Predictive Monitoring and Maintenance System for Industrial Robots Based on Digital Twins

Abstract

This invention relates to the field of industrial robot data processing technology, specifically to a predictive monitoring and maintenance system for industrial robots based on digital twins. The system includes: a data acquisition module for acquiring multiple core data sets; a data processing module for determining the initial reference level of each piece of real data, correcting the initial reference level to obtain an effective reference level; a data calculation module for determining the abnormal latency of each piece of twin data and calculating the effective reference level of each piece of twin data; a machine learning module for extracting data features from each piece of core data, setting weight values ​​for the data features, and inputting the data features and weight values ​​of each piece of core data into a machine learning model to obtain status monitoring results; and a monitoring and maintenance module for performing predictive monitoring and maintenance of key parts of the industrial robot based on the status monitoring results. This invention enables accurate and reliable predictive monitoring and maintenance of industrial robots.

Description

Technical Field

[0001] This invention relates to the field of industrial robot data processing technology, and specifically to a predictive monitoring and maintenance system for industrial robots based on digital twins. Background Technology

[0002] Digital twins are a technology that creates virtual models of physical entities using digital means and utilizes real-time data for simulation, monitoring, and optimization. In the context of digital twins, the condition monitoring and maintenance of critical components of industrial robots is of paramount importance.

[0003] In related technologies, the monitoring and maintenance of key parts of industrial robots usually relies solely on real data collected by data acquisition devices such as sensors. However, a single data source is insufficient to fully reflect the health status of key parts of industrial robots, and the real data collected by data acquisition devices such as sensors is easily affected by various factors, resulting in insufficient accuracy and reliability in the monitoring and maintenance process of key parts of industrial robots. Summary of the Invention

[0004] To address the technical problems of traditional monitoring and maintenance solutions, such as limited data sources and susceptibility to interference from various factors, leading to insufficient accuracy and reliability in the monitoring and maintenance of critical components of industrial robots, this invention aims to provide a predictive monitoring and maintenance system for industrial robots based on digital twins. The specific technical solution adopted is as follows:

[0005] A predictive monitoring and maintenance system for industrial robots based on digital twins, the system comprising:

[0006] The data acquisition module is used to acquire multiple core data points of key parts of the industrial robot; each of these core data points includes both real data and twin data.

[0007] The data processing module is used to determine the initial reference value and missing core degree of each piece of real data, and to correct the initial reference value of each piece of real data based on the missing core degree to obtain the effective reference value of each piece of real data.

[0008] The data calculation module is used to determine the abnormal latency of each twin data item, and calculate the effective reference degree of each twin data item based on the abnormal latency of each twin data item and the effective reference degree of the corresponding real data item.

[0009] The machine learning module is used to extract the data features of each core data item, and set the weight value of the corresponding data feature according to the effective reference degree of the real data and the effective reference degree of the twin data in each core data item. The data features and weight values ​​of each core data item are input into the pre-built machine learning model to obtain the state monitoring results output by the machine learning model.

[0010] The monitoring and maintenance module is used to perform predictive monitoring and maintenance on key parts of the industrial robot based on the status monitoring results.

[0011] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins is provided, wherein the data processing module determines the initial reference level for each piece of real data, including:

[0012] Determine the theoretical quantity, actual quantity, and time series of real data in each core data item, as well as the time series of twin data;

[0013] The ratio of the actual quantity to the theoretical quantity is obtained by dividing the actual quantity by the theoretical quantity.

[0014] Calculate the dynamic time-normalized distance between the time series of the real data and the time series of the twin data in each core data item;

[0015] Add the dynamic time-normalized distance to 1 and take the reciprocal to obtain the distance correlation value;

[0016] The initial reference value for each piece of real data is calculated by multiplying the quantity ratio by the distance correlation value.

[0017] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins includes a data processing module that determines the degree of missing core data for each piece of real data, including:

[0018] Determine the volatility instability of each element within the time series of each real data item;

[0019] Based on the aforementioned volatility instability, clustering is performed on all data elements in the time series of each real data item to generate multiple clusters;

[0020] Determine the volatility instability and missing data for each cluster;

[0021] Based on the volatility and instability of each cluster corresponding to each piece of real data and the amount of missing data, the degree of missing core of each piece of real data is calculated.

[0022] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins determines the volatility instability of each element within a time series of real data, including:

[0023] Determine the data value, slope value, and corresponding data value and slope value of each surrounding data point within the time series of each real data point;

[0024] Calculate the absolute value of the difference between the slope value of each element data and the slope value of each surrounding data to obtain the slope difference value of each element data under each surrounding data.

