Control method of industrial robot and related device
By acquiring the current pending tasks and target status information of the industrial robot, target operating parameter information is generated, which solves the problem of inaccurate configuration of industrial robot operating parameters and improves control accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DEXFORCE TECH CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-16
Smart Images
Figure CN122008260B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of industrial control technology, specifically to a control method and related devices for an industrial robot. Background Technology
[0002] As a core piece of equipment in the field of intelligent manufacturing, industrial robots typically require manual configuration of their operating parameters during operation. For example, the operating parameters of the industrial robot, such as the recognition accuracy during operation, are configured manually through the configuration interface on the display. Due to manual processing, it is impossible to accurately determine the relevant transaction attributes of the industrial robot when it is performing tasks, resulting in low accuracy in configuring operating parameters, which in turn leads to low accuracy in controlling the industrial robot. Summary of the Invention
[0003] This application provides a control method and related apparatus for an industrial robot. It can generate corresponding target operating parameter information by combining the current pending task and target state information of the industrial robot, and finally use the target operating parameter information for control, thereby improving the accuracy of target operating parameter determination and thus improving the accuracy of industrial robot control.
[0004] A first aspect of this application provides a control method for an industrial robot, the method comprising:
[0005] Obtain the current pending tasks of the industrial robot, and obtain the target state information of the industrial robot;
[0006] Determine the application scenario information based on the current pending transactions;
[0007] Determine the operational accuracy information and operational configuration information based on the application scenario information;
[0008] The reference operating parameter information is determined based on the operating accuracy information and the operating configuration information;
[0009] Generate parameter correction information based on the target state information;
[0010] The reference operating parameter information is corrected using the parameter correction information to obtain the target operating parameter information;
[0011] The industrial robot is controlled based on the target operating parameter information.
[0012] In one possible implementation, obtaining the target state information of the industrial robot includes:
[0013] Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset;
[0014] Perform data balancing on the anomalous data in the anomalous dataset to obtain the target dataset;
[0015] Feature extraction is performed on the target dataset to obtain the target feature vector;
[0016] The state of the industrial robot is predicted based on the target feature vector and the real-time operating data of the industrial robot to obtain the target state information.
[0017] In one possible implementation, the step of performing data balancing on the anomalous data in the anomalous dataset to obtain the target dataset includes:
[0018] The abnormal dataset is denoised to obtain the first dataset.
[0019] Cluster the first data in the first dataset to obtain M subsets.
[0020] Determine the classification size for each of the M subsets to obtain a set of classification size information;
[0021] The target dataset is obtained by performing data balancing on the M subsets based on the classification size information in the classification size information set.
[0022] In one possible implementation, a method for performing data balancing on M subsets of datasets based on classification size information in a classification size information set to obtain a target dataset includes:
[0023] Determine the target test set corresponding to each of the M subsets of data, thus obtaining the M target test sets;
[0024] Calculate the misclassification rate for each of the M target test sets to obtain the M target misclassification rate sets;
[0025] Calculate the average misclassification error rate for each subset of data based on the target misclassification rates in the M target misclassification rate sets, and obtain the M average misclassification error rates;
[0026] Based on the average misclassification error rate of M data sets and the classification size information in the classification size information set, determine the number of running data sets that need to be synthesized for each of the M data sets, and obtain the first set of data sets.
[0027] Based on the first quantity in the first quantity set, data is synthesized from each of the M subsets to obtain the target dataset.
[0028] In one possible implementation, a method for determining the target test set corresponding to each of M subsets of datasets to obtain M target test sets includes:
[0029] In each of the M subsets, a predetermined number of running data points are selected as the training set for the corresponding subset, resulting in M reference training sets.
[0030] Determine the test set corresponding to each of the M reference training sets to obtain the M reference test sets;
[0031] Iterative linear classification training is performed on each of the M reference test sets to obtain M target test sets.
[0032] In one possible implementation, a method for extracting features from a target dataset to obtain a target feature vector includes:
[0033] Extract the temporal feature information of the target dataset to obtain the first feature vector;
[0034] Based on the first feature vector, the associated features of the target dataset are extracted to obtain the target feature vector.
[0035] A second aspect of this application provides a control device for an industrial robot, the device comprising:
[0036] The acquisition unit is used to acquire the current pending tasks of the industrial robot and to acquire the target state information of the industrial robot.
[0037] The determining unit is configured to: determine application scenario information based on the current pending transaction; determine operating accuracy information and operating configuration information based on the application scenario information; determine reference operating parameter information based on the operating accuracy information and operating configuration information; generate parameter correction information based on the target status information; and perform correction processing on the reference operating parameter information using the parameter correction information to obtain the target operating parameter information.
[0038] The control unit is used to control the industrial robot according to the target operating parameter information.
[0039] In one possible implementation, the acquisition unit is specifically used for: obtaining the target state information of the industrial robot.
[0040] Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset;
[0041] Perform data balancing on the anomalous data in the anomalous dataset to obtain the target dataset;
[0042] Feature extraction is performed on the target dataset to obtain the target feature vector;
[0043] The state of the industrial robot is predicted based on the target feature vector and the real-time operating data of the industrial robot to obtain the target state information.
