Robot anomaly diagnosis method and system based on physical adjacent edge residual difference baseline

Abstract

The application discloses a robot anomaly diagnosis method and system based on physical adjacent edge residual difference baseline, obtains measurement and corresponding reference joint data, acquires residual sequence according to constructed joint topology graph, the joint topology graph contains graph nodes and graph edges, and calculates the current edge difference value based on the node residual statistical feature vector extracted from the residual sequence, calculates the edge residual difference feature of the graph edge based on the current edge difference value and the healthy edge difference value, generates the adjacency matrix according to the joint topology graph, calculates the node feature matrix according to the node residual statistical feature vector, processes the edge residual difference feature of the graph edge, obtains the edge feature matrix, processes the adjacency, node and edge feature matrices, generates the stabilized residual joint graph feature, calculates the anomaly score and the joint level contribution value, and generates the industrial robot health monitoring output according to the score and the contribution value. The application generates the stabilized residual joint graph feature, determines the anomaly level by using the anomaly score, and realizes the node and coupling deviation evaluation positioning.

Description

Technical Field

[0001] This invention relates to the fields of industrial robot condition monitoring, control cycle signal processing, equipment anomaly diagnosis and intelligent maintenance, and in particular to a robot anomaly diagnosis method and system based on the baseline of physical adjacent edge residual difference. Background Technology

[0002] Industrial robots, with their high repeatability, flexibility, and continuous operation capabilities, are widely used in industrial scenarios such as automobile manufacturing, welding, assembly, material handling, painting, grinding, and loading / unloading. Multi-jointed serial industrial robots typically include a controller, servo driver, encoder, servo motor, reducer, end effector, and related sensing units for current, torque, and temperature. During operation, the robot continuously generates various control signals and equipment status signals, including target joint position, actual joint position, target joint speed, actual joint speed, target torque, measured torque, motor current, joint temperature, control cycle timestamps, and servo alarm status.

[0003] However, under the complex operating conditions of industrial sites, existing methods for diagnosing anomalies in industrial robots face the following core bottlenecks: First, the diagnostic stability is insufficient under small sample anomaly events. Real-world anomaly events in industrial robots are typically limited in number, and different trajectory segments, speed levels, load conditions, and end-effector actions significantly alter the amplitude and timing of control signals for each joint. Directly relying on splicing raw multi-joint signals, single-joint thresholds, single residual amplitudes, or discrete classification labels for diagnosis can easily misjudge changes in operating conditions as anomalies, or fail to generate stable early warning outputs during the propagation phase of minor anomalies.

[0004] Second, the physical adjacency relationships in mechanical serial connections are not fully utilized. In serial industrial robots, there are motion correlations, torque couplings, and error propagation relationships between adjacent joints. Existing methods, even when using graph models, often treat the graph structure as a general data correlation graph or feature connection relationship, making it difficult to characterize the residual coupling relationships under healthy operating conditions for each physically adjacent edge in a mechanical serial structure. Therefore, when the residual relationships between adjacent joints deviate from the healthy coupling state, existing methods do not easily provide physically meaningful edge-level anomaly representations.

[0005] Third, there is a disconnect between maintenance and troubleshooting outputs and on-site needs. Industrial sites do not always require a direct conclusion that "a certain component is damaged." They also require consistent outputs of event-level anomaly levels, the order of joints to be investigated, suspected anomaly signal channels, and maintenance prompts. Simply outputting the anomaly category or the joint with the largest residual error easily overlooks the coupling and propagation relationships between adjacent joints, making it difficult for maintenance personnel to quickly locate the troubleshooting path based on diagnostic results.

[0006] Existing technologies attempt to address these issues through health template residuals, statistical thresholds, multi-source time series models, machine learning classifiers, or graph neural networks. However, health template residuals typically focus on single-joint deviations and do not establish edge-level baselines for health residual differences between adjacent physical joints; while multi-source time series models or ordinary graph models can fuse more signals, their graph edges often lack clear mechanical and physical meanings; and classification models rely on the number of anomalous labels and lack stability under small sample anomalous events.