[0025] Calculate the absolute value of the difference between the data value of each element and the data value of each surrounding data to obtain the numerical difference value of each element under each surrounding data.

[0026] Multiply the slope difference value corresponding to each surrounding data point by the numerical difference value to obtain the difference product value corresponding to each surrounding data point;

[0027] The mean of the product of the differences of all surrounding data for each element is used to obtain the volatility instability of each element in the time series of each real data item.

[0028] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins calculates the degree of missing core data for each piece of real data based on the volatility instability and data missing amount of each cluster corresponding to each piece of real data, including:

[0029] The summation of the fluctuation instability of all clusters corresponding to each real data point yields the cluster fluctuation summation value.

[0030] The influence of missing values ​​is obtained by dividing the volatility instability of each cluster by the sum of the cluster volatility.

[0031] Multiply the number of missing data points for each cluster by the impact of the missing values ​​to obtain the cluster missing importance for each cluster;

[0032] The mean of the cluster missing importance of all clusters corresponding to each real data item is calculated and normalized to obtain the degree of missing core of each real data item.

[0033] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins includes a data processing module that, based on the degree of missing core data, corrects the initial reference value of each piece of real data to obtain an effective reference value for each piece of real data, including:

[0034] The effective reference value of each piece of real data is obtained by multiplying the degree of missing core of each piece of real data by the initial reference value.

[0035] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins includes a data calculation module that determines the abnormal latency of each piece of twin data, including:

[0036] Obtain the acquisition time of the last element in the time series of each real data item, the generation time of the first element in the time series of each twin data item, the theoretical data volume of each real data item, and the data generation volume of each twin data item;

[0037] Calculate the absolute value of the difference between the acquisition time of the last element in the time series of each real data item and the generation time of the first element in the time series of the corresponding twin data item to obtain the time difference value of each twin data item;

[0038] Calculate the absolute value of the difference between the theoretical data volume of each real data item and the data generation volume of the corresponding twin data item to obtain the data volume difference value of each twin data item;

[0039] Based on the time difference value and the data volume difference value of each twin data item, the abnormal delay degree of each twin data item is calculated.

[0040] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins calculates the abnormal delay degree of each piece of twin data based on the time difference value and the data volume difference value of each piece of twin data, including:

[0041] The time difference value of each twin data point is divided by a preset time difference threshold to obtain the time difference ratio.

[0042] The ratio of the data volume difference value of each twin data item is obtained by dividing the data volume difference value by the preset data volume difference threshold.

[0043] The abnormal delay degree of each twin data item is calculated by multiplying the time difference ratio by the data volume difference ratio.

[0044] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins is provided. The data calculation module calculates the effective reference degree of each piece of twin data based on the abnormal latency of each piece of twin data and the effective reference degree of the corresponding real data, including:

[0045] Based on the abnormal latency of each twin data point, determine the first intermediate parameter value for each twin data point;

[0046] Multiply the first intermediate parameter value by a preset adjustment coefficient to obtain the second intermediate parameter value for each twin data item;

[0047] The effective reference value of each twin data point is calculated by subtracting the effective reference value of the corresponding real data point from the second intermediate parameter value of the twin data point.

[0048] According to the present invention, a predictive monitoring and maintenance system for industrial robots based on digital twins is provided, wherein the monitoring and maintenance module includes:

[0049] The data comparison unit is used to compare the predicted values ​​of each state in the state monitoring results with their respective state reference values ​​to obtain the comparison results.

[0050] The monitoring and maintenance unit is used to generate a monitoring and maintenance plan for predictive monitoring and maintenance of key parts of the industrial robot when at least one of the state prediction values ​​in the comparison results deviates from the corresponding state reference value.

[0051] The present invention has the following beneficial effects:

[0052] The data acquisition module acquires multiple core data points for key components of the industrial robot, including real data and twin data. The data processing module determines the initial reference value and missing core value for each real data point. Based on the missing core value, the initial reference value for each real data point is corrected to obtain the effective reference value for each real data point. Then, the data calculation module determines the abnormal latency for each twin data point. Based on the abnormal latency for each twin data point and the effective reference value for the corresponding real data point, the effective reference value for each twin data point is calculated. The machine learning module extracts the data features for each core data point and sets the weight values ​​for the corresponding data features according to the effective reference values ​​of the real data and twin data points in each core data point. The data features and weight values ​​for each core data point are input into a pre-built machine learning model to obtain the status monitoring results output by the machine learning model. Finally, the monitoring and maintenance module performs predictive monitoring and maintenance on the key components of the industrial robot based on the status monitoring results. Since the core data used in the monitoring and maintenance of key parts of industrial robots includes not only real data but also twin data generated through digital twin technology, the data sources are more diverse. By calculating the effective reference value of real data and twin data, more accurate data support can be provided for the generation of condition monitoring results. Combined with machine learning models, the condition monitoring results of key parts of industrial robots can be obtained efficiently and accurately, thereby enabling more precise and reliable predictive monitoring and maintenance of key parts of industrial robots. Attached Figure Description

