A method for determining operating failure of industrial robots

Abstract

The present invention discloses an industrial robot operation fault judgment method, which relates to the technical field of fault judgment. By comprehensively analyzing trajectory data to be executed by the robot and corresponding load data, the influence of the robot operation state on the observation time window is evaluated; and the influence of the robot operation state on the observation time window is divided into influence levels, which are divided into low-level influence and high-level influence, and corresponding solutions are enabled to improve the accuracy of the robot operation fault judgment. In this way, it can be judged according to the actual situation whether it is necessary to appropriately increase the size of the observation time window, reduce the missed judgment and inaccuracy of the fault judgment, thereby reducing the safety hazards of equipment and reducing the impact on production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of fault determination, and in particular to a method for determining an industrial robot operation fault. Background Art

[0002] Industrial robots are devices that can perform automated tasks and are widely used in various production links of the manufacturing industry. They usually complete end-of-line operations through robotic arms and can perform a variety of tasks. By programming and inputting various operating data, industrial robots can efficiently complete various tasks while meeting high precision and high efficiency requirements, significantly improving production efficiency.

[0003] During operation, when the operating trajectory of an industrial robot suddenly deviates from the set standard trajectory and the deviation value exceeds the preset allowable range of the system, in order to ensure continuous production and timely detect whether there is an operating failure, the system will set an observation time window in the following period of time to continuously monitor the degree of trajectory deviation; through the observation time window, the system can determine whether the deviation is a systematic accidental error or an actual failure, and take corresponding measures; the advantage of this is that it can effectively distinguish between accidental errors and failures, discover potential problems in a timely manner, and avoid adverse consequences caused by equipment damage or production interruptions. At the same time, this method can also reduce safety hazards caused by small errors and further improve the operating reliability and work efficiency of industrial robots.

[0004] In actual monitoring, the observation time window is usually set based on historical experience. For example, if the trajectory deviation usually disappears within a certain number of seconds, the observation time window can be set to the corresponding number of seconds. However, in actual operation scenarios, the observation time required under different operating conditions may be different. If it is not possible to determine whether the size of the observation time window needs to be appropriately increased based on the actual situation, it may lead to missed and inaccurate fault determination, thereby increasing equipment safety risks and affecting production efficiency. Summary of the invention

[0005] The purpose of the present invention is to solve the above-mentioned problems and provide a method for determining an industrial robot operation fault.

[0006] The present invention provides a method for determining an operation fault of an industrial robot, the method comprising:

[0007] Obtain the trajectory data and corresponding load data that the robot is about to perform, analyze the three-dimensional position of the landing point of the robot's operation action according to the trajectory data, calculate the trajectory complexity coefficient, and evaluate the complexity of the landing point of the robot's operation action;

[0008] Analyze the load degree of the robot's upcoming operation according to the load data, calculate the load coefficient, and evaluate the impact of the robot's upcoming operation on the robot's own load;

[0009] Comprehensively analyze the complexity of the robot's upcoming operation point and the impact of the robot's upcoming operation on the robot's own load, and evaluate the impact of the robot's operating state on the originally set observation time window;

[0010] The impact of the robot's operating state on the observation time window is divided into low-level impact and high-level impact, and corresponding solutions are enabled.

[0011] Optionally, the trajectory complexity coefficient is obtained by:

[0012] Get the three-dimensional coordinates and corresponding timestamps of each trajectory point to be executed by the robot, and mark the coordinates of each trajectory point as , and record the corresponding timestamp as ;in, , Indicates the total number of trajectory point coordinates;

[0013] Calculate the first-order derivative of each trajectory point, that is, the corresponding velocity vector , the calculation formula is: ;

[0014] Calculate the second-order derivative of each trajectory point, that is, the corresponding acceleration vector , the calculation formula is: ;

[0015] The calculation trajectory is The curvature at each time point , the calculation formula is: ; In the formula, is the modulus of the second-order derivative of the trajectory, is the modulus of the first-order derivative of the trajectory;

[0016] Calculate the trajectory complexity coefficient, the calculation formula is: , where is the trajectory complexity coefficient.

