A stamping die assembly control method and system

Abstract

The application provides a stamping die assembly control method and system, and relates to the technical field of stamping die assembly. The method comprises the following steps: obtaining part type information of a part to be assembled; obtaining key feature information of an assembled component and key feature information of the part to be assembled according to the part type information; calculating an actual relative pose of the part to be assembled relative to the assembled component; calculating a relative pose deviation of the part to be assembled relative to the assembled component according to ideal relative pose data and the actual relative pose; generating a compensation instruction of robot alignment movement according to the relative pose deviation; and controlling the robot to perform alignment operation on the part to be assembled according to the compensation instruction. The method aims to solve the alignment problem caused by part diversity, frequent switching and error accumulation in the flexible assembly of stamping dies, and realizes high-precision and flexible accurate alignment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of stamping die assembly, in particular to a stamping die assembly control method and system. BACKGROUND

[0002] An automatic stamping die assembly production line undertakes flexible production tasks of multiple types of dies. The same production line needs to handle the assembly of dies of different types, which include parts of different categories, shapes, sizes, and weights. The robot system needs to identify the type of the part currently needing assembly and call the corresponding assembly strategy and parameters according to the characteristics of the part. In actual production, the parts of different types of dies may be frequently switched for assembly. This frequent switching requires the robot system to quickly adapt to the new part characteristics and assembly requirements and reduce the preparation time caused by switching. The system needs to quickly load and switch the sensor data processing model and alignment control algorithm corresponding to the current part, involving sensor parameter reconfiguration, feature recognition threshold adjustment, force control parameter setting, etc. The loading and switching process is time-consuming or the newly loaded model deviates from the actual part, which may cause low efficiency or insufficient accuracy when attempting alignment for the first time, or even alignment failure.

[0003] In addition, the assembly process of a stamping die is usually multi-step and involves the sequential assembly of multiple parts. The assembly accuracy of the previous step directly affects the alignment and fitting of the subsequent parts. Even if the previous step is completed within its own requirements, the accumulated small errors may cause the ideal assembly position of the subsequent parts to deviate. A simple alignment strategy based on a CAD model fails because it assumes that all previous components are installed in the ideal position. Due to error accumulation, the robot system cannot simply rely on the preset CAD model position when assembling subsequent parts, but needs to perceive the actual state of the already assembled components in real time and adjust the target alignment position and pose of the current part to be assembled accordingly. This requires a multi-sensor system to not only perceive the pose of the current part to be assembled, but also to be able to perceive key features on the already assembled components through vision or touch, calculate the relative position and pose relationship between the part to be assembled and the already assembled components, and control the alignment based on this relative relationship.

[0004] The above-mentioned various factors superimpose, and the traditional feedback control is difficult to guarantee the high precision and robustness of complex assembly tasks. For example, when the ideal relative position deviates due to errors in the previous step, relying only on visual recognition or force feedback for alignment may not be able to accurately find the best insertion direction or be inefficient. When different types of dies are frequently switched, manually setting and optimizing parameters for each part and each assembly step is a huge workload and is difficult to cover all potential situations.

[0005] Currently, there is no effective technical solution to the above problems. SUMMARY

[0006] The application aims to provide a stamping die assembly control method and system, aiming to solve the alignment problem caused by the diversity of parts, frequent switching and accumulation of previous errors in the flexible assembly of stamping dies, and realize high-precision and flexible accurate alignment.

[0007] In a first aspect, the application provides a stamping die assembly control method applied to an automatic assembly production line using a robot, comprising the following steps:

[0008] Obtain part type information of the part to be assembled;

[0009] According to the part type information, obtain the characteristic parameters of the assembled component, the characteristic parameters of the part to be assembled and the ideal relative pose data;

[0010] According to the characteristic parameters of the assembled component, obtain the key feature information of the assembled component;

[0011] According to the characteristic parameters of the part to be assembled, obtain the key feature information of the part to be assembled;

[0012] According to the key feature information of the assembled component and the key feature information of the part to be assembled, calculate the actual relative pose of the part to be assembled relative to the assembled component;

[0013] According to the ideal relative pose data and the actual relative pose, calculate the relative pose deviation of the part to be assembled relative to the assembled component;

[0014] According to the relative pose deviation, generate compensation instructions for robot alignment movement;

[0015] According to the compensation instructions, control the robot to perform alignment operation on the part to be assembled.

[0016] The stamping die assembly control method provided by the application can realize real-time sensing of the actual relative position and attitude relationship between the part to be assembled and the assembled component, and perform pose compensation control based on the relative relationship. The whole process is driven by part type identification to load parameters, obtain the key feature information of the assembled component and the part to be assembled, calculate the actual relative pose between them, compare it with the ideal relative pose, calculate the deviation, and convert the deviation into robot movement compensation instructions to guide the robot to complete accurate alignment.

[0017] In a second aspect, the application provides a stamping die assembly control system applied to an automatic assembly production line using a robot, comprising:

[0018] A first acquisition module for acquiring part type information of the part to be assembled;

[0019] The second acquisition module is used for acquiring the characteristic parameters of the corresponding assembled component, the characteristic parameters of the to-be-assembled part and the ideal relative pose data according to the part type information.

[0020] The third acquisition module is used for acquiring the key characteristic information of the assembled component according to the characteristic parameters of the assembled component.

[0021] The fourth acquisition module is used for acquiring the key characteristic information of the to-be-assembled part according to the characteristic parameters of the to-be-assembled part.

[0022] The first calculation module is used for calculating the actual relative pose of the to-be-assembled part relative to the assembled component according to the key characteristic information of the assembled component and the key characteristic information of the to-be-assembled part.

[0023] The second calculation module is used for calculating the relative pose deviation of the to-be-assembled part relative to the assembled component according to the ideal relative pose data and the actual relative pose.

[0024] The generation module is used for generating a compensation instruction of the robot alignment motion according to the relative pose deviation.

[0025] The control module is used for controlling the robot to perform the alignment operation on the to-be-assembled part according to the compensation instruction.

[0026] As can be seen from the above, the stamping die assembly control method provided by the application effectively solves the alignment problem caused by the cumulative error of the previous assembly in the flexible assembly of the stamping die, significantly improves the assembly accuracy and success rate, and enhances the adaptability of the system to the frequent switching of multiple models of molds and improves the flexibility of the production line. The method does not depend on the absolute accuracy of the global coordinate system, reduces the strict requirements on the robot calibration and tool positioning, and improves the robustness of the system in the actual production environment.

[0027] Other features and advantages of the application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application as described in the written description and from the practice of the embodiments of the application. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 A flow chart of the stamping die assembly control method provided by the embodiment of the application.

[0029] Figure 2 A structural schematic diagram of the stamping die assembly control system provided by the embodiment of the application.

[0030] Label explanation:

[0031] 100. First acquisition module; 200. Second acquisition module; 300. Third acquisition module; 400. Fourth acquisition module; 500. First calculation module; 600. Second calculation module; 700. Generation module; 800. Control module. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0033] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0034] Reference Appendix Figure 1 This invention provides a stamping die assembly control method, applied to automated assembly lines using robots, comprising the following steps:

[0035] Obtain the part type information of the parts to be assembled;

[0036] Based on the part type information, obtain the characteristic parameters of the corresponding assembled parts, the characteristic parameters of the parts to be assembled, and the ideal relative pose data.

[0037] Based on the characteristic parameters of the assembled components, obtain the key characteristic information of the assembled components;

[0038] Based on the feature parameters of the parts to be assembled, obtain the key feature information of the parts to be assembled.

[0039] Based on the key feature information of the assembled components and the key feature information of the parts to be assembled, calculate the actual relative pose of the parts to be assembled relative to the assembled components.

[0040] Based on the ideal relative pose data and the actual relative pose, calculate the relative pose deviation of the part to be assembled relative to the assembled parts.

[0041] According to the relative pose deviation, a compensation instruction of the robot alignment movement is generated;

[0042] According to the compensation instruction, the robot is controlled to perform the alignment operation on the to-be-assembled part.

[0043] The part type information refers to data used to uniquely identify the type, model or specification of the to-be-assembled part, which can be implemented by using digital coding, string identifier or database index, such as part number, material code, which is mainly to enable the system to identify the part currently needing to be processed, and accordingly load or call corresponding process parameters, feature data and assembly strategy.