[0053] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a system architecture diagram of a predictive monitoring and maintenance system for industrial robots based on digital twins, provided in one embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram showing the distribution of all elements in the time series of the m-th real data item;

[0056] Figure 3 for Figure 2 The diagram shows the clustering results of all elements in the time series. Detailed Implementation

[0057] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a predictive monitoring and maintenance system for industrial robots based on digital twins proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0059] The following is in conjunction with the appendix Figures 1 to 3 Specifically, this invention provides a specific solution for a predictive monitoring and maintenance system for industrial robots based on digital twins.

[0060] Please see Figure 1 This illustrates a system architecture diagram of a predictive monitoring and maintenance system for industrial robots based on digital twins, according to an embodiment of the present invention. Figure 1 As shown in the figure, the predictive monitoring and maintenance system for industrial robots based on digital twins provided in this embodiment of the invention specifically includes:

[0061] The data acquisition module 110 is used to acquire multiple core data points of key parts of the industrial robot; each core data point includes real data and twin data.

[0062] In this embodiment, the key part can be the joint of the industrial robot. The joint usually adopts a complex mechatronics design and is mainly composed of servo motors and reducers.

[0063] In practical applications, several core data points can characterize the health status of critical components of an industrial robot. Real data refers to data collected through sensors and other data acquisition devices, while digital twin data is generated using digital twin technology. In this embodiment, real data can include vibration data, motor current, and torque—data that reflects the health status of critical components. The statistical period for each piece of real data is a fixed value, such as being collected every 3 seconds. Digital twin data is virtual data corresponding to the real data such as vibration data, motor current, and torque.

[0064] The data processing module 120 is used to determine the initial reference value and missing core degree of each piece of real data, and to correct the initial reference value of each piece of real data based on the missing core degree, so as to obtain the effective reference value of each piece of real data.

[0065] In this embodiment, the initial reference level characterizes the preliminary referenceability of each piece of real data during subsequent monitoring and maintenance of key parts of the industrial robot. The missing core level characterizes the degree to which missing data in each piece of real data affects the accuracy of the subsequent representation of the state of key parts.

[0066] The data calculation module 130 is used to determine the abnormal latency of each twin data item, and calculate the effective reference degree of each twin data item based on the abnormal latency of each twin data item and the effective reference degree of the corresponding real data item.

[0067] Understandably, the anomaly latency can characterize the degree of delay in the twin data generation process caused by abnormal circumstances. The effective reference value of twin data can characterize its reference value in subsequent monitoring and maintenance of key parts of industrial robots.

[0068] The machine learning module 140 is used to extract data features for each core data item, and set the weight values ​​of the corresponding data features according to the effective reference degree of the real data and the effective reference degree of the twin data in each core data item. The data features and weight values ​​of each core data item are input into the pre-built machine learning model to obtain the state monitoring results output by the machine learning model.

[0069] In this embodiment, the condition monitoring results can characterize the operating status of key parts of the industrial robot, and can provide effective data basis for predictive monitoring and maintenance of key parts.

[0070] The monitoring and maintenance module 150 is used to perform predictive monitoring and maintenance on key parts of the industrial robot based on the condition monitoring results.

[0071] The solution provided in this embodiment relies on core data from both real data and digital twin data generated through digital twin technology, making the data sources more diverse. By calculating the effective reference value of real data and digital twin data, more accurate data support can be provided for the generation of condition monitoring results. Combined with machine learning models, the condition monitoring results of key parts of industrial robots can be obtained efficiently and accurately, thereby enabling more precise and reliable predictive monitoring and maintenance of key parts of industrial robots.

[0072] In one embodiment, the data processing module determines the initial reference level for each piece of real data, specifically including:

[0073] The first step is to determine the theoretical quantity, actual quantity, and time series of real data in each core data item, as well as the time series of twin data.

[0074] The second step is to divide the actual quantity by the theoretical quantity to obtain the quantity ratio.

[0075] The third step is to calculate the dynamic time-normalized distance between the time series of the real data and the time series of the twin data in each core data item.

[0076] The fourth step is to add the dynamic time-normalized distance to 1 and take the reciprocal to obtain the distance correlation value.

