A method and system for joint health monitoring of an industrial robot

Abstract

This invention discloses a method and system for monitoring the joint health status of industrial robots, relating to the field of equipment condition monitoring technology. The method includes: acquiring and preprocessing actual torque data and preset torque data for each joint of the target robot; determining the monitoring starting point and constructing a real-time anomaly monitoring sequence; calculating the anomaly values ​​of each joint in real time based on the real-time anomaly monitoring sequence and constructing an anomaly value sequence; calculating the real-time fault impact degree of each joint based on the anomaly value sequence; acquiring historical torque data of robots of the same model, calculating the baseline fault impact degree corresponding to each joint, and obtaining the real-time fault impact weight of each joint based on the real-time fault impact degree; and weighting the corresponding anomaly values ​​based on the real-time fault impact weight to calculate the joint health evaluation value of the target robot in real time. This invention can effectively distinguish between instantaneous torque disturbances and actual structural degradation, achieving accurate assessment of joint health status.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to a method and system for monitoring the joint health status of industrial robots. Background Technology

[0002] Industrial robots, as core equipment in modern manufacturing, are widely used in automated production lines. Their joint systems (including rotary joints and kinetic joints) endure repetitive movements at high frequencies and high loads for extended periods, making them prone to structural deterioration issues such as wear, loosening, and fatigue. The health of each rotary joint, operating under high loads for extended periods, directly determines the precision and lifespan of the entire industrial robot. Therefore, real-time monitoring of the joint health status of industrial robots is crucial for preventing structural failures, providing early warnings of faults, and reducing losses caused by unexpected production line downtime. However, existing methods for monitoring the joint health status of industrial robots mostly remain at the level of single-joint torque threshold alarms, failing to consider the chain effects of multi-joint linkage coupling, load disturbances, and the amplification of upper-level joint failures. These methods not only cannot distinguish between instantaneous torque disturbances and actual structural deterioration but also lack mechanisms for tracing the source of linkage anomalies, making it difficult to identify chain patterns where "slight loosening of lower-level joints leads to synchronous deterioration of upper-level joints."

[0003] Specifically, traditional single-point threshold alarm methods often rely on preset torque thresholds for anomaly detection, triggering an alarm when the actual torque exceeds the threshold. This monitoring and alarm method not only struggles to identify early, subtle fault characteristics but also fails to effectively distinguish between transient torque peaks caused by load disturbances (such as material jamming or external impacts) and persistent torque anomalies caused by mechanical wear or structural deterioration (such as loosening or deformation), easily leading to false alarms or missed alarms. Secondly, the robotic arm of an industrial robot is a coupled system with multiple joints. During movement, these joints are strongly coupled, and minor anomalies in lower-level joints propagate upwards through the kinematic chain, amplifying their impact based on dynamic relationships, causing synchronous anomalies in upper-level joints. Existing monitoring methods treat each joint as an independent monitoring unit, lacking quantitative analysis of the inter-joint linkages, making it difficult to identify fault propagation paths and pinpoint the root cause joint of the fault.

[0004] Furthermore, in actual operation, the impact of transient anomalies and persistent failures on the health status of industrial robot arms varies, and the importance of different joints within the overall machine differs; for example, failures in the base joint have a greater impact. Most existing monitoring methods do not consider the duration of anomalies and treat all joint anomalies equally, failing to differentiate assessments based on their physical location and impact weight. This dilutes or obscures early failure characteristics of critical, underlying joints, making it easy for monitoring systems to miss early structural failures and miss the optimal maintenance window. Maintenance remains primarily based on shutdowns after failures, resulting in significant losses from unexpected production line downtime and increased high maintenance costs. Summary of the Invention