[0007] The purpose of this invention is to overcome the instability of existing industrial robot anomaly diagnosis methods in scenarios involving small sample anomalies, changes in operating conditions, and propagation of adjacent joint couplings. It provides a robot anomaly diagnosis method and system based on a baseline of residual differences between physically adjacent edges. This invention does not focus on ordinary residual classification or ordinary graph model structures. Instead, it establishes a healthy residual difference baseline for each physically adjacent graph edge in the mechanically connected structure of the industrial robot. It then uses the standardized deviation of the statistical difference in the residuals of adjacent joints within the current event window relative to this healthy edge baseline to characterize the abnormal deviation of the residual relationship between adjacent joints relative to the healthy coupling state. This results in the output of event-level anomaly level, the order of joints to be investigated, suspected anomaly signal channels, or maintenance prompts. Summary of the Invention

[0008] To address the problems in related technologies, this application provides a robot anomaly diagnosis method and system based on the baseline of physical adjacent edge residual difference, which solves the problems mentioned in the background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline, comprising the following steps: Step S1: Acquire joint control signals of multiple serial joints within the event window to be diagnosed according to the industrial robot control cycle; the joint control signals include measurement feedback data and reference joint data corresponding to the measurement feedback data; Step S2: Construct a joint topology graph based on the mechanical serial structure of the industrial robot. The joint topology graph includes graph nodes and graph edges. Step S3: Based on the measurement feedback data of the joints corresponding to the graph nodes and the reference joint data, construct the residual sequence of the physical joints within the event window to be diagnosed; Step S4: Extract the node residual statistical feature vector based on the residual sequence; Step S5: Read the node residual statistical feature vectors of adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and estimate the mean healthy edge difference and scale parameter of the graph edge based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed; standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge; Step S6: Generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stable node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the graph edge order, and generate stable residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. Step S7: Calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stabilized residual joint graph features. Compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level. Assign the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship. Calculate the joint-level contribution value together with the stable node features of the corresponding joint. Step S8: Generate industrial robot health monitoring output based on the anomaly level and the joint-level contribution value.

[0010] Furthermore, the reference joint data includes at least one of the following: expected state data, torque reference data, or normal template data estimated from a normal operation window outside the event window to be diagnosed.

[0011] Furthermore, the specific process of step S3 is as follows: Measurement feedback data includes , , and ; For the first Each physical joint in the control cycle The The actual measured value of the joint-like control signal; For the first Each physical joint in the control cycle The actual position of the joint; For the first Each physical joint in the control cycle The actual speed of the joint; For the first Each physical joint in the control cycle The measured torque or estimated torque; When the reference joint data is normal template data, based on The normal template data is represented as follows: ; In the formula, For the first On the first physical joint Joint-like control signals in corresponding and Normal template data below; for and The sample set of normal operation control cycles; This refers to the phase or trajectory segment of the work cycle; For speed settings or load conditions; based on The residual sequence of normal template data is represented as follows: ; In the formula, For the first On the first physical joint Joint-like control signals in the control cycle Normal template data residual sequence; At this point, the calculated As a residual sequence within the event window to be diagnosed; When the reference joint data includes desired state data or torque reference data, based on , and calculate , and ; For the first Each physical joint in the control cycle Position tracking residual; For the first Each physical joint in the control cycle Speed ​​tracking residual; For the first Each physical joint in the control cycle The torque reference residual; Calculated , and These are respectively used as the residual sequences and input into step S4.

[0012] Furthermore, the specific process of step S4 is as follows: The scaled residuals of the residual sequence are represented as follows: ; In the formula, Let be the scaled residual of the control signal of the k-th type of joint on the i-th physical joint during the control period u; The mean residual value of the control signal residual of the k-th type of joint on the i-th physical joint is estimated by the normal operation window. The scale parameter is the scale parameter obtained by estimating the residual of the control signal of the k-th type of joint on the i-th physical joint using the normal operation window. To prevent constants with a denominator of zero; right Extract the statistical features of the node residuals; these features include the absolute scaled residual mean, the root mean square of the scaled residuals, the standard deviation of the scaled residuals, the maximum absolute scaled residual, the quantiles, and the residual energy. These features are then arranged in a predetermined order to form a node residual statistical feature vector. ; Let be the node residual statistical feature vector of the i-th physical joint within the event window t to be diagnosed.