[0077] The fifth step is to multiply the quantity ratio by the distance correlation value to calculate the initial reference value for each piece of real data.

[0078] In this embodiment, it is assumed that M core data items for key parts of the industrial robot are acquired, each of which has both real data and corresponding twin data. Since twin data is generated based on real data, the real data may be missing during acquisition due to sensor malfunctions, connection failures, or other reasons. However, twin data, because it integrates data completion technology, is relatively complete when generated from real data.

[0079] Understandably, the less complete a piece of real data is for an industrial robot, the lower its reliability for subsequent monitoring and maintenance of its critical components. Since twin data is generated from real data, the similarity between the same piece of real data and its corresponding twin data is high. Therefore, by combining the completeness of the real data itself with the similarity to the corresponding twin data, the initial reliability of each piece of real data can be determined.

[0080] Specifically, the initial reference value of the m-th real data item can be represented as follows:

[0081]

[0082] Where, θ m This represents the initial reference level of the m-th real data item. This represents the theoretical number of the m-th actual data item. L represents the actual number of data points in the m-th data category, DTW() represents the dynamic time warping distance calculation function, and L... m This represents a time series of the m-th real data item. This represents the time series of the m-th twin data item. The +1 is to avoid the case where the denominator is 0.

[0083] Understandably, the DTW() function mainly calculates the DTW (Dynamic Time Warping) distance between two time series. The core idea is to find an optimal warping path through dynamic programming to non-linearly align the two time series, allowing them to "bend" and "stretch" on the time axis, thereby more accurately measuring the similarity between the two time series, which is to calculate the distance between them.

[0084] In practical applications, the data of critical components of industrial robots, such as joints, fluctuates significantly under high-speed movement, heavy load conditions, and high-precision machining tasks. If data is missing or the degree of data loss is too great, the status of these critical components will be inaccurate during subsequent monitoring and maintenance of the industrial robot.

[0085] Therefore, to determine the reliability of each piece of real data, in addition to considering the completeness of the real data itself and its similarity to the corresponding twin data, it is also necessary to consider the location of missing data. The processing tasks corresponding to the locations of missing data vary, and the degree of impact on the state of critical parts of the industrial robot also differs. Therefore, the degree of core missing data for each piece of real data can be determined by calculating its location.

[0086] In one embodiment, the data processing module determines the degree of missing core information for each piece of real data, specifically including:

[0087] First, determine the volatility of each element in the time series of each piece of real data.

[0088] In a specific implementation, the volatility instability of each element within the time series of each piece of real data is determined, specifically including:

[0089] The first step is to determine the data value, slope value, and corresponding data value and slope value of each surrounding data in the time series of each real data item.

[0090] In this embodiment, all element data in the time series of each real data item can be categorized according to their volatility instability, grouping unstable element data into the same category. Specifically, the volatility instability of each element data item can be determined by obtaining the volatility difference between each element data item in the time series and its two or more surrounding data items. The number of surrounding data items for a single element data item can be taken as an empirical value of 10, but the implementer can set it according to specific circumstances without any specific limitation. Surrounding data refers to data adjacent to the element data in the time series.

[0091] Specifically, in this embodiment, the total amount of surrounding data of the p-th element in the time series of the m-th real data item is denoted as I, and then the slope value C of the p-th element in the time series of the m-th real data item is calculated. m,p And calculate the slope value C of the i-th surrounding data corresponding to the p-th element of the m-th real data in the time series. m,pi Let D be the data value of the p-th element in the time series of the m-th real data item. m,p Let D be the data value of the i-th surrounding data corresponding to the p-th element in the time series of the m-th real data. m,pi .

[0092] In practical applications, the slope value of each element and the slope value of each surrounding data can be obtained using any slope determination scheme. For example, one can first fit the time series of the m-th true data and determine the function expression of the fitted curve, and then obtain the slope value by differentiation based on the function expression and the data values ​​of the element or surrounding data. Alternatively, the slope value can be estimated using the central difference method, forward difference method, or backward difference method based on the known data values ​​of each element in the time series of the m-th true data. The specific slope determination scheme can be reasonably selected according to actual needs, and no specific limitation is made here.

[0093] The second step is to calculate the absolute value of the difference between the slope value of each element data and the slope value of each surrounding data, so as to obtain the slope difference value of each element data under each surrounding data.

[0094] The third step is to calculate the absolute value of the difference between the data value of each element and the data value of each surrounding data, so as to obtain the numerical difference value of each element under each surrounding data.