[0005] To achieve a comprehensive and accurate assessment of the joint health status of industrial robots and to address the problem that traditional monitoring methods often miss early structural faults, this invention provides a method and system for monitoring the joint health status of industrial robots, the technical solution of which is as follows: In a first aspect, the present invention provides a method for monitoring the joint health status of an industrial robot, comprising the following steps: acquiring actual torque data and preset torque data of each joint of the target robot and preprocessing them, determining the monitoring starting point and constructing a real-time anomaly monitoring sequence; calculating the anomaly value of each joint in real time based on the real-time anomaly monitoring sequence and constructing an anomaly value sequence for each joint; calculating the real-time fault impact degree of each joint based on the anomaly value sequence of each joint; acquiring historical torque data of the same model of robot, calculating the baseline fault impact degree corresponding to each joint, and obtaining the real-time fault impact weight of each joint based on the real-time fault impact degree; and weighting the corresponding anomaly values ​​based on the real-time fault impact weight to calculate the joint health evaluation value of the target robot in real time. In this process, the entire operation of the target robot from its initial position to its reset after completing an action is considered as a complete sampling cycle. The actual torque data of each joint of the target robot is collected in real time and uploaded to the robot control terminal. The actual torque data and preset torque data of each joint of the target robot are directly read from the robot control terminal when the target robot performs an action.

[0006] Preferably, the absolute difference between the actual torque data and the corresponding preset torque data is used as the torque deviation at the corresponding sampling time. The tolerance of torque deviation for each joint is set according to the processing accuracy requirements of the actual application scenario. At the current sampling time, when the torque deviation of any joint exceeds the tolerance range, the current sampling time is used as the monitoring starting point for all joints. The actual torque data and preset torque data of each joint are normalized. The normalized actual torque data starting from the monitoring starting point is used to form the real-time anomaly monitoring sequence of the target robot in the current sampling period. The real-time anomaly monitoring sequence of each joint starting from the monitoring starting point is updated in real time based on the real-time collected actual torque data.

[0007] Preferably, any joint of the target robot is selected as the target joint. Based on the real-time anomaly monitoring sequence corresponding to the target joint at the current sampling time, the absolute difference between the normalized actual torque data and the normalized preset torque data at each sampling time in the real-time anomaly monitoring sequence is calculated. The mean of each absolute difference is used as the torque deviation degree of the target joint at the current sampling time. The number of sampling points in the real-time anomaly monitoring sequence is mapped using a hyperbolic tangent function. The mapped value is used as the anomaly duration feature of the target joint at the current sampling time. The product between the torque deviation degree and the anomaly duration feature is used as the anomaly value of the target joint at the current sampling time. Similarly, the anomaly values ​​of each joint at the current sampling time are obtained. Starting from the monitoring start point, the anomaly values ​​of each joint are recorded and formed into anomaly value sequences for each joint. The anomaly values ​​and anomaly value sequences are updated synchronously in real time based on the changes in the sampling time.

[0008] Preferably, any joint of the target robot is selected as the target joint, and the outlier sequence of each joint at the current sampling time is extracted. The Pearson correlation coefficient between the outlier sequence of the target joint and the outlier sequences of the other joints is calculated in turn. When the Pearson correlation coefficient is greater than 0, the value of the Pearson correlation coefficient is used as the synchronization coefficient between the corresponding joint and the target joint. When the Pearson correlation coefficient is less than or equal to 0, the corresponding synchronization coefficient is set to 0. The maximum and minimum values ​​of each synchronization coefficient corresponding to the target joint are extracted and the range is calculated. The natural exponential function is used to reverse map the range, and the mapped value is used as the real-time fault impact degree of the target joint at the current sampling time. Similarly, the real-time fault impact degree of each joint at the current sampling time is obtained, and the real-time fault impact degree is updated synchronously in real time based on the changes in the sampling time.

[0009] Preferably, under the same application scenario, historical torque data of the same model robot is acquired from multiple historical sampling periods, including historical actual torque data and historical preset torque data of each joint. Based on the same processing method, the monitoring starting point of the historical torque data in each historical sampling period is obtained, and the last sampling point in each historical sampling period is taken as the monitoring endpoint. The data sequence of a certain joint from the monitoring starting point to the monitoring endpoint is taken as a historical anomaly monitoring sequence of the corresponding joint, thereby obtaining multiple historical anomaly monitoring sequences of each joint. Based on the same calculation process, multiple historical anomaly values ​​and historical anomaly value sequences of each joint are obtained, and then the impact degree of multiple historical faults of each joint is calculated.