[0017] Optionally, the load factor is obtained by:

[0018] The weight and torque of the robot itself are obtained when the robot is to perform each action, and are expressed as data points for each action. The corresponding data points of all actions constitute a data set; the data set is clustered to determine the load factor of each action; specifically:

[0019] The steps to cluster the data set according to the K-means clustering method are as follows:

[0020] Step 1: Use the elbow method to determine the optimal number of clusters K for the data set;

[0021] Step 2: Randomly select K data points in the data set as the initial cluster centers. For each data point, calculate its Euclidean distance to each initial cluster center, traverse the K initial cluster centers, and assign it to the cluster corresponding to the nearest initial cluster center;

[0022] Step 3: After traversing all data points, the initial clusters are obtained. For each initial cluster, the mean of the data points in it is calculated to obtain a new cluster center.

[0023] Step 4: Repeat steps 2 and 3 until the cluster center no longer changes, and obtain the final cluster and the final cluster center;

[0024] The final clusters and final cluster centers are mapped to two-dimensional coordinates to obtain the load factor of each action.

[0025] Optionally, the final clusters and the final cluster centers are mapped to two-dimensional coordinates to obtain the load factor of each action including:

[0026] Calculate the weight coefficient of each final cluster, the calculation formula is: ,in Expressed as The weight coefficients of the final clusters, Represented as the number of all data points in the dataset, Indicates The number of data points in the final clusters;

[0027] Calculate the distance of each data point in each final cluster to the coordinate origin on the two-dimensional coordinate to obtain the preliminary coordinate distance of the data point, and multiply the preliminary coordinate distance of the data point by the weight coefficient corresponding to the final cluster in which the data point is located to obtain the actual coordinate distance of the data point;

[0028] The actual coordinate distance of each data point is used as the load coefficient of the corresponding action, and the mean of the load coefficients of all actions is calculated as the load coefficient.

[0029] Optionally, the complexity of the landing point of the robot's upcoming operation action and the impact of the robot's upcoming operation action on the robot's own load are comprehensively analyzed to evaluate the impact of the robot's operation state on the observation time window:

[0030] The trajectory complexity coefficient and the load coefficient are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the influence value label of the observation time window for each group of first eigenvectors as the prediction target, and minimizes the sum of prediction errors of the influence value labels of all observation time windows as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The influence value of the observation time window is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0031] Optionally, the influence degree of the robot operation state on the observation time window is divided into influence levels, and is divided into low-level influence and high-level influence. Specifically:

[0032] The impact value of the observation time window is compared with the preset standard threshold. If the impact value of the observation time window is less than the preset standard threshold, it means that the robot's next operation has a weak impact on the size of the observation time window. At this time, it is classified as a low-level impact. The observation time window continues to monitor and observe according to the originally set time window to determine whether the industrial robot operation fails.

[0033] If the impact value of the observation time window is not less than the preset standard threshold, it means that the robot's subsequent operation has a strong impact on the size of the observation time window. At this time, it is classified as a high-level impact, and the length of the observation time window is increased. The robot is monitored and observed to determine whether there is a fault in the operation of the industrial robot.

[0034] Optionally, monitoring, observing and determining whether an industrial robot operation fails specifically includes:

[0035] Continuously obtain the trajectory deviation within the observation time window to obtain a trajectory deviation set; compare each trajectory deviation in the trajectory deviation set with the preset trajectory deviation threshold. If any trajectory deviation is not less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has a fault; if all trajectory deviations are less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has no fault.

[0036] Beneficial effects of the present invention:

[0037] 1. The present invention analyzes the three-dimensional position of the point where the robot is about to perform an operation through trajectory data, calculates the trajectory complexity coefficient, and evaluates the complexity of the point where the robot is about to perform an operation, which can improve the probability of early fault detection. In particular, in high-complexity operations, fault signals may gradually appear. By accurately calculating the trajectory complexity coefficient, excessive monitoring of simple tasks can be avoided, resources can be saved, and the observation time window can be adjusted according to the trajectory complexity to improve the reliability of fault detection;

[0038] 2. The present invention analyzes the load degree of the robot's upcoming operation action through load data, calculates the load coefficient, and evaluates the impact of the robot's upcoming operation action on the robot itself.