[0044] In one possible implementation, in terms of performing data balancing on the anomalous data in the anomalous dataset to obtain the target dataset, the acquisition unit is specifically used for:
[0045] The abnormal dataset is denoised to obtain the first dataset.
[0046] Cluster the first data in the first dataset to obtain M subsets.
[0047] Determine the classification size for each of the M subsets to obtain a set of classification size information;
[0048] The target dataset is obtained by performing data balancing on the M subsets based on the classification size information in the classification size information set.
[0049] In one possible implementation, regarding the data balancing of M subsets based on classification size information in the classification size information set to obtain the target dataset, the acquisition unit is specifically used for:
[0050] Determine the target test set corresponding to each of the M subsets of data, thus obtaining the M target test sets;
[0051] Calculate the misclassification rate for each of the M target test sets to obtain the M target misclassification rate sets;
[0052] Calculate the average misclassification error rate for each subset of data based on the target misclassification rates in the M target misclassification rate sets, and obtain the M average misclassification error rates;
[0053] Based on the average misclassification error rate of M data sets and the classification size information in the classification size information set, determine the number of running data sets that need to be synthesized for each of the M data sets, and obtain the first set of data sets.
[0054] Based on the first quantity in the first quantity set, data is synthesized from each of the M subsets to obtain the target dataset.
[0055] In one possible implementation, after determining the target test set corresponding to each of the M subsets of datasets, M target test sets are obtained. The acquisition unit is specifically used for:
[0056] In each of the M subsets, a predetermined number of running data points are selected as the training set for the corresponding subset, resulting in M reference training sets.
[0057] Determine the test set corresponding to each of the M reference training sets to obtain the M reference test sets;
[0058] Iterative linear classification training is performed on each of the M reference test sets to obtain M target test sets.
[0059] In one possible implementation, regarding the feature extraction of the target dataset to obtain the target feature vector, the acquisition unit is specifically used for:
[0060] Extract the temporal feature information of the target dataset to obtain the first feature vector;
[0061] Based on the first feature vector, the associated features of the target dataset are extracted to obtain the target feature vector.
[0062] A third aspect of this application provides an industrial robot, including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the first aspect of this application.
[0063] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of this application.
[0064] A fifth aspect of this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of this application. The computer program product may be a software installation package.
[0065] Implementing the embodiments of this application has at least the following beneficial effects:
[0066] By acquiring the current pending tasks of the industrial robot and its target state information, application scenario information is determined based on the current pending tasks. Operating accuracy information and operating configuration information are determined based on the application scenario information. Reference operating parameter information is determined based on the operating accuracy information and the operating configuration information. Parameter correction information is generated based on the target state information. The reference operating parameter information is then corrected using the parameter correction information to obtain the target operating parameter information. The industrial robot is then controlled based on the target operating parameter information. Therefore, the target operating parameter information can be generated by combining the current pending tasks and target state information of the industrial robot. Finally, the target operating parameter information is used for control, improving the accuracy of target operating parameter determination and thus improving the accuracy of industrial robot control. Attached Figure Description
[0067] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0068] Figure 1 This application provides a flowchart illustrating a control method for an industrial robot.
[0069] Figure 2 This is a schematic diagram of the structure of an industrial robot provided in an embodiment of this application;
[0070] Figure 3 This application provides a schematic diagram of the structure of a control device for an industrial robot. Detailed Implementation
[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0072] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.
[0073] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.
[0074] To better understand the industrial robot control method provided in this application, a brief introduction to existing industrial robot control methods is given below. Typically, manual configuration of the corresponding operating parameters is required, for example, through a configuration interface on a display. These parameters, such as recognition accuracy during operation, are manually configured. However, this manual processing can lead to inaccurate determination of the attributes of relevant transactions during robot processing, resulting in low accuracy in configuring operating parameters and consequently, low accuracy in controlling the industrial robot.
[0075] To address the aforementioned technical problems, this application provides a control method for an industrial robot. This method combines the current pending tasks and target state information of the industrial robot to generate corresponding target operating parameter information. Finally, the target operating parameter information is used for control, thereby improving the accuracy of target operating parameter determination and thus improving the accuracy of industrial robot control.
[0076] Please see Figure 1 , Figure 1 This application provides a flowchart illustrating a control method for an industrial robot. For example... Figure 1 As shown, the method includes:
[0077] 101. Obtain the current pending tasks of the industrial robot and obtain the target state information of the industrial robot.
[0078] The currently pending task includes its corresponding application scenario information. For example, the application scenario information may include depalletizing, disordered grasping, ordered loading / unloading, or single-target precision positioning. Specifically, it may involve grasping large objects, and the processing time for this grasping task is usually relatively long. The operational configuration and accuracy required for industrial robots to perform tasks will differ depending on the application scenario. The currently pending task can be obtained through manual user input or by receiving instructions.
[0079] When acquiring target state information, state prediction can be performed by combining historical and current operating data of the industrial robot. Current operating data can be understood as the relevant operational information of the industrial robot after it is powered on. For example, it can be data extracted from internal or external mechanical sensors, temperature sensors, vibration sensors, and acceleration sensors. The data types in historical and current operating data correspond to each other.