[0010] Preferably, a Gaussian distribution is used to statistically process the impact of multiple historical faults on each joint, and the impact is filtered based on the three sigma principle to obtain the impact of historical faults within three standard deviations for each joint. The average of the impact of historical faults on a selected joint is used as the baseline impact of the corresponding joint, and the baseline impact of faults on each joint is obtained in the same way. The cumulative value of the baseline impact of faults on each joint is used as the comprehensive benchmark, and the ratio between the real-time impact of faults on each joint and the comprehensive benchmark is used as the real-time impact weight of the corresponding joint at the current sampling time. The real-time impact weight is updated synchronously in real time based on the changes at the sampling time.

[0011] Preferably, the product of the outlier value of each joint at the current sampling time and the corresponding real-time fault impact weight is used as the weighted outlier value of the corresponding joint at the current sampling time. The average of the weighted outlier values ​​of each joint is used as the comprehensive joint anomaly degree of the target robot at the current sampling time. The value of 1 minus the comprehensive joint anomaly degree is used as the joint health evaluation value of the target robot at the current sampling time. Based on the processing accuracy requirements of the actual application scenario, a joint health threshold is set. When the joint health evaluation value is less than the joint health threshold, it is determined that the target robot at the current sampling time has a risk of joint structure deterioration and an early warning operation is performed.

[0012] Secondly, the present invention provides a joint health status monitoring system for industrial robots, which is used to implement the aforementioned joint health status monitoring method for industrial robots. The system includes: a processor, a memory, a communication interface, a data acquisition device, and a robot control terminal. The processor stores computer program instructions for implementing the aforementioned joint health status monitoring method for industrial robots. The communication interface is communicatively connected to the robot control terminal, and the robot control terminal is communicatively connected to the data acquisition device.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention filters out instantaneous torque disturbances by introducing outlier calculations to characterize persistent deviations, thereby effectively distinguishing instantaneous torque disturbances from actual structural degradation to avoid false alarms. Then, through the synchronization analysis of multi-joint linkage coupling, a fault chain effect tracing mechanism is established, which can identify the chain effect of faults propagating upwards from the underlying joints and accurately locate the root cause joint. Simultaneously, based on historical torque data, the impact of historical faults is obtained, and a weighted health evaluation model based on real-time fault impact weights is constructed, highlighting the warning weight of high-risk underlying joints, significantly improving the early fault identification rate and the warning sensitivity for early structural faults in key underlying joints. Compared with existing technologies, this invention achieves a transformation from "single-point threshold alarm" to "comprehensive trend warning" and from "post-event downtime maintenance" to "early predictive maintenance," effectively reducing losses from unexpected production line downtime, lowering maintenance costs, and extending equipment lifespan. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating an implementation method for monitoring the joint health status of an industrial robot according to an embodiment of the present invention. Figure 2 This is a structural block diagram of a joint health status monitoring system for industrial robots according to an embodiment of the present invention. Detailed Implementation

[0015] The technical features of the present invention will be further described in detail below with reference to the accompanying drawings so that those skilled in the art can understand them.

[0016] A method for monitoring the joint health status of industrial robots, the implementation process is as follows: Figure 1 As shown, the specific implementation steps are as follows: Step S1: Obtain the actual torque data and preset torque data of each joint of the target robot and perform preprocessing, determine the monitoring starting point and construct a real-time anomaly monitoring sequence.

[0017] Because robot joints are prone to structural degradation such as wear, loosening, and fatigue due to prolonged operation, resulting in persistent torque anomalies, but during robot operation, transient torque peaks may occur due to load disturbances such as material jamming and external impacts. To effectively distinguish between transient torque disturbances and actual structural degradation, it is necessary to periodically collect torque data during each complete execution of the target robot's action. In real-world scenarios, a continuous small anomaly may be more indicative of a fault than a transient large peak. Therefore, judging whether a joint's torque anomaly is a transient disturbance or a persistent anomaly based on both the magnitude and duration of the anomaly can help mitigate the problem of false alarms caused by transient interference.

