the present application also discloses another specific implementation method of a digital twin system training method for a robot.

[0079] Compared with the digital twin system training method of the robot disclosed in the aforementioned embodiment, the digital twin system training method of the robot in the present application further includes step S5 after steps S1 to S4 are executed: according to the evaluation results of the training effect, the training effect is divided into high training effect, general training effect and low training effect. When the training effect is a general training effect, the digital twin system is evaluated for fault risk prediction through fuzzy rules, and the training effect is dynamically adjusted according to the results of the fault risk prediction evaluation. Step S5 is specifically described as follows.

[0080] According to the evaluation results, the training effect of the robot's digital twin system is divided into high training effect, general training effect, and low training effect. According to the classification of the training effect, the digital twin system is continuously optimized to improve its stability. The obtained effect evaluation coefficient of the robot's digital twin system training is compared with the gradient standard threshold. The gradient standard threshold includes the first standard threshold and the second standard threshold, and the first standard threshold is less than the second standard threshold. The effect evaluation coefficient of the robot's digital twin system training is compared with the first standard threshold and the second standard threshold respectively.

[0081] Specifically, if the effect evaluation coefficient is greater than the second standard threshold, the training effect of the robot's digital twin system is classified as a high training effect. Regular system maintenance and updates are required, including software updates, installation of security patches, optimization of system configurations, etc., to ensure that the system is always in the best state. If the effect evaluation coefficient is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the training effect of the robot's digital twin system is classified as a general training effect; a detailed diagnosis and analysis of the digital twin system is performed to identify the performance bottlenecks and problems of the digital twin system, and targeted improvement measures are taken. If the effect evaluation coefficient is less than the first standard threshold, the training effect of the robot's digital twin system is classified as a low training effect. Major improvements and optimizations of the digital twin system are required in a targeted manner. Conduct a comprehensive review and reconstruction of the digital twin system, and redesign the digital twin system architecture or key components to meet actual needs and improve the scalability and stability of the digital twin system.

[0082] In this embodiment, by executing the same standard operating steps on the robot and its corresponding digital twin system and analyzing its behavioral output data, it is determined whether the responses of the robot and the digital twin system corresponding to the robot are consistent when executing the same operating steps. If inconsistency occurs, the stability of the digital twin system data processing is evaluated by analyzing the usage of the system resources of the digital twin under different load conditions. Subsequently, the data transmission delay is measured using the ping command and its fluctuation is analyzed to evaluate the real-time performance of data transmission. The stability of the digital twin system resources and the real-time performance of data transmission are comprehensively analyzed to evaluate the training effect, and the training effect is divided into three categories: high, general, and low. Then, the digital twin system is continuously optimized according to the evaluation results to improve the stability and training effect of the digital twin system.

[0083] When the training effect is classified as a general training effect, the digital twin system is subjected to a fault risk prediction and evaluation, and the training is dynamically adjusted according to the evaluation results. Specifically, in the present application, a fault risk prediction and evaluation is performed on the digital twin system through fuzzy rules, and the training is dynamically adjusted according to the evaluation results, and specifically: when the training effect of the digital twin system of the robot is classified as a general training effect, that is, the effect evaluation coefficient of the training of the digital twin system of the robot generated within a fixed time is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, and the effect evaluation coefficient generated in the subsequent fixed time period is used to establish a data set, and the effect evaluation coefficient in the data set and the effect evaluation coefficient reference threshold are used as input items, and include the first input item and the second input item. First input item: effect evaluation coefficient (such as high, medium, low); second input item: effect evaluation coefficient reference threshold (such as high, medium, low); output item: digital twin system operation status (such as normal, faulty).

[0084] Defining fuzzy rules means that if the effect evaluation coefficient is high and the effect evaluation coefficient reference threshold is low, then the operation state of the digital twin system is normal operation; if the effect evaluation coefficient is medium and the effect evaluation coefficient reference threshold is medium, then the operation state of the digital twin system is normal operation; if the effect evaluation coefficient is low and the effect evaluation coefficient reference threshold is high, then the operation state of the digital twin system fails. The effect evaluation coefficient and the effect evaluation coefficient reference threshold are converted into membership values ​​in the fuzzy set respectively. The defined fuzzy rules are used to infer the fuzzified input data to obtain the fuzzy results of the output items. The fuzzification is defuzzified by the centroid method, and the fuzzified results are converted into the specific operation state of the digital twin system, and the training is dynamically adjusted, wherein the training is dynamically adjusted according to the defuzzification results. Including: if the operation state of the digital twin system is normal operation, the current training state is continued; if the operation state of the digital twin system is a fault, corresponding countermeasures are taken according to the specific situation, such as suspending the training, increasing monitoring, repairing the digital twin system, etc., to ensure the smooth progress of the training and the learning effect.

[0085] In this embodiment, in the training of the digital twin system of the robot, if the training effect is at a general level, fault risk prediction evaluation and dynamic adjustment will be performed. The digital twin system is evaluated by fuzzy rules, and the training effect and the reference threshold are fuzzified and reasoned to determine the operating status of the digital twin system. If the digital twin system operates normally, the current training status is maintained; if the digital twin system fails, measures are taken according to the situation, such as suspending the training or repairing the digital twin system, to ensure that the training can be carried out effectively and avoid the collapse of the digital twin system.

[0086] This embodiment has the same technical solution as that contained in the aforementioned first specific implementation mode, which is shown in the first specific implementation mode and will not be described in detail here.