[0080] 102. Based on the current pending transaction and the target status information, determine the target operating parameter information of the industrial robot.
[0081] Specifically, a method for determining the target operating parameters of an industrial robot based on the current pending transaction and the target state information includes:
[0082] A1. Determine the application scenario information based on the current pending transaction;
[0083] A2. Determine the operating accuracy information and operating configuration information based on the application scenario information;
[0084] A3. Determine reference operating parameter information based on the operating accuracy information and the operating configuration information;
[0085] A4. Generate parameter correction information based on the target state information;
[0086] A5. The reference operating parameter information is corrected using the parameter correction information to obtain the target operating parameter information.
[0087] Different pending transactions have corresponding application scenario information. This can be achieved by extracting keywords from the transactions, looking up the keywords in a table, and then obtaining the application scenario information. A mapping table between keywords and application scenario information is maintained, allowing for lookups to retrieve the application scenario information for the current pending transaction. This mapping table is obtained through system configuration.
[0088] Specifically, different application scenarios have their corresponding operational accuracy and configuration information. For example, in the application scenario of unordered grasping, the operational accuracy will be lower than that in a monocular precise positioning scenario. To complete the tasks corresponding to the application scenario, parameter configuration is required to obtain operational configuration information. This operational configuration information can also be obtained by looking up a pre-built mapping table. The operational configuration information can include relevant configurations of operational parameters, such as the grasping parameters, motion parameters, and rotation angle during grasping of the industrial robot.
[0089] The runtime parameters generated from the runtime configuration information can be fine-tuned using runtime accuracy information to obtain reference runtime parameter information. Specifically, the corresponding data can be extracted from the runtime configuration information to obtain runtime parameter information. Then, combined with the runtime accuracy, parameters that do not conform to the runtime accuracy can be adjusted to the parameter range corresponding to the runtime accuracy, thus obtaining the runtime data in the reference runtime parameter information. When adjusting parameters that do not conform to the runtime accuracy to the parameter range corresponding to the runtime accuracy, it can be done by adjusting the parameters that do not conform to the runtime accuracy to the median value of the parameter range corresponding to the runtime accuracy.
[0090] Because operating parameters need to adapt to different operating states to improve the accuracy of operation control, correction coefficients can be generated based on target state information. These correction coefficients are correlated with the target state information. The better the target state information indicates the industrial robot's state, the better its operating conditions, and therefore, the smaller the correction force of the correction coefficient (i.e., the smaller the adjustment force to the parameters in the reference operating parameter information). Conversely, the worse the target state information indicates the industrial robot's state, the worse its operating conditions, and therefore, the larger the correction force of the correction coefficient (i.e., the larger the adjustment force to the parameters in the reference operating parameter information). The adjustment force can be understood as the degree of matching between the parameter and the operating state when adjusting the parameter; the smaller the adjustment force, the higher the matching degree, and vice versa. Therefore, the matching degree between the target operating parameter information and the target state information after correction using the correction parameters will be higher than a preset matching degree threshold, thereby improving the accuracy of subsequent control. The preset matching degree threshold is set through empirical values or historical data.
[0091] Specifically, taking the scenario of unordered grasping of automotive parts as an example, the current task to be processed is 'grabbing Class A parts'. The system determines the application scenario as 'unordered grasping' by looking up the keywords 'grab' and 'Class A parts' in the table. At this time, the grasping speed in the reference operating parameters is set to high-speed grasping and the accuracy is set to medium. During actual grasping, the image information of grasping Class A parts can be extracted, and the position to be grasped can be obtained by combining the image information. The grasping position and the reference operating parameters are combined to process the grasping of Class A parts. If the target status information shows that the robot's end effector has slight vibration (poor status), a correction coefficient with a large correction force is generated to reduce the grasping speed and improve the grasping accuracy, sacrificing some efficiency for a higher grasping success rate.
[0092] 103. Control the industrial robot according to the target operating parameter information.
[0093] The target operating parameter information is used to control the industrial robot to execute the currently pending task.
[0094] In this example, by acquiring the current pending tasks of the industrial robot and the target state information of the industrial robot, the target operating parameter information of the industrial robot is determined based on the current pending tasks and the target state information. The industrial robot is then controlled based on the target operating parameter information. Therefore, the corresponding operating parameter information can be generated by combining the current pending tasks and the target state information of the industrial robot. Finally, the operating parameter information is used for control, which improves the accuracy of the operating parameter determination and thus improves the accuracy of the control of the industrial robot.
[0095] In one possible implementation, a method for obtaining target state information of the industrial robot includes:
[0096] B1. Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset.
[0097] Specifically, this can be achieved by extracting historical operating data of the industrial robot from a pre-set historical database to obtain a historical dataset; then, by extracting unbalanced operating data from the historical operating dataset and identifying the unbalanced operating data as anomalous data to obtain an anomalous dataset. The pre-set historical database includes operating data of the industrial robot at various times, extracted from internal or external mechanical sensors, temperature sensors, vibration sensors, and acceleration sensors.