[0018] Specifically, a complete sampling cycle is defined as one operation of the target robot from its initial position to its reset after completion. A dynamic torque sensor is used to synchronously collect the actual torque data of each joint of the target robot in real time at a fixed sampling frequency and upload it to the robot control terminal. The robot control terminal directly reads the actual torque data and preset torque data of each joint when the target robot performs its actions. The absolute difference between the actual torque data and the corresponding preset torque data is taken as the torque deviation at the corresponding sampling time. Based on the machining accuracy requirements of the actual application scenario, the tolerance for torque deviation of each joint is set. At the current sampling time, if the torque deviation of any joint exceeds the tolerance range, the current sampling time is taken as the monitoring starting point for all joints. The normalized actual torque data starting from the monitoring starting point forms the real-time anomaly monitoring sequence of the target robot for the current sampling cycle. The real-time anomaly monitoring sequence of each joint is updated synchronously in real time based on the real-time collected actual torque data.

[0019] The actual torque data is collected starting from the initial moment of the target robot's current action, and the sampling frequency can be set to 10Hz. For the implementation of normalization, the maximum and minimum values ​​of the torque of each joint are first set according to the actual application scenario and the rated parameters of the target robot. Then, the Min-Max normalization method is used to normalize the actual torque data and preset torque data of each joint, thereby eliminating the influence of dimensions and allowing the data of different joints to be compared on the same scale.

[0020] Furthermore, the tolerance for torque deviation refers to the torque fluctuation error that each joint of the target robot can tolerate. The higher the machining accuracy requirement, the lower the tolerance value should be. The tolerance may vary for different types of joints. An exemplary value for the tolerance can be set to 0.03 times the corresponding preset torque data. When the deviation between the real-time torque of any joint and the preset torque exceeds the preset tolerance range, it indicates that the corresponding joint may be in an abnormal state. Based on the multi-joint linkage coupling characteristics of the robot, it can be assumed that all joints of the target robot show abnormal signs at the corresponding sampling time. Therefore, the corresponding sampling time can be used as the monitoring starting point for all joints to begin abnormal monitoring of all joints for a complete working cycle.

[0021] Step S2: Based on the real-time anomaly monitoring sequence, calculate the anomaly values ​​of each joint in real time and construct the anomaly value sequence of each joint.

[0022] To measure the degree of abnormality of each joint and quantify the abnormal characteristics of each joint at the current sampling time, it is necessary to comprehensively consider two dimensions: the magnitude of the abnormality and the duration of the abnormality.

[0023] Specifically, any joint of the target robot is selected as the target joint. Based on the real-time anomaly monitoring sequence corresponding to the target joint at the current sampling time, the absolute difference between the normalized actual torque data and the normalized preset torque data at each sampling time in the real-time anomaly monitoring sequence is calculated. The mean of each absolute difference is taken as the torque deviation degree of the target joint at the current sampling time. The hyperbolic tangent function is used to map the number of sampling points in the real-time anomaly monitoring sequence. The mapped value is taken as the anomaly duration feature of the target joint at the current sampling time. The product between the torque deviation degree and the anomaly duration feature is taken as the anomaly value of the target joint at the current sampling time. Similarly, the anomaly values ​​of each joint at the current sampling time are obtained. Starting from the monitoring start point, the anomaly values ​​of each joint are recorded and formed into anomaly value sequences for each joint. The anomaly values ​​and anomaly value sequences are updated synchronously in real time based on the changes in the sampling time.

[0024] The outlier value of the target joint at the current sampling time is r, and its calculation formula is as follows:

[0025] In the formula, j represents the number of sampling points in the real-time anomaly monitoring sequence corresponding to the current sampling time. Represents the hyperbolic tangent function. This represents the value of the i-th sampling point within the real-time anomaly monitoring sequence, i.e., the normalized actual torque data corresponding to the i-th sampling time within the real-time anomaly monitoring sequence. This represents the normalized preset torque data corresponding to the i-th sampling point within the real-time anomaly monitoring sequence; when i is 1, it represents the monitoring start point.