[0044] The feature parameters of the assembled component and the feature parameters of the to-be-assembled part refer to a set of data describing the geometric shape, physical property or spatial position of the assembled component and the to-be-assembled part, which can be implemented by using three-dimensional model data, point cloud data, CAD model data or raw data collected by sensors, such as image data obtained by a vision sensor, force / torque data obtained by a force sensor, which is mainly to provide basic information for identifying and extracting key features of the assembled component.

[0045] The ideal relative pose data refers to the spatial position and attitude relationship data of the to-be-assembled part relative to the assembled component in the ideal assembly state, which can be implemented by using a combination of homogeneous transformation matrix, Euler angle and translation vector or a combination of quaternion and translation vector, such as theoretical relative pose calculated by a CAD model, which is mainly to provide a target reference for measuring the deviation between the actual assembly state and the ideal state.

[0046] The key feature information of the assembled component and the key feature information of the to-be-assembled part refer to geometric or physical feature data extracted from the feature parameters of the assembled component and the to-be-assembled part, which is crucial for determining the spatial position and attitude, which can be implemented by using feature point coordinates, feature line equations, feature surface normal vectors or feature region descriptors, such as hole center coordinates obtained by visual recognition, edge line segment parameters, which is mainly to represent the state of the assembled component and the to-be-assembled part in the actual space.

[0047] The actual relative pose refers to the position and attitude relationship of the to-be-assembled part relative to the assembled component in the current actual space calculated according to the key feature information of the assembled component and the key feature information of the to-be-assembled part, which can be implemented by using a combination of homogeneous transformation matrix, Euler angle and translation vector or a combination of quaternion and translation vector, such as relative transformation calculated by point cloud registration or feature matching algorithm, which is mainly to reflect the real spatial relationship in the assembly site, including the error possibly introduced by previous assembly.

[0048] The relative pose deviation refers to the difference between the ideal relative pose data and the actual relative pose, and is used to quantify the degree of deviation of the to-be-assembled part from the ideal assembly position and attitude. It can be represented by a displacement deviation vector and a rotation deviation vector, such as the Euclidean distance and the angle difference between the ideal pose and the actual pose. The main purpose is to provide the basis for generating robot compensation instructions.

[0049] The compensation instruction of the robot alignment motion refers to the motion instruction generated according to the relative pose deviation, which is used to correct the motion of the robot end effector for the alignment operation of the to-be-assembled part. It can be implemented by robot joint space displacement, robot Cartesian space displacement or force / torque control instruction, such as robot end moving ΔX along X axis and rotating Δθ around Z axis. The main purpose is to guide the robot to accurately adjust the to-be-assembled part to the ideal assembly position and attitude relative to the assembled component.

[0050] The core innovation of the present application is to perceive and extract the key feature information of the assembled component and the to-be-assembled part in real time, calculate the actual relative pose between them, compare it with the preset ideal relative pose, obtain the accurate relative pose deviation, and generate the robot compensation instruction based on the deviation. In this way, the problem that the traditional method only relies on the preset position or simple feedback to cope with the frequent switching of part models and the accumulation of previous assembly errors is overcome, and the effects of improving the automation assembly precision and robustness are achieved.

[0051] Specifically, the overall working principle of the present solution is as follows: First, the system acquires the type information of the current part to be assembled, in order to identify the part and load the corresponding process data. Next, according to the part type information, the system acquires the feature parameters and ideal relative pose data of the assembled components and the parts to be assembled related to the current assembly task. These data are pre-configured and used to guide the subsequent perception and calculation process. Then, the system collects the actual state data of the assembled components and the parts to be assembled using sensors (such as visual sensors, force sensors, etc.), and extracts the key feature information of each from it. These key feature information reflects the position and attitude of the parts in the actual space. Based on the extracted key feature information of the assembled components and the parts to be assembled, the system calculates the actual relative pose of the parts to be assembled relative to the assembled components. This actual relative pose is calculated according to the real perception data on site, and contains the cumulative error caused by the previous assembly. Subsequently, the system compares the calculated actual relative pose with the pre-set ideal relative pose data, and calculates the relative pose deviation between the two. This deviation quantifies the degree of deviation of the parts to be assembled from the ideal assembly position and attitude. Finally, the system generates compensation instructions for correcting the robot motion according to the calculated relative pose deviation, and controls the robot to execute these compensation movements to perform accurate alignment operations on the parts to be assembled. The whole process forms a closed-loop control based on actual perception and deviation compensation, enabling the robot to dynamically adjust the alignment strategy according to the actual situation on site and effectively cope with various uncertainties.

[0052] As a preferred embodiment, the scheme of the present application is implemented as follows: the system obtains the model information of the parts to be assembled through reading the production plan or sensor recognition. According to the model information, the system retrieves the CAD model data of the assembled parts (such as the mold base plate) and the parts to be assembled (such as the guide pillar), the ideal relative pose (such as the theoretical insertion position and direction of the guide pillar relative to the base plate), and the parameter configuration for feature extraction and pose calculation from the database. The system uses the visual sensor installed at the end of the robot or on the production line to capture images of the assembled parts and the parts to be assembled. Through image processing algorithms, the key feature points (such as the center of the positioning hole) on the assembled parts and the key feature points (such as the center of the bottom of the guide pillar) on the parts to be assembled are extracted from the captured images. Based on these extracted key feature point coordinates, the system calculates the actual relative pose of the parts to be assembled relative to the assembled parts using a pose solving algorithm (such as the PnP algorithm or the iterative closest point algorithm), represented as a homogeneous transformation matrix. The system compares the calculated actual relative pose matrix with the ideal relative pose matrix stored in the database, and calculates the displacement deviation vector and the rotation deviation vector between the two. According to the calculated displacement and rotation deviation, the system generates compensation motion instructions for the end effector of the robot in the Cartesian space, such as the translation amount along the X, Y, Z axes and the rotation amount around the Rx, Ry, Rz axes. Finally, the system sends these compensation instructions to the robot controller to control the robot to perform accurate alignment and insertion operations on the parts to be assembled.

[0053] Through the above scheme, the present application can perceive the actual state of the assembly site in real time, calculate the real relative pose between the parts to be assembled and the assembled parts, and perform accurate compensation based on the actual deviation, thereby effectively solving the problems of reduced robot alignment accuracy and efficiency caused by parameter mismatch due to frequent switching of part models and accumulation of previous assembly errors in automatic stamping die assembly, and improving the success rate and flexibility of automatic assembly.

[0054] In some embodiments, the step of calculating the actual relative pose of the parts to be assembled relative to the assembled parts based on the key feature information of the assembled parts and the key feature information of the parts to be assembled comprises:

[0055] A1. According to the part type information, obtain the pose calculation parameters related to the assembled parts and the parts to be assembled;

[0056] A2. Based on the key feature information of the assembled parts and the key feature information of the parts to be assembled, establish a total set of feature correspondence relationships between the key feature information of the assembled parts and the key feature information of the parts to be assembled; the total set of feature correspondence relationships includes a plurality of feature correspondence relationships;

[0057] A3. From the total set of feature correspondence, a preset number of feature correspondences are randomly selected as a subset of feature correspondences;

[0058] A4. According to the subset of feature correspondences and the pose calculation parameters, a candidate relative pose of the to-be-assembled part relative to the assembled part is calculated;

[0059] A5. Based on the candidate relative pose, consistency checking is performed on all feature correspondences in the total set of feature correspondences to determine all feature correspondences that meet the candidate relative pose;

[0060] A6. Steps A3-A5 are repeatedly executed until, after repeated execution for a preset number of times, a candidate relative pose that meets the largest number of feature correspondences is identified and taken as the actual relative pose of the to-be-assembled part relative to the assembled part.

