A Multi-Axis Hole Robot Assembly Method Based on Anomaly Recognition and Retreat Reconstruction

Abstract

This invention discloses a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction, relating to the fields of robot precision assembly and intelligent manufacturing technology. The method includes: extracting contact feature vectors from the assembly process observation vectors during the robot's insertion process based on pre-alignment results; weighted summing of the values ​​of all elements in the contact feature vectors to obtain a contact anomaly discrimination index; when the contact anomaly discrimination index is greater than or equal to a preset anomaly threshold, determining the robot's retraction direction and distance based on the contact anomaly discrimination index and the contact feature vectors, and executing a retraction action; generating realignment control quantities and reinsertion reference path parameters for reassembly recovery; and controlling the robot to perform reassembly recovery according to the realignment control quantities and reinsertion reference path parameters until assembly is completed or anomaly recovery termination conditions are met. This method improves the overall success rate of the assembly process.

Description

Technical Field

[0001] This application relates to the field of robot precision assembly and intelligent manufacturing technology, and in particular to a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction. Background Technology

[0002] In high-end equipment manufacturing, automated assembly of complex structural components, and flexible intelligent manufacturing production lines, multi-axis hole assembly is a typical high-precision contact assembly task. Compared to single-axis single-hole assembly, micro-clearance multi-axis hole assembly not only places higher demands on the positional and orientation accuracy of the robot's end effector, but also involves significant geometric coupling, force coupling, and insertion synchronization constraints among multiple hole-axis pairs. During assembly, if a local hole-axis pair experiences eccentric contact, unilateral interference, local jamming, or multi-hole asynchrony due to radial deviation, orientation deviation, or insufficient local geometric margin, it often further leads to force imbalance, insertion blockage, and assembly interruption in the remaining hole-axis pairs.

[0003] Existing technologies use auxiliary structures to adjust the pose of the front and rear shells and dynamically correct assembly deviations after detection until the assembly stop is perfectly aligned with the shell surface, enabling assembly correction for specific structural components. However, in complex multi-degree-of-freedom pose adjustment scenarios, there are difficulties in pose adjustment and insufficient adaptability, resulting in a low overall success rate in the assembly process. Summary of the Invention

[0004] Therefore, it is necessary to provide a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction to address the aforementioned technical problems. This method improves the overall success rate of the assembly process.

[0005] The following technical solution is adopted in this specification: This specification provides a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction, including: Based on the initial pose information of the robot end effector, the axis array workpiece, and the hole array workpiece, the overall position error and attitude error of the axis array workpiece relative to the hole array workpiece are solved, and the robot is driven to complete the pre-alignment before multi-axis hole assembly based on the overall position error and attitude error. During the insertion process performed by the robot based on the pre-alignment results, an assembly process observation vector is constructed based on the real-time collected robot contact state data, and a contact feature vector representing the current contact state is calculated based on the assembly process observation vector; the assembly process observation vector includes contact force components, contact torque components, position increment, attitude increment and current insertion depth in each direction; The values ​​of all elements in the contact feature vector are weighted and summed to obtain the contact anomaly detection index. When the contact anomaly detection index is greater than or equal to the preset anomaly threshold, the robot is controlled to perform a retreat action based on the contact anomaly detection index, the assembly process observation vector, and the contact feature vector. Based on the yielding action, the attitude difference between the front and rear axis array workpieces and the hole array workpieces is completed, and the re-alignment control quantity and re-insertion reference path parameters are generated. Based on the realignment control quantity and the reinsertion reference path parameters, the robot is controlled to perform recovery assembly until the assembly is completed or the abnormal recovery termination condition is met.

[0006] Optionally, the assembly process observation vector includes k Time Robot End x Contact force components in the direction, k Time Robot End y Contact force components in the direction, k Time Robot End z Contact force components in the direction, k Time Robot End x Contact torque components in the direction, k Time Robot End y Contact torque components in the direction, k Time Robot End z Contact torque components in the direction, within the current sampling period x Position increment of direction, within the current sampling period y Position increment of direction, within the current sampling period z Position increment of direction, within the current sampling period x The attitude increment in the direction, within the current sampling period y The attitude increment in the direction, within the current sampling period z The orientation increment and current insertion depth; k This refers to any point in the assembly process.

[0007] Optionally, the contact feature vector includes normalized normal contact force characteristics, normalized tangential contact force characteristics, normalized tangential contact torque characteristics, normalized normal contact force change rate characteristics, normalized tangential contact torque change rate characteristics, normalized radial deviation characteristics, normalized attitude deviation characteristics, normalized local geometric margin characteristics, multi-hole axial asynchronous discrete characteristics, multi-hole synchronization index, and normalized insertion velocity decay characteristics; the contact feature vector characterizing the current contact state is calculated based on the assembly process observation vector, specifically including: For any moment in the assembly process k ,according to k Time Robot End x direction,y direction and z The contact force components in the direction determine the normal contact force and tangential contact force vectors; according to k Time Robot End x direction, y direction and z The contact torque components in the directional direction are used to determine the tangential contact torque vector; The first k Normal contact force at time minus the first k After the normal contact force at time -1, divide it by the sampling period to obtain the rate of change of the normal contact force; The first k The magnitude of the tangential contact moment vector minus the first k After taking the magnitude of the tangential contact torque vector at time -1, divide it by the sampling period to obtain the rate of change of the tangential contact torque. The first k Insertion depth at time minus the first k After determining the insertion depth at time -1, divide by the sampling period to obtain the insertion velocity. The magnitudes of the normal contact force, tangential contact force vector, tangential contact moment vector, normal contact force rate of change, tangential contact moment rate of change, radial deviation, attitude deviation, degree of local interference, and insertion velocity are normalized respectively, and the normalization results are integrated to obtain the contact feature vector.

[0008] Optionally, the contact force vector includes x Contact force components in the direction, y Contact force components in the direction and z The contact force component in the direction; based on the contact anomaly discrimination index and the assembly process observation vector, the robot is controlled to perform a yielding action, specifically including: Based on the contact force vector, determine the local contact normal unit vector; The yield direction is obtained by weighting and summing the local contact normal unit vector, the error yield direction unit vector, the axial yield unit vector, and the attitude correction direction unit vector, and then dividing by the magnitude of the weighted sum. The normalized attitude deviation feature, normalized local geometric margin feature, and contact anomaly discrimination index are weighted and summed to obtain the yield distance; the coefficient of the normalized attitude deviation feature is the contact anomaly category coefficient, the coefficient of the normalized local geometric margin feature is the local geometric margin yield amplitude parameter, and the coefficient of the contact anomaly discrimination index is the anomaly degree yield amplitude parameter. Multiply the yield direction and the yield distance to obtain the yield control amount; The robot is controlled to perform a retreat action based on the retreat control variable.

[0009] Optionally, the method further includes: After each yielding action is completed, the yielding amplitude parameter vector is updated based on the contact anomaly discrimination index, posture difference, porous synchronization index, and the updated normal contact force after yielding, and is used to calculate the next yielding distance. The yielding amplitude parameter vector is obtained by integrating the anomaly degree yielding amplitude parameter, the local geometric margin yielding amplitude parameter, and the contact anomaly category coefficient.

[0010] Optionally, based on the attitude differences between the front and rear axis array workpieces and the hole array workpieces after the retraction action, re-alignment control quantities and re-insertion reference path parameters are generated, specifically including: Based on the attitude differences, a local hole-axis matching residual matrix is ​​constructed; Based on the local hole-axis matching residual matrix, the multi-hole synchronization index updated after yielding, and the normal contact force updated after yielding, determine the position error vector and attitude error vector updated after yielding. Based on the position error vector and attitude error vector updated after the yielding, the local pose compensation amount for restoring the assembly is calculated. Based on the local pose compensation, the realignment control quantity and reinsertion reference path parameters are calculated.

[0011] Optionally, the local hole-shaft matching residual matrix is: ; in, For the local hole-shaft matching residual matrix, After making concessions The first axis to be inserted and the first Local matching residuals between target holes N This represents the number of axes to be inserted in the axis array. After yielding The first axis to be inserted and the first The local matching residuals between the target holes are: ; in, For the first The reference center position of the axis to be inserted. For the first The reference center position of each target hole For the first Local attitude parameters of the axis to be inserted. For the first Local attitude parameters of the target hole, For the first The first axis to be inserted and the first Local geometric margin between target holes , , These are the residual weighting coefficients. max To obtain the maximum value, The Euclidean norm represents the magnitude of the difference; The local pose compensation amount is: ; in, For the first k Local pose compensation amount at time. For the first k The position error vector updated after each stepback. For the first k The attitude error vector updated after each step back. For the first k The objective function for local state reconstruction at time t is For the first k Local state reconstruction objective function at time step gradient, This is the position error compensation gain matrix. Attitude error compensation gain matrix, Reconstruct the gradient compensation gain matrix of the objective function; The re-insertion reference path parameters are: ; ; in, To re-insert reference path parameters, , The robot's end-effector pose at the end of the yielding motion. Let the target insertion direction be the unit vector. This is the path advancement function along the target insertion direction. To advance the current assembly and reassembly phase to its target depth, This is the horizontal buffer adjustment coefficient; The repositioning control quantity is: ; in, For the first k The realignment control quantity at any given moment. To reposition the control gain matrix.