[0039] High load increases the risk of failure. By analyzing load changes, early warning can be given and the observation time window can be adjusted to ensure that potential fault signals are captured. Dynamic adjustment of the observation window helps to accurately determine faults under high load and avoid premature or late processing. At the same time, the load factor can also optimize maintenance efficiency, extend the service life of the robot, and provide a decision-making basis for system optimization. The robot's performance and stability can be improved by adjusting the operation path or control algorithm.

[0040] 3. The present invention conducts a comprehensive analysis on the complexity of the landing point of the robot's upcoming operation action and the impact of the robot's upcoming operation action on the robot's own load, thereby evaluating the impact of the robot's upcoming operation action on the robot's own load. It can determine whether it is necessary to appropriately increase the size of the observation time window based on the actual situation, reduce missed fault judgments and inaccuracies, thereby reducing equipment safety hazards and reducing the impact on production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with the accompanying drawings.

[0042] Figure 1 The present invention is a flow chart of a method for determining an operation fault of an industrial robot. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0045] The embodiment of the present invention provides a method for determining an operation fault of an industrial robot. Figure 1 , Figure 1 A flowchart of an industrial robot operation fault determination method provided by an embodiment of the present invention. The method comprises the following steps:

[0046] Obtain the trajectory data and corresponding load data that the robot is about to perform, analyze the three-dimensional position of the landing point of the robot's operation action according to the trajectory data, calculate the trajectory complexity coefficient, and evaluate the complexity of the landing point of the robot's operation action;

[0047] Analyze the load degree of the robot's upcoming operation according to the load data, calculate the load coefficient, and evaluate the impact of the robot's upcoming operation on the robot's own load;

[0048] Comprehensively analyze the complexity of the robot's upcoming operation point and the impact of the robot's upcoming operation on the robot's own load, and evaluate the impact of the robot's operating state on the originally set observation time window;

[0049] The impact of the robot's operating status on the observation time window is divided into low-level impact and high-level impact, and corresponding solutions are enabled to improve the accuracy of robot operation fault judgment.

[0050] In one embodiment, the three-dimensional position of the robot to be operated is analyzed according to the trajectory data, the trajectory complexity coefficient is calculated, and the complexity of the landing point of the robot to be operated is evaluated;

[0051] It should be noted that when observing and judging whether an industrial robot has an operational failure, the complexity of the three-dimensional position of the landing point of the operating action that needs to be executed in the setting of the next observation time window is an important factor affecting the length of the observation time window; the trajectory complexity coefficient is used to measure the complexity of the trajectory of the landing point of the operating action executed by the robot. This coefficient evaluates the complexity of the trajectory operation by analyzing the three-dimensional coordinates of the trajectory to be operated; a high trajectory complexity coefficient often indicates that the operation of the industrial robot is too complex. The robot needs to deal with more local changes and details when performing tasks, and needs to frequently adjust the path or posture to cope with delicate or complex operation requirements; in this case, the robot may face higher stress concentration, joint wear, decreased sensor accuracy or increased control system load during operation, and these potential hidden dangers usually require a certain amount of time to accumulate before they can manifest as detectable faults. If the observation time window is too short, it may not cover enough operating cycles, thereby missing key fault signals and failing to capture all potential faults of the robot.

[0052] Analyzing the effect of trajectory complexity coefficient on evaluating the influence of observation time window size has the following advantages:

[0053] Improve the probability of early fault detection: The trajectory complexity coefficient reflects the operational complexity of the robot when performing tasks. For highly complex operations, the robot needs to deal with more sophisticated control and path adjustments, which means that potential fault signals may gradually appear over multiple cycles. By extending the observation time window, it can ensure that the system has enough time to capture these tiny but gradually accumulated fault signals, so that problems can be discovered and handled earlier. This delay effect is particularly important in complex operations because some faults take a certain amount of time to accumulate before becoming obvious.

[0054] Avoiding over-monitoring and wasting resources: Although more complex trajectory operations require longer observation time windows, by accurately calculating the trajectory complexity coefficient, it is possible to avoid setting all operations to too long observation windows. For simpler tasks, the trajectory complexity coefficient is low, which means that the robot operation is more stable and can be effectively monitored in a short time window. This precise adjustment enables flexible switching between complex and simple tasks, avoiding over-monitoring of simple tasks, saving computing and storage resources, and not losing critical monitoring opportunities.