[0095] The fourth step is to multiply the slope difference value and the numerical difference value corresponding to each surrounding data point to obtain the difference product value corresponding to each surrounding data point.

[0096] The fifth step is to calculate the average of the product of the differences of all surrounding data for each element, thus obtaining the volatility instability of each element in the time series of each real data item.

[0097] When the slope and data values ​​of the p-th element in the time series of the m-th real data differ significantly from those of its surrounding data, it indicates that the fluctuation of the p-th element is abnormal. In this embodiment, the instability of the p-th element in the time series of the m-th real data can be expressed as follows:

[0098]

[0099] Where, α m,p The normalization function is denoted by norm(), and I represents the total amount of data surrounding the p-th element in the time series of the m-th real data. m,p C represents the slope of the p-th element in the time series of the m-th real data. m,pi D represents the slope value of the i-th surrounding data corresponding to the p-th element of the time series of the m-th real data. m,p D represents the data value of the p-th element in the time series of the m-th real data item. m,pi This represents the data value of the i-th surrounding data corresponding to the p-th element in the time series of the m-th real data item.

[0100] Then, based on the volatility instability, clustering is performed on all data elements in the time series of each real data item to generate multiple clusters.

[0101] In this embodiment, the k-means clustering method can be used to cluster all elements in the time series of the m-th real data item. Figure 2 This example illustrates the distribution of all elements in the time series of the m-th real data item. During clustering, the distance between different data elements is represented by the difference in volatility instability, and the number of clusters is determined using the silhouette coefficient method. Assume that N clusters are obtained after clustering. The volatility instability of the real data in each cluster is similar. Different data elements in each cluster correspond to different processing tasks. Figure 3 An example is shown Figure 2 The clustering results for all elements in the time series are shown.

[0102] Subsequently, the volatility instability and data missing amount for each cluster were determined.

[0103] In this embodiment, the average value of the fluctuation instability of all element data in each cluster is taken as the fluctuation instability of each cluster. The larger the value of the fluctuation instability of each cluster, the higher the fluctuation instability of the real data in that cluster. If the real data in that cluster is missing in the time series, it will have a greater impact on the accuracy of the state representation of the key parts of the industrial robot.

[0104] In practical applications, a sliding window-based method can be used to calculate all missing values ​​in the time series of each real data item. Then, the number of missing values ​​in the real data of each cluster within the corresponding range of the time series can be calculated, which is the amount of missing data.

[0105] Finally, based on the volatility and instability of each cluster corresponding to each piece of real data and the amount of missing data, the degree of missing core for each piece of real data is calculated.

[0106] In a specific implementation, based on the volatility and instability of each cluster corresponding to each piece of real data and the amount of missing data, the degree of missing core for each piece of real data is calculated, specifically including:

[0107] The first step is to sum the fluctuation instability of all clusters corresponding to each piece of real data to obtain the cluster fluctuation summation value.

[0108] The second step is to divide the fluctuation instability of each cluster by the sum of the cluster fluctuations to obtain the impact of missing values.

[0109] The third step is to multiply the number of missing data points for each cluster by the impact of the missing values ​​to obtain the cluster missing importance for each cluster.

[0110] The fourth step is to calculate the mean of the cluster missing importance of all clusters corresponding to each real data item and then normalize it to obtain the degree of missing cores for each real data item.

[0111] In this embodiment, the degree of missing core data for the m-th real data item can be represented as follows:

[0112]

[0113] Where, β m The degree of missing core information in the m-th data point is represented by norm(), where norm() is the normalization function, N represents the number of clusters obtained from the clustering, and E represents the number of clusters obtained from the clustering. n D represents the number of missing data in the nth cluster. n D represents the fluctuation instability corresponding to the nth cluster. j This represents the fluctuation instability corresponding to the j-th cluster. This represents the sum of cluster fluctuations for N clusters. This represents the cluster missing importance of the nth cluster.

[0114] In one embodiment, the data processing module corrects the initial reference value of each piece of real data based on the degree of missing core data, thereby obtaining the effective reference value of each piece of real data, specifically including:

[0115] The effective reference value of each piece of real data is obtained by multiplying the degree of missing core of each piece of real data by the initial reference value.

[0116] In this embodiment, the effective reference value of the m-th real data item can be expressed as follows:

[0117]

[0118] in, θ represents the effective reference value of the m-th real data item. m β represents the initial reference value of the m-th real data item. m This indicates the degree of missing core information in the m-th data item.