[0013] Furthermore, the specific process of step S5 is as follows: The current edge difference value is calculated as follows: ; In the formula, This represents the current edge difference value between the i-th physical joint and the (i+1)-th physical joint. is the node residual statistical feature vector of the (i+1)th physical joint within the event window t to be diagnosed; The healthy edge difference value is calculated as follows: ; In the formula, For normal operation window The Middle The difference in healthy edges of physically adjacent graph edges under φ and m; For normal operation window The statistical feature vector of the node residuals of the i-th physical joint; For normal operation window The node residual statistical feature vector of the (i+1)th physical joint; Represents the L2 norm; The mean and scale parameters of the healthy edge differences are expressed as follows: ; ; In the formula, For the first Mean difference of healthy edges for physically adjacent graph edges under φ and m; This is the set of normal operating windows under φ and m; For the first The scale parameters for the healthy edge difference of physically adjacent graph edges under φ and m; The edge residual difference feature is represented as follows: ; In the formula, The first event in the event window to be diagnosed t Characteristics of edge residual differences between physically adjacent graph edges; For the first Edge weights of physically adjacent graph edges; The phase or trajectory segment of the work cycle corresponding to the event window t to be diagnosed; The speed range or load condition corresponding to the event window t to be diagnosed; When e represents the index of the physically adjacent graph edge, the first... Edge residual difference characteristics of physically adjacent graph edges Recorded as ; The edge residual difference feature is the e-th physically adjacent graph edge in the event window t to be diagnosed.

[0014] Furthermore, the stabilized node features described in step S6 Statistical eigenvectors of nodal residuals After robust scaling and amplitude limiting, the result is expressed as: ; In the formula, The stable node features of the i-th physical joint under the event window t to be diagnosed; This is the amplitude limiting function operator; Features corresponding to the i-th physical joint The median estimated from the normal operating window; Features corresponding to the i-th physical joint The absolute deviation of the median; The amplitude limiting threshold; For robust scaling.

[0015] Furthermore, the joint-level contribution value is calculated as follows: ; ; In the formula, The unnormalized joint-level contribution value of the i-th physical joint in the event window t to be diagnosed; Let L1 norm be the stable node feature of the i-th physical joint; Contribute weights to edge features; The set of graph edges connected to the i-th physical joint; The edge contribution allocation coefficient is assigned to the i-th physical joint for the e-th physically adjacent graph edge. The normalized joint-level contribution value of the i-th physical joint in the event window t to be diagnosed; For the first The unnormalized joint-level contribution value of each physical joint under the event window t to be diagnosed.

[0016] A robot anomaly diagnosis system based on physical adjacent edge residual difference baseline is applied to a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline, including: The joint control signal acquisition module is used to acquire joint control signals of multiple serial joints within a diagnostic event window according to the industrial robot control cycle; the joint control signals include measurement feedback data and reference joint data corresponding to the measurement feedback data; The joint topology graph construction module is used to construct a joint topology graph based on the mechanical serial structure of an industrial robot. The joint topology graph includes graph nodes and graph edges. The residual sequence construction module is used to construct the residual sequence of the physical joint within the event window to be diagnosed based on the measurement feedback data of the joint corresponding to the graph node and the reference joint data. The node residual statistical feature vector extraction module is used to extract node residual statistical feature vectors based on the residual sequence. The edge residual difference feature construction module is used to read the node residual statistical feature vectors of the adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed, to estimate the mean healthy edge difference and scale parameter of the graph edge; and to standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge. The stabilized residual joint graph feature generation module is used to generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stabilized node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the order of graph edges, and generate stabilized residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. The anomaly level and contribution value calculation module is used to calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stable residual joint graph features, compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level, and allocate the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship, and calculate the joint-level contribution value together with the stable node features of the corresponding joint. The health monitoring output generation module is used to generate industrial robot health monitoring output based on the abnormality level and the joint-level contribution value.

[0017] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline.

[0018] A non-volatile computer storage medium storing computer-executable instructions that execute a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline.

[0019] Compared with existing technologies, the present invention has the following advantages: This invention constructs a joint topology graph, calculates the current edge difference value using the node residual statistical feature vectors of the graph nodes at both ends of the graph edge, estimates the mean and scale parameters of healthy edge differences based on the healthy edge difference values ​​of the same graph edge within the normal operation window, standardizes the current edge difference value to construct edge residual difference features, distinguishes between operating condition changes and true anomalies, and improves the diagnostic stability of small samples.