[0061] Pose computation parameters refer to configuration data used to guide the pose computation process, which can be set in the form of algorithm selection, iteration number, convergence threshold, error tolerance range, etc. Key feature information refers to representative geometric or physical attribute data extracted from the assembled component or the to-be-assembled part, which can be obtained in the form of key points in point cloud data, corner points or edge features in images, or contact point coordinates collected by tactile sensors, etc. Feature correspondence total set refers to the preliminary matching set established between the key feature information of the assembled component and the to-be-assembled part, which can be stored in the form of a list or data structure containing multiple feature correspondences. Feature correspondence refers to the possible correspondence between a key feature in the assembled component and a key feature in the to-be-assembled part, which can be represented in the form of a pair of key feature indices or coordinate pairs. Feature correspondence subset refers to a part of feature correspondences randomly selected from the feature correspondence total set, which can be constructed in the form of a list or array containing a preset number of feature correspondences. The preset number refers to the minimum number of feature correspondences required for pose computation, which can be determined according to the requirements of the pose computation algorithm used (e.g., three-point perspective, quaternion method, etc.). Candidate relative pose refers to the preliminary pose estimate of the to-be-assembled part relative to the assembled component calculated based on the feature correspondence subset, which can be represented in the form of a rotation matrix and a translation vector, a quaternion and a translation vector, or Euler angles and a translation vector, etc. Consistency check refers to the process of evaluating whether each feature correspondence in the feature correspondence total set is consistent with the current candidate relative pose, which can be implemented in the form of calculating the re-projection error or transformation error of the feature correspondence under the candidate relative pose, and comparing it with the set threshold. All feature correspondences consistent with the candidate relative pose refer to the set of feature correspondences that are determined to be consistent with the current candidate relative pose after consistency check, which can be stored in the form of a list or set. The preset number of times refers to the total number of times the random selection, candidate pose calculation, and consistency check processes are repeated, which can be set according to the desired robustness and computational efficiency. The candidate relative pose with the most feature correspondences refers to the candidate relative pose with the most feature correspondences consistent with the consistency check among all generated candidate relative poses after repeating the process for the preset number of times. The actual relative pose refers to the final determined precise spatial relationship of the to-be-assembled part relative to the assembled component, which can be represented in the form of a rotation matrix and a translation vector, a quaternion and a translation vector, or Euler angles and a translation vector, etc.

[0062] The scheme introduces a mechanism of random sampling and iterative consistency check to construct a robust actual relative pose calculation process. First, the parameters related to pose calculation are obtained according to the current part type information, which enables the subsequent calculation process to be adjusted according to the characteristics of different parts, improving the relevance of the calculation. Then, based on the key feature information of the assembled components and the to-be-assembled parts, the feature correspondence set between them is established, which is the basis for pose calculation. However, considering the errors that may exist in the actual scene, this set may contain a large number of incorrect correspondences. In order to overcome the influence of incorrect correspondences, the scheme uses an iterative approach. In each iteration, a predetermined number of feature correspondences are randomly selected from the set that may contain incorrect correspondences as a subset. This subset is usually the minimum number of correspondences required for pose calculation. The purpose of random selection is to have a higher probability of obtaining a subset containing only correct correspondences in multiple attempts. Then, based on the current selected feature correspondence subset and the obtained pose calculation parameters, a candidate relative pose is calculated, which is a preliminary estimate based on the current subset. Based on the candidate relative pose, all correspondences in the total set of feature correspondences are checked for consistency to determine whether each correspondence is consistent with the current candidate pose, thereby determining all feature correspondences that support the candidate pose. By checking all correspondences in the total set, the reliability of the current candidate pose can be evaluated, and a set of possible correct correspondences can be identified. Repeat the above random sampling, candidate pose calculation and consistency check process for a predetermined number of times to generate multiple candidate poses and their corresponding support sets. Finally, the candidate pose with the most supported feature correspondences is selected as the actual relative pose of the to-be-assembled part relative to the assembled component. This is because a correct pose will usually be supported by most of the correct correspondences in the total set, while an incorrect pose will have fewer supporters. This iterative random sampling and consistency checking approach can effectively filter out the most reliable pose estimate from a set containing a large number of incorrect correspondences. The scheme combined with the key feature information of the assembled components and the to-be-assembled parts can effectively overcome the problem of abnormal feature space distribution caused by the accumulation of previous assembly errors or perception uncertainty, and even in the case of a large number of incorrect correspondences in the key feature information, an accurate and robust actual relative pose can be obtained, providing a reliable basis for subsequent generation of compensation instructions.

[0063] In one embodiment, the stamping die assembly control method can be performed in the following way when calculating the actual relative pose of a part to be assembled with respect to an assembled component. First, the system reads the pose calculation parameters related to the type of part to be assembled, such as punch or die, from a pre-set database according to the current type of part to be assembled, which can include the upper limit of the number of iterations of the RANSAC algorithm for point cloud registration, the inlier distance threshold, and the type of algorithm for solving the pose (for example, the method based on SVD decomposition). Then, the system obtains the key feature information of the assembled component (for example, the die base) and the part to be assembled (for example, the punch), which can be obtained by three-dimensional scanning or visual recognition in the previous step, expressed as a set of key points in two sets of three-dimensional point cloud data. Based on the two sets of key point sets, the system establishes a total set of feature correspondence between all possible key point pairs, for example, each key point in the assembled component is matched with each key point in the part to be assembled to form a set containing a large number of potential corresponding relationships, which may contain many false matching pairs. Then, the system enters an iterative loop and repeats a pre-set number of times (for example, 1000 times). In each iteration, the system randomly selects a pre-set number of feature correspondence subsets from the total set of feature correspondences, for example, if a three-point-based pose solving algorithm is used, 3 pairs of feature correspondences are randomly selected. Using this randomly selected feature correspondence subset and the previously obtained pose calculation parameters, the system calculates a candidate relative pose, for example, a rigid transformation matrix that transforms the key points of the part to be assembled into the coordinate system of the assembled component. Based on the calculated candidate relative pose, the system performs consistency checking on all feature correspondences in the total set of feature correspondences. For each pair of feature correspondences in the total set, the system applies it to the candidate relative pose for transformation and calculates the distance or error between the transformed points and the corresponding points, and if the error is less than the set inlier distance threshold, it is considered that this pair of feature correspondences is consistent with the current candidate pose. The system counts the number of all feature correspondences consistent with the current candidate pose. After repeating the above iteration process for a pre-set number of times, the system compares all the candidate relative poses generated by iteration, selects the one with the most supported feature correspondences, and determines it as the actual relative pose of the part to be assembled with respect to the assembled component.

[0064] By obtaining pose calculation parameters according to the part type information, the pose calculation process can be more targeted. The feature correspondence relationship total set is established based on the key feature information of the assembled components and the to-be-assembled parts, which provides basic data for pose calculation. A preset number of feature correspondence relationships are randomly selected from the feature correspondence relationship total set as a subset, and a candidate relative pose is calculated according to the subset, which is an effective pose estimation strategy. The consistency of all feature correspondence relationships in the total set is tested based on the candidate relative pose, which can identify the correct correspondence relationship that supports the current pose. The random selection, candidate pose calculation and consistency test process are repeatedly executed, and the candidate relative pose that meets the most feature correspondence relationships is identified as the actual relative pose. This iterative optimization process can effectively eliminate the influence of false correspondence relationships, significantly improve the accuracy and robustness of actual relative pose calculation, and solve the problem of abnormal feature space distribution caused by the accumulation of previous errors, thereby providing a reliable basis for subsequent generation of compensation instructions.

[0065] In some embodiments, the specific steps in step A5 include:

[0066] A51. According to the part type information, obtain the basic consistency threshold, feature correspondence relationship reliability evaluation method and threshold adjustment rule related to consistency test;

[0067] A52. Traverse each feature correspondence relationship in the feature correspondence relationship total set, and execute the following steps A521-A524 for each feature correspondence relationship:

[0068] A521. For the currently traversed feature correspondence relationship, use the obtained feature correspondence relationship reliability evaluation method to evaluate the reliability of the current feature correspondence relationship, and obtain a reliability evaluation result;

[0069] A522. Based on the obtained threshold adjustment rule, determine the consistency test threshold for the current feature correspondence relationship according to the obtained basic consistency threshold and the reliability evaluation result;

[0070] A523. Calculate the consistency measure of the current feature correspondence relationship and the candidate relative pose according to the candidate relative pose;

[0071] A524. Determine whether the consistency measure of the current feature correspondence relationship meets the determined consistency test threshold; if it meets, the current feature correspondence relationship is determined to meet the candidate relative pose;

[0072] A53. After traversing all feature correspondence relationships in the feature correspondence relationship total set, obtain all feature correspondence relationships that meet the candidate relative pose.