[0098] Imbalanced operational data can be understood as follows: The historical database contains operational data of the robot from time tn to time t. If, at a certain time between tn and t, the number of normal data points collected (meeting a preset data size, set through experience and historical data) does not meet this preset data size, then the data points that do not meet the preset data size (specifically, lower than the preset data size) are identified as imbalanced operational data. For example, if the data size is 10 data points of different types, but only 8 data points of different types are collected, then this data is considered imbalanced.
[0099] B2. Perform data balancing on the abnormal data in the abnormal dataset to obtain the target dataset.
[0100] Specifically, the abnormal data in the abnormal dataset can be divided into a majority class abnormal dataset and a minority class abnormal dataset according to a preset data quantity threshold. Noisy data in the abnormal dataset can be removed according to the sample centers corresponding to the majority class abnormal dataset and the minority class abnormal dataset to obtain the first dataset. The first dataset can be clustered to obtain M sub-datasets. The classification size corresponding to each sub-dataset can be determined. Data balancing can be performed on each sub-dataset according to the classification size information corresponding to each sub-dataset to obtain the target dataset.
[0101] B3. Extract features from the target dataset to obtain the target feature vector.
[0102] Specifically, this can be achieved by extracting the temporal features between each running data point in the target dataset, and then extracting the correlation features between each running data point based on the temporal features, thus obtaining the target feature vector.
[0103] B4. Based on the target feature vector and the real-time operating data of the industrial robot, predict the state of the industrial robot to obtain the target state information.
[0104] Specifically, this can be achieved by extracting the state vector space corresponding to the industrial robot to obtain a reference state vector space; inputting the target feature vector into an MLP (Multi-Layer Perceptron), which updates the reference state vector based on the input target feature vector to obtain the target state vector space; using internal or external mechanical sensors, temperature sensors, vibration sensors, and acceleration sensors to collect real-time data from the industrial robot to obtain real-time operating data; and taking the state in the target state vector space that best matches (highest matching degree) the real-time operating data as the operating state of the industrial robot to obtain the target state information.
[0105] Among them, the target state vector in the target state vector space can be the operating condition of the industrial robot, and different target state vectors have their corresponding operating conditions.
[0106] In one possible implementation, the step of performing data balancing on the anomalous data in the anomalous dataset to obtain the target dataset includes:
[0107] C1. Denoise the abnormal dataset to obtain the first dataset;
[0108] C2. Cluster the first data in the first dataset to obtain M subsets.
[0109] C3. Determine the classification size for each of the M subsets to obtain a set of classification size information.
[0110] C4. Perform data balancing on the M subsets based on the classification size information in the classification size information set to obtain the target dataset.
[0111] Specifically, this can be achieved by using a general regression / classification algorithm (such as KNN (K-Nearest Neighbors)) to identify noisy data in anomaly datasets, thus obtaining suspected noisy datasets; extracting the number of data points corresponding to each type of anomaly in the anomaly dataset, thus obtaining a data quantity set; and classifying the anomaly data in the anomaly dataset into minority class anomaly datasets and majority class anomaly datasets based on the data quantity set and a preset data quantity threshold. The types of anomaly datasets include runtime, location, vibration, temperature, speed, and torque; the preset data quantity threshold can be determined through empirical or historical values.
[0112] Calculate the sample centers corresponding to the minority class anomaly dataset to obtain the first sample centers; calculate the sample centers corresponding to the majority class anomaly dataset to obtain the second sample centers; calculate the Euclidean distance between each suspected noise data in the suspected noise dataset and the first sample centers to obtain the first Euclidean distance set; calculate the Euclidean distance between each suspected noise data in the suspected noise dataset and the second sample centers to obtain the second Euclidean distance set.
[0113] Calculate the Euclidean distance between each element in the suspected noise dataset and its corresponding element in the minority class anomaly dataset, obtaining N sets of third Euclidean distances. Calculate the Euclidean distance between each element in the suspected noise dataset and its corresponding element in the majority class anomaly dataset, obtaining N sets of fourth Euclidean distances. Based on the elements in the N sets of third Euclidean distances, determine a preset number of nearest neighbors in the minority class anomaly dataset corresponding to each element in the suspected noise dataset, obtaining the minority class nearest neighbor anomaly dataset. Based on the elements in the N sets of fourth Euclidean distances, determine a preset number of nearest neighbors in the majority class anomaly dataset corresponding to each element in the suspected noise dataset, obtaining the majority class nearest neighbor anomaly dataset. Here, N represents the number of elements in the suspected noise dataset; the preset number of nearest neighbors can be determined through empirical values or historical data.
[0114] The density of the nearest neighbor domain corresponding to each element in the suspected noise dataset is calculated based on the elements in the minority class nearest neighbor anomaly dataset, resulting in a first nearest neighbor domain density information set; the density of the nearest neighbor domain corresponding to each element in the suspected noise dataset is calculated based on the elements in the majority class nearest neighbor dataset, resulting in a second nearest neighbor domain density information set.