[0012] Optionally, the criterion for assembly completion is determined jointly based on position error, attitude error, insertion depth, contact force, and multi-hole synchronization; the function corresponding to the assembly completion criterion is: ; in, This is a sign that assembly is complete. Insert a depth threshold for the target. To accommodate the reconstructed radial position error vector, The allowable threshold for radial position error. To reconstruct the attitude error vector after yielding, The attitude error allowable threshold, To accommodate the normal contact force after reconstruction, The normal contact force stability threshold, This refers to the asynchronous discrete amount of the porous structure along its axial direction. This is the allowable threshold for axial synchronization; The termination condition for anomaly recovery is determined by a combination of the number of anomaly recovery attempts, the cumulative yield, and the severity of the anomaly; the discriminant function corresponding to the termination condition is: ; in, This is a sign indicating the termination of abnormal recovery. This represents the number of yield recovery operations performed in the current assembly task. The maximum number of recoveries allowed. For the first The retreat distance of the second retreat action The cumulative maximum allowable setback distance, To identify abnormal contact indicators, The length of the statistical window for anomaly severity. This is the threshold for abnormal termination.

[0013] Optionally, the weight vector of the elements in the contact feature vector. It is determined using an adaptive update method; The first contact feature vector l The update expression for the weight vector of each element is: ; in, Weight vector The Middle The element in the first... The weight coefficients corresponding to each time point Weight vector The Middle The element in the first... The weighting coefficients corresponding to each discrete sampling time. Update the step size for weights. To prevent positive numbers with a denominator of zero, , Contact feature vector The One portion, Contact feature vector The One portion, The subscript is used for summation.

[0014] Optionally, the method further includes: The normalized radial deviation feature, normalized tangential contact moment feature, and normalized tangential contact force feature are weighted and summed to obtain the eccentric contact anomaly score; The normalized local geometric margin feature, normalized tangential contact force feature, and normalized normal contact force rate of change feature are weighted and summed to obtain the unilateral interference anomaly score. The stuck anomaly score is obtained by weighted summation of the normalized normal contact force characteristics, normalized insertion velocity decay characteristics, and normalized normal contact force change rate characteristics. The normalized attitude deviation feature, normalized tangential contact torque feature, and normalized tangential contact torque change rate feature are weighted and summed to obtain the attitude mismatch anomaly score. The pore synchronization index, the discrete characteristics of pore axial asynchrony, and the normalized radial deviation characteristics are weighted and summed to obtain the pore asynchrony anomaly score. The contact anomaly category corresponding to the maximum value of the anomaly scores among the eccentric contact anomaly score, unilateral interference anomaly score, jamming anomaly score, attitude mismatch anomaly score, and multi-pore asynchronous anomaly score is determined as the current contact anomaly category; the contact anomaly categories include eccentric contact anomaly, unilateral interference anomaly, jamming anomaly, attitude mismatch anomaly, and multi-pore asynchronous anomaly.

[0015] This specification provides a multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction, including: The pre-alignment module is used to solve the overall position error and attitude error of the axis array workpiece relative to the hole array workpiece based on the initial pose information of the robot end effector, axis array workpiece and hole array workpiece, and drive the robot to complete the pre-alignment before multi-axis hole assembly based on the overall position error and attitude error. The extraction module is used to construct an assembly process observation vector based on the real-time collected robot contact state data during the insertion process based on the pre-alignment result of the robot, and to calculate a contact feature vector representing the current contact state based on the assembly process observation vector; the assembly process observation vector includes contact force components, contact torque components, position increment, attitude increment and current insertion depth in each direction; The yielding module is used to perform a weighted summation of the values ​​of all elements in the contact feature vector to obtain a contact anomaly detection index. When the contact anomaly detection index is greater than or equal to a preset anomaly threshold, the robot is controlled to perform a yielding action based on the contact anomaly detection index, the assembly process observation vector, and the contact feature vector. The generation module is used to complete the attitude difference between the front and rear axis array workpieces and the hole array workpieces based on the yielding action, and generate the re-alignment control quantity and re-insertion reference path parameters. The assembly module is used to control the robot to perform recovery assembly based on the realignment control quantity and the reinsertion reference path parameters until the assembly is completed or the abnormal recovery termination condition is met.

[0016] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction.

[0017] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction.

[0018] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In the multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction provided in this specification, real-time collected contact state data is used to construct an assembly process observation vector, and contact feature vectors are extracted from the assembly process observation vector. This achieves multi-source characterization of the contact state during the assembly of micro-gap multi-axis holes, enabling continuous observation of the anomaly formation process. The values ​​of all elements in the contact feature vector are weighted and summed to obtain a contact anomaly discrimination index. When the contact anomaly discrimination index is greater than or equal to a preset anomaly threshold, the robot is controlled to perform a retraction action based on the contact anomaly discrimination index, the assembly process observation vector, and the contact feature vector, making the retraction control targeted and adaptive. After the retraction action is completed, the realignment control quantity and re-insertion reference path parameters are generated based on the posture differences between the front and rear axis array workpieces and the hole array workpieces after the retraction action. The current local assembly state parameters are reconstructed after the anomaly, enabling local recovery assembly after the anomaly. According to the realignment control quantity and the re-insertion reference path parameters, the robot is controlled to perform recovery assembly until the assembly is completed or the anomaly recovery termination condition is reached. This is achieved through the generation of the re-insertion path and the joint setting of the assembly completion criterion and the anomaly recovery termination condition.

[0019] This method constructs an assembly process observation vector through multi-source contact features, enabling continuous monitoring and accurate identification of micro-gap assembly anomalies. A yielding mechanism is triggered based on a weighted contact anomaly identification index, and the yielding direction and distance are adaptively generated using the contact feature vector, improving response specificity. After the yielding action is executed, the realignment control quantity and re-insertion reference path parameters are generated based on the posture differences between the front and rear axis array workpieces and the hole array workpieces, driving pose compensation, realignment, and path replanning. This enables autonomous recovery assembly after anomalies, significantly improving the overall success rate of the assembly process. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This document provides a schematic diagram of a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction. Figure 2 This invention provides a robot contact anomaly identification and retraction reconstruction method for micro-gap multi-axis hole assembly; Figure 3 A schematic diagram of a multi-axis hole assembly object provided by the present invention; Figure 4 This is a schematic diagram of the axis array coordinate system provided by the present invention; Figure 5 This is a schematic diagram of the aperture array coordinate system provided by the present invention; Figure 6 A schematic diagram illustrating the assembly process observation vector and contact feature vector construction process provided by this invention; Figure 7 This is a schematic diagram of contact anomaly identification and classification provided by the present invention; Figure 8 This is a schematic diagram of adaptive yield control provided by the present invention; Figure 9 This is a schematic diagram of local state reconstruction provided by the present invention; Figure 10 A schematic diagram illustrating the re-insertion path generation and remaining hole-axis pair matching sequence optimization provided by the present invention; Figure 11 This is a schematic diagram of the closed-loop execution control for resuming assembly in this invention; Figure 12 This specification provides a schematic diagram of a multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction. Figure 13 This specification provides a schematic diagram of a computer device for implementing a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction.

[0021] Explanation of reference numerals in the attached drawings: 1. Industrial computer; 2. Robot controller; 3. Robot body; 4. End effector; 5. Torque sensor; 6. Vision inspection unit; 7. Axis array workpiece; 8. Hole array workpiece; 9. Assembly platform; 10. Axis; 11. Hole. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0023] Devices such as desktop computers, servers, and laptops are capable of executing the solutions described in this manual. For ease of explanation, the following description will focus on servers as the primary execution method.

[0024] Online identification, anomaly classification, directional retreat, and local state reconstruction after retreat in the assembly process of micro-gap multi-axis holes are of great significance for improving the stability, recovery efficiency, and overall success rate of the robot assembly process.

[0025] Several existing robot assembly control methods have studied problems such as shaft-hole assembly, compliant assembly, and assembly deviation adjustment. Chinese Patent No.: CN110238839A, Title: A Non-Model Robot Multi-Hole Assembly Control Method Optimized by Environmental Prediction. This invention optimizes the parameters of the non-model control algorithm using deep reinforcement learning, which can shorten the operation time of existing assembly schemes and improve the control performance of multi-axis hole assembly to a certain extent. However, this method requires extensive training, and the algorithm implementation process is highly dependent on data and training samples. Furthermore, relying on force sensors to continuously contact the workpiece for assembly adjustment not only increases the cost of the assembly system but also easily causes wear or even damage to the workpiece surface during continuous contact, making it unsuitable for assembly scenarios with soft materials or fragile parts. Chinese Patent No.: CN111319042A, Title: A Robot Compliant Assembly Control Method Based on Forgetting Factor Dynamic Parameters. This invention uses visual positioning to obtain the assembly trajectory and dynamically adjusts the impedance position value through a forgetting factor function to reduce impedance force and suppress large oscillations generated during assembly. This solution can improve the stability of compliant assembly to some extent, but it mainly addresses general compliant assembly control problems. It lacks effective mechanisms to handle issues such as local interference, synchronization mismatch, and jamming in multi-axis hole coupling assembly tasks. Chinese Patent No.: CN104029388A, Title: A Method for Assembling Control of Front and Rear Shells of a Router. This invention uses an auxiliary structure to adjust the pose of the front and rear shells and dynamically corrects them after detecting assembly deviations until the assembly stop is perfectly aligned with the shell surface. This solution can achieve assembly correction for specific structural components, but it relies heavily on specific auxiliary structures and specific assembly conditions, resulting in weak versatility. Furthermore, it faces difficulties in adjusting pose and lacks adaptability in complex multi-degree-of-freedom pose adjustment scenarios.