[0073] The base consistency threshold can be a pre-set, general threshold for a specific part type as a reference for subsequent threshold adjustment. The feature correspondence reliability evaluation method can be a technique for quantifying the confidence of a single feature correspondence, which can be evaluated based on the similarity between feature descriptors, the consistency of local geometry, or using a machine learning model, for example. The reliability evaluation result can be a numerical value, a score, or a level, reflecting the likelihood that the correspondence is a correct correspondence. The threshold adjustment rule can be a function, a lookup table, or a set of logical conditions, used to calculate the specific inspection threshold for the current feature correspondence based on the base consistency threshold and the reliability evaluation result. The consistency inspection threshold is a judgment standard dynamically determined according to the reliability of each feature correspondence itself. The consistency measure can be the spatial distance between the feature points of the to-be-assembled part after the candidate relative pose transformation and the corresponding feature points of the already-assembled component, or other indicators that can measure the geometric consistency of the correspondence with the candidate pose.

[0074] The scheme is aimed at the consistency check of feature correspondence in the process of calculating the actual relative pose. By customizing the strategy according to the part type and adjusting the check standard according to the reliability of each correspondence itself, this method can cope with the problem of false correspondence caused by surface characteristics (such as reflection, oil stains, scratches, texture), improve the accuracy and robustness of consistency check. Specifically, first, according to the part type information, the basic consistency threshold related to consistency check, the feature correspondence reliability evaluation method and the threshold adjustment rule are obtained, which provides the basic parameters and methods for subsequent differentiated processing of different feature correspondences. These parameters are obtained according to the part type information, which reflects the adaptability to different part characteristics and assembly requirements. Then, each feature correspondence in the total set of feature correspondences is traversed to ensure that each potential correspondence is analyzed and checked in detail. During the traversal process, for the currently traversed feature correspondence, the reliability of the current feature correspondence is evaluated using the obtained feature correspondence reliability evaluation method, and the reliability evaluation result is obtained. This quantifies the quality or confidence of each feature correspondence itself, providing a key basis for subsequent threshold adjustment. Then, based on the obtained threshold adjustment rule, the consistency check threshold for the current feature correspondence is determined according to the obtained basic consistency threshold and the reliability evaluation result. This is the core of the scheme, which no longer uses a fixed threshold, but dynamically determines a consistency check threshold that is most suitable for the current feature correspondence according to the reliability of each feature correspondence itself (reflected through the evaluation result), combined with the pre-set threshold adjustment rule and the basic threshold. For example, for feature correspondences with high reliability, a stricter threshold can be used; for feature correspondences with low reliability, the threshold can be appropriately relaxed. This dynamic adjustment makes the consistency check more flexible and accurate. Subsequently, the consistency measure of the current feature correspondence and the candidate relative pose is calculated according to the candidate relative pose. This is the standard calculation process of consistency check, which is used to measure the matching degree of the current feature correspondence and the candidate pose. Determine whether the consistency measure of the current feature correspondence meets the determined consistency check threshold. The key here is to determine whether it meets the threshold dynamically determined for the current feature correspondence. By comparing with this dynamic threshold, the true "inliers" that meet the current candidate pose can be more accurately identified, avoiding false positives or false negatives caused by fixed thresholds, and improving the accuracy of inlier identification. If it meets, the current feature correspondence is determined to meet the candidate relative pose. Finally, after traversing all feature correspondences in the total set of feature correspondences, all feature correspondences that meet the candidate relative pose are obtained. This set is a more accurate inlier set obtained after dynamic threshold check, providing a more reliable data basis for subsequent evaluation of the pros and cons of the candidate pose.Through the above steps, the scheme overcomes the instability of feature extraction and matching due to changes in the surface characteristics of the parts, resulting in false correspondence. By customizing the strategy according to the part type and adjusting the inspection standard according to the reliability of each correspondence itself, the false correspondence problem caused by surface characteristics can be addressed, improving the accuracy and robustness of consistency inspection.

[0075] As a specific implementation, it can be applied to the alignment process of guide pillars and guide sleeves in stamping die assembly. Assuming that the part to be assembled is a guide pillar and the assembled component is a guide sleeve on a die seat. When calculating the actual relative pose of the guide pillar relative to the die seat, consistency verification of feature correspondence (e.g., correspondence of point features on the guide pillar and point features on the guide sleeve) is needed. First, according to the part type information (guide pillar), the system obtains parameters related to consistency verification. For example, the basic consistency threshold can be set to 0.2mm, the feature correspondence reliability evaluation method can use a combination evaluation based on local surface normal consistency and feature descriptor distance, and the threshold adjustment rule can be set as a piecewise function that scales the basic threshold by different degrees according to the reliability evaluation result. Then, all established feature correspondences are traversed. For a specific correspondence, such as the correspondence of a point on the top of the guide pillar and a point on the inner wall of the guide sleeve, the reliability is evaluated using the obtained evaluation method, resulting in a reliability evaluation result, such as 0.7 (indicating that this correspondence has a 70% chance of being correct). Then, according to the obtained threshold adjustment rule, the basic consistency threshold 0.2mm and the reliability evaluation result 0.7 are combined to calculate the consistency verification threshold for this correspondence. For example, if the rule specifies that a reliability of 0.7 corresponds to a threshold scaling factor of 1.1, then the dynamic threshold is determined to be 0.2mm*1.1=0.22mm. Subsequently, according to the current candidate relative pose, the consistency measure of this feature correspondence is calculated, for example, the point on the guide pillar is transformed to the die seat coordinate system through the candidate pose, and the distance between it and the corresponding point on the guide sleeve is calculated, resulting in 0.18mm. Determine whether the consistency measure 0.18mm meets the determined threshold 0.22mm. Since 0.18mm is less than 0.22mm, the correspondence is determined to be consistent with the candidate relative pose. Repeat this process until all feature correspondences are verified, and finally obtain the set of all feature correspondences that meet the current candidate relative pose.

[0076] By the technical solution, the application can customize the consistency inspection strategy according to different part types, and dynamically adjust the inspection standard according to the reliability of each feature correspondence. Therefore, even in the case that the feature extraction and matching are unstable due to the change of part surface characteristics, a large number of false correspondences are generated, the true inliers can be more accurately identified, the accuracy and robustness of calculating the actual relative pose based on the random sample consensus method are improved, and reliable pose information is provided for subsequent high-precision robot alignment assembly.

[0077] In some embodiments, the specific steps in step A522 include:

[0078] Obtaining the feature type involved in the current feature correspondence;

[0079] According to the feature type, determining the threshold value adjustment sub-rule corresponding to the feature type from the obtained threshold value adjustment rule;

[0080] According to the determined threshold value adjustment sub-rule, the obtained basic consistency threshold value and the reliability evaluation result, determining the consistency inspection threshold value for the current feature correspondence.

[0081] The feature type refers to the category of geometric elements constituting the feature correspondence, such as point, line, surface, circle, cylinder, etc. The threshold value adjustment sub-rule refers to a set of threshold value adjustment logic, parameters or functions defined for a specific feature type, which is a component of the overall threshold value adjustment rule set.