[0115] Based on the elements in the first Euclidean distance set, the second Euclidean distance set, the first nearest neighbor density information set, and the second nearest neighbor density information set, calculate the anomaly score corresponding to each element in the suspected noise dataset to obtain an anomaly score information set. Identify the anomaly score information in the anomaly score information set that is greater than or equal to a preset anomaly score threshold as target anomaly score information to obtain a target anomaly score information set. Identify the element in the suspected noise dataset corresponding to each element in the target anomaly score information set as noise data to obtain a noise dataset. Remove elements from the noise dataset from the anomaly dataset to obtain the first dataset. The preset anomaly score threshold can be determined through empirical values or historical values.
[0116] Specifically, the anomaly score for each element in the suspected noise dataset can be calculated using the following formula: elements from the first Euclidean distance set, elements from the second Euclidean distance set, elements from the first nearest neighbor density information set, and elements from the second nearest neighbor density information set. This yields the anomaly score information set.
[0117]
[0118]
[0119]
[0120] In the formula This represents the abnormal rating information in the abnormal rating information set; This represents the adaptive weight corresponding to the Euclidean distance metric. The adaptive weights corresponding to the index representing the density information of the nearest neighbor region; This represents the i-th first Euclidean distance in the first set of Euclidean distances; This represents the i-th second Euclidean distance in the set of second Euclidean distances; This represents the i-th density information of the first nearest neighbor in the first nearest neighbor density information set; This represents the i-th density information of the second nearest neighbor in the set of second nearest neighbor density information; This indicates that x is normalized, scaling its range to a specified value. ,when hour, This indicates that the first Euclidean distance in the first set of Euclidean distances is normalized. hour, This indicates that the second Euclidean distances in the second Euclidean distance set are normalized. hour, This indicates that the density information of the first nearest neighbor in the first nearest neighbor density information set is normalized. hour, This indicates that the density information of the second nearest neighbor in the second nearest neighbor density information set is normalized. The importance coefficient representing the density information of the nearest neighbor region can be determined by historical or empirical values, and its value range is [missing value]. ,make sure The value is always greater than The larger the importance coefficient, the higher the importance of the nearest neighbor density information; conversely, the smaller the importance coefficient, the lower the importance of the nearest neighbor density information. max() takes the maximum value, and min() takes the minimum value.
[0121] After obtaining the first dataset, the AHC (Agglomerative Hierarchical Clustering) algorithm can be used to cluster the elements in the first dataset to obtain M sub-datasets.
[0122] After obtaining M subsets of data, we can extract the number of data points in each subset of the M subsets to obtain a data quantity information set; then, we can determine the elements in the data quantity information set as the classification size of the corresponding subset of the M subsets to obtain a classification size information set.
[0123] After obtaining the classification size information set, the target test set corresponding to each of the M subsets can be determined. Based on the misclassification rate corresponding to the target test set and the elements in the classification size information set, the number of running data to be synthesized for each of the M subsets can be determined. Then, data synthesis can be performed on each of the M subsets according to the required number of running data to be synthesized, thus obtaining the target dataset.
[0124] In this example, the Euclidean distance between the suspected noise data and the sample center in the suspected noise dataset is calculated, as well as the nearest neighbor density corresponding to the suspected noise data in the suspected noise dataset. At the same time, the anomaly score of the suspected noise data is calculated based on the Euclidean distance and the nearest neighbor density. Suspected noise data with an anomaly score higher than the preset anomaly score threshold is identified as noise data. This accurately identifies and removes noise data from the abnormal dataset, thereby improving the accuracy of predicting the state of industrial robots.
[0125] In one possible implementation, a method for performing data balancing on K subsets of a dataset based on classification size information in a classification size information set to obtain a target dataset includes:
[0126] D1. Determine the target test set corresponding to each of the M subsets of data to obtain the M target test sets;
[0127] D2. Extract the target misclassification rate corresponding to each of the M target test sets to obtain the M target misclassification rates;
[0128] D3. Calculate the M average misclassification error rates for each subset of data based on the misclassification rates among the M target misclassification rates;
[0129] D4. Determine the number of running data to be synthesized for each of the M subsets based on the elements in the M average misclassification error rates and the elements in the classification size information set, and obtain the first quantity set;
[0130] D5. Based on the first quantity in the first quantity set, perform data synthesis on each of the M subsets to obtain the target dataset.
[0131] Specifically, this can be achieved by randomly selecting a predetermined number of data points from each of the M subsets as the training set, and using the remaining data points as the test set. The misclassification rate for each test set is calculated, and iterative linear classification training is performed on the test sets based on the misclassification rate to obtain M target test sets. The predetermined number of data points for random selection is set using empirical values or historical data.
[0132] After obtaining the M target test sets, the M target misclassification rates can be obtained by extracting the minority class misclassification rate and the majority class misclassification rate corresponding to each target test set in the M target test sets during iterative classification training.
[0133] After obtaining M target misclassification rates, the average misclassification error rate can be obtained by calculating the ratio between the sum of the majority class misclassification rate and the minority class misclassification rate corresponding to each of the M target misclassification rate sets and the misclassification rate category (e.g., the target misclassification rate includes the minority class misclassification rate and the majority class misclassification rate, and the misclassification rate category is 2).