[0020] This invention generates stable residual joint graph features through adjacency matrix, node feature matrix, and edge feature matrix. It uses the stable node features and edge residual difference features to calculate anomaly scores, compares the anomaly scores with anomaly thresholds to determine the anomaly level, and assigns the edge residual difference features to the joints at both ends of the corresponding graph edge. Together with the stable node features, it calculates the joint-level contribution value, thereby realizing a comprehensive evaluation of the node's own residual and the deviation of adjacent coupling, as well as the location of suspected faults.

[0021] This invention generates stable residual joint graph features, uses anomaly scoring to determine anomaly levels, and calculates joint-level contribution values ​​by combining edge residual difference features with stable node features, thereby achieving node and coupling deviation assessment and positioning. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0023] like Figure 1 As shown, the present invention provides a technical solution: a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline, comprising: Step S1: Acquire joint control signals of multiple serial joints within the event window to be diagnosed according to the industrial robot control cycle; the joint control signals include measurement feedback data and reference joint data corresponding to the measurement feedback data; Step S2: Construct a joint topology graph based on the mechanical serial structure of the industrial robot. The joint topology graph includes graph nodes and graph edges. Step S3: Based on the measurement feedback data of the joints corresponding to the graph nodes and the reference joint data, construct the residual sequence of the physical joints within the event window to be diagnosed; Step S4: Extract the node residual statistical feature vector based on the residual sequence; Step S5: Read the node residual statistical feature vectors of adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and estimate the mean healthy edge difference and scale parameter of the graph edge based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed; standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge; Step S6: Generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stable node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the graph edge order, and generate stable residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. Step S7: Calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stabilized residual joint graph features. Compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level. Assign the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship. Calculate the joint-level contribution value together with the stable node features of the corresponding joint. Step S8: Generate industrial robot health monitoring output based on the anomaly level and the joint-level contribution value.

[0024] The reference joint data includes at least one of the following: expected state data, torque reference data, or normal template data estimated from a normal operation window outside the event window to be diagnosed.

[0025] In step S2, the joint topology diagram is constructed as follows: (1); In the formula, This is a joint topology diagram; A set of graph nodes; It is the set of graph edges; This refers to the graph node corresponding to the nth physical joint; Let be the physically adjacent graph edges between the i-th physical joint and the (i+1)-th physical joint in the mechanical serial structure; This represents the total number of physically adjacent graph edges.

[0026] Among them, measurement feedback data includes , , and ; This represents the actual measured value of the control signal for the k-th type of joint of the i-th physical joint during control cycle u. Let be the actual position of the i-th physical joint during control cycle u; Let be the actual velocity of the i-th physical joint during control cycle u; The measured or estimated torque of the i-th physical joint during control cycle u; In step S3, when the reference joint data is normal template data, based on The normal template data is represented as follows: (2); In the formula, This is the normal template data for the k-th type joint control signal on the i-th physical joint under the corresponding ϕ and m; ϕ represents the sample set of normal operation control cycles under ϕ and m; ϕ represents the phase or trajectory segment of the operation cycle; m represents the speed gear or load condition. based on The residual sequence of normal template data is represented as follows: (3); In the formula, The normal template data residual sequence of the control signal of the k-th type of joint on the i-th physical joint during the control period u; At this point, the result calculated by formula (3) The residual sequence is input to step S4.

[0027] In step S3, when the reference joint data includes desired state data or torque reference data, based on , and calculate , and Represented as: (4); (5); (6); In the formula, For the position tracking residual of the i-th physical joint during control cycle u; Let i be the target position of the i-th physical joint during control cycle u; The velocity tracking residual of the i-th physical joint during control cycle u; Let be the target velocity of the i-th physical joint during control cycle u; The torque reference residual of the i-th physical joint during control cycle u; This is the torque reference value for the i-th physical joint during control cycle u.

[0028] Calculate using formulas 4 to 6 , and These are respectively used as the residual sequences and input into step S4.