[0026] In summary, while existing technologies have improved robot assembly from the perspectives of reinforcement learning-based optimized control, compliant impedance adjustment, vision-guided posture adjustment, and assembly deviation correction, they still have the following shortcomings: First, the granularity of contact anomaly identification during micro-gap multi-axis hole assembly is insufficient, making it difficult to effectively distinguish between different abnormal states such as eccentric contact, unilateral interference, local jamming, posture mismatch, and multi-hole asynchrony. Second, after anomalies occur, most existing solutions lack directional retreat strategies and local state reconstruction mechanisms that match the anomaly category, making it difficult to retain existing favorable assembly states. Third, there is a lack of systematic closed-loop processing methods for the remaining hole-axis constraint relationships, local matching residuals, and re-insertion path adjustment under multi-axis and multi-hole coupled assembly conditions. Based on this background, to address the problems of insufficient anomaly identification capability, weak targeting of recovery strategies, and low local recovery efficiency in existing technologies during micro-gap multi-axis hole assembly, this invention proposes a robot contact anomaly identification and retreat reconstruction method for micro-gap multi-axis hole assembly.

[0027] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic diagram of a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction, as described in this specification. The method specifically includes the following steps: S101: Based on the initial pose information of the robot end effector, the axis array workpiece, and the hole array workpiece, solve the overall position error and attitude error of the axis array workpiece relative to the hole array workpiece, and drive the robot to complete the pre-alignment before multi-axis hole assembly according to the overall position error and attitude error.

[0029] Figure 2 This invention provides a robot contact anomaly identification and retraction reconstruction method for micro-gap multi-axis hole assembly, such as... Figure 2 As shown, the following steps are taken: establish the coordinate system of the assembly object and complete the pre-alignment; construct the assembly process observation vector and contact feature vector; calculate the contact anomaly discrimination index and classify the anomaly categories; determine the retreat direction and retreat distance according to the anomaly category; construct the local hole-shaft matching residual matrix and the state reconstruction objective function; and realize closed-loop control based on the assembly completion criterion and the anomaly recovery termination condition.

[0030] This method mainly includes six stages: assembly object modeling and pre-alignment, contact state observation and feature construction, contact anomaly identification and classification, adaptive yield control, local state reconstruction, and assembly recovery and closed-loop execution. The six stages are completed sequentially according to the assembly execution order, and a closed loop is formed in the assembly recovery stage with anomaly identification, yield control, and state reconstruction as the core.

[0031] Figure 3 This is a schematic diagram of a multi-axis hole assembly object provided by the present invention, as shown below. Figure 3 As shown, it includes an industrial control computer 1, a robot controller 2, a robot body 3, an end effector 4, a torque sensor 5, a vision inspection unit 6, an axis array workpiece 7, a hole array workpiece 8, and an assembly platform 9. The industrial control computer 1 includes: assembly process observation, contact anomaly identification, adaptive yield control, local state reconstruction, and resumption of assembly execution.

[0032] Figure 4 This is a schematic diagram of the axis array coordinate system provided by the present invention, as shown below. Figure 4 As shown, axis 10, axis array coordinate system .

[0033] Figure 5 This is a schematic diagram of the aperture array coordinate system provided by the present invention, as shown below. Figure 5 As shown, hole 11, hole array coordinate system .

[0034] First, establish the robot's base coordinate system. End effector coordinate system 1 / 2 axis array coordinate system Hole Array Coordinate System In the formula, the robot's base coordinate system... A global reference position fixedly set on the robot base or assembly unit, serving as a unified reference coordinate system for the entire assembly process; end effector coordinate system. Fixed to the robot's end effector gripper center, used to characterize the real-time position and orientation of the end effector during assembly; axis array coordinate system The geometric reference position fixed to the axis array workpiece is used to describe the overall pose of multiple axes to be inserted; hole array coordinate system. The geometric reference position is fixed to the workpiece of the hole array and is used to describe the overall pose relationship of multiple target holes.

[0035] The robot obtains the initial pose information of the robot end effector, the axis array workpiece, and the hole array workpiece, establishes the robot base coordinate system, the end effector coordinate system, the axis array coordinate system, and the hole array coordinate system, calculates the overall position error and overall attitude error of the axis array workpiece relative to the hole array workpiece, and controls the robot to complete the pre-alignment before multi-axis hole assembly based on the overall position error and overall attitude error.

[0036] The homogeneous pose matrix of the robot end effector in the robot base coordinate system is expressed by formula (1): (1); in, Let be the homogeneous pose matrix of the robot's end effector in the robot's base coordinate system. Let be the rotation matrix of the end effector relative to the robot's base coordinate system. is the position vector of the end effector relative to the robot's base coordinate system.

[0037] The homogeneous pose matrix of the axis array workpiece and the hole array workpiece in the robot base coordinate system is expressed by formula (2): (2); in, Indicates the axis array workpiece in the robot base coordinate system The homogeneous pose matrix below, Indicates the axis array workpiece in the robot base coordinate system The rotation matrix below, Indicates the axis array workpiece in the robot base coordinate system The position vector below; Indicates the workpiece with hole array in the robot base coordinate system The homogeneous pose matrix below, Indicates the workpiece with hole array in the robot base coordinate system The rotation matrix below, Indicates the workpiece with hole array in the robot base coordinate system The position vector below.

[0038] In actual implementation, and The initial pose of the workpiece can be obtained through a vision inspection unit, a calibration unit, or a tooling reference relationship. For example, the vision inspection unit first acquires the initial image information of the axis array workpiece and the hole array workpiece, and then calculates the initial pose of the axis array workpiece and the hole array workpiece relative to the robot's base coordinate system by combining the pre-completed camera extrinsic parameter calibration and workpiece reference calibration. Alternatively, the initial pose of the workpiece can be directly determined through locating pins, reference blocks, or other known mechanical reference relationships in the tooling fixture. This invention does not limit the specific method of obtaining the initial pose, as long as the pose reference information required for subsequent assembly can be obtained.

[0039] After obtaining the above pose relationship, the relative pose matrix of the axis array workpiece with respect to the hole array workpiece is expressed as formula (3): (3); in, Let be the relative pose matrix of the axis array workpiece relative to the hole array workpiece. This indicates the positional deviation of the workpiece in the axis array coordinate system within the hole array coordinate system. This represents the orientation deviation of the workpiece in the axis array coordinate system under the hole array coordinate system. To further quantify the overall mismatch, the overall position error vector is defined as Equation (4): (4); in, This is the overall position error vector. These represent the overall positional deviations of the axis array workpiece relative to the hole array workpiece in the three coordinate axes. Let be the position vector of the axis array workpiece in the robot's base coordinate system. Let be the position vector of the hole array workpiece in the robot's base coordinate system.

[0040] The robot first uses the overall position error vector and the overall attitude error vector Pre-alignment control is completed, moving the entire axis array workpiece to a position close to the assembly window of the hole array workpiece. Pre-alignment refers to minimizing the macroscopic position and orientation of the axis array workpiece and the hole array workpiece before formal contact assembly, thereby reducing the impact and interference risks caused by significant mismatches during the subsequent contact assembly stage. After pre-alignment, the robot's end effector enters the approach assembly stage, preparing to establish hole-axis contact.

[0041] To describe the local fit state during multi-axis hole assembly, assume that there are a total of [number missing] shafts in the shaft array. There are [number] shafts to be inserted, and the hole array contains [number] holes. The target hole, the first The reference center position of each shaft to be inserted is denoted as , No. The reference center position of each target hole is denoted as Then the first The local relative deviation vector of a pair of holes is defined by formula (5): (5); in, For the first The local relative deviation vector of each hole axis pair .

[0042] The local relative deviation vector is decomposed into radial deviation components and axial deviation components, which are denoted by formula (6): (6); in, Indicates the first The degree of local mismatch between the hole and shaft pair in the transverse plane For the first The shaft pairs are in the transverse plane Local deviation components in direction, For the first The shaft pairs are in the transverse plane Local deviation components in direction, For the first Each hole shaft pair Axial deviation component in the direction, Indicates the first The relative deviation of the hole shaft pair in the axial direction.

[0043] For micro-gap assembly scenarios, let the first... The effective radial clearance of each hole shaft pair is Then its local geometric margin is defined by formula (7): (7); in, For the first Local geometric margin of a hole-shaft pair.