[0055] Improve the intelligent level of decision-making: The trajectory complexity coefficient provides an intelligent decision-making basis for setting the observation time window. The system can automatically adjust the length of the observation time window according to the complexity of the trajectory that the robot is about to execute, thereby avoiding manual intervention and excessive empirical judgment while ensuring monitoring accuracy. This automated and intelligent adjustment not only improves the timeliness of fault detection, but also can respond flexibly according to changes in tasks, improving the automation level and reliability of the system.

[0056] The method for obtaining the trajectory complexity coefficient is:

[0057] Before an industrial robot performs a task, the worker usually sets the landing point and corresponding time of each operation action, that is, the operation trajectory. By using offline programming software (such as ABB's RobotStudio, KUKA's SimPro, Fanuc's ROBOGUIDE, etc.), the robot operation task can be simulated and the coordinates of the operation path can be generated; the operation trajectory is generated by a motion planning algorithm, and these trajectory coordinates are calculated by a predetermined path generation algorithm (such as based on geometric planning, sampling algorithm, etc.). These algorithms generate the trajectory coordinates of the robot's end effector and the time corresponding to each trajectory coordinate by considering the robot's workspace, task requirements, collision avoidance constraints, etc.

[0058] Get the three-dimensional coordinates and corresponding timestamps of each trajectory point to be executed by the robot, and mark the coordinates of each trajectory point as , and record the corresponding timestamp as ;in, , , Indicates the total number of trajectory point coordinates;

[0059] Calculate the first-order derivative of each trajectory point, that is, the corresponding velocity vector , the calculation formula is: ;

[0060] Calculate the second-order derivative of each trajectory point, that is, the corresponding acceleration vector , the calculation formula is: ;

[0061] The calculation trajectory is The curvature at each time point , the calculation formula is: ; In the formula, is the modulus of the second-order derivative of the trajectory (acceleration vector), is the modulus of the first-order derivative of the trajectory (velocity vector);

[0062] Calculate the trajectory complexity coefficient, the calculation formula is: , where is the trajectory complexity coefficient.

[0063] It should be noted that by calculating the velocity, acceleration and curvature of the trajectory points, the robot's motion changes during the task can be accurately reflected. By comprehensively considering the rate of change of the trajectory (velocity, acceleration) and the curvature of the trajectory (curvature), the complexity of the robot's motion can be more comprehensively described. This helps to more accurately assess the robot's load when the trajectory complexity is high, because highly complex trajectories are often accompanied by greater loads and control difficulties. Using these physical characteristics to calculate the load factor can more reasonably reflect the potential failure risk of the robot under high load conditions and provide a more accurate monitoring and early warning mechanism.

[0064] In one embodiment, the load degree of the robot to be operated is analyzed according to the load data, the load coefficient is calculated, and the load influence degree of the robot to be operated on the robot itself is evaluated;

[0065] It should be noted that when observing and judging whether an industrial robot has an operational failure, the degree of influence of the robot's load on the robot itself when setting the next observation time window is an important factor affecting the length of the observation time window; the load factor is used to measure the influence of the operation that the end effector of the industrial robot will perform on the industrial robot itself. This factor evaluates the load of the industrial robot by analyzing the weight and torque that the industrial robot bears when performing the operation; a high load factor often means that the industrial robot bears a large weight and torque when performing the operation, which means that the robot needs higher power and more frequent adjustments when operating. Long-term high-load operation may cause wear of the robot joints, reduced motion accuracy, and even system overload, thereby increasing the risk of failure. Therefore, the higher the load factor, the greater the operating pressure and potential failure possibility faced by the robot when completing the task. Usually, a longer observation time window is required to ensure sufficient time to capture the fault signal and perform corresponding processing.