[0029] In step S4, to reduce the dimensional differences between different physical joints, different joint control signal channels, and different operating conditions, the residual sequence can first be scaled and the residual representation can be expressed as follows: (7); In the formula, Let be the scaled residual of the control signal of the k-th type of joint on the i-th physical joint during the control period u; The mean residual value of the control signal residual of the k-th type of joint on the i-th physical joint is estimated by the normal operation window. The scale parameter is the scale parameter obtained by estimating the residual of the control signal of the k-th type of joint on the i-th physical joint using the normal operation window. To prevent constants with a denominator of zero; Among them, for Extract the statistical features of the node residuals; the statistical features of the node residuals include the absolute scaled residual mean, the root mean square of the scaled residuals, the standard deviation of the scaled residuals, the maximum absolute scaled residuals, the quantiles, and the residual energy, and form the node residual statistical feature vector by combining the absolute scaled residual mean, the root mean square of the scaled residuals, the standard deviation of the scaled residuals, the maximum absolute scaled residuals, the quantiles, and the residual energy in a predetermined order.

[0030] The residual energy of the i-th physical joint within the event window t to be diagnosed is represented as: (8); In the formula, The residual energy of the i-th physical joint within the event window t to be diagnosed; The total number of control cycles contained in the event window t to be diagnosed; The set of control cycles corresponding to the event window t to be diagnosed; The statistical feature vector of the constituent node residuals is represented as follows: (9); In the formula, Let be the node residual statistical feature vector of the i-th physical joint within the event window t to be diagnosed; The mean of the absolute scaled residuals of the i-th physical joint within the event window t to be diagnosed; The root mean square of the scaled residual of the i-th physical joint within the event window t to be diagnosed; The mean of the scaled residuals of the i-th physical joint within the event window t to be diagnosed; The standard deviation of the scaled residuals of the i-th physical joint within the event window t to be diagnosed; The maximum absolute scaled residual of the i-th physical joint within the event window t to be diagnosed; The absolute scaled residual 95th quantile of the i-th physical joint within the event window t to be diagnosed; The statistical components in the nodal residual statistical eigenvector are represented as follows: (10); (11); (12); (13); (14); (15); In the formula, This is the square root operator; The maximum value operator; This is a function operator for taking the 95th percentile.

[0031] In step S5, the current edge difference value is calculated as follows: (16); In the formula, This represents the current edge difference value between the i-th physical joint and the (i+1)-th physical joint. is the node residual statistical feature vector of the (i+1)th physical joint within the event window t to be diagnosed; The current edge difference value can also be calculated using formula (17) and expressed as follows: (17); In the formula, Let be the weight value of the p-th feature dimension; Let p be the specific value of the p-th feature dimension of the statistical feature vector of the node residuals of the i-th physical joint within the event window t to be diagnosed. This is the specific value of the p-th feature dimension of the statistical feature vector of the node residuals of the (i+1)-th physical joint within the event window t to be diagnosed.

[0032] The healthy edge difference value calculated in step S5 is expressed as follows: (18); In the formula, For normal operation window The Middle The difference in healthy edges of physically adjacent graph edges under φ and m; For normal operation window The statistical feature vector of the node residuals of the i-th physical joint; For normal operation window The node residual statistical feature vector of the (i+1)th physical joint; This represents the L2 norm.

[0033] The mean and scale parameters of the healthy edge differences are expressed as follows: (19); (20); In the formula, For the first Mean difference of healthy edges for physically adjacent graph edges under φ and m; This is the set of normal operating windows under φ and m; For the first The scale parameters for the healthy edge difference of physically adjacent graph edges under φ and m.

[0034] In step S5, the edge residual difference feature is constructed as follows: (twenty one); In the formula, The first event in the event window to be diagnosed t Characteristics of edge residual differences between physically adjacent graph edges; For the first Edge weights of physically adjacent graph edges; The phase or trajectory segment of the work cycle corresponding to the event window t to be diagnosed; The speed range or load condition corresponding to the event window t to be diagnosed; When e represents the index of the physically adjacent graph edge, the first... Edge residual difference characteristics of physically adjacent graph edges Recorded as ; The edge residual difference feature is the e-th physically adjacent graph edge in the event window t to be diagnosed.

[0035] The stabilized residual joint graph features in step S6 are represented as follows: (twenty two); In the formula, The stable residual articulation features in the event window t to be diagnosed; It is an adjacency matrix; The node feature matrix; Let be the edge feature matrix.