[0044] when When, it indicates that the current local radial deviation is still within the assemblable geometric tolerance range of the hole-shaft pair; when When, it indicates that the current local hole-shaft pair is in a critical contact state; when This indicates that local interference has occurred between the hole and shaft pairs. Local geometric margin. It can serve as an important criterion for subsequent contact anomaly identification and backoff control, reflecting which hole-shaft pair first approaches the mismatch boundary or interferes.

[0045] To characterize the synchronous insertion state of multiple hole shaft pairs during the assembly process, a multi-hole synchronization index is also defined. In a preferred embodiment, the porous synchronization index is given by formula (8): (8); in, As an indicator of pore synchronization, For the first The average value of the relative axial deviation of each hole shaft at each discrete sampling time. , For the first The first hole shaft pair in the... The relative deviation along the axial direction at each discrete sampling moment For normalization, To prevent positive integers with a denominator of zero, the closer the synchronization index is to 1, the more consistent the axial assembly process of multiple hole-shaft pairs; the smaller the synchronization index, the more obvious the asynchronous phenomenon among multiple holes. This index will be used in subsequent anomaly identification and local state reconstruction calculations.

[0046] S102: During the insertion process performed by the robot based on the pre-alignment result, an assembly process observation vector is constructed based on the real-time collected robot contact state data, and a contact feature vector representing the current contact state is calculated based on the assembly process observation vector; the assembly process observation vector includes contact force components, contact torque components, position increment, attitude increment and current insertion depth in each direction.

[0047] In one exemplary embodiment, the assembly process observation vector includes k Time Robot End x Contact force components in the direction, k Time Robot End y Contact force components in the direction, k Time Robot End z Contact force components in the direction, k Time Robot End x Contact torque components in the direction, k Time Robot End y Contact torque components in the direction, k Time Robot End z Contact torque components in the direction, within the current sampling period x Position increment of direction, within the current sampling periody Position increment of direction, within the current sampling period z Position increment of direction, within the current sampling period x The attitude increment in the direction, within the current sampling period y The attitude increment in the direction, within the current sampling period z The orientation increment and current insertion depth; k This refers to any point in the assembly process.

[0048] Specifically, Figure 6 This is a schematic diagram illustrating the assembly process observation vector and contact feature vector construction process provided by the present invention, as shown below. Figure 6 As shown, the robot drives the array of workpieces along a preset approach direction to gradually approach the array of holes, and collects state variables such as contact force, contact torque, displacement increment, attitude increment, and insertion depth in real time during each discrete sampling period. Figure 6 The contact force refers to the contact force component, the contact torque refers to the contact torque component, and the displacement increment includes the change in end-effector position and the change in end-effector attitude. An assembly process observation vector is constructed based on state variables such as contact force, contact torque, displacement increment, attitude increment, and insertion depth. A sliding window is used to filter the observation vector. Figure 6 The observation vector in the diagram is the assembly process observation vector; the normal contact force, tangential contact force, and tangential contact torque are calculated; the local interference degree, multi-hole synchronization, and axial asynchronous discrete quantities are calculated; each characteristic quantity is normalized and a characteristic vector is constructed, which is the contact characteristic vector. Each characteristic quantity includes the normal contact force, the magnitude of the tangential contact force vector, the magnitude of the tangential contact torque vector, the rate of change of the normal contact force, the rate of change of the tangential contact torque, the radial deviation, the attitude deviation, the degree of local interference, and the insertion speed.

[0049] The observation vector for the assembly process is given by formula (9): (9); in, Indicates that the robot's end effector is at x Contact force components in the direction, Indicates that the robot's end effector is at y Contact force components in the direction, Indicates that the robot's end effector is at z Contact force components in the direction, Indicates that the robot's end effector is at x Contact torque components in the direction, Indicates that the robot's end effector is at y Contact torque components in the direction, Indicates that the robot's end effector is at z Contact torque components in the direction, Indicates the current sampling periodx The change in the position of the end point in the direction. Indicates the current sampling period y The change in the position of the end point in the direction. Indicates the current sampling period z The change in the position of the end point in the direction. , , This represents the change in end-effector attitude within the current sampling period. This indicates the insertion depth of the end or shaft array relative to the reference assembly plane during the current assembly process.

[0050] By incorporating the above quantities into the observation vector, the coupled changes of "force-displacement-attitude-depth" during the assembly process can be fully described.

[0051] In an exemplary embodiment, the contact feature vector includes normalized normal contact force characteristics, normalized tangential contact force characteristics, normalized tangential contact torque characteristics, normalized normal contact force change rate characteristics, normalized tangential contact torque change rate characteristics, normalized radial deviation characteristics, normalized attitude deviation characteristics, normalized local geometric margin characteristics, porous axial asynchronous discrete characteristics, porous synchronization index, and normalized insertion velocity decay characteristics. The contact feature vector characterizing the current contact state is calculated based on the assembly process observation vector, specifically including: for any moment in the assembly process... k ,according to k Time Robot End x direction, y direction and z Determine the normal and tangential contact force vectors based on the contact force components along the direction; according to k Time Robot End x direction, y direction and z The contact torque components in the directional direction are used to determine the tangential contact torque vector; the first... k Normal contact force at time minus the first k After the normal contact force at time -1, divide it by the sampling period to obtain the rate of change of the normal contact force; then... k The magnitude of the tangential contact moment vector minus the first k After determining the magnitude of the tangential contact torque vector at time -1, divide it by the sampling period to obtain the rate of change of the tangential contact torque; k Insertion depth at time minus the first kAfter determining the insertion depth at time -1, divide by the sampling period to obtain the insertion velocity. Normalize the magnitudes of the normal contact force, tangential contact force vector, tangential contact moment vector, normal contact force rate of change, tangential contact moment rate of change, radial deviation, attitude deviation, degree of local interference, and insertion velocity respectively, and integrate the normalization results to obtain the contact feature vector.

[0052] Specifically, since the raw observation data during the assembly contact stage may be affected by factors such as sampling noise, transient disturbances, and structural vibrations, in this embodiment, such as Figure 4 As shown, the observation vector for the assembly process A sliding window filter is applied to improve the stability of subsequent anomaly detection. The filtered assembly process observation vector is expressed as formula (10): (10); in, This is the filtered assembly process observation vector. The length of the sliding window. This is the observation vector for the assembly process. For the first Assembly process observation vector at each discrete sampling time. This is used for summation indexing. This process effectively reduces occasional fluctuations within a single sampling period, making contact feature extraction more stable.

[0053] In the contact feature extraction process, let the target insertion direction unit vector be Equation (11): (11); in, The target insertion direction is the unit vector.

[0054] The contact force vector is given by formula (12): (12); The contact torque vector is given by formula (13): (13); in, The contact force vector, This is the contact torque vector.

[0055] The normal contact force is then defined by formula (14): (14); in, It is the normal contact force.

[0056] The tangential contact force vector is given by formula (15): (15); in, This is the tangential contact force vector.

[0057] The magnitude of the tangential contact force vector is given by formula (16): (16); in, Let || be the magnitude of the tangential contact force vector, and || represent the magnitude of the orientation quantity.

[0058] The tangential contact torque vector is expressed by formula (17): (17); in, This is the tangential contact torque vector.

[0059] The magnitude of the tangential contact moment vector is expressed by formula (18): (18); in, Let be the magnitude of the tangential contact torque vector.

[0060] Furthermore, to describe the trend of contact state changes, the rate of change of normal contact force is defined as formula (19): (19); in, The normal contact force is the rate of change. For the first k Normal contact force at moment, For the first k- Normal contact force at moment 1 The sampling period.

[0061] The rate of change of tangential contact torque is given by formula (20): (20); in, For the first k The rate of change of tangential contact torque at time t, For the first k The magnitude of the tangential contact moment vector at time t. For the first k- The magnitude of the tangential contact moment vector at time 1. The sampling period.

[0062] The rate of change of normal contact force is used to characterize how fast the degree of contact compression changes, while the rate of change of tangential contact torque is used to characterize the degree of change of contact deflection tendency or local friction tendency.

[0063] Meanwhile, the insertion speed is defined as formula (21): (twenty one); in, For insertion speed, For the first k Insertion depth at time step For the first k- Insertion depth at time 1.

[0064] When the insertion speed decreases significantly or approaches zero, while the normal contact force and tangential contact torque continue to increase, it usually indicates that there is a tendency for local jamming or rigid blockage in the assembly process.

[0065] To unify the representation of various heterogeneous state variables into feature inputs usable for anomaly identification, normalization processing was performed on the normal contact force, tangential contact force, tangential contact torque, rate of change of normal contact force, rate of change of tangential contact torque, radial deviation, attitude deviation, degree of local interference, and insertion velocity attenuation. This was combined with the discrete quantities of axial asynchronous quantities in the multi-hole structure. and synchronization indicators The contact feature vector is constructed as shown in formula (22): (twenty two); in, For contact feature vectors, To normalize the contact force characteristics, To normalize the tangential contact force characteristics, Normalized tangential contact moment characteristics To represent the normalized normal contact force rate of change characteristic, Normalized tangential contact torque rate of change characteristics Normalized radial deviation characteristics Normalized attitude deviation characteristics Normalized local geometric margin features Porous axial asynchronous discrete characteristics Pore ​​synchronization index, Normalized insertion rate decay characteristics.