[0026] For the outlier calculation formula, the multiplication factor in the first term represents the duration of the anomaly at the current sampling time of the target joint. As the number of sampling points j in the real-time anomaly monitoring sequence increases, the anomaly duration increases. The hyperbolic tangent function, which monotonically increases from 0 and infinitely approaches 1, normalizes the duration of the abnormality. The multiplication factor in the latter term represents the degree of torque deviation of the target joint at the current sampling time, which represents the average absolute torque deviation from the monitoring start point to the current sampling time. The larger this value is, the greater the average deviation of the actual torque of the target joint from the normal preset torque level.

[0027] Furthermore, outliers are the product of the duration of the outlier and the degree of torque deviation. Therefore, the larger the outlier, the greater the probability that the target joint has experienced a significant torque deviation for a long time and that the target joint has undergone actual structural deterioration. Moreover, the source of the fault can be determined based on the magnitude of the outliers of each joint. The joint with the largest outlier is most likely to be the root cause of the fault or the joint most severely affected by the fault.

[0028] Step S3: Calculate the real-time fault impact of each joint based on the outlier sequence of each joint.

[0029] The joints on an industrial robot are interconnected mechanisms. A failure in one joint, especially a bottom joint, can be transmitted to other joints through the kinematic chain, causing a chain reaction and resulting in synchronous abnormalities in the interconnected joints. When a joint in a different location experiences an abnormality, the impact on the overall industrial robot varies, and the importance of different joints in the whole machine also differs. For example, a failure in the base joint has a greater impact. Therefore, it is necessary to quantify the real-time impact of each joint's failure to provide a data foundation for subsequently constructing real-time failure impact weights.

[0030] Specifically, any joint of the target robot is selected as the target joint. The outlier sequence of each joint at the current sampling time is extracted. The Pearson correlation coefficient between the outlier sequence of the target joint and the outlier sequences of the other joints is calculated in turn. When the Pearson correlation coefficient is greater than 0, the value of the Pearson correlation coefficient is used as the synchronization coefficient between the corresponding joint and the target joint. When the Pearson correlation coefficient is less than or equal to 0, the corresponding synchronization coefficient is set to 0. The maximum and minimum values ​​of the synchronization coefficients corresponding to the target joint are extracted and the range of the synchronization coefficients is calculated. The natural exponential function is used to reverse map the range, and the mapped value is used as the real-time fault impact degree of the target joint at the current sampling time. Similarly, the real-time fault impact degree of each joint at the current sampling time is obtained, and the real-time fault impact degree is updated synchronously in real time based on the changes in the sampling time.

[0031] Wherein, if joint p is selected as the target joint, the synchronization coefficient between the target joint and joint k at the current sampling time is: The calculation formula is as follows:

[0032] In the formula, This represents the sequence of outliers at joint p at the current sampling time. This represents the sequence of outliers at joint k at the current sampling time. The function representing the calculation of the Pearson correlation coefficient.

[0033] PPMCC (Pearson-product-moment-correlation coefficient) is used to measure the degree and direction of the linear correlation between the trends of change of two variables; if the abnormality of joint k is highly synchronized with the abnormality of the target joint, that is, they increase and decrease simultaneously, then... A value close to 1 indicates that the abnormality of joint k is likely caused by a fault in the target joint; if it is asynchronous or unrelated, then... The value is close to 0, indicating that the abnormality of joint k is less affected by the failure of the target joint.

[0034] The larger the value of the real-time fault impact of the target joint at the current sampling moment, the greater the impact of the target joint on the target robot as a whole and the other joints. The fault effect of the target joint will spread evenly and synchronously to most of the linked joints. At this time, the synchronization coefficients of all joints except the target joint are very high and the values ​​are close. The smaller the range corresponding to the target joint, the higher the synchronization between the target joint and the other joints. Conversely, the larger the range corresponding to the target joint, the smaller the value of the real-time fault impact, indicating that the target joint has a smaller impact on the target robot as a whole and the other joints.