[0082] The present scheme aims to more accurately determine the consistency check threshold of each feature correspondence when performing feature correspondence consistency check. A technical means of fine threshold adjustment based on feature type is proposed. By identifying the feature type associated with the current feature correspondence, and selecting a specific threshold adjustment rule according to the type, the characteristics of different types of features can be more fully considered, thereby improving the accuracy and robustness of consistency check. Specifically, first, the feature type involved in the current feature correspondence is obtained, which identifies the type of feature that the current feature correspondence is based on, such as point, line, surface, hole, etc. Different types of features may have different error distributions in actual measurement, or different sensitivity to translation and rotation when calculating pose. Then, according to the feature type, the threshold adjustment sub-rule corresponding to the feature type is determined from the obtained threshold adjustment rules. This step shows that the system's preset threshold adjustment rules are not single, but contain multiple sub-rules for different feature types. By using the feature type information obtained in the previous step, the system can accurately find the specific adjustment logic or parameters suitable for the current feature type from the rule set. For example, the sub-rule for point features may focus more on position error adjustment, while the sub-rule for line or surface features may focus more on direction or normal error adjustment. This feature type-based sub-rule selection makes the threshold adjustment process more targeted and better reflects the actual characteristics of this type of feature. Finally, according to the determined threshold adjustment sub-rule, the obtained basic consistency threshold, and the reliability evaluation result, the consistency check threshold for the current feature correspondence is determined. This step is the process of finally calculating the threshold. It combines three pieces of information: a general basic consistency threshold, a reliability evaluation result of the quality or confidence of the current feature correspondence itself, and most importantly, a specific adjustment sub-rule for the current feature type. By applying this specific sub-rule, the system can calculate a more reasonable and more realistic consistency check threshold according to the inherent properties of this type of feature, combined with the basic threshold and the reliability evaluation result. For example, for a certain feature type with high measurement accuracy and good reliability evaluation result, its corresponding sub-rule may set the final threshold more strictly; while for another feature type with lower measurement accuracy or general reliability evaluation result, its corresponding sub-rule may set the final threshold relatively loose. This threshold determination method that combines feature type, basic threshold, and reliability evaluation result makes the consistency check more accurate in distinguishing inliers that meet the candidate pose from outliers that do not meet it, thereby improving the accuracy and robustness of actual pose calculation.The fine threshold determination method is applied to the consistency checking process of all feature correspondences in the total set of feature correspondences, which can more accurately identify the inlier set supporting the current candidate pose, so as to more likely identify the candidate pose representing the actual relative pose with the largest number of feature correspondences in the iterative process of repeatedly performing random sampling and consistency checking.

[0083] By determining the corresponding threshold adjustment sub-rule according to the feature type, and determining the consistency checking threshold for the current feature correspondence based on the sub-rule, the basic consistency threshold and the reliability evaluation result, the scheme can more accurately reflect the error characteristics and influence on pose calculation of different types of features, thereby improving the accuracy and robustness of feature correspondence consistency checking, especially in processing complex assembly scenes containing multiple geometric types of features, which helps to more reliably calculate the actual relative pose of the to-be-assembled part relative to the assembled part.

[0084] In some embodiments, the step of calculating the relative pose deviation of the to-be-assembled part relative to the assembled part according to the ideal relative pose data and the actual relative pose comprises:

[0085] According to the part type information, the deviation calculation parameters related to the ideal relative pose data and the actual relative pose are obtained; the deviation calculation parameters include sensitivity weights of different pose components and deviation representation conversion rules;

[0086] According to the ideal relative pose data and the actual relative pose, the initial relative pose deviation of the to-be-assembled part relative to the assembled part is calculated;

[0087] According to the deviation calculation parameters, based on the initial relative pose deviation, the weighted processing is performed on different pose components of the initial relative pose deviation by applying the sensitivity weights of different pose components;

[0088] According to the deviation calculation parameters, based on the weighted processed initial relative pose deviation, the weighted processed initial relative pose deviation is converted into a deviation representation corresponding to the part type by applying the deviation representation conversion rules;

[0089] The converted deviation representation is taken as the relative pose deviation of the to-be-assembled part relative to the assembled part.

[0090] The deviation calculation parameter refers to a set of configuration data used to guide the relative pose deviation calculation process, which can be implemented in a configuration file, a database record, or a data structure in memory. The sensitivity weight of different pose components refers to a numerical factor that quantifies the influence of pose deviation in different degrees of freedom on the final assembly result, which can be implemented in a vector, a matrix, or a lookup table. The deviation representation conversion rule refers to a mapping or algorithm that defines how to convert one form of pose deviation data into another form, which can be implemented in a mathematical formula, a lookup table, or a software function. The initial relative pose deviation refers to the original, unprocessed pose difference obtained by directly comparing the ideal pose and the actual pose, which can be implemented in a homogeneous transformation matrix, a combination of Euler angles and translation vectors, or a combination of quaternions and translation vectors. The weighted processing refers to a mathematical operation that adjusts the importance of each component of the original data according to the pre-set weight factor, which can be implemented in vector multiplication, matrix multiplication, or component multiplication. The converted deviation representation refers to the pose deviation data expressed in a form more suitable for a specific application (such as robot control) after being processed by a specific rule, which can be implemented in translation and rotation components in a specific coordinate system, deviation components based on mating features, or standardized deviation values.

[0091] The present solution aims to provide a more refined and targeted relative pose deviation calculation method to address the issue that simple unit pose differences cannot fully reflect the actual assembly requirements, thereby improving the precision and effectiveness of robot alignment compensation. First, according to the part type information, obtain the deviation calculation parameters related to the ideal relative pose data and the actual relative pose. These parameters include the sensitivity weights of different pose components and the deviation representation conversion rules. This step is the basis for realizing customized deviation calculation, by obtaining specific parameters related to the current type of parts to be assembled, the subsequent deviation processing can be optimized for the characteristics and assembly requirements of different parts. The sensitivity weights reflect the different influence degrees of different pose components (such as translation deviations along X, Y, Z axes, and rotation deviations around X, Y, Z axes) on the final assembly effect or success rate, while the deviation representation conversion rules define how to convert the original pose deviation into a form more suitable for guiding the robot alignment. Then, according to the ideal relative pose data and the actual relative pose, calculate the initial relative pose deviation of the part to be assembled relative to the assembled component. This is the starting point of deviation calculation, by comparing the ideal state and the actual perceived state, an original, unprocessed pose difference is obtained. Then, based on the initial relative pose deviation, apply the sensitivity weights of different pose components to the different pose components of the initial relative pose deviation through the weighted processing of the deviation calculation parameters obtained. This step uses sensitivity weights to assign different importance to each component of the initial deviation, highlighting the key deviation components that have a greater impact on assembly, and weakening the components with less impact. This solves the problem that simple unit pose differences cannot distinguish the importance of different components, so that the subsequent compensation instructions can focus more on correcting those deviations that are crucial to assembly precision and success rate. Further, based on the weighted initial relative pose deviation, convert the weighted initial relative pose deviation to the deviation representation corresponding to the part type by applying the deviation representation conversion rules according to the deviation calculation parameters obtained. This step uses the deviation representation conversion rules to convert the weighted deviation into a form more suitable for the current part type and assembly task. For example, for pin-hole fitting, the deviation may need to be converted into deviation along the pin axis direction and perpendicular to the pin axis direction; for plane fitting, it may need to be converted into the gap between planes and angular deviation. This conversion makes the deviation information more intuitive and easier for the robot control system to understand and use to generate accurate alignment adjustment instructions, improving the effectiveness of compensation. Finally, the converted deviation representation is taken as the relative pose deviation of the part to be assembled relative to the assembled component. This final deviation representation form is optimized for part type, pose component sensitivity, and fitting requirements, and can more accurately reflect the key deviations that need to be compensated in actual assembly, providing a reliable input for subsequent generation of efficient and accurate robot alignment compensation instructions.In this way, when calculating the relative pose deviation, the scheme no longer simply obtains the original difference, but combines the characteristics of the parts and the assembly requirements to process and convert the deviation in a targeted manner, making the calculated deviation information more instructive, so as to generate more accurate and effective robot alignment compensation instructions, overcoming the problem that the simple unit pose difference in the prior art cannot fully guide the complex assembly alignment.

[0092] For example, assuming that the part to be assembled is a pin that needs to be inserted into a hole in an already assembled component. First, the system obtains the part type information of the part, such as "positioning pin A". According to the part type, the system obtains the corresponding deviation calculation parameters from the preset database or configuration file. These parameters may specify that for the insertion of "positioning pin A", the translational deviation sensitivity along the pin axis direction is low, while the translational deviation perpendicular to the pin axis direction and the rotational deviation around the axis perpendicular to the pin axis direction are high. At the same time, the parameters also specify the deviation representation conversion rule, which requires converting the six-degree-of-freedom deviation into translational deviation along the pin axis direction, two-dimensional translational deviation perpendicular to the pin axis direction, and rotational deviation around the pin axis direction. Then, the system calculates the initial six-degree-of-freedom relative pose deviation according to the ideal relative pose data (such as the ideal relative position of the pin and the hole in the CAD model) and the actual relative pose (such as the actual position of the pin relative to the hole measured by a vision sensor). Then, the system applies the obtained sensitivity weights to weight process each component of the initial deviation, for example, the translational deviation component in the vertical direction is multiplied by a larger weight value, while the translational deviation component along the axis direction is multiplied by a smaller weight value. Finally, the system applies the obtained deviation representation conversion rule to convert the weighted six-degree-of-freedom deviation into translational deviation along the pin axis direction, two-dimensional translational deviation perpendicular to the pin axis direction, and rotational deviation around the pin axis direction. This converted deviation representation, such as "vertical deviation [0.1mm, 0.05mm], axial deviation [0.2mm], angle deviation [0.5 degrees]", is output as the final relative pose deviation, which is used to generate robot alignment compensation instructions.