[0119] In practical applications, it is generally assumed that the effective reference value of each piece of real data and the effective reference value of the corresponding twin data are the same. However, since twin data is generated based on real data, the generation of twin data can be slow when there are network delays or excessive data volume. But the status monitoring of key parts of industrial robots is real-time. If the real data of key parts of industrial robots is abnormal, and the twin data has not been generated in time or is incomplete, the effective reference value of that piece of twin data will be low.

[0120] If the validity reference value of the twin data is not reset at this time, the validity reference value of this core data will be too low, which will lead to inaccurate status monitoring results of key parts of the industrial robot.

[0121] In practical applications, the acquisition time (timestamp information) of each element in each piece of real data can be obtained from the backend management system, along with the generation time of the corresponding twin data. There is a fixed waiting time between the generation of real data and the generation of twin data. After this waiting time, the status monitoring system for key parts of the industrial robot begins operation. If a delay occurs, the difference between the twin data generation time and the real data acquisition time will be greater, and the amount of twin data will also decrease. Therefore, by observing the difference between the twin data generation time and the real data acquisition time, as well as the difference in their data volume, the degree of abnormal delay in each piece of twin data can be determined.

[0122] In one embodiment, the data calculation module determines the abnormal latency of each twin data item, specifically including:

[0123] The first step is to obtain the acquisition time of the last element in the time series of each real data item, the generation time of the first element in the time series of each twin data item, the theoretical data volume of each real data item, and the data generation volume of each twin data item.

[0124] In this embodiment, the acquisition time of the last element in the time series of the m-th real data can be denoted as T. m1 The generation time of the first element in the time series of the m-th twin data can be denoted as: After the waiting time ends, the amount of data generated for the m-th twin data item can be denoted as: The theoretical data size of the m-th real data item can be denoted as:

[0125] The second step is to calculate the absolute value of the difference between the acquisition time of the last element in the time series of each real data item and the generation time of the first element in the time series of the corresponding twin data item, so as to obtain the time difference value of each twin data item.

[0126] The third step is to calculate the absolute value of the difference between the theoretical data volume of each real data item and the data generation volume of the corresponding twin data item, so as to obtain the data volume difference value of each twin data item.

[0127] The fourth step is to calculate the abnormal latency of each twin data item based on the time difference value and data volume difference value of each twin data item.

[0128] In one embodiment, based on the time difference value and data volume difference value of each twin data item, the abnormal latency of each twin data item is calculated, specifically including:

[0129] First, the time difference value of each twin data item is divided by a preset time difference threshold to obtain the time difference ratio.

[0130] Then, the difference in data volume for each twin data item is divided by a preset data volume difference threshold to obtain the data volume difference ratio.

[0131] Finally, the time difference ratio is multiplied by the data volume difference ratio to calculate the abnormal latency of each twin data item.

[0132] In this embodiment, the abnormal delay of the m-th twin data item can be expressed as follows:

[0133]

[0134] Where, γ m T represents the abnormal latency of the m-th twin data item. m1This represents the time when the last element of the m-th data item in the time series was retrieved. This represents the generation time of the first element in the time series of the m-th twin data item. This represents the amount of data generated for the m-th twin data item. Let T represent the theoretical data volume of the m-th actual data item, T represent the preset time difference threshold, and A represent the preset data volume difference threshold.

[0135] In one embodiment, the data calculation module calculates the effective reference degree of each twin data item based on the abnormal latency of each twin data item and the effective reference degree of the corresponding real data item, including:

[0136] First, based on the abnormal latency of each twin data point, the first intermediate parameter value for each twin data point is determined.

[0137] Then, the first intermediate parameter value is multiplied by a preset adjustment coefficient to obtain the second intermediate parameter value for each twin data item.

[0138] Finally, the effective reference value of each piece of real data is calculated by subtracting the second intermediate parameter value of the corresponding twin data.

[0139] It is understandable that when the abnormal latency of the twin data is large, the effective reference value of the twin data will also be relatively large. Specifically, when the abnormal latency of the twin data is greater than 0, it indicates that the twin data has experienced abnormal latency. In this case, the effective reference value of the twin data should be adjusted downward based on the effective reference value of the real data. In this embodiment, the effective reference value of the m-th twin data item can be expressed as follows:

[0140]

[0141] Where, ρ m This indicates the effective reference value of the m-th twin data item. γ represents the effective reference value of the m-th real data item. m This represents the abnormal delay of the m-th twin data item, and F represents the preset adjustment coefficient, which is a fixed value.