[0134] After obtaining M average misclassification error rates, the following steps can be taken: Extract minority class running data from each of the M subsets to obtain M minority class running datasets; extract the number of elements from each of the M minority class running datasets to obtain M minority class classification dimensions; extract majority class running data from each of the M subsets to obtain M majority class running datasets; extract the number of elements from each of the M majority class running datasets to obtain M majority class classification dimensions; calculate the difference between each element in the M majority class classification dimensions and the corresponding element in the M minority class classification dimensions, and use this difference as the number of running data points to be synthesized from each of the M subsets. The process involves: obtaining a reference quantity set; extracting the minority class misclassification rate from the M target misclassification rates to obtain M reference misclassification rates; using a common mutual influence calculation method (such as the Pearson correlation coefficient method) to extract the mutual influence between the M average misclassification error rates and the classification size information in the classification size information set to obtain a first mutual influence set; calculating the correction coefficients corresponding to the elements in the reference quantity set based on the average misclassification error rates among the M average misclassification error rates, the elements in the first mutual influence set, and the reference misclassification rates among the M reference misclassification rates to obtain a correction coefficient set; and correcting the corresponding reference quantities in the reference quantity set based on the elements in the correction coefficient set to obtain a first quantity set.
[0135] Specifically, the number of running data points needed for each of the M subsets can be determined using the following formula, based on elements from the M average misclassification error rates and the set of classification size information, thus obtaining the first set of data points:
[0136]
[0137]
[0138] In the formula This represents the i-th first quantity in the first quantity set; This represents the i-th correction coefficient in the set of correction coefficients; This represents the i-th majority class classification size among M majority class classification sizes; This represents the i-th minority class classification size among M minority class classification sizes; This represents the i-th reference misclassification rate among M reference misclassification rates; The first mutual influence degree in the first mutual influence degree set can be understood as the mutual influence degree between the i-th average misclassification error rate in the M average misclassification error rates and the i-th classification size information in the classification size information set. It is used to amplify the minority class misclassification rate in the target misclassification rate, thereby balancing the impact of the imbalance of classification size information on the linear classifier. Let i represent the i-th average misclassification error rate among M average misclassification error rates; Indicates a small compensation term. =0.0001, used to avoid the denominator being 0 due to an excessively small average misclassification error rate.
[0139] Since balancing the datasets in the M subsets is achieved by supplementing the minority class datasets in each subset so that the classification size of the supplemented minority class datasets is equal to that of the majority class datasets, the balancing of the datasets in the M subsets can be accomplished by synthesizing the minority class datasets from the M subsets.
[0140] After obtaining the first set of data, a general data synthesis method can be used to synthesize the first set of data for the minority class dataset in each of the M subsets, thus obtaining the target dataset.
[0141] In this example, the number of data points to be synthesized for each subset is accurately calculated using the classification size and average misclassification error rate corresponding to the M subsets, resulting in a first set of data points. Data is then synthesized for the minority class subsets in each subset based on the first number in the first set of data points, thereby improving the accuracy in balancing outlier datasets and consequently improving the accuracy in predicting the state of industrial robots.
[0142] In one possible implementation, a method for determining the target test set corresponding to each of M subsets of datasets to obtain M target test sets includes:
[0143] E1. Select a preset number of running data sets from each of the M subsets as the training set for the corresponding subset, and obtain M reference training sets.
[0144] E2. Determine the test set corresponding to each of the M reference training sets to obtain the M reference test sets;
[0145] E3. Perform iterative linear classification training on each of the M reference test sets to obtain M target test sets.
[0146] Specifically, this can be achieved by randomly selecting a predetermined number of running data points from each of the M subsets of data to serve as the training set for each of the M subsets, thus obtaining M reference training sets.
[0147] After obtaining M reference training sets, M reference test sets can be obtained by using the remaining running data from each of the M subsets (excluding the running data in the reference training sets) as the test set for each of the M subsets. The preset number of random selections is set based on empirical values or historical data.
[0148] After obtaining M reference test sets, a general linear classifier (such as a logistic regression classifier, linear support vector machine, or Naive Bayes classifier) can be trained using the running data from the M reference training sets to obtain a first linear classifier. The linear classifier is then used to classify the running data in the M reference test sets, and the misclassification rate for each of the M reference test sets is calculated, resulting in M reference misclassification rate sets. Elements in each of the M reference training sets are then adjusted to obtain M first training sets and M first test sets. The first linear classifier is then trained using the running data from the first training sets to obtain a second linear classifier. Finally, the second linear classifier is used to classify the running data in the M first test sets. Calculate the misclassification rate for each of the M first test sets to obtain M sets of first misclassification rates. Determine if the first difference between each first misclassification rate in the M sets of first misclassification rates and the corresponding reference misclassification rate in the M sets of reference misclassification rates is equal to or less than a preset difference threshold. If the first difference is less than the preset difference threshold, the first test set in the M sets of first test sets is determined as the target test set, resulting in M target test sets. If the first difference is greater than or equal to the preset difference threshold, the elements in the M first training sets are repeatedly adjusted until the difference between the misclassification rate of the first test set corresponding to the adjusted first training set and the misclassification rate corresponding to the first test set corresponding to the previous adjusted first test set is less than the preset difference threshold, resulting in M target test sets. The target test set refers to the test dataset that enables the classifier to achieve the optimal classification effect after iterative optimization. When the difference between the misclassification rates of two adjacent iterations is less than the preset difference threshold (e.g., 0.001), the classifier is considered to have achieved a stable classification effect, and the corresponding test set at this time is the target test set.