[0093] By obtaining and applying different deviation calculation parameters according to the part type information, including sensitivity weights of different pose components and deviation representation conversion rules, the scheme can perform targeted weighting processing and representation conversion on the initial relative pose deviation. This makes the calculated relative pose deviation more accurately reflect the key deviation components that have the greatest impact on assembly accuracy under specific part types and fitting methods, and converts the deviation information into a form more suitable for guiding the robot to accurately align. Therefore, the scheme overcomes the problem that the simple unit pose difference cannot fully consider the importance of different deviation components and the differences in fitting methods, and can generate more optimized and accurate robot alignment compensation instructions, thereby improving the precision and efficiency of automatic assembly of stamping dies.

[0094] In some embodiments, according to the deviation calculation parameter, the step of converting the weighted initial relative pose deviation into a deviation representation corresponding to the part type based on the weighted initial relative pose deviation by applying a deviation representation conversion rule includes:

[0095] According to the part type information, the mating feature definition, the deviation representation model, and the tolerance information related to the current part mating mode are obtained;

[0096] According to the mating feature definition, the key mating geometric features on the assembled component and the to-be-assembled part are identified;

[0097] According to the identified key mating geometric features and the weighted initial relative pose deviation, the relative deviation between the key mating geometric features is calculated;

[0098] According to the deviation representation model and the tolerance information, the calculated relative deviation between the key mating geometric features is converted into a deviation representation corresponding to the part type, reflecting the mating state and the tolerance allowance.

[0099] The mating manner refers to the expected connection or mating type between the to-be-assembled part and the already-assembled part, such as interference fit, clearance fit, transition fit, bolt connection, pin positioning, etc., which can be implemented by using predefined enumerated values, mating standard codes or text descriptions. The mating feature definition refers to the rules or templates describing the key geometric features and their mutual relationships that are relied on to implement a specific mating manner, such as defining that two planes need to be parallel and the distance needs to be within a specific range, or two cylindrical surfaces need to be coaxial, which can be implemented by using a structured data format, a set of geometric constraint rules or a preset feature template. The deviation representation model refers to the way or algorithm for quantifying and formatting the relative deviation between key mating geometric features, such as representing the deviation as an offset relative to the tolerance center, a margin ratio relative to the tolerance boundary, or a displacement and rotation vector in a specific coordinate system, which can be implemented by using mathematical formulas, lookup tables or data structure definitions. The tolerance information refers to the allowable deviation range or limit related to a specific mating feature, which is used to evaluate whether the mating state meets the design requirements, which can be implemented by using numerical ranges, upper and lower limit values or tolerance level standards. The key mating geometric feature refers to the geometric element that plays a decisive role in the process of implementing part mating, such as a plane, a hole, a shaft, an edge, a corner point, etc., the relative position and attitude of these features directly affect the success and final precision of assembly, which can be implemented by using geometric primitive parameters, feature point coordinate sets or specific identifiers in CAD models. The relative deviation refers to the difference in position or attitude between two geometric features, such as the distance or angle between two planes, the offset or inclination between two axes, which can be represented in the form of displacement vector, rotation matrix or Euler angle. The deviation representation refers to the quantitative result obtained by converting the calculated relative deviation according to the specific model and tolerance information, which can reflect the current mating state and tolerance margin, which can be implemented by using numerical values, state flags, color coding or specific format data packets.

[0100] The present scheme elaborates how to convert the weighted initial relative pose deviation into a deviation representation corresponding to the part type, reflecting the fitting state and tolerance margin, so as to more accurately quantify the deviation that has the greatest impact on the current fitting, and provide a basis for subsequent generation of more targeted compensation instructions that can cope with different fitting mode requirements. First, according to the part type information, the fitting feature definition, deviation representation model and tolerance information related to the current part fitting mode are obtained. This step is the basis for customized processing of different part types and fitting modes. By obtaining the fitting feature definition, it is clear which geometric features are the key to achieving fitting; by obtaining the deviation representation model, it is determined how to quantify and represent the deviation between these key features; by obtaining the tolerance information, it provides a standard for judging whether the deviation is acceptable. Obtaining these specific parameters according to the part type information ensures that the subsequent deviation conversion process can fully consider the specific requirements and characteristics of the current assembly task. Second, according to the fitting feature definition, the key fitting geometric features on the assembled component and the part to be assembled are identified. After the fitting feature definition is clear, this step finds the specific geometric features on the actual assembled component and the part to be assembled that meet the definition. These features are the physical carriers of actual fitting and deviation. Identifying these key features according to the fitting feature definition ensures that the subsequent calculated deviation is for the actual fitting site, rather than a general overall deviation. Then, according to the identified key fitting geometric features and the weighted initial relative pose deviation, the relative deviation between the key fitting geometric features is calculated. After identifying the key fitting features, this step uses the overall relative pose deviation that has been calculated and weighted, and applies it to these key features to calculate the specific relative position and attitude deviation between them. For example, the gap or angle deviation between two fitting planes, or the distance and angle deviation between two fitting axes. Calculating the relative deviation between the identified key fitting geometric features converts the overall pose deviation into a more physically meaningful deviation quantity related to specific fitting, laying the foundation for subsequent refined deviation representation. Finally, according to the deviation representation model and the tolerance information, the calculated relative deviation between the key fitting geometric features is converted into a deviation representation corresponding to the part type, reflecting the fitting state and tolerance margin. This is the final conversion step. It uses the previously obtained deviation representation model and tolerance information to further process and represent the calculated relative deviation between the key fitting geometric features. The deviation representation model guides how to quantify and format these deviations, for example, it can be represented as a proportion relative to the tolerance range, a flag indicating whether it exceeds the tolerance range, or in a certain specific vector or matrix form that is more suitable for robot control. The tolerance information is used to evaluate whether the deviation is within the allowed range, and this evaluation result is incorporated into the final deviation representation, for example, by color coding, numerical proportion or state flag to reflect the tolerance margin.According to the bias representation model and the tolerance information, the conversion is performed, so that the final bias representation not only contains the numerical value of the bias, but also intuitively reflects the current fitting state and the distance from the tolerance boundary, so that the bias which has the greatest impact on the current fitting can be more accurately quantified, and more targeted compensation instructions which can cope with different fitting mode requirements are generated for the robot to provide direct and rich input information. Through the above steps, on the basis of calculating the weighted initial relative pose bias, the influence of the bias on the specific fitting feature is further analyzed, and combined with the part type and fitting requirement, it is converted into a more instructive bias representation. This conversion process considers the sensitivity and attention point difference of different fitting modes to the bias, so that the final bias representation can more accurately reflect the "pain point" of the actual assembly, thereby providing a solid foundation for subsequent generation of differentiated and more effective compensation strategies. Compared with the scheme of only weighting the overall pose bias, it can more finely guide the robot to perform the alignment operation, especially in complex or high-precision fitting scenarios, it can significantly improve the success rate and precision of assembly.