[0142] In practical applications, the effective reference degree of real data and twin data in each core data item can be calculated through the above analysis. Then, the data features of each core data item can be extracted. Specifically, this embodiment uses the MRMR (Minimum Redundancy Maximum Relevance) method to extract the data features of each core data item. The MRMR method is a feature selection algorithm whose core idea is to find a subset of features that are most correlated with the target variable while having the least redundancy. Here, correlation refers to the degree of association between a feature and the target variable. The higher the correlation, the more valuable the feature is for predicting the target variable. Redundancy refers to the degree of similarity between features. Features with high redundancy contain a lot of duplicate information. Retaining multiple such features has limited improvement on model performance and may even increase computational complexity.

[0143] After obtaining the data features of each core data item, the weight value of the data features of each core data item can be set according to the effective reference degree of the real data and twin data in each core data item. Subsequently, the data features and weight values ​​of each core data item are input into the pre-built machine learning model, and then the status monitoring results of key parts of the industrial robot are output.

[0144] In this embodiment, the machine learning model can adopt a PNN (Product-based Neural Network) architecture. The PNN architecture uses a multi-layer neural network, which can explicitly perform cross-operations on data features, effectively capturing second-order and higher-order interactions between data features. Compared to traditional neural networks, it can better handle complex relationships in sparse data. In practical applications, the ReLU function can be used as the activation function.

[0145] In some embodiments, the process of setting the weight value of the data features of each core data item based on the effective reference degree of the real data and the effective reference degree of the twin data in each core data item can be implemented through the following process:

[0146] First, calculate the average effective reference value of the real data and the effective reference value of the twin data in each core data item to obtain the comprehensive reference value of each core data item.

[0147] Then, the overall reference scores of all core data are summed to obtain the overall reference summation value.

[0148] Subsequently, the comprehensive reference value is obtained by dividing the comprehensive reference degree of each core data item by the sum of comprehensive reference values.

[0149] Finally, the comprehensive reference ratio corresponding to each core data item is used as the weight value of the data feature of each core data item.

[0150] In one embodiment, the monitoring and maintenance module specifically includes:

[0151] The data comparison unit is used to compare the predicted values ​​of each state in the state monitoring results with their corresponding state reference values ​​to obtain the comparison results.

[0152] The monitoring and maintenance unit is used to generate a monitoring and maintenance plan for predictive monitoring and maintenance of key parts of the industrial robot when at least one of the predicted state values ​​in the comparison results deviates from the corresponding state reference value.

[0153] In this embodiment, taking the joint as a key component as an example, the status monitoring results can specifically include various predicted status values ​​such as joint parameters, motor parameters, and mechanical structure parameters. Specifically, joint parameters include joint angles, joint velocities, joint accelerations, and joint torques—parameters that characterize the joint's motion state. Motor parameters include operating current, operating voltage, operating temperature, and operating speed—parameters that characterize the motor's operating condition. Mechanical structure parameters include mechanical vibration, mechanical noise, mechanical component clearances, and wear levels—parameters that characterize the health status of the mechanical structure.

[0154] In practical applications, a corresponding state reference value can be set for each state prediction value in the state monitoring results. Then, each state prediction value is compared with its corresponding state reference value. If any one or more state prediction values ​​deviate from the state reference value, it indicates that there is a health hazard in the key parts of the industrial robot, which requires timely maintenance.

[0155] In some embodiments, an allowable deviation between each state prediction value and the state reference value can be set. When the difference between any state prediction value and the state reference value exceeds the preset allowable deviation, it can be determined that the state prediction value deviates from the state reference value.

[0156] In this embodiment, a monitoring and maintenance plan can be pre-established for each state prediction value that deviates from the state reference value, and a monitoring and maintenance decision table can be generated. Subsequently, the corresponding monitoring and maintenance plan can be extracted from the monitoring and maintenance decision table based on the comparison results.

[0157] In summary, this embodiment determines the initial reference value of each piece of real data by analyzing the missing data of key parts of the industrial robot and the similarity with the corresponding twin data. Then, it corrects the initial reference value of each piece of real data by analyzing the missing data locations. Subsequently, it determines the effective reference value of the twin data by analyzing the delay of the twin data corresponding to each piece of real data. This provides accurate and reliable data basis for subsequent use of machine learning models to monitor the status of key parts of the industrial robot.