[0066] Analyzing the effect of load factor on evaluating the effect of observation time window size has the following advantages:

[0067] Improve the timeliness of fault warning: The load factor can directly reflect the working intensity of the robot when performing tasks. A higher load factor means that the robot may face greater torque and weight during operation, thereby increasing the risk of failure. By analyzing the changes in the load factor, the potential failure risk of the robot can be predicted in advance, providing an important basis for adjusting the observation time window, ensuring sufficient time for fault warning, and avoiding missing fault signals;

[0068] Optimize the dynamic adjustment of the observation time window: The load factor is closely related to the robot's workload. Under high load conditions, the robot needs more time to stabilize its state when performing tasks, so increasing the observation time window helps capture more potential fault signals. Through dynamic analysis of the load factor, the size of the observation time window can be flexibly adjusted according to the actual load conditions of the task, thereby achieving more accurate fault judgment and prediction;

[0069] Improve fault prevention and maintenance efficiency: Long-term high-load operation not only increases the wear of robot joints, but may also lead to decreased motion accuracy or overload of the control system. In the setting of the observation time window, the load factor is an important influencing factor, which can help extend the observation window under high load conditions, so as to timely discover and take preventive measures before potential faults develop into serious stages. By evaluating the load factor, the robot can be maintained and repaired at the appropriate time, thereby extending the service life of the robot and improving the overall operating efficiency;

[0070] Provide decision-making basis for system optimization: A high load factor usually means that the robot needs more energy and adjustment frequency to maintain normal operation, which may affect the robot's working cycle and accuracy. By comprehensively analyzing the impact of the load factor on the length of the observation time window, it can provide guidance for the design and optimization of the robot system, help adjust the operation path, reduce the load, or improve the robot control algorithm, thereby improving the robot's stability and performance.

[0071] The load factor is obtained as follows:

[0072] It should be noted that, usually, during the task planning stage, the staff will preset the load parameters for each operation action, and use load analysis tools or simulation software (such as ABB's RobotStudio, KUKA's SimPro, etc.) to simulate and calculate the load weight and torque corresponding to each operation action in combination with the task requirements;

[0073] The weight and torque of the robot itself are obtained when the robot is about to perform each action, and are expressed as data points for each action. The data points of each action can be expressed as a two-dimensional vector, namely [weight, torque]. The corresponding data points of all actions constitute a data set; cluster the data set to determine the load coefficient of each action; the specific method is:

[0074] The steps to cluster the data set according to the K-means clustering method are as follows:

[0075] Step 1: Use the elbow method to determine the optimal number of clusters K for the data set;

[0076] Step 2: Randomly select K data points in the data set as the initial cluster centers. For each data point, calculate its Euclidean distance to each initial cluster center, traverse the K initial cluster centers, and assign it to the cluster corresponding to the nearest initial cluster center;

[0077] Step 3: After traversing all data points, the initial clusters are obtained. For each initial cluster, the mean of the data points in it is calculated to obtain a new cluster center.

[0078] Step 4: Repeat steps 2 and 3 until the cluster center no longer changes, and obtain the final cluster and the final cluster center;

[0079] The final clusters and final cluster centers are mapped to two-dimensional coordinates to obtain the load factor of each action.

[0080] In one implementation, according to the above method, the optimal clustering results of the data set can be obtained. These clustering results can be used to identify different clusters in the data set, as well as the center and characteristics of each cluster. Such clustering results are helpful for further data analysis and interpretation.

[0081] It should be noted that clustering the data set by K-means clustering is simple and efficient, and can quickly process large-scale data sets; K-means clustering is a commonly used unsupervised learning algorithm, which is used to divide the data set into pre-defined K clusters; the goal of this algorithm is to divide the data points into K clusters, so that each data point belongs to the cluster represented by the cluster center closest to it, and the data points within the cluster are as similar as possible, while the data points between different clusters may not be similar;

[0082] It should be noted that weight and torque present a common linear relationship. When the weight and torque of each action are greater, the load coefficient of the corresponding action will be greater.

[0083] It should be noted that the data set is clustered according to the K-means clustering method:

[0084] Step 1: Use the elbow method to determine the optimal number of clusters K for the data set; the elbow method is one of the common methods used to determine the optimal number of clusters K in K-means clustering. The basic idea is to find an "elbow point" by observing the relationship between the number of clusters K and the total internal deviation square sum (WCSS) of the clustering results. The K value corresponding to the "elbow point" is the position where the WCSS decreases significantly after the number of clusters increases to a certain extent. WCSS refers to the sum of the squares of the distances from each data point to the center of the cluster to which it belongs, which indicates the compactness of the clustering results.