[0066] S103: The values ​​of all elements in the contact feature vector are weighted and summed to obtain the contact anomaly discrimination index. When the contact anomaly discrimination index is greater than or equal to the preset anomaly threshold, the robot is controlled to perform a retreat action based on the contact anomaly discrimination index, the assembly process observation vector and the contact feature vector.

[0067] In an exemplary embodiment, the contact anomaly categories include eccentric contact anomaly, unilateral interference anomaly, jamming anomaly, attitude mismatch anomaly, and multi-pore asynchronous anomaly. The contact anomalies are classified based on contact feature vectors to obtain the current contact anomaly category. Specifically, this includes: weighted summation of normalized radial deviation features, normalized tangential contact torque features, and normalized tangential contact force features to obtain an eccentric contact anomaly score; weighted summation of normalized local geometric margin features, normalized tangential contact force features, and normalized normal contact force rate of change features to obtain a unilateral interference anomaly score; and weighted summation of normalized normal contact force features and normalized insertion... The speed decay characteristic and the normalized normal contact force change rate characteristic are weighted and summed to obtain the jamming anomaly score; the normalized attitude deviation characteristic, the normalized tangential contact torque characteristic, and the normalized tangential contact torque change rate characteristic are weighted and summed to obtain the attitude mismatch anomaly score; the multi-pore synchronization index, the multi-pore axial asynchronous discrete characteristic, and the normalized radial deviation characteristic are weighted and summed to obtain the multi-pore asynchronous anomaly score; the anomaly category corresponding to the maximum value of the anomaly score among the eccentric contact anomaly score, the unilateral interference anomaly score, the jamming anomaly score, the attitude mismatch anomaly score, and the multi-pore asynchronous anomaly score is determined as the current contact anomaly category.

[0068] Specifically, in obtaining the contact feature vector Then, the contact anomaly identification and classification stage begins. Figure 7 This is a schematic diagram of contact anomaly identification and classification provided by the present invention, such as... Figure 7 As shown, input contact features, which are contact feature vectors; calculate contact anomaly discrimination index; compare the contact anomaly discrimination index with the threshold and complete the anomaly judgment; calculate the anomaly scores for eccentric contact, unilateral interference, local jamming, attitude mismatch, and multi-pore asynchronous anomaly respectively; output the current contact anomaly category.

[0069] The main tasks at this stage are: first, to determine whether the current assembly state has deviated from the normal contact assembly state; and second, after confirming that a contact anomaly has occurred, to further identify the dominant category of the current anomaly in order to provide targeted input for subsequent retreat control. Unlike existing technologies that rely solely on a single force threshold to determine whether a collision has occurred, this invention fuses the force state, attitude state, interference state, insertion state, and multi-hole synchronization state to achieve finer-grained identification of contact anomalies.

[0070] First, based on the contact feature vector The contact anomaly discrimination index is constructed as formula (23): (twenty three); in, To identify abnormal contact indicators, For contact anomaly discrimination weight vector, The first contact feature vector The weight of each component, The first contact feature vector One portion, This is the contact feature vector.

[0071] Contact Anomaly Detection Indicators This index is used to comprehensively measure the deviation of the current assembly state from the normal assembly state. When the normal contact force, tangential contact force, tangential contact torque, degree of local interference, attitude deviation, and degree of asynchronous movement of the porous structure all increase simultaneously, the contact anomaly detection index is used. The corresponding increase.

[0072] Define the contact anomaly detection function as formula (24): (twenty four); in, For contact anomaly detection function, This is the threshold for detecting abnormal contact.

[0073] When contact anomaly detection function When the current assembly contact state is within acceptable limits, the robot can continue performing the current insertion action; when the contact anomaly detection function is activated... When this occurs, it indicates that a contact anomaly has occurred in the current assembly state, and the robot stops rigidly advancing and enters the anomaly classification and retreat control stage.

[0074] After determining that a contact anomaly has occurred, in order to further identify the source of the anomaly, scoring functions were constructed for eccentric contact anomaly, unilateral interference anomaly, local jamming anomaly, attitude mismatch anomaly, and multi-pore asynchronous anomaly.

[0075] The formula for calculating the eccentric contact anomaly score is formula (25): (25); Among them, among them, For the first k Eccentric contact anomaly score at any given moment These are the weighting coefficients for the normalized radial deviation characteristic. These are the weighting coefficients for normalizing the tangential contact moment characteristics. These are the weighting coefficients for normalizing the tangential contact force characteristics. For the first k Normalized radial deviation characteristics at time t. For the first k The characteristics of the normalized tangential contact torque at any given time. For the first k The normalized tangential contact force characteristics at time t, in this scoring function, It mainly reflects the degree of eccentricity mismatch between the hole and shaft in the lateral direction. and This mainly reflects the increasing trend of tangential moment and tangential force caused by eccentric contact. Therefore, when local eccentricity is significant and accompanied by a large tangential force, It will achieve a large value.

[0076] The formula for calculating the unilateral interference anomaly score is formula (26): (26); in, For the first k Unilateral interference anomaly score at time point, For the first k Normalized local geometric margin characteristics at time step For the first k Normalized tangential contact force characteristics at time intervals For the first k Characteristics of the normalized normal contact force change rate at time t. These are the weighting coefficients for the normalized local geometric margin features. These are the weighting coefficients for normalizing the tangential contact force characteristics. This is a weighting coefficient representing the normalized rate of change of normal contact force. When significant interference occurs on one side of the hole-shaft pair and a sudden increase in normal force occurs... The value is relatively large.

[0077] The formula for calculating the sluggishness score is formula (27): (27); in, For the first k Timing-based stuttering anomaly score, For the first k Normalized normal contact force characteristics at time t. For the first k Normalized insertion rate decay characteristics at time points. For the first k Characteristics of the normalized normal contact force change rate at time t. Weighting coefficients for the normalized normal contact force characteristics. Weighting coefficients for normalized insertion rate decay characteristics. Weighting coefficients for the normalized normal contact force change rate characteristic; when the normal contact force remains large, the insertion velocity decreases significantly, and the normal force still shows an increasing trend, the probability of local jamming is considered to be high.

[0078] The formula for calculating the posture mismatch anomaly score is formula (28): (28); in, For the first k At any given moment, the postural mismatch score is given. For the first k Normalized attitude deviation characteristics at any given time. For the first k The characteristics of the normalized tangential contact torque at any given time. For the first k Characteristics of the normalized rate of change of tangential contact torque at any given time. These are the weighting coefficients for the normalized attitude deviation characteristics. These are the weighting coefficients for normalizing the tangential contact moment characteristics. The weighting coefficients represent the normalized tangential contact torque rate of change characteristic. When there is a significant tilt or local angular mismatch between the shaft array workpiece and the hole array workpiece, it usually causes a large attitude error and tangential torque, which in turn leads to... Increase.

[0079] The formula for calculating the multi-pore asynchronous anomaly score is formula (29): (29); in, For the first k Timing-based scoring of multi-hole asynchronous anomalies. For the first k The time-varying synchronicity index of the multi-hole system. For the first k The asynchronous discrete characteristics of the porous axial direction at different times. For the first k Normalized radial deviation characteristics at time t. The weighting coefficients for the pore synchronization index are: The weighting coefficients represent the axial asynchronous discrete characteristics of the porous structure. These are the weighting coefficients for normalizing the radial deviation characteristics. When some hole-shaft pairs have been clearly inserted, while others remain in shallow contact or interference, It is often quite large.

[0080] The current contact anomaly category is determined by a classification function, which is formula (30): (30); in, This indicates the dominant contact anomaly category label at the current moment. For the first j Anomaly scoring function, max This indicates taking the maximum value.

[0081] In other words, at each sampling time, the five anomaly scores are compared, and the anomaly category with the highest score is taken as the dominant basis for the current retreat control. In this way, the robot can select different retreat directions and retreat amplitudes for different anomaly states, instead of uniformly adopting a simple overall retreat strategy.

[0082] The first contact feature vector l The update expression for the weight vector of each element is given by formula (31): (31); in, Weight vector The Middle The element in the first... The weight coefficients corresponding to each time point Weight vector The Middle The element in the first... The weighting coefficients corresponding to each discrete sampling time. Update the step size for weights. To prevent positive numbers with a denominator of zero, , Contact feature vector The One portion, Contact feature vector The One portion, The subscript is used for summation.

[0083] In one exemplary embodiment, the contact force vector includes x Contact force components in the direction, y Contact force components in the direction and z The system determines the contact force component in the direction of contact. Based on contact anomaly detection indices and assembly process observation vectors, the robot is controlled to perform a retreating action. Specifically, this includes: determining the local contact normal unit vector based on the contact force vector; weighting and summing the local contact normal unit vector, error retreating direction unit vector, axial retreating unit vector, and attitude correction direction unit vector, and dividing by the magnitude of the weighted sum to obtain the retreating direction; weighting and summing the normalized attitude deviation feature, normalized local geometric margin feature, and contact anomaly detection indices to obtain the retreating distance; the coefficients of the normalized attitude deviation feature are contact anomaly category coefficients, the coefficients of the normalized local geometric margin feature are local geometric margin retreating amplitude parameters, and the coefficients of the contact anomaly detection indices are anomaly degree retreating amplitude parameters; multiplying the retreating direction and retreating distance yields the retreating control quantity; and controlling the robot to perform a retreating action based on the retreating control quantity.