[0035] Step S4: Obtain historical torque data of the same model of robot, calculate the baseline fault impact degree of each joint, and obtain the real-time fault impact weight of each joint based on the real-time fault impact degree.

[0036] Since the effects of different joints are not consistent, a failure in the bottom joint can lead to more serious abnormalities in the upper joints. Therefore, abnormalities in the bottom joints need to be given relatively higher attention. Conversely, the top joints have a relatively smaller impact on the health assessment, so abnormalities in the top joints can be given relatively lower attention. Historical torque data can be used to learn the criticality of each joint, and when constructing the joint health assessment mechanism for industrial robots, different weights can be given to the abnormal values ​​of each joint.

[0037] Specifically, in the same application scenario, historical torque data from multiple historical sampling periods of the same model robot are acquired, including historical actual torque data and historical preset torque data for each joint. Based on the same processing method, the monitoring starting point of the historical torque data in each historical sampling period is obtained, and the last sampling point in each historical sampling period is taken as the monitoring endpoint. The data sequence of a certain joint from the monitoring starting point to the monitoring endpoint is taken as a historical anomaly monitoring sequence for the corresponding joint, thereby obtaining multiple historical anomaly monitoring sequences for each joint. Based on the same calculation process, multiple historical anomaly values ​​and historical anomaly value sequences for each joint are obtained, and then the impact degree of multiple historical faults of each joint is calculated.

[0038] In addition, a Gaussian distribution was used to statistically process the impact of multiple historical faults on each joint, and the impact was filtered based on the three sigma principle to obtain the impact of historical faults within three standard deviations for each joint. The average of the impact of historical faults on a selected joint was used as the baseline impact of the corresponding joint, and the baseline impact of faults on each joint was obtained in the same way. The cumulative value of the baseline impact of faults on each joint was used as the comprehensive benchmark, and the ratio of the real-time impact of faults on each joint to the comprehensive benchmark was used as the real-time impact weight of the corresponding joint at the current sampling time. The real-time impact weight of faults was updated synchronously in real time based on the changes at the sampling time.

[0039] Wherein, the real-time fault impact weight of the nth joint at the current sampling time is: The calculation formula is as follows:

[0040] In the formula, This indicates the real-time fault impact level of the nth joint at the current sampling time. This represents the set of baseline fault impact levels corresponding to each joint of the target robot. This represents the comprehensive baseline, which is the cumulative value of the impact of all baseline faults within the set.

[0041] Step S5: Weight the corresponding outliers based on the real-time fault impact weight, and calculate the joint health evaluation value of the target robot in real time.

[0042] Specifically, the product of the outlier value of each joint at the current sampling time and the corresponding real-time fault impact weight is used as the weighted outlier value of the corresponding joint at the current sampling time. The average of the weighted outlier values ​​of each joint is used as the comprehensive joint anomaly degree of the target robot at the current sampling time. The value of 1 minus the comprehensive joint anomaly degree is used as the joint health evaluation value of the target robot at the current sampling time. Based on the processing accuracy requirements of the actual application scenario, a joint health threshold is set. When the joint health evaluation value is less than the joint health threshold, it is determined that the target robot at the current sampling time has a risk of joint structural deterioration and an early warning operation is performed.

[0043] The joint health evaluation value of the target robot at the current sampling time is U, and its calculation formula is as follows:

[0044] In the formula, m represents the total number of joints of the target robot. This represents the real-time fault impact weight of the nth joint at the current sampling time. This represents the outlier value of the nth joint at the current sampling time; the range of joint health evaluation values ​​is... The joint health threshold can be set to 0.7.

[0045] In addition to directly determining fault warnings based on joint health thresholds, a graded warning mechanism can also be set up based on the range of joint health evaluation values. This allows for the real-time determination of the target robot's current risk level based on the joint health evaluation value at the current sampling time, and the execution of corresponding warning strategies.