[0085] Step 2: According to the optimal number of clusters K, randomly select K data points from the data points as the initial cluster centers of clustering. The initial cluster centers refer to the K cluster centers randomly selected before starting the clustering process. These initial cluster centers are the starting points when the algorithm starts.

[0086] Step 3: For each data point in the data set, calculate the distance between it and each initial cluster center using the Euclidean distance. For example, calculate the data point and the initial cluster centers The distance between , and its calculation formula is ,in, Represents data points In the The values ​​in the clusters, Represents the initial cluster center In the The value of the clusters, Represented as data points and the initial cluster centers The number of features is calculated, and the data points are assigned to the cluster to which the nearest cluster center belongs. Euclidean distance refers to the distance between two points in Euclidean space, and is also one of the most commonly used distance metrics. In K-means clustering, Euclidean distance is usually used to calculate the distance between data points and cluster centers.

[0087] Step 4: After traversing all data points, we get the initial clusters. For each initial cluster, we calculate the mean of the data points in it. The calculation formula is: ,in Expressed as The set of data points in a cluster, Expressed as The new cluster centers of the clusters;

[0088] Step 5: Repeat steps 3 and 4 until the cluster center no longer changes, and obtain the final cluster and the final cluster center.

[0089] It should be specifically explained that the step of assigning a data point to the cluster to which the nearest cluster center belongs is to sort the distances between the data point and each cluster center from large to small, and select the smallest distance to assign the data point to the cluster to which it belongs.

[0090] In one embodiment, the final clusters and the final cluster centers are mapped to two-dimensional coordinates to obtain the load coefficients of each action, including:

[0091] Calculate the weight coefficient of each final cluster, the calculation formula is: ,in Expressed as The weight coefficients of the final clusters, Represented as the number of all data points in the dataset, Indicates The number of data points in the final clusters;

[0092] Calculate the distance of each data point in each final cluster to the coordinate origin on the two-dimensional coordinate to obtain the preliminary coordinate distance of the data point, and multiply the preliminary coordinate distance of the data point by the weight coefficient corresponding to the final cluster in which the data point is located to obtain the actual coordinate distance of the data point;

[0093] The actual coordinate distance of each data point is used as the load coefficient of the corresponding action, and the mean of the load coefficients of all actions is calculated as the load coefficient.

[0094] In one implementation, by considering the weight and torque of each action, the robot operation task is converted into specific numerical data for cluster analysis, which can accurately reflect the load degree of different actions during execution, thereby providing a more intuitive basis for evaluating the possible failure risk of the robot. Secondly, the elbow method is used to determine the optimal clustering number K, which ensures the rationality of the clustering results and avoids the analysis error caused by too many or too few clusters. This method can effectively distinguish the load characteristics of different operation actions, so that the robot can obtain a longer observation time window when performing high-load tasks to more fully capture potential fault signals. This method more finely reflects the influence of each operation action through the introduction of weight coefficients, making the load coefficient calculation more accurate and comprehensive, which helps to improve the accuracy and timeliness of fault prediction, thereby enhancing the reliability and safety of the robot system.

[0095] It should be noted that, because weight and torque are selected as data points for clustering, these two data points can now be used as coordinates and mapped to the two-dimensional coordinate axis, and each action will be mapped to a point on the two-dimensional coordinate axis; after clustering is completed, all data points will be distributed in different final clustering clusters on the two-dimensional coordinate axis, the closer the final clustering cluster is to the origin, and the smaller the load coefficient of the action corresponding to the data point it includes, the smaller the impact on the observation window size setting, and vice versa, the greater the impact on the observation window size setting; at the same time, the weight coefficient of each final clustering cluster is calculated to correct the distance from the data point to the origin, so that the different importance of each final clustering cluster can be ensured, and the influence of noise and outliers on the clustering results can be reduced to a certain extent, so that the load coefficient of each action calculated in the end is more accurate.

[0096] In one embodiment, the complexity of the landing point of the robot's upcoming operation action and the impact of the robot's upcoming operation action on the robot's own load are comprehensively analyzed to evaluate the impact of the robot's operation state on the observation time window;

[0097] The trajectory complexity coefficient and the load coefficient are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the influence value label of the observation time window for each group of first eigenvectors as the prediction target, and minimizes the sum of prediction errors of the influence value labels of all observation time windows as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The influence value of the observation time window is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0098] The method for obtaining the influence value of the observation time window is as follows: from the first feature vector training data of the trained machine learning model, obtain the corresponding function expression: ; In the formula, is the output function of the model, is the trajectory complexity coefficient, is the load factor, is the influence value of the observation time window.