[0084] After determining the current contact anomaly category, the adaptive yield control phase begins. Figure 8 This is a schematic diagram of the adaptive yield control provided by the present invention, as shown below. Figure 8 As shown, the local contact normal is extracted and the error retreat direction, axial retreat direction, and attitude correction direction are determined; the retreat direction vector is constructed by fusing information from various aspects; the retreat distance is calculated based on the degree of anomaly, the degree of local interference, and the attitude deviation; the retreat control quantity is generated and the robot is controlled to perform the retreat action; the local assembly state is updated after the retreat.

[0085] The core of this stage is to adaptively generate a retreat direction and retreat distance that match the current anomaly category based on the current local contact state, its own geometric mismatch state, and its posture mismatch state. This allows the robot to effectively resolve the current abnormal contact without excessively disrupting the established favorable local assembly relationship.

[0086] First, based on the filtered contact force vector The local contact normal unit vector is calculated using formula (32): (32); in, The local contact normal unit vector, Here is the filtered contact force vector, and || is the modulus length. To prevent positive constants with a denominator of zero, the local contact normal unit vector is used to characterize the dominant force direction of the current abnormal contact. In yield control, it can be used to guide the robot to prioritize releasing the local contact clamping state along the opposite direction of the abnormal contact.

[0087] Secondly, based on the local hole shaft radial deviation, construct the unit vector of the error relief direction. ,in The error backoff direction unit vector is obtained by normalizing the current local radial deviation vector, and is used to characterize the backoff direction for reducing the current lateral mismatch between the hole and shaft; the axial backoff unit vector is constructed according to the direction opposite to the target insertion direction. This is used to reduce contact compression by axial release when necessary; then, based on the filtered attitude error vector... The unit vector for attitude correction direction is determined by formula (33): (33); In the formula, The unit vector for attitude correction direction. This is the filtered attitude error vector. This is the attitude error mapping matrix, used to map the current attitude error vector to the attitude correction direction or attitude correction amount of the end effector. To prevent positive constants with a denominator of zero, the attitude correction direction unit vector is used to compensate for attitude mismatch during the retreat process.

[0088] Local contact normal unit vector Error backoff direction unit vector axial yield unit vector and attitude correction direction unit vector Perform weighted fusion, the first k The vector corresponding to the retreat direction at time t is given by formula (34): (34); in, For the first k The vector corresponding to the retreat direction at time . The current local contact normal unit vector, Let be the unit vector of the error backoff direction determined by the local radial deviation. The axial clearance unit vector is opposite to the target insertion direction. Let be the unit vector of the attitude correction direction determined by the attitude deviation. These are the weighting coefficients for the current local contact normal unit vector. These are the weighting coefficients for the unit vector of the error backoff direction determined by the local radial deviation. The weighting coefficient for the axial clearance unit vector opposite to the target insertion direction. The weighting coefficients for the unit vector of attitude correction direction determined by attitude deviation. , , , These represent the weighting coefficients for the local contact normal, error relief direction, axial relief direction, and attitude correction direction, respectively.

[0089] The aforementioned weighting coefficients can be preset or adjusted online according to the current anomaly category. For example, for unilateral interference anomalies, the weights of the local contact normal and error backoff direction can be appropriately increased; for attitude mismatch anomalies, the weight of the attitude correction direction can be appropriately increased; and for local jamming anomalies, the weight of the axial backoff direction can be appropriately increased.

[0090] Yield distance The degree of anomaly, the degree of local interference, and the degree of attitude deviation are all jointly determined. k The yield distance at any given time is given by formula (35): (35); in, For the first k The distance of retreat at all times, For the first k The anomaly level at any given time, yield amplitude parameter. For the first k The local geometric margin yield magnitude parameter at time t. For the first k Contact anomaly category at any given time The corresponding yield amplitude parameter, To identify abnormal contact indicators, To normalize the local geometric margin features, This represents the normalized attitude deviation characteristics.

[0091] In an exemplary embodiment, the method further includes: after each yielding action is completed, updating the yielding amplitude parameter vector based on the contact anomaly discrimination index, posture difference, porous synchronization index and the normal contact force updated after yielding, for the calculation of the next yielding distance; the yielding amplitude parameter vector is obtained by integrating the anomaly degree yielding amplitude parameter, the local geometric margin yielding amplitude parameter and the contact anomaly category coefficient.

[0092] Specifically, , , To match the current anomaly category The corresponding yield amplitude parameter. Through this expression, the yield distance can be increased with the increase of the severity of contact anomaly, the degree of local interference, and the degree of attitude mismatch, thereby ensuring that more severe anomalies can obtain sufficient release space.

[0093] The update expression for the yield magnitude parameter vector is given by formula (36): (36); in, For the first k The yield magnitude parameter vector at time +1 For the first k The yield magnitude parameter vector at time step, Update the step size for the parameters. To restore the performance cost function, , , , To restore the weighting coefficients of the performance cost function, the yield magnitude parameter vector includes anomaly degree yield magnitude parameters, local geometric margin yield magnitude parameters, and contact anomaly categories.

[0094] The performance recovery cost function is given by formula (37): (37); in, , , , The weighting coefficients for the performance cost function.

[0095] By using the performance cost function, the degree of anomaly, local residuals, porosity synchronization, and contact state in the current restoration assembly process can be comprehensively characterized. Online parameter updates based on this function enable the method of this invention to have better adaptability under different assembly conditions.

[0096] The yield control amount can be obtained from the yield direction vector and the yield distance. The yield control amount is given by formula (38): (38); in, To yield control quantity.

[0097] The robot performs a retreating action based on the retreating control quantity. In a preferred embodiment, the retreating action can be performed in a single retreat or in multiple small steps. If the multiple small steps method is used, the contact anomaly judgment index can be recalculated after each small step retreat to determine whether the current anomaly has been alleviated. In this way, excessive retreat at one time can be avoided from damaging the favorable local assembly state.

[0098] S104: Based on the yielding action, complete the attitude difference between the front and rear axis array workpieces and the hole array workpieces, and generate the re-alignment control quantity and re-insertion reference path parameters.

[0099] After the yielding action is completed, the local state reconstruction phase begins. Figure 9 This is a schematic diagram of local state reconstruction provided by the present invention, such as... Figure 9 As shown, a local hole-axis matching residual matrix is ​​constructed; a local state reconstruction objective function is established; the local pose compensation amount is solved; and the re-alignment control amount and re-insertion reference path parameters are generated.

[0100] A key feature of this invention is that, after an anomaly occurs, the entire assembly is not directly reset and restarted. Instead, the current local assembly state is reconstructed using the new state after the retreat, thereby preserving the advantageous local alignment relationships that have already been formed as much as possible and improving the efficiency of reassembly.

[0101] In an exemplary embodiment, based on the attitude difference between the front and rear axis array workpieces and the hole array workpieces after the yielding action, realignment control quantities and re-insertion reference path parameters are generated. Specifically, this includes: constructing a local hole-axis matching residual matrix based on the attitude difference; determining the position error vector and attitude error vector updated after the yielding based on the local hole-axis matching residual matrix, the multi-hole synchronization index updated after the yielding, and the normal contact force updated after the yielding; calculating the local pose compensation amount for reassembly based on the position error vector and attitude error vector updated after the yielding; and calculating the realignment control quantities and re-insertion reference path parameters based on the local pose compensation amount.

[0102] The residual matrix for local hole-shaft matching is given by formula (39): (39); in, For the hole shaft matching residual matrix, After making concessions The first axis to be inserted and the first Local matching residuals between target holes N This represents the number of axes to be inserted into the axis array.

[0103] After yielding The first axis to be inserted and the first The local matching residual between the target holes is given by formula (40): (40); in, For the first The reference center position of the axis to be inserted. For the first The reference center position of each target hole For the first Local attitude parameters of the axis to be inserted. For the first Local attitude parameters of the target hole, For the first The first axis to be inserted and the first Local geometric margin between target holes , , These are the residual weighting coefficients. max To obtain the maximum value, Let Euclidean norm represent the magnitude of the difference.

[0104] Obtain the local state reconstruction objective function; solve the local state reconstruction objective function based on the current local assembly state parameters to obtain the first... k The updated position error vector after the step-back at time step and the first step-back step-up step k The attitude error vector updated after the step-by-step retreat; based on the first step-by-step... k The updated position error vector after the step-back at time step and the first step-back step-up step k The attitude error vector is updated after each step back, and the local pose compensation amount used to restore assembly is calculated. Based on the local pose compensation amount, the re-alignment control amount and re-insertion reference path parameters are calculated.

[0105] The objective function for local state reconstruction is formula (41); (41); in, For the first k The objective function for local state reconstruction at time t is For the first k The position error vector updated after each stepback. For the first k The attitude error vector updated after each step back. For the first k The pore synchronization index is updated after each yield point. For the first k The normal contact force is updated after each retreat. , , , , All of these are weight coefficients of the objective function.