[0149] In this example, each of the M subsets is split into a test set and a training set. The linear classifier is trained using the running data in the training set. The misclassifications of the linear classifier when classifying data in the test set are extracted and used as a classification performance indicator. The elements in the training set are adjusted to repeatedly train the linear classifier until the classification performance of the linear classifier reaches its optimal value (the first difference is less than a preset difference threshold). This improves the accuracy of classifying abnormal data in abnormal datasets using the linear classifier, and thus improves the accuracy of predicting the state of industrial robots.
[0150] In one possible implementation, a method for extracting features from a target dataset to obtain a target feature vector includes:
[0151] F1. Extract the temporal feature information of the target dataset to obtain the first feature vector;
[0152] F2. Based on the first feature vector, perform correlation feature extraction on the target dataset to obtain the target feature vector.
[0153] Specifically, this can be achieved by inputting the running data from the target dataset into a BLSTM network (Bidirectional Long Short-Term Memory), using the BLSTM network to extract the temporal features between each running data in the target dataset, and obtaining a first feature vector set; then concatenating the elements in the first feature vector set to obtain a first feature vector.
[0154] After obtaining the first feature vector, the first feature vector and the running data in the target dataset can be input into a CNN (Convolutional Neural Network). The CNN network extracts spatial features between the running data in the target dataset based on the first feature vector, resulting in a spatial feature information set. The elements in the spatial feature information set are concatenated to obtain a spatial feature vector. The CNN network is then used to extract correlation features from the running data in the target dataset based on the spatial feature vector, resulting in a target feature vector set. Finally, the elements in the target feature vector set are concatenated to obtain the target feature vector.
[0155] For examples consistent with the above embodiments, please refer to... Figure 2 , Figure 2 This application provides a schematic diagram of the structure of an industrial robot, as shown in the embodiment. Figure 2As shown, it includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps.
[0156] Obtain the current pending tasks of the industrial robot, and obtain the target state information of the industrial robot;
[0157] Based on the current pending task and the target status information, the target operating parameter information of the industrial robot is determined;
[0158] The industrial robot is controlled based on the target operating parameter information.
[0159] The above primarily describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, an industrial robot includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0160] This application embodiment can divide the industrial robot into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.
[0161] For those consistent with the above, please refer to Figure 3 , Figure 3 This application provides a schematic diagram of the structure of a control device for an industrial robot. For example... Figure 3 As shown, the device includes:
[0162] The acquisition unit 301 is used to acquire the current pending tasks of the industrial robot and to acquire the target state information of the industrial robot.
[0163] The determining unit 302 is configured to: determine application scenario information based on the current pending transaction; determine running accuracy information and running configuration information based on the application scenario information; determine reference running parameter information based on the running accuracy information and running configuration information; generate parameter correction information based on the target state information; and perform correction processing on the reference running parameter information using the parameter correction information to obtain the target running parameter information.
[0164] The control unit 303 is used to control the industrial robot according to the target operating parameter information.
[0165] In one possible implementation, the acquisition unit 301 is specifically used for: acquiring the target state information of the industrial robot.
[0166] Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset;
[0167] Perform data balancing on the anomalous data in the anomalous dataset to obtain the target dataset;
[0168] Feature extraction is performed on the target dataset to obtain the target feature vector;
[0169] The state of the industrial robot is predicted based on the target feature vector and the real-time operating data of the industrial robot to obtain the target state information.
[0170] In one possible implementation, in terms of performing data balancing on the anomalous data in the anomalous dataset to obtain the target dataset, the acquisition unit 301 is specifically used for:
[0171] The abnormal dataset is denoised to obtain the first dataset.
[0172] Cluster the first data in the first dataset to obtain M subsets.
[0173] Determine the classification size for each of the M subsets to obtain a set of classification size information;
[0174] The target dataset is obtained by performing data balancing on the M subsets based on the classification size information in the classification size information set.
[0175] In one possible implementation, in obtaining the target dataset by performing data balancing on M subsets of the dataset based on the classification size information in the classification size information set, the acquisition unit 301 is specifically used for:
[0176] Determine the target test set corresponding to each of the M subsets of data, thus obtaining the M target test sets;
[0177] Calculate the misclassification rate for each of the M target test sets to obtain the M target misclassification rate sets;
[0178] Calculate the average misclassification error rate for each subset of data based on the target misclassification rates in the M target misclassification rate sets, and obtain the M average misclassification error rates;
[0179] Based on the average misclassification error rate of M data sets and the classification size information in the classification size information set, determine the number of running data sets that need to be synthesized for each of the M data sets, and obtain the first set of data sets.