[0099] In one embodiment, the influence of the robot operation state on the observation time window is divided into low-level influence and high-level influence, and corresponding solutions are enabled to improve the accuracy of robot operation fault determination;

[0100] The impact value of the observation time window is compared with the preset standard threshold. If the impact value of the observation time window is less than the preset standard threshold, it means that the robot's next operation has a weak impact on the size of the observation time window. At this time, it is classified as a low-level impact. The observation time window continues to monitor and observe according to the originally set time window to determine whether the industrial robot operation fails.

[0101] If the impact value of the observation time window is not less than the preset standard threshold, it means that the robot's subsequent operation has a strong impact on the size of the observation time window. At this time, it is classified as a high-level impact, and the length of the observation time window is increased. The robot is monitored and observed to determine whether there is a fault in the operation of the industrial robot.

[0102] It should be noted that the originally set time window can identify the time point or precursor of the common occurrence of faults by analyzing past fault cases and abnormal events; for example, assuming that the trajectory deviation disappears within 30 seconds, the observation time window can be set to 30 seconds. Therefore, based on the statistical analysis of historical data, a reasonable observation time window can be determined to ensure that the occurrence of abnormalities or faults can be captured;

[0103] It should be noted that increasing the length of the observation time window to two to three times the original time window can effectively balance monitoring accuracy and resource consumption. The advantage of this setting is that extending the monitoring time can provide more time to capture abnormalities that may occur in the robot under high load or complex trajectories, but will not excessively increase the computing and storage burden of the system. By appropriately increasing the observation time window, it is possible to ensure timely warnings in the early stages of potential failures, while avoiding unnecessary waste of resources and ensuring monitoring efficiency and system stability.

[0104] In one embodiment, monitoring, observing and determining whether an industrial robot operation fails includes:

[0105] Continuously obtain the trajectory deviation within the observation time window to obtain a trajectory deviation set; compare each trajectory deviation in the trajectory deviation set with the preset trajectory deviation threshold. If any trajectory deviation is not less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has a fault; if all trajectory deviations are less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has no fault, and the previous operation trajectory suddenly deviates from the set standard trajectory and the deviation value exceeds the preset allowable range of the system, which is just an accidental system error.

[0106] In one implementation, by continuously monitoring and comparing trajectory deviations, it is possible to detect in real time whether an industrial robot has experienced anomalies or failures during the execution of a task. If the trajectory deviation exceeds the preset threshold, timely identification and determination of the robot operation failure can effectively avoid potential damage to the robot or shutdown during the production process, thereby improving production efficiency and equipment reliability. At the same time, setting a reasonable trajectory deviation threshold and comparison mechanism can filter out accidental or temporary deviations, avoid overreaction, and ensure that the system is more stable and does not produce misjudgments. This method makes the monitoring system more intelligent and flexible, and can detect potential risks in a timely manner and avoid the expansion of losses, which helps to ensure the efficient operation of the production line and the long-term use of equipment.