[0087] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent implementation methods or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

[0088] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0089] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A robot digital twin system training method, characterized in that: The following steps are involved: S1: respectively executing and obtaining the behavior output data of the robot and the digital twin system corresponding to the robot when performing the same operation steps, and using the behavior output data as input, determining whether the responses of the digital twin system and the robot are consistent through the isolation forest algorithm; S2: When the responses of the digital twin system and the robot are inconsistent, the system resource usage of the digital twin system under different load conditions is analyzed to determine the degree of abnormality of the system resources in processing the behavior output data and to evaluate the stability of the system resources in processing the behavior output data; S3: Use the ping command to measure the behavior output data transmission delay from the robot to the digital twin system, analyze the delay fluctuation of the behavior output data transmission, and evaluate the real-time performance of the behavior output data transmission; S4: Analyze the stability of the system resources in processing the behavior output data and the real-time transmission of the behavior output data to evaluate the training effect of the robot's digital twin system.

2. The digital twin system training method according to claim 1, characterized in that: The step S1 is specifically as follows: When the robot and the digital twin system perform the same operation steps, the behavior output data is preprocessed, and the behavior output data includes one of position data, speed data or posture data or a combination of any several data; Taking the pre-processed behavior output data as input, performing anomaly detection on the pre-processed behavior output data by using an isolation forest algorithm to obtain an anomaly detection result; Based on the anomaly detection results, outliers in the behavioral output data of the robot and the digital twin system are identified.

3. The digital twin system training method according to claim 2, characterized in that: The method of identifying abnormal values ​​in the behavior output data of the robot and the digital twin system according to the abnormality detection result is specifically: The difference sd of the number of outliers between the robot and the digital twin system is calculated by the following formula: sd=|n actual -n virtua |; Where n actual is the number of outliers of the robot, n virtua is the number of outliers in the digital twin system; When the difference sd is greater than or equal to the preset standard difference, the digital twin system and the robot respond inconsistently; when the difference sd is less than the preset standard difference, the digital twin system and the robot respond consistently.

4. The digital twin system training method according to claim 1, characterized in that: The determination of the abnormality of the system resources in processing the behavior output data in step S2 is specifically as follows: Analyze the deviation of the average response time of the output data of the system resource processing behavior of the digital twin system to obtain the response time deviation index WD; The specific method for obtaining the response time deviation index WD is: Collect response time series of output data of system resource processing behavior; Select a prior distribution p(θ), where the parameter (θ) represents the distribution parameter of the response time; Construct the likelihood function p({t e }|θ) as a given parameter (θ), and the system resource processing behavior output data response time t is calculated by the following formula e The probability of occurrence p({t e }|μ,σ 2 ): Wherein, μ is the average value of the response time series of the output data of the system resource processing behavior, σ is the standard deviation of the response time series of the output data of the system resource processing behavior, and the response time series is {t1, t2, t3, ..., t e ,...,t q }, where e=1, 2, 3, ..., q, q is a positive integer greater than 0; Use Bayesian theorem to update the prior distribution p(θ) to obtain the output data response time t of system resource processing behavior e The parameter distribution of The new response time distribution p(t new ∣{t e }), p(t new ∣{t e })=∫p(t new ∣p(θ){t e })dθ; According to the new response time distribution p(t new ∣{t e }) and calculate the response time deviation index WD by the following formula: In the formula, σ new is the standard deviation of the predicted new response time series.

5. The digital twin system training method according to claim 4, characterized in that: Also includes: The obtained response time deviation index WD is compared with the standard response time deviation index threshold pre-set in the historical data. If the response time deviation index WD is greater than or equal to the standard response time deviation index threshold, a system resource data processing abnormal signal is generated; if the response time deviation index WD is less than the standard response time deviation index threshold, a system resource data processing normal signal is generated.

6. The digital twin system training method according to claim 5, characterized in that: The step S3 is specifically as follows: Execute the ping command to obtain the RTT value sequence of the data packets formed by each output data transmission, and select a fixed-size window N, which represents the moving average window size; Starting from the Nth data packet, calculate the average value of the first N data packets as the Nth moving average value, and move the window backward in sequence. Each time a new data packet is added, calculate the average value of all data packets in the current window to obtain the moving average value DG of the RTT of the next data packet. The moving average Where x is the total number of packets, RTT j is the RTT value of the jth data packet; Perform a difference operation on each moving average value DG and the RTT value of the data packet to obtain the fluctuation value sequence of each data packet RTT, and calculate the standard deviation GK of the fluctuation value sequence.

7. The digital twin system training method according to claim 6, characterized in that: The step S3 further comprises: The RTT fluctuation value of the data packet is compared with the RTT fluctuation value of the data packet under the standard state in the historical data. If the RTT fluctuation value of the data packet is greater than or equal to the RTT fluctuation value of the data packet under the standard state in the historical data, the output behavior outputs a data transmission abnormality signal; if the RTT fluctuation value of the data packet is less than the RTT fluctuation value of the data packet under the standard state in the historical data, the output behavior outputs a data transmission normal signal.

8. The digital twin system training method according to claim 7, characterized in that: The step S4 further comprises: The response time deviation index WD and the data packet RTT fluctuation value are normalized, and the training effect evaluation coefficient GH is calculated by the normalized response time deviation index WD and the data packet RTT fluctuation value; The training effect evaluation coefficient Where a1 and a2 are the proportional coefficients of the response time deviation index WD and the standard deviation GK of the fluctuation value sequence, respectively, and a2>a1>0.

9. The digital twin system training method according to claim 8, characterized in that: It also includes step S5: according to the evaluation results of the training effect, the training effect is divided into high training effect, general training effect and low training effect. When the training effect is a general training effect, the digital twin system is evaluated for fault risk prediction through fuzzy rules, and the training effect is dynamically adjusted according to the results of the fault risk prediction evaluation.

10. The digital twin system training method according to claim 1, characterized in that: The same operating steps are standard operating steps predefined by the robot and the digital twin system corresponding to the robot.

Images