[0106] By solving the objective function of local state reconstruction, the local pose compensation amount, realignment control amount, and reinsertion reference path parameters used for reassembly are obtained.

[0107] The local pose compensation amount is given by formula (42): (42); in, For the first k Local pose compensation amount at time. For the first k The position error vector updated after each stepback. For the first k The attitude error vector updated after each step back. For the first k The objective function for local state reconstruction at time t is For the first k Local state reconstruction objective function at time step gradient, This is the position error compensation gain matrix. Attitude error compensation gain matrix, The gradient compensation gain matrix of the reconstructed objective function is used; the local pose compensation is used to re-correct the end position and attitude after the retreat, so that the robot can re-enter the recovery assembly stage based on the current better local state.

[0108] Figure 10 This is a schematic diagram illustrating the re-insertion path generation and remaining hole-axis pair matching sequence optimization provided by the present invention, as shown below. Figure 10 As shown, the optimal matching sequence of the remaining hole-shaft pairs is solved; a re-insertion reference path is generated based on the local pose compensation; the re-insertion reference path is used to perform re-alignment; and the re-assembly process is advanced.

[0109] During the reassembly phase, the first step is to use the local hole-shaft matching residual matrix. Determine the optimal matching order for the remaining hole-shaft pairs. The optimal matching order is given by formula (43): (43); in, For the remainder The set of all permutations of hole-shaft pairs to be assembled. By solving for the optimal matching order, it is possible to redetermine which hole-shaft pairs are more suitable for priority reassembly after the yielding, thereby avoiding continued assembly along the original fixed order and causing further jamming.

[0110] After obtaining the optimal matching order, the re-insertion reference path parameters are constructed based on the local pose compensation amount and the target insertion direction. The re-insertion reference path parameters are given by formula (44): (44); in, To re-insert reference path parameters, s To re-insert the path parameters of the reference path, , The robot's end-effector pose at the end of the yielding motion. Let the target insertion direction be the unit vector. This is the path advancement function along the target insertion direction. To advance the current assembly and reassembly phase to its target depth, This is the horizontal buffer adjustment coefficient.

[0111] The path advancement function is given by formula (45): (45); in, To advance the current assembly and reassembly phase to its target depth, This is the horizontal buffer adjustment coefficient.

[0112] The first term in the path advancement function The second item reflects the gradual advancement along the target insertion direction. This incorporates lateral buffering compensation along the middle of the path. By setting lateral buffering, the rigid impact at the initial stage of reassembly can be reduced, allowing the robot to first make a flexible, buffered approach before gradually resuming normal insertion.

[0113] The repositioning control quantity is given by formula (46): (46); in, For the first k The realignment control quantity at any given moment. To reposition the control gain matrix, For the first k Local pose compensation amount at time.

[0114] S105: Based on the realignment control quantity and the reinsertion reference path parameters, control the robot to perform recovery assembly until the assembly is completed or the abnormal recovery termination condition is met.

[0115] Figure 11 This is a schematic diagram of the closed-loop execution control for resuming assembly in this invention, as shown below. Figure 11 As shown, during the closed-loop execution of the reassembly process, the robot performs the reassembly action according to the current reinsertion reference path and continuously monitors whether the assembly completion state has been reached. The criterion for assembly completion is determined jointly based on position error, attitude error, insertion depth, contact force, and multi-hole synchronization. The function corresponding to the assembly completion criterion is formula (47): (47); in, This is a sign that assembly is complete. Insert a depth threshold for the target. To accommodate the reconstructed radial position error vector, The allowable threshold for radial position error. To reconstruct the attitude error vector after yielding, The attitude error allowable threshold, To accommodate the normal contact force after reconstruction, The normal contact force stability threshold, This refers to the asynchronous discrete amount of the porous structure along its axial direction. This is the allowable threshold for axial synchronization; when When the current assembly has met the comprehensive requirements of depth, position, attitude, force, and synchronization, the multi-axis hole assembly is considered complete.

[0116] Meanwhile, to prevent the robot from repeatedly performing retreat and recovery assembly anomalies indefinitely under abnormal conditions where recovery is not feasible, the recovery termination condition is determined by combining the number of abnormal recovery attempts, the cumulative retreat amount, and the severity of the anomaly; the discriminant function corresponding to the abnormal recovery termination condition is formula (48): (48); in, This is a sign indicating the termination of abnormal recovery. This represents the number of yield recovery operations performed in the current assembly task. The maximum number of recoveries allowed. For the first The retreat distance of the second retreat action The cumulative maximum allowable setback distance, To identify abnormal contact indicators, The length of the statistical window for anomaly severity. This is the threshold for abnormal termination. When... If the error occurs, it indicates that the current anomaly has exceeded the preset recovery capability boundary, and the robot stops the recovery assembly process and outputs an anomaly termination signal.

[0117] The following closed loop is formed during the assembly recovery phase: when assembly is incomplete and the abnormal recovery termination condition is not met, the robot continues to execute the cycle of "contact state observation—contact anomaly identification—adaptive retreat—local state reconstruction—assembly recovery"; when the assembly completion criterion is met, an assembly completion signal is output and the current assembly task ends; when the abnormal recovery termination condition is met, an abnormal termination signal is output and the current assembly task stops. In this way, the present invention can form a stable abnormal handling and assembly recovery closed loop in complex micro-gap multi-axis hole assembly scenarios.

[0118] This invention forms a complete technical solution suitable for the assembly process of multi-axis holes with micro-gap by modeling the assembly object, observing and constructing contact states, identifying and classifying contact anomalies, adaptive yield control, reconstructing local states, and resuming assembly in a closed loop. Compared with existing technologies, this invention can more accurately identify abnormal contact states during the assembly process and implement targeted yield control and local state reconstruction according to different abnormal states, thereby improving the efficiency of resuming assembly after anomalies, reducing ineffective retreat and repeated adjustments, and improving the stability and success rate of the overall assembly process.

[0119] When applying the multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction provided in this manual, it is not necessary to follow the... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.

[0120] The above describes one or more embodiments of a multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction provided in this specification. Based on the same idea, this specification also provides a corresponding multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction, such as... Figure 12 As shown.

[0121] Figure 12 A schematic diagram of a multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction provided in this specification includes: The pre-alignment module 1201 is used to solve the overall position error and attitude error of the axis array workpiece relative to the hole array workpiece based on the initial pose information of the robot end effector, axis array workpiece and hole array workpiece, and drive the robot to complete the pre-alignment before multi-axis hole assembly based on the overall position error and attitude error.

[0122] The extraction module 1202 is used to construct an assembly process observation vector based on the real-time collected robot contact state data during the insertion process based on the pre-alignment result of the robot, and to calculate a contact feature vector representing the current contact state based on the assembly process observation vector; the assembly process observation vector includes contact force components, contact torque components, position increment, attitude increment and current insertion depth in each direction.

[0123] The retreat module 1203 is used to perform a weighted summation of the values ​​of all elements in the contact feature vector to obtain a contact anomaly discrimination index. When the contact anomaly discrimination index is greater than or equal to a preset anomaly threshold, the robot is controlled to perform a retreat action based on the contact anomaly discrimination index, the assembly process observation vector, and the contact feature vector.

[0124] The generation module 1204 is used to complete the attitude difference between the front and rear axis array workpieces and the hole array workpieces based on the yielding action, and generate the re-alignment control quantity and re-insertion reference path parameters.

[0125] Assembly module 1205 is used to control the robot to perform recovery assembly based on the realignment control quantity and reinsertion reference path parameters until the assembly is completed or the abnormal recovery termination condition is reached.

[0126] Specific limitations regarding the multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction can be found in the limitations of the multi-axis hole robot assembly method based on anomaly recognition and retraction reconstruction described above, and will not be repeated here. Each module in the aforementioned multi-axis hole robot assembly device based on anomaly recognition and retraction reconstruction can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0127] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A multi-axis hole robot assembly method based on anomaly recognition and yielding reconstruction is provided.