[0158] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0159] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A digital-twin-based predictive monitoring maintenance system for industrial robots, characterized in that, The system comprises: a data acquisition module configured to acquire a plurality of core data of key parts of an industrial robot, wherein each of the core data comprises real data and twin data; a data processing module configured to determine an initial reference degree and a missing core degree of each of the real data respectively, and correct the initial reference degree of each of the real data based on the missing core degree to obtain an effective reference degree of each of the real data; a data calculation module configured to determine an abnormal delay degree of each of the twin data, and calculate an effective reference degree of each of the twin data based on the abnormal delay degree of each of the twin data and the effective reference degree of the corresponding real data; a machine learning module configured to extract data features of each of the core data, set weight values of the data features according to the effective reference degrees of the real data and the effective reference degrees of the twin data, and input the data features and the weight values of each of the core data into a pre-constructed machine learning model to obtain a state monitoring result output by the machine learning model; a monitoring and maintenance module configured to perform predictive monitoring and maintenance on the key parts of the industrial robot according to the state monitoring result; the initial reference degree of the mth real data is represented as follows: wherein, denotes an initial reference degree of the mth real data, denotes a theoretical quantity of the mth real data, denotes an actual quantity of the mth real data, denotes a dynamic time warping distance calculation function, denotes a time series of the mth real data, denotes a time series of the mth twin data; the missing core degree of the mth real data is represented as follows: wherein, denotes the missing core degree of the mth real data, denotes a normalization function, N denotes the number of class clusters obtained by clustering, denotes the data missing amount of the nth class cluster, denotes the fluctuation instability corresponding to the nth class cluster, denotes the fluctuation instability corresponding to the jth class cluster, denotes the cluster fluctuation summation value of the N class clusters, denotes the cluster missing importance of the nth class cluster; the effective reference degree of the mth twin data is represented as follows: wherein, represents the effective reference degree of the mth item of twin data, represents the effective reference degree of the mth item of true data, represents the abnormal delay degree of the mth item of twin data, represents a preset adjustment coefficient.

2. The digital-twin-based predictive monitoring and maintenance system for industrial robots according to claim 1, characterized in that, the fluctuation instability degree of each element data in the time series sequence of each of the real data is determined, including: the data value, the slope value of each element data, and the data value and the slope value of each surrounding data corresponding to each element data in the time series sequence of each of the real data are determined; the absolute value of the difference between the slope value of each element data and the slope value of each surrounding data is calculated to obtain the slope difference value of each surrounding data under each element data; the absolute value of the difference between the data value of each element data and the data value of each surrounding data is calculated to obtain the numerical difference value of each surrounding data under each element data; the difference product value corresponding to each surrounding data is obtained by multiplying the slope difference value and the numerical difference value corresponding to each surrounding data; the fluctuation instability degree of each element data in the time series sequence of each of the real data is obtained by averaging the difference product values corresponding to all surrounding data under each element data.

3. The digital-twin-based predictive monitoring and maintenance system for industrial robots according to claim 1, characterized in that, The data processing module corrects the initial reference degree of each of the real data based on the missing core degree to obtain the effective reference degree of each of the real data, including: the effective reference degree of each of the real data is obtained by multiplying the missing core degree and the initial reference value.

4. The digital-twin-based predictive monitoring and maintenance system for industrial robots according to claim 1, characterized in that, The data calculation module determines the abnormal delay degree of each of the twin data, including: the acquisition time of the last element data in the time series sequence of each of the real data, the generation time of the first element data in the time series sequence of each of the twin data, the theoretical data amount of each of the real data, and the data generation amount of each of the twin data are acquired; the absolute value of the difference between the acquisition time of the last element data in the time series sequence of each of the real data and the generation time of the first element data in the time series sequence of the corresponding twin data is calculated to obtain the time difference value of each of the twin data; Calculate the absolute value of the difference between the theoretical data volume of each item of real data and the data generation volume of the corresponding item of twin data, to obtain the data volume difference value of each item of twin data; Based on the time difference value and the data volume difference value of each item of twin data, the abnormal delay degree of each item of twin data is calculated.

5. The digital-twin-based predictive monitoring and maintenance system of industrial robots according to claim 4, characterized in that, Based on the time difference value and the data volume difference value of each item of twin data, the abnormal delay degree of each item of twin data is calculated, including: Divide the time difference value of each item of twin data by the preset time difference threshold to obtain the time difference ratio value; Divide the data volume difference value of each item of twin data by the preset data volume difference threshold to obtain the data volume difference ratio value; Multiply the time difference ratio value and the data volume difference ratio value to calculate the abnormal delay degree of each item of twin data.

6. The digital-twin-based predictive monitoring and maintenance system of industrial robots according to claim 1, characterized in that, The monitoring and maintenance module comprises: A data comparison unit is configured to compare each state prediction value in the state monitoring result with the corresponding state reference value to obtain a comparison result; A monitoring and maintenance unit is configured to generate a monitoring and maintenance scheme for predictive monitoring and maintenance of the key parts of the industrial robot when at least one state prediction value in the comparison result deviates from the corresponding state reference value.

Images