[0180] Based on the first quantity in the first quantity set, data is synthesized from each of the M subsets to obtain the target dataset.
[0181] In one possible implementation, after determining the target test set corresponding to each of the M subsets of datasets to obtain the M target test sets, the acquisition unit 301 is specifically used for:
[0182] In each of the M subsets, a predetermined number of running data points are selected as the training set for the corresponding subset, resulting in M reference training sets.
[0183] Determine the test set corresponding to each of the M reference training sets to obtain the M reference test sets;
[0184] Iterative linear classification training is performed on each of the M reference test sets to obtain M target test sets.
[0185] In one possible implementation, regarding the feature extraction of the target dataset to obtain the target feature vector, the acquisition unit 301 is specifically used for:
[0186] Extract the temporal feature information of the target dataset to obtain the first feature vector;
[0187] Based on the first feature vector, the associated features of the target dataset are extracted to obtain the target feature vector.
[0188] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the industrial robot control methods described in the above method embodiments.
[0189] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the industrial robot control methods described in the above method embodiments.
[0190] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0191] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0192] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.
[0193] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0194] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.
[0195] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0196] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.
[0197] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A control method for an industrial robot, characterized in that, The method includes: Obtain the current pending tasks of the industrial robot, and obtain the target state information of the industrial robot; Determine the application scenario information based on the current pending transactions; Determine the operational accuracy information and operational configuration information based on the application scenario information; The reference operating parameter information is determined based on the operating accuracy information and the operating configuration information; Generate parameter correction information based on the target state information; The reference operating parameter information is corrected using the parameter correction information to obtain the target operating parameter information; The industrial robot is controlled based on the target operating parameter information.
2. The control method for an industrial robot according to claim 1, characterized in that, Obtaining the target state information of the industrial robot includes: Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset; Perform data balancing on the anomalous data in the anomalous dataset to obtain the target dataset; Feature extraction is performed on the target dataset to obtain the target feature vector; The state of the industrial robot is predicted based on the target feature vector and the real-time operating data of the industrial robot to obtain the target state information.
3. The control method for an industrial robot according to claim 2, characterized in that, The process of performing data balancing on the abnormal data in the abnormal dataset to obtain the target dataset includes: The abnormal dataset is denoised to obtain the first dataset. Cluster the first data in the first dataset to obtain M subsets. Determine the classification size for each of the M subsets to obtain a set of classification size information; The target dataset is obtained by performing data balancing on the M subsets based on the classification size information in the classification size information set.
4. The control method for an industrial robot according to claim 3, characterized in that, The process of balancing M subsets of data based on classification size information in the classification size information set to obtain the target dataset includes: Determine the target test set corresponding to each of the M subsets of data, thus obtaining the M target test sets; Calculate the misclassification rate for each of the M target test sets to obtain the M target misclassification rate sets; Calculate the average misclassification error rate for each subset of data based on the target misclassification rates in the M target misclassification rate sets, and obtain the M average misclassification error rates; Based on the average misclassification error rate of M data sets and the classification size information in the classification size information set, determine the number of running data sets that need to be synthesized for each of the M data sets, and obtain the first set of data sets. Based on the first quantity in the first quantity set, data is synthesized from each of the M subsets to obtain the target dataset.
5. The control method for an industrial robot according to claim 4, characterized in that, The process of determining the target test set corresponding to each of the M subsets of data to obtain the M target test sets includes: In each of the M subsets, a predetermined number of running data points are selected as the training set for the corresponding subset, resulting in M reference training sets. Determine the test set corresponding to each of the M reference training sets to obtain the M reference test sets; Iterative linear classification training is performed on each of the M reference test sets to obtain M target test sets.
6. The control method for an industrial robot according to claim 5, characterized in that, The step of extracting features from the target dataset to obtain the target feature vector includes: Extract the temporal feature information of the target dataset to obtain the first feature vector; Based on the first feature vector, the associated features of the target dataset are extracted to obtain the target feature vector.
7. A control device for an industrial robot, characterized in that, The device includes: The acquisition unit is used to acquire the current pending tasks of the industrial robot and to acquire the target state information of the industrial robot. The determining unit is configured to: determine application scenario information based on the current pending transaction; determine operation accuracy information and operation configuration information based on the application scenario information; determine reference operation parameter information based on the operation accuracy information and operation configuration information; generate parameter correction information based on the target status information; and perform correction processing on the reference operation parameter information using the parameter correction information to obtain target operation parameter information. The control unit is used to control the industrial robot according to the target operating parameter information.
8. The control device for an industrial robot according to claim 7, characterized in that, In acquiring the target state information of the industrial robot, the acquisition unit is specifically used for: Extract abnormal data from the historical operating data of industrial robots to obtain an abnormal dataset; Perform data balancing on the anomalous data in the anomalous dataset to obtain the target dataset; Feature extraction is performed on the target dataset to obtain the target feature vector; The state of the industrial robot is predicted based on the target feature vector and the real-time operating data of the industrial robot to obtain the target state information.
9. An industrial robot, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the steps of the control method for the industrial robot as described in any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the control method for the industrial robot as described in any one of claims 1 to 6.