[0128] This instruction manual also provides Figure 13 The schematic diagram of the computer device shown is as follows: Figure 13 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A multi-axis hole robot assembly method based on anomaly recognition and yielding reconstruction is provided.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0130] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff, characterized in that, include: Based on the initial pose information of the robot end effector, the axis array workpiece, and the hole array workpiece, the overall position error and attitude error of the axis array workpiece relative to the hole array workpiece are calculated, and the robot is driven to complete the pre-alignment before multi-axis hole assembly according to the overall position error and the attitude error. During the insertion process performed by the robot based on the pre-alignment result, an assembly process observation vector is constructed based on the real-time collected robot contact state data, and a contact feature vector representing the current contact state is calculated based on the assembly process observation vector. The assembly process observation vector includes contact force components, contact torque components, position increment, attitude increment, and current insertion depth in each direction; The values ​​of all elements in the contact feature vector are weighted and summed to obtain a contact anomaly detection index. When the contact anomaly detection index is greater than or equal to a preset anomaly threshold, the robot is controlled to perform a retreat action based on the contact anomaly detection index, the assembly process observation vector, and the contact feature vector. Based on the yielding action, the attitude difference between the front and rear axis array workpieces and the hole array workpieces is completed, and the re-alignment control quantity and re-insertion reference path parameters are generated. Based on the realignment control quantity and the reinsertion reference path parameters, the robot is controlled to perform recovery assembly until the assembly is completed or the abnormal recovery termination condition is met. The process of generating realignment control quantities and re-insertion reference path parameters based on the attitude differences between the front and rear axis array workpieces and the hole array workpieces after the yielding action specifically includes: constructing a local hole-axis matching residual matrix based on the attitude differences; determining the position error vector and attitude error vector updated after the yielding based on the local hole-axis matching residual matrix, the multi-hole synchronization index updated after the yielding, and the normal contact force updated after the yielding; calculating the local pose compensation amount for reassembly based on the position error vector and attitude error vector updated after the yielding; and calculating the realignment control quantities and re-insertion reference path parameters based on the local pose compensation amount. The local hole-shaft matching residual matrix is: ； in, For the local hole-shaft matching residual matrix, After making concessions The first axis to be inserted and the first Local matching residuals between target holes N This represents the number of axes to be inserted in the axis array. After yielding The first axis to be inserted and the first The local matching residuals between the target holes are: ； in, For the first The reference center position of the axis to be inserted. For the first The reference center position of each target hole For the first Local attitude parameters of the axis to be inserted. For the first Local attitude parameters of the target hole, For the first The first axis to be inserted and the first Local geometric margin between target holes , , These are the residual weighting coefficients. max To obtain the maximum value, The Euclidean norm represents the magnitude of the difference; The local pose compensation amount is: ； in, For the first k Local pose compensation amount at time. For the first k The position error vector updated after each stepback. For the first k The attitude error vector updated after each step back. For the first k The objective function for local state reconstruction at time t is For the first k Local state reconstruction objective function at time step gradient, This is the position error compensation gain matrix. This is the attitude error compensation gain matrix. To reconstruct the gradient compensation gain matrix of the objective function; The re-insertion reference path parameters are: ； ； in, To re-insert reference path parameters, , The robot's end-effector pose at the end of the yielding motion. Let the target insertion direction be the unit vector. This is the path advancement function along the target insertion direction. To advance the current assembly and reassembly phase to its target depth, This is the horizontal buffer adjustment coefficient. s To re-insert the path parameters of the reference path, for The sine value; The realignment control amount is: ； in, For the first k The realignment control quantity at any given moment. To reposition the control gain matrix.

2. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 1, characterized in that, The assembly process observation vector includes k Time Robot End x Contact force components in the direction, k Time Robot End y Contact force components in the direction, k Time Robot End z Contact force components in the direction, k Time Robot End x Contact torque components in the direction, k Time Robot End y Contact torque components in the direction, k Time Robot End z Contact torque components in the direction, within the current sampling period x Position increment of direction, within the current sampling period y Position increment of direction, within the current sampling period z Position increment of direction, within the current sampling period x The attitude increment in the direction, within the current sampling period y The attitude increment in the direction, within the current sampling period z The orientation increment and current insertion depth; k This refers to any point in the assembly process.

3. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 2, characterized in that, The contact feature vector includes normalized normal contact force characteristics, normalized tangential contact force characteristics, normalized tangential contact torque characteristics, normalized normal contact force change rate characteristics, normalized tangential contact torque change rate characteristics, normalized radial deviation characteristics, normalized attitude deviation characteristics, normalized local geometric margin characteristics, multi-hole axial asynchronous discrete characteristics, multi-hole synchronization index, and normalized insertion velocity decay characteristics; the calculation of the contact feature vector characterizing the current contact state based on the assembly process observation vector specifically includes: For any moment in the assembly process k ,according to k Time Robot End x direction, y direction and z The contact force components in the direction determine the normal contact force and tangential contact force vectors; according to k Time Robot End x direction, y direction and z The contact torque components in the directional direction are used to determine the tangential contact torque vector; The first k Normal contact force at time minus the first k After the normal contact force at time -1, divide it by the sampling period to obtain the rate of change of the normal contact force; The first k The magnitude of the tangential contact moment vector minus the first k After taking the magnitude of the tangential contact torque vector at time -1, divide it by the sampling period to obtain the rate of change of the tangential contact torque. The first k Insertion depth at time minus the first k After determining the insertion depth at time -1, divide by the sampling period to obtain the insertion velocity. The normal contact force, the magnitude of the tangential contact force vector, the magnitude of the tangential contact moment vector, the rate of change of the normal contact force, the rate of change of the tangential contact moment, the radial deviation, the attitude deviation, the degree of local interference, and the insertion speed are normalized respectively, and the normalization results are integrated to obtain the contact feature vector.

4. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 3, characterized in that, The contact force vector includes x Contact force components in the direction, y Contact force components in the direction and z The contact force component in the direction; Based on the contact anomaly detection index and the assembly process observation vector, the robot is controlled to perform a retreating action, specifically including: Based on the contact force vector, determine the local contact normal unit vector; The yield direction is obtained by weighting and summing the local contact normal unit vector, the error yield direction unit vector, the axial yield unit vector, and the attitude correction direction unit vector, and then dividing by the magnitude of the weighted sum. The normalized attitude deviation feature, normalized local geometric margin feature, and contact anomaly discrimination index are weighted and summed to obtain the yield distance; the coefficient of the normalized attitude deviation feature is the contact anomaly category coefficient, the coefficient of the normalized local geometric margin feature is the local geometric margin yield amplitude parameter, and the coefficient of the contact anomaly discrimination index is the anomaly degree yield amplitude parameter. Multiply the yielding direction and the yielding distance to obtain the yielding control amount; The robot is controlled to perform a retreat action based on the retreat control quantity.

5. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 4, characterized in that, The method further includes: After each yielding action is completed, the yielding amplitude parameter vector is updated based on the contact anomaly discrimination index, posture difference, porous synchronization index, and the normal contact force updated after the yielding, for use in calculating the next yielding distance; the yielding amplitude parameter vector is obtained by integrating the anomaly degree yielding amplitude parameter, the local geometric margin yielding amplitude parameter, and the contact anomaly category coefficient.

6. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 3, characterized in that, The criterion for assembly completion is determined jointly based on position error, attitude error, insertion depth, contact force, and multi-hole synchronization; the function corresponding to the assembly completion criterion is: ； in, This is a sign that assembly is complete. Insert a depth threshold for the target. To accommodate the reconstructed radial position error vector, The allowable threshold for radial position error. To reconstruct the attitude error vector after yielding, The attitude error allowable threshold, To accommodate the normal contact force after reconstruction, The normal contact force stability threshold, This refers to the asynchronous discrete amount of the porous structure along its axial direction. This is the allowable threshold for axial synchronization; The termination condition for abnormal recovery is determined by a combination of the number of abnormal recovery attempts, the cumulative yield amount, and the severity of the abnormality; the discriminant function corresponding to the termination condition for abnormal recovery is: ； in, This is a sign indicating the termination of abnormal recovery. This represents the number of yield recovery operations performed in the current assembly task. The maximum number of recoveries allowed. For the first The retreat distance of the second retreat action The cumulative maximum allowable setback distance, To identify abnormal contact indicators, The length of the statistical window for anomaly severity. This is the threshold for abnormal termination.

7. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 3, characterized in that, The weight vector of the elements in the contact feature vector It is determined using an adaptive update method; The first contact feature vector l The update expression for the weight vector of each element is: ； in, Weight vector The Middle The element in the first... The weight coefficients corresponding to each time point Weight vector The Middle The element in the first... The weighting coefficients corresponding to each discrete sampling time. Update the step size for weights. To prevent positive numbers with a denominator of zero, , Contact feature vector The One portion, Contact feature vector The One portion, The subscript is used for summation.

8. The multi-axis hole robot assembly method based on anomaly recognition and reconstructive backoff as described in claim 3, characterized in that, The method further includes: The normalized radial deviation feature, normalized tangential contact moment feature, and normalized tangential contact force feature are weighted and summed to obtain the eccentric contact anomaly score; The normalized local geometric margin feature, normalized tangential contact force feature, and normalized normal contact force rate of change feature are weighted and summed to obtain the unilateral interference anomaly score. The stuck anomaly score is obtained by weighted summation of the normalized normal contact force characteristics, normalized insertion velocity decay characteristics, and normalized normal contact force change rate characteristics. The normalized attitude deviation feature, normalized tangential contact torque feature, and normalized tangential contact torque change rate feature are weighted and summed to obtain the attitude mismatch anomaly score. The pore synchronization index, the discrete characteristics of pore axial asynchrony, and the normalized radial deviation characteristics are weighted and summed to obtain the pore asynchrony anomaly score. The contact anomaly category corresponding to the maximum value of the anomaly scores among the eccentric contact anomaly score, the unilateral interference anomaly score, the jamming anomaly score, the attitude mismatch anomaly score, and the multi-pore asynchronous anomaly score is determined as the current contact anomaly category; the contact anomaly category includes eccentric contact anomaly, unilateral interference anomaly, jamming anomaly, attitude mismatch anomaly, and multi-pore asynchronous anomaly.

Images