[0046] This invention also discloses a joint health status monitoring system for industrial robots, used to implement the aforementioned joint health status monitoring method for industrial robots. The system structure is as follows: Figure 2 As shown, it includes: a processor, a memory, a communication interface, a data acquisition device, and a robot control terminal. The processor stores computer program instructions for implementing the above-mentioned method for monitoring the joint health status of an industrial robot. The communication interface is connected to the robot control terminal, and the robot control terminal is connected to the data acquisition device.

[0047] Specifically, the data acquisition device is a dynamic torque sensor installed on each joint, and the robot control terminal is an operation control platform used to set the parameters of the target robot's actions, such as the preset torque. The actual torque data collected in real time by the data acquisition device is first uploaded to the robot control terminal, and then the actual torque data and preset torque data of each joint of the target robot are directly read from the robot control terminal.

[0048] The embodiments included in this invention are merely descriptions of preferred embodiments of the invention and are not limited to the precise structures described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection. Any variations and improvements made by those skilled in the art to the technical solutions of this invention without departing from the design concept of this invention should fall within the scope of protection of this invention.

Claims

1. A method for monitoring the joint health status of industrial robots, characterized in that: Acquire the actual torque data and preset torque data of each joint of the target robot and perform preprocessing to determine the monitoring starting point and construct a real-time anomaly monitoring sequence; Based on the real-time anomaly monitoring sequence, the abnormal values ​​of each joint are calculated in real time and the abnormal value sequence of each joint is constructed; based on the abnormal value sequence of each joint, the real-time fault impact of each joint is calculated. Obtain historical torque data of the same model of robot, calculate the baseline fault impact degree of each joint, and obtain the real-time fault impact weight of each joint based on the real-time fault impact degree. Based on the real-time fault impact weight, the corresponding outlier values ​​are weighted and the joint health evaluation value of the target robot is calculated in real time. In this process, the entire operation of the target robot from its initial position to its reset after completing an action is considered as a complete sampling cycle. The actual torque data of each joint of the target robot is collected in real time and uploaded to the robot control terminal. The actual torque data and preset torque data of each joint of the target robot are directly read from the robot control terminal when the target robot performs an action.

2. The method for monitoring the joint health status of an industrial robot according to claim 1, characterized in that, The process of acquiring and preprocessing the actual torque data and preset torque data of each joint of the target robot includes: using the absolute difference between the actual torque data and the corresponding preset torque data as the torque deviation at the corresponding sampling time; setting the tolerance of torque deviation for each joint according to the processing accuracy requirements of the actual application scenario; at the current sampling time, when the torque deviation of any joint exceeds the tolerance range, using the current sampling time as the monitoring starting point for all joints; normalizing the actual torque data and preset torque data of each joint; forming a real-time anomaly monitoring sequence for the target robot in the current sampling period from the normalized actual torque data starting from the monitoring starting point; and updating the real-time anomaly monitoring sequence of each joint in real time based on the real-time acquired actual torque data.

3. The method for monitoring the joint health status of an industrial robot according to claim 2, characterized in that, The real-time calculation of outliers for each joint and the construction of an outlier sequence for each joint include: selecting any joint of the target robot as the target joint; based on the real-time anomaly monitoring sequence corresponding to the target joint at the current sampling time, calculating the absolute difference between the normalized actual torque data and the normalized preset torque data at each sampling time within the real-time anomaly monitoring sequence; using the mean of each absolute difference as the torque deviation degree of the target joint at the current sampling time; mapping the number of sampling points within the real-time anomaly monitoring sequence using a hyperbolic tangent function; using the mapped value as the anomaly duration feature of the target joint at the current sampling time; and using the product between the torque deviation degree and the anomaly duration feature as the outlier of the target joint at the current sampling time. Similarly, the outliers of each joint at the current sampling time are obtained; starting from the monitoring starting point, the outliers of each joint are recorded and formed into an outlier sequence for each joint, and the outliers and outlier sequences are updated synchronously in real time based on changes in the sampling time.