[0036] Among them, the stabilization node features mentioned in step S6 Statistical eigenvectors of nodal residuals After robust scaling and amplitude limiting, it is represented as: (twenty three); In the formula, The stable node features of the i-th physical joint under the event window t to be diagnosed; This is the amplitude limiting function operator; Features corresponding to the i-th physical joint The median estimated from the normal operating window; Features corresponding to the i-th physical joint The absolute deviation of the median; The amplitude limiting threshold; For robust scaling.

[0037] The abnormal score mentioned in step S7 is represented as follows: (twenty four); In the formula, Anomaly score for the event window t to be diagnosed; The maximum outlier weight for a single node; The operator for finding the maximum value of the i-th physical joint; The weight of the maximum coupling anomaly term in the graph edge; The operator for finding the maximum value of the e-th physically adjacent graph edge; The weight of the overall global anomaly item; An operator for calculating the average value of the i-th physical joint.

[0038] The abnormal threshold is represented as: (25); In the formula, This is the abnormal threshold; for quantiles; For normal operation window Abnormal scores; A set of windows that are functioning normally; when Greater than When the event window to be diagnosed is determined to be an abnormal event window, and the abnormality score is used as the basis for the determination. Determine the level of abnormality.

[0039] In step S7, the joint-level contribution value is calculated as follows: (26); (27); In the formula, The unnormalized joint-level contribution value of the i-th physical joint in the event window t to be diagnosed; Let L1 norm be the stable node feature of the i-th physical joint; Contribute weights to edge features; The set of graph edges connected to the i-th physical joint; The edge contribution allocation coefficient is assigned to the i-th physical joint for the e-th physically adjacent graph edge. The normalized joint-level contribution value of the i-th physical joint in the event window t to be diagnosed; For the first The unnormalized joint-level contribution value of each physical joint under the event window t to be diagnosed.

[0040] In step S2 of this embodiment, the industrial robot is represented as a joint topology graph G=(V,E). Each graph node corresponds to a physical joint of the industrial robot, and each graph edge corresponds to the motion relationship or torque coupling relationship between adjacent physical joints in the mechanical serial structure. For a six-axis serial robot, the joint topology graph can be represented as J1-J2-J3-J4-J5-J6; J1 is the base rotation joint (waist, rotating around the Z-axis); J2 is the upper arm pitch joint (shoulder); J3 is the forearm pitch joint (elbow); J4 is the wrist rotation joint (forearm rotation); J5 is the wrist swing joint (wrist pitch); and J6 is the end flange rotation joint (wrist rotation).

[0041] A robot anomaly diagnosis system based on physical adjacent edge residual difference baseline is applied to a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline, including: The joint control signal acquisition module is used to acquire joint control signals of multiple serial joints within a diagnostic event window according to the industrial robot control cycle; the joint control signals include measurement feedback data and reference joint data corresponding to the measurement feedback data; The joint topology graph construction module is used to construct a joint topology graph based on the mechanical serial structure of an industrial robot. The joint topology graph includes graph nodes and graph edges. The residual sequence construction module is used to construct the residual sequence of the physical joint within the event window to be diagnosed based on the measurement feedback data of the joint corresponding to the graph node and the reference joint data. The node residual statistical feature vector extraction module is used to extract node residual statistical feature vectors based on the residual sequence. The edge residual difference feature construction module is used to read the node residual statistical feature vectors of the adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed, to estimate the mean healthy edge difference and scale parameter of the graph edge; and to standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge. The stabilized residual joint graph feature generation module is used to generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stabilized node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the order of graph edges, and generate stabilized residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. The anomaly level and contribution value calculation module is used to calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stable residual joint graph features, compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level, and allocate the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship, and calculate the joint-level contribution value together with the stable node features of the corresponding joint. The health monitoring output generation module is used to generate industrial robot health monitoring output based on the abnormality level and the joint-level contribution value.

[0042] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline.