[0107] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for determining an industrial robot operation fault, characterized in that: The following steps are involved: Obtain the trajectory data and corresponding load data that the robot is about to perform, analyze the three-dimensional position of the landing point of the robot's operation action according to the trajectory data, calculate the trajectory complexity coefficient, and evaluate the complexity of the landing point of the robot's operation action; Analyze the load degree of the robot's upcoming operation according to the load data, calculate the load coefficient, and evaluate the impact of the robot's upcoming operation on the robot's own load; Comprehensively analyze the complexity of the robot's upcoming operation point and the impact of the robot's upcoming operation on the robot's own load, and evaluate the impact of the robot's operating state on the originally set observation time window; The impact of the robot's operating status on the observation time window is classified into low-level impact and high-level impact, and corresponding solutions are activated; The method for obtaining the trajectory complexity coefficient is: Get the three-dimensional coordinates and corresponding timestamps of each trajectory point to be executed by the robot, and mark the coordinates of each trajectory point as , and record the corresponding timestamp as ;in, , Indicates the total number of trajectory point coordinates; Calculate the first-order derivative of each trajectory point, that is, the corresponding velocity vector , the calculation formula is: ； Calculate the second-order derivative of each trajectory point, that is, the corresponding acceleration vector , the calculation formula is: ； The calculation trajectory is The curvature at each time point , the calculation formula is: ; In the formula, is the modulus of the second-order derivative of the trajectory, is the modulus of the first-order derivative of the trajectory; Calculate the trajectory complexity coefficient, the calculation formula is: , where is the trajectory complexity coefficient; A comprehensive analysis is conducted on the complexity of the landing point of the robot's upcoming operation and the impact of the robot's upcoming operation on the robot's own load, and the impact of the robot's operating state on the observation time window is evaluated as follows: The trajectory complexity coefficient and the load coefficient are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the influence value label of the observation time window for each group of first eigenvectors as the prediction target, and minimizes the sum of prediction errors of the influence value labels of all observation time windows as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The influence value of the observation time window is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

2. The method for determining an industrial robot operation fault according to claim 1, characterized in that: The load factor is obtained as follows: The weight and torque of the robot itself are obtained when the robot is to perform each action, and are expressed as data points for each action. The corresponding data points of all actions constitute a data set; the data set is clustered to determine the load factor of each action; specifically: The steps to cluster the data set according to the K-means clustering method are as follows: Step 1: Use the elbow method to determine the optimal number of clusters K for the data set; Step 2: Randomly select K data points in the data set as the initial cluster centers. For each data point, calculate its Euclidean distance to each initial cluster center, traverse the K initial cluster centers, and assign it to the cluster corresponding to the nearest initial cluster center; Step 3: After traversing all data points, the initial clusters are obtained. For each initial cluster, the mean of the data points in it is calculated to obtain a new cluster center. Step 4: Repeat steps 2 and 3 until the cluster center no longer changes, and obtain the final cluster and the final cluster center; The final clusters and final cluster centers are mapped to two-dimensional coordinates to obtain the load factor of each action.

3. The method for determining an industrial robot operation fault according to claim 2, characterized in that: Mapping the final clusters and the final cluster centers to two-dimensional coordinates, the load coefficients of each action are obtained: Calculate the weight coefficient of each final cluster, the calculation formula is: ,in Expressed as The weight coefficients of the final clusters, Represented as the number of all data points in the dataset, Indicates The number of data points in the final clusters; Calculate the distance of each data point in each final cluster to the coordinate origin on the two-dimensional coordinate to obtain the preliminary coordinate distance of the data point, and multiply the preliminary coordinate distance of the data point by the weight coefficient corresponding to the final cluster in which the data point is located to obtain the actual coordinate distance of the data point; The actual coordinate distance of each data point is used as the load coefficient of the corresponding action, and the mean of the load coefficients of all actions is calculated as the load coefficient.

4. The method for determining an industrial robot operation fault according to claim 1, characterized in that: The influence of the robot operation state on the observation time window is divided into low-level influence and high-level influence. The impact value of the observation time window is compared with the preset standard threshold. If the impact value of the observation time window is less than the preset standard threshold, it means that the robot's next operation has a weak impact on the size of the observation time window. At this time, it is classified as a low-level impact. The observation time window continues to monitor and observe according to the originally set time window to determine whether the industrial robot operation fails. If the impact value of the observation time window is not less than the preset standard threshold, it means that the robot's subsequent operation has a strong impact on the size of the observation time window. At this time, it is classified as a high-level impact, and the length of the observation time window is increased. The robot is monitored and observed to determine whether there is a fault in the operation of the industrial robot.

5. The method for determining an industrial robot operation fault according to claim 4, characterized in that: Monitoring and observing to determine whether an industrial robot operation fails is as follows: Continuously obtain the trajectory deviation within the observation time window to obtain a trajectory deviation set; compare each trajectory deviation in the trajectory deviation set with the preset trajectory deviation threshold. If any trajectory deviation is not less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has a fault; if all trajectory deviations are less than the preset trajectory deviation threshold, it is determined that the industrial robot operation has no fault.

Images