4. The method for monitoring the joint health status of an industrial robot according to claim 2, characterized in that, The calculation of the real-time fault impact of each joint includes: selecting any joint of the target robot as the target joint, extracting the outlier sequence of each joint at the current sampling time, calculating the Pearson correlation coefficient between the outlier sequence of the target joint and the outlier sequences of the other joints in turn, and when the Pearson correlation coefficient is greater than 0, using the value of the Pearson correlation coefficient as the synchronization coefficient between the corresponding joint and the target joint; when the Pearson correlation coefficient is less than or equal to 0, the corresponding synchronization coefficient is set to 0; extracting the maximum and minimum values ​​of each synchronization coefficient corresponding to the target joint and calculating the range, using the natural exponential function to perform inverse mapping on the range, and using the mapped value as the real-time fault impact of the target joint at the current sampling time. Similarly, the real-time fault impact of each joint at the current sampling time is obtained, and the real-time fault impact is updated synchronously in real time based on the changes in the sampling time.

5. A method for monitoring the joint health status of an industrial robot according to any one of claims 2 to 4, characterized in that, The process of acquiring historical torque data of the same model robot and calculating the baseline fault impact degree of each joint includes: acquiring historical torque data of the same model robot for multiple historical sampling periods under the same application scenario, including historical actual torque data and historical preset torque data of each joint; acquiring the monitoring starting point of historical torque data in each historical sampling period based on the same processing method, and taking the last sampling point in each historical sampling period as the monitoring endpoint; taking the data sequence of a certain joint from the monitoring starting point to the monitoring endpoint as a historical anomaly monitoring sequence of the corresponding joint, thereby obtaining multiple historical anomaly monitoring sequences of each joint; and obtaining multiple historical anomaly values ​​and historical anomaly value sequences of each joint based on the same calculation process, thereby calculating the impact degree of multiple historical faults of each joint.

6. The method for monitoring the joint health status of an industrial robot according to claim 5, characterized in that, The method of obtaining the real-time fault impact weight of each joint based on the real-time fault impact degree includes: statistically processing the multiple historical fault impact degrees of each joint using a Gaussian distribution, and filtering them based on the three sigma principle to obtain the historical fault impact degree within three standard deviations corresponding to each joint; taking the average of the historical fault impact degree of a selected joint as the baseline fault impact degree corresponding to the corresponding joint; similarly obtaining the baseline fault impact degree corresponding to each joint; taking the cumulative value of the baseline fault impact degree corresponding to each joint as the comprehensive benchmark; and taking the ratio between the real-time fault impact degree of each joint and the comprehensive benchmark as the real-time fault impact weight of the corresponding joint at the current sampling time. The real-time fault impact weight is updated synchronously in real time based on the changes in the sampling time.

7. A method for monitoring the joint health status of an industrial robot according to any one of claims 1 to 4 or claim 6, characterized in that, The step of weighting outliers based on real-time fault impact weights and calculating the joint health evaluation value of the target robot in real time includes: multiplying the outlier value of each joint at the current sampling time with the corresponding real-time fault impact weight as the weighted outlier value of the corresponding joint at the current sampling time; using the average of the weighted outliers of each joint as the comprehensive joint anomaly degree of the target robot at the current sampling time; and subtracting the comprehensive joint anomaly degree from 1 as the joint health evaluation value of the target robot at the current sampling time. Based on the processing accuracy requirements of the actual application scenario, a joint health threshold is set. When the joint health evaluation value is less than the joint health threshold, it is determined that the target robot at the current sampling time has a risk of joint structural deterioration and an early warning operation is executed.

8. A joint health status monitoring system for industrial robots, characterized in that: The device includes a processor, a memory, a communication interface, a data acquisition device, and a robot control terminal. The processor stores computer program instructions for implementing the joint health status monitoring method for an industrial robot as described in any one of claims 1 to 7. The communication interface is communicatively connected to the robot control terminal, and the robot control terminal is communicatively connected to the data acquisition device.

Images