[0101] In one embodiment, according to the part type information, for example, it is identified that the current part to be assembled is a guide pillar in a stamping die, the system can obtain parameters related to the fitting mode of the guide pillar (for example, clearance fit with a guide sleeve) from a preset database or configuration file. These parameters can include fitting feature definition, for example, defining the outer cylindrical surface of the guide pillar and the inner cylindrical surface of the guide sleeve as the key fitting features; deviation representation model, for example, defining the relative deviation between the cylindrical surfaces as the radial offset and axial offset, and further quantifying it as the proportion relative to the fitting clearance; and tolerance information, for example, the radial and axial tolerance range of the guide pillar and guide sleeve fitting. Then, according to the obtained fitting feature definition, the point cloud data of the assembled guide sleeve and the to-be-assembled guide pillar obtained by the visual sensor, the inner cylindrical surface of the guide sleeve and the outer cylindrical surface of the guide pillar are identified by the point cloud processing algorithm. Then, according to the identified inner cylindrical surface of the guide sleeve, the outer cylindrical surface of the guide pillar and the previously calculated and weighted processed overall initial relative pose deviation of the guide pillar relative to the guide sleeve, the relative deviation between the two cylindrical surfaces is calculated, for example, the distance and angle between their center axes, and the axial distance between the end faces. Finally, according to the obtained deviation representation model and tolerance information, the calculated cylindrical surface relative deviation is converted into a deviation representation reflecting the fitting state and tolerance margin. For example, the axial distance is converted into a percentage relative to the radial tolerance range, the axial distance is converted into a percentage relative to the axial tolerance range, and combined with the tolerance information to determine whether it is out of range, and finally these information can be integrated into a structured data package containing radial margin percentage, axial margin percentage and whether it is out of range, as the final deviation representation. This representation directly reflects the "urgency" or "looseness" in the radial and axial directions when the guide pillar is inserted into the guide sleeve, and whether there is a risk of interference or excessive clearance, providing a direct basis for generating fine compensation motion instructions for the robot that take into account the radial and axial differences.

[0102] Through the above technical means, the present scheme can convert abstract pose deviation into deviation representation with more physical meaning and guiding value according to the characteristics of different part types and fitting modes. This deviation representation not only quantifies the relative deviation between the key fitting features, but also incorporates tolerance information, directly reflecting the current fitting state and tolerance margin. This makes the subsequent generated robot compensation instructions more accurate in addressing key issues in actual assembly, such as prioritizing the resolution of radial deviation that may cause interference, or adjusting the axial position to meet specific fitting depth requirements. Therefore, the present scheme can significantly improve the precision and robustness of robot positioning operations in the stamping die flexible assembly process, effectively addressing the challenges brought by different part switching and error accumulation, and improving the overall assembly efficiency and success rate.

[0103] The accompanying drawings are referred to in Figure 2The application provides a stamping die assembly control system, which is applied to an automatic assembly production line using a robot and comprises the following:

[0104] A first acquisition module 100 is configured to acquire part type information of a part to be assembled.

[0105] A second acquisition module 200 is configured to acquire feature parameters of an assembled component corresponding to the part type information, feature parameters of the part to be assembled and ideal relative pose data according to the part type information.

[0106] A third acquisition module 300 is configured to acquire key feature information of the assembled component according to the feature parameters of the assembled component.

[0107] A fourth acquisition module 400 is configured to acquire key feature information of the part to be assembled according to the feature parameters of the part to be assembled.

[0108] A first calculation module 500 is configured to calculate an actual relative pose of the part to be assembled relative to the assembled component according to the key feature information of the assembled component and the key feature information of the part to be assembled.

[0109] A second calculation module 600 is configured to calculate a relative pose deviation of the part to be assembled relative to the assembled component according to the ideal relative pose data and the actual relative pose.

[0110] A generation module 700 is configured to generate a compensation instruction of a robot alignment movement according to the relative pose deviation.

[0111] A control module 800 is configured to control the robot to perform an alignment operation on the part to be assembled according to the compensation instruction.

[0112] In some embodiments, the first calculation module 500 is configured to perform the following when calculating the actual relative pose of the part to be assembled relative to the assembled component according to the key feature information of the assembled component and the key feature information of the part to be assembled:

[0113] A1. acquiring pose calculation parameters related to the assembled component and the part to be assembled according to the part type information;

[0114] A2. establishing a feature correspondence relationship total set between the key feature information of the assembled component and the key feature information of the part to be assembled based on the key feature information of the assembled component and the key feature information of the part to be assembled; the feature correspondence relationship total set comprises a plurality of feature correspondence relationships;

[0115] A3. randomly selecting a preset number of feature correspondence relationships from the feature correspondence relationship total set as a feature correspondence relationship subset;

[0116] A4. Calculate a candidate relative pose of the part to be assembled relative to the assembled component according to the subset of feature correspondence and the pose calculation parameter;

[0117] A5. Perform consistency check on all feature correspondences in the total set of feature correspondences based on the candidate relative pose, and determine all feature correspondences that conform to the candidate relative pose;

[0118] A6. Repeat steps A3-A5 until a candidate relative pose with the largest number of feature correspondences is identified as the actual relative pose of the part to be assembled relative to the assembled component after repeating a preset number of times.

[0119] In some embodiments, the first calculation module 500 performs, when used for performing consistency check on all feature correspondences in the total set of feature correspondences based on the candidate relative pose, and determining all feature correspondences that conform to the candidate relative pose:

[0120] A51. Obtain, according to the part type information, a basic consistency threshold, a feature correspondence reliability evaluation method, and a threshold adjustment rule related to the consistency check;

[0121] A52. Traverse each feature correspondence in the total set of feature correspondences, and perform the following steps A521-A524 on each feature correspondence:

[0122] A521. For the currently traversed feature correspondence, evaluate the reliability of the current feature correspondence using the obtained feature correspondence reliability evaluation method to obtain a reliability evaluation result;

[0123] A522. Determine, according to the obtained basic consistency threshold and the reliability evaluation result, a consistency check threshold for the current feature correspondence based on the obtained threshold adjustment rule;

[0124] A523. Calculate, according to the candidate relative pose, a consistency measure of the current feature correspondence and the candidate relative pose;

[0125] A524. Determine whether the consistency measure of the current feature correspondence meets the determined consistency check threshold; if so, determine the current feature correspondence as conforming to the candidate relative pose;

[0126] A53. After traversing all feature correspondences in the total set of feature correspondences, obtain all feature correspondences that conform to the candidate relative pose.

[0127] In some embodiments, the first calculation module 500 performs, when used for determining, according to the obtained basic consistency threshold and the reliability evaluation result, a consistency check threshold for the current feature correspondence based on the obtained threshold adjustment rule:

[0128] obtaining a feature type involved in the current feature correspondence;

[0129] determining, according to the feature type, a threshold adjustment sub-rule corresponding to the feature type from the obtained threshold adjustment rule;

[0130] determining, according to the determined threshold adjustment sub-rule, the obtained basic consistency threshold, and the reliability evaluation result, a consistency checking threshold for the current feature correspondence.

[0131] In some embodiments, the second computing module 600 performs, when calculating the relative pose deviation of the to-be-assembled part relative to the assembled component according to the ideal relative pose data and the actual relative pose:

[0132] obtaining deviation calculation parameters related to the ideal relative pose data and the actual relative pose according to the part type information; the deviation calculation parameters include sensitivity weights of different pose components and a deviation representation conversion rule;

[0133] calculating an initial relative pose deviation of the to-be-assembled part relative to the assembled component according to the ideal relative pose data and the actual relative pose;

[0134] performing, according to the deviation calculation parameters, weighted processing on different pose components of the initial relative pose deviation based on the initial relative pose deviation by applying the sensitivity weights of different pose components;

[0135] performing, according to the deviation calculation parameters, conversion of the weighted processed initial relative pose deviation into a deviation representation corresponding to the part type based on the weighted processed initial relative pose deviation by applying the deviation representation conversion rule;

[0136] taking the converted deviation representation as the relative pose deviation of the to-be-assembled part relative to the assembled component.

[0137] In some embodiments, the second computing module 600 performs, when converting the weighted processed initial relative pose deviation into a deviation representation corresponding to the part type based on the weighted processed initial relative pose deviation by applying the deviation representation conversion rule according to the deviation calculation parameters:

[0138] obtaining a fitting feature definition, a deviation representation model, and tolerance information related to the current part fitting mode according to the part type information;

[0139] identifying key fitting geometric features on the assembled component and the to-be-assembled part according to the fitting feature definition;

[0140] According to the identified key mating geometric features and the initial relative pose deviation after the weighting processing, the relative deviation between the key mating geometric features is calculated;

[0141] According to the deviation representation model and the tolerance information, the calculated relative deviation between the key mating geometric features is converted into a deviation representation corresponding to the part type, reflecting the mating state and the tolerance allowance.