[0043] A non-volatile computer storage medium storing computer-executable instructions that execute a robot anomaly diagnosis method based on physical adjacent edge residual difference baseline.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robot anomaly diagnosis method based on baseline of residual difference between physically adjacent edges, characterized in that, Includes the following steps: Step S1: Acquire joint control signals of multiple serial joints within the event window to be diagnosed according to the industrial robot control cycle; the joint control signals include measurement feedback data and reference joint data corresponding to the measurement feedback data; Step S2: Construct a joint topology graph based on the mechanical serial structure of the industrial robot. The joint topology graph includes graph nodes and graph edges. Step S3: Based on the measurement feedback data of the joints corresponding to the graph nodes and the reference joint data, construct the residual sequence of the physical joints within the event window to be diagnosed; Step S4: Extract the node residual statistical feature vector based on the residual sequence; Step S5: Read the node residual statistical feature vectors of adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and estimate the mean healthy edge difference and scale parameter of the graph edge based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed; standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge; Step S6: Generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stable node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the graph edge order, and generate stable residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. Step S7: Calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stabilized residual joint graph features. Compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level. Assign the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship. Calculate the joint-level contribution value together with the stable node features of the corresponding joint. Step S8: Generate industrial robot health monitoring output based on the anomaly level and the joint-level contribution value.

2. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 1, characterized in that: The reference joint data includes at least one of the following: expected state data, torque reference data, or normal template data estimated from a normal operation window outside the event window to be diagnosed.

3. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 2, characterized in that: The specific process of step S3 is as follows: Measurement feedback data includes , , and ; For the first Each physical joint in the control cycle The The actual measured value of the joint-like control signal; For the first Each physical joint in the control cycle The actual position of the joint; For the first Each physical joint in the control cycle The actual speed of the joint; For the first Each physical joint in the control cycle The measured torque or estimated torque; When the reference joint data is normal template data, based on The normal template data is represented as follows: ； In the formula, For the first On the first physical joint Joint-like control signals in corresponding and Normal template data below; for and The sample set of normal operation control cycles; This refers to the phase or trajectory segment of the work cycle; For speed settings or load conditions; based on The residual sequence of normal template data is represented as follows: ； In the formula, For the first On the first physical joint Joint-like control signals in the control cycle Normal template data residual sequence; At this point, the calculated As a residual sequence within the event window to be diagnosed; When the reference joint data includes desired state data or torque reference data, based on , and calculate , and ; For the first Each physical joint in the control cycle Position tracking residual; For the first Each physical joint in the control cycle Speed ​​tracking residual; For the first Each physical joint in the control cycle The torque reference residual; Calculated , and These are respectively used as the residual sequences and input into step S4.

4. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 3, characterized in that: The specific process of step S4 is as follows: The scaled residuals of the residual sequence are represented as follows: ； In the formula, Let be the scaled residual of the control signal of the k-th type of joint on the i-th physical joint during the control period u; The mean residual value of the control signal residual of the k-th type of joint on the i-th physical joint is estimated by the normal operation window. The scale parameter is the scale parameter obtained by estimating the residual of the control signal of the k-th type of joint on the i-th physical joint using the normal operation window. To prevent constants with a denominator of zero; right Extract the statistical features of the node residuals; these features include the absolute scaled residual mean, the root mean square of the scaled residuals, the standard deviation of the scaled residuals, the maximum absolute scaled residual, the quantiles, and the residual energy. These features are then arranged in a predetermined order to form a node residual statistical feature vector. ; Let be the node residual statistical feature vector of the i-th physical joint within the event window t to be diagnosed.

5. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 4, characterized in that: The specific process of step S5 is as follows: The current edge difference value is calculated as follows: ； In the formula, This represents the current edge difference value between the i-th physical joint and the (i+1)-th physical joint. is the node residual statistical feature vector of the (i+1)th physical joint within the event window t to be diagnosed; The healthy edge difference value is calculated as follows: ； In the formula, For normal operation window The Middle The difference in healthy edges of physically adjacent graph edges under φ and m; For normal operation window The statistical feature vector of the node residuals of the i-th physical joint; For normal operation window The node residual statistical feature vector of the (i+1)th physical joint; Represents the L2 norm; The mean and scale parameters of the healthy edge differences are expressed as follows: ； ； In the formula, For the first Mean difference of healthy edges for physically adjacent graph edges under φ and m; This is the set of normal operating windows under φ and m; For the first The scale parameters for the healthy edge difference of physically adjacent graph edges under φ and m; The edge residual difference feature is represented as follows: ； In the formula, The first event in the event window to be diagnosed t Characteristics of edge residual differences between physically adjacent graph edges; For the first Edge weights of physically adjacent graph edges; The phase or trajectory segment of the work cycle corresponding to the event window t to be diagnosed; The speed range or load condition corresponding to the event window t to be diagnosed; When e represents the index of the physically adjacent graph edge, the first... Edge residual difference characteristics of physically adjacent graph edges Recorded as ; The edge residual difference feature is the e-th physically adjacent graph edge in the event window t to be diagnosed.

6. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 5, characterized in that: The stabilized node features described in step S6 Statistical eigenvectors of nodal residuals After robust scaling and amplitude limiting, the result is expressed as: ； In the formula, The stable node features of the i-th physical joint under the event window t to be diagnosed; This is the amplitude limiting function operator; Features corresponding to the i-th physical joint The median estimated from the normal operating window; Features corresponding to the i-th physical joint The absolute deviation of the median; The amplitude limiting threshold; For robust scaling.

7. The robot anomaly diagnosis method based on physical adjacent edge residual difference baseline according to claim 6, characterized in that: The joint-level contribution value is calculated as follows: ； ； In the formula, The unnormalized joint-level contribution value of the i-th physical joint in the event window t to be diagnosed; Let L1 norm be the stable node feature of the i-th physical joint; Contribute weights to edge features; The set of graph edges connected to the i-th physical joint; The edge contribution allocation coefficient is assigned to the i-th physical joint for the e-th physically adjacent graph edge. The normalized joint-level contribution value of the i-th physical joint under the event window t to be diagnosed; For the first The unnormalized joint-level contribution value of each physical joint under the event window t to be diagnosed.

8. A robot anomaly diagnosis system based on physical adjacent edge residual difference baseline, applied to the robot anomaly diagnosis method based on physical adjacent edge residual difference baseline as described in any one of claims 1-7, characterized in that, include: The joint control signal acquisition module is used to acquire joint control signals of multiple serial joints within the event window to be diagnosed according to the industrial robot control cycle. The joint control signal includes measurement feedback data and reference joint data corresponding to the measurement feedback data; The joint topology graph construction module is used to construct a joint topology graph based on the mechanical serial structure of an industrial robot. The joint topology graph includes graph nodes and graph edges. The residual sequence construction module is used to construct the residual sequence of the physical joint within the event window to be diagnosed based on the measurement feedback data of the joint corresponding to the graph node and the reference joint data. The node residual statistical feature vector extraction module is used to extract node residual statistical feature vectors based on the residual sequence. The edge residual difference feature construction module is used to read the node residual statistical feature vectors of the adjacent graph nodes at both ends of the graph edge to calculate the current edge difference value of the graph edge within the event window to be diagnosed; and based on the healthy edge difference value of the same graph edge in the normal operation window outside the event window to be diagnosed, to estimate the mean healthy edge difference and scale parameter of the graph edge; and to standardize the current edge difference value using the mean healthy edge difference and scale parameter to construct the edge residual difference feature of the graph edge. The stabilized residual joint graph feature generation module is used to generate an adjacency matrix based on the joint topology graph, process the node residual statistical feature vector to obtain stabilized node features and form a node feature matrix, form an edge feature matrix by arranging the edge residual difference features of each graph edge in the order of graph edges, and generate stabilized residual joint graph features from the adjacency matrix, node feature matrix and edge feature matrix. The anomaly level and contribution value calculation module is used to calculate the anomaly score of the event window to be diagnosed based on the stable node features and edge residual difference features in the stable residual joint graph features, compare the anomaly score with the anomaly threshold determined by the normal operation window outside the event window to be diagnosed to determine the anomaly level, and allocate the edge residual difference features to the joints at both ends of the corresponding graph edge according to the graph edge connection relationship, and calculate the joint-level contribution value together with the stable node features of the corresponding joint. The health monitoring output generation module is used to generate industrial robot health monitoring output based on the abnormality level and the joint-level contribution value.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute the robot anomaly diagnosis method based on physical adjacent edge residual difference baseline as described in any one of claims 1-7.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer can execute instructions to perform the robot anomaly diagnosis method based on the residual difference baseline of physical adjacent edges as described in any one of claims 1-7.

Images