[0142] In this article, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations.

[0143] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A stamping die assembly control method, applied to an automated assembly line using robots, characterized in that, Includes the following steps: Obtain the part type information of the parts to be assembled; Based on the part type information, obtain the feature parameters of the corresponding assembled parts, the feature parameters of the parts to be assembled, and the ideal relative pose data. Based on the characteristic parameters of the assembled components, obtain the key characteristic information of the assembled components; Based on the feature parameters of the parts to be assembled, obtain the key feature information of the parts to be assembled. Based on the key feature information of the assembled components and the key feature information of the parts to be assembled, calculate the actual relative pose of the parts to be assembled relative to the assembled components. Based on the ideal relative pose data and the actual relative pose, calculate the relative pose deviation of the part to be assembled relative to the assembled parts. Based on the relative pose deviation, compensation commands for the robot's alignment motion are generated. According to the compensation instructions, control the robot to perform alignment operations on the parts to be assembled; Based on the key feature information of the assembled components and the key feature information of the part to be assembled, the steps for calculating the actual relative pose of the part to be assembled with respect to the assembled components include: A1. Based on the part type information, obtain the pose calculation parameters related to the assembled parts and the parts to be assembled; A2. Based on the key feature information of assembled components and the key feature information of parts to be assembled, establish a set of feature correspondences between the key feature information of assembled components and the key feature information of parts to be assembled; the set of feature correspondences includes multiple feature correspondences. A3. Randomly select a preset number of feature correspondences from the total set of feature correspondences as a subset of feature correspondences; A4. Calculate the candidate relative pose of the part to be assembled relative to the assembled parts based on the feature correspondence subset and pose calculation parameters; A5. Based on the candidate relative pose, perform a consistency check on all feature correspondences in the feature correspondence set to determine all feature correspondences that conform to the candidate relative pose; A6. Repeat steps A3-A5 until the preset number of repetitions are performed, identify the candidate relative pose with the most matching feature correspondences and use it as the actual relative pose of the part to be assembled relative to the assembled component. The specific steps in step A5 include: A51. Based on the part type information, obtain the basic consistency threshold, feature correspondence reliability assessment method, and threshold adjustment rules related to consistency inspection; A52. Traverse every feature correspondence in the set of feature correspondences, and perform the following steps A521-A524 for each feature correspondence: A521. For the feature correspondences currently being traversed, the reliability of the current feature correspondences is evaluated using the reliability evaluation method of the obtained feature correspondences, and the reliability evaluation result is obtained. A522. Based on the obtained threshold adjustment rules, determine the consistency check threshold for the current feature correspondence according to the obtained basic consistency threshold and reliability assessment results; A523. Based on the candidate relative pose, calculate the consistency measure between the current feature correspondence and the candidate relative pose; A524. Determine whether the consistency measure of the current feature correspondence meets the determined consistency test threshold; if it does, then determine the current feature correspondence as a candidate relative pose. A53. After traversing all feature correspondences in the feature correspondence set, all feature correspondences that match the candidate relative pose are obtained.

2. The stamping die assembly control method according to claim 1, characterized in that, The specific steps in step A522 include: Obtain the feature types involved in the current feature correspondence; Based on the feature type, determine the threshold adjustment sub-rule corresponding to the feature type from the obtained threshold adjustment rules; Based on the determined threshold adjustment sub-rules, the obtained basic consistency threshold, and the reliability assessment results, determine the consistency check threshold for the current feature correspondence.

3. The stamping die assembly control method according to claim 1, characterized in that, The steps for calculating the relative pose deviation of the part to be assembled relative to the assembled components, based on the ideal relative pose data and the actual relative pose, include: Based on the part type information, obtain the deviation calculation parameters related to the ideal relative pose data and the actual relative pose; the deviation calculation parameters include the sensitivity weights of different pose components and the deviation representation conversion rules; Based on the ideal relative pose data and the actual relative pose, calculate the initial relative pose deviation of the part to be assembled relative to the assembled parts. Based on the deviation calculation parameters, and based on the initial relative pose deviation, the different pose components of the initial relative pose deviation are weighted by applying the sensitivity weights of different pose components. Based on the deviation calculation parameters, and using the weighted initial relative pose deviation, the weighted initial relative pose deviation is converted into a deviation representation corresponding to the part type by applying the deviation representation conversion rule. The converted deviation is represented as the relative pose deviation of the part to be assembled relative to the assembled components.

4. The stamping die assembly control method according to claim 3, characterized in that, Based on the deviation calculation parameters, and using the weighted initial relative pose deviation, the steps to convert the weighted initial relative pose deviation into a deviation representation corresponding to the part type by applying deviation representation conversion rules include: Based on the part type information, obtain the mating feature definition, deviation representation model and tolerance information related to the current part's mating method; Based on the definition of mating features, identify the key mating geometric features on assembled components and parts to be assembled; Based on the identified key mating geometric features and the weighted initial relative pose deviation, the relative deviation between the key mating geometric features is calculated. Based on the deviation representation model and tolerance information, the calculated relative deviations between key mating geometric features are converted into deviation representations that correspond to the part type and reflect the mating state and tolerance margin.

5. The stamping die assembly control method according to claim 4 further includes an assembly control system, applied to an automated assembly production line using robots, characterized in that, include: The first acquisition module is used to acquire the part type information of the parts to be assembled; The second acquisition module is used to acquire the feature parameters of the assembled parts, the feature parameters of the parts to be assembled, and the ideal relative pose data according to the part type information. The third acquisition module is used to acquire key feature information of the assembled parts based on the feature parameters of the assembled parts; The fourth acquisition module is used to acquire key feature information of the parts to be assembled based on the feature parameters of the parts to be assembled. The first calculation module is used to calculate the actual relative pose of the part to be assembled relative to the assembled part based on the key feature information of the assembled component and the key feature information of the part to be assembled. The second calculation module is used to calculate the relative pose deviation of the part to be assembled relative to the assembled parts based on the ideal relative pose data and the actual relative pose. The generation module is used to generate compensation instructions for the robot's alignment motion based on the relative pose deviation. The control module is used to control the robot to perform alignment operations on the parts to be assembled, according to the compensation instructions.

6. The stamping die assembly control method according to claim 5, characterized in that, The first calculation module is executed when calculating the actual relative pose of the part to be assembled relative to the assembled part based on the key feature information of the assembled component and the key feature information of the part to be assembled: A1. Based on the part type information, obtain the pose calculation parameters related to the assembled parts and the parts to be assembled; A2. Based on the key feature information of assembled components and the key feature information of parts to be assembled, establish a set of feature correspondences between the key feature information of assembled components and the key feature information of parts to be assembled; the set of feature correspondences includes multiple feature correspondences. A3. Randomly select a preset number of feature correspondences from the total set of feature correspondences as a subset of feature correspondences; A4. Calculate the candidate relative pose of the part to be assembled relative to the assembled parts based on the feature correspondence subset and pose calculation parameters; A5. Based on the candidate relative pose, perform a consistency check on all feature correspondences in the feature correspondence set to determine all feature correspondences that conform to the candidate relative pose; A6. Repeat steps A3-A5 until the preset number of repetitions are completed, identify the candidate relative pose with the highest number of matching feature correspondences and use it as the actual relative pose of the part to be assembled relative to the assembled component.

7. The stamping die assembly control method according to claim 6, characterized in that, The first calculation module is executed when performing a consistency check on all feature correspondences in the total set of feature correspondences based on candidate relative poses, and determining all feature correspondences that conform to the candidate relative poses: A51. Based on the part type information, obtain the basic consistency threshold, feature correspondence reliability assessment method, and threshold adjustment rules related to consistency inspection; A52. Traverse every feature correspondence in the set of feature correspondences, and perform the following steps A521-A524 for each feature correspondence: A521. For the feature correspondences currently being traversed, the reliability of the current feature correspondences is evaluated using the reliability evaluation method of the obtained feature correspondences, and the reliability evaluation result is obtained. A522. Based on the obtained threshold adjustment rules, determine the consistency check threshold for the current feature correspondence according to the obtained basic consistency threshold and reliability assessment results; A523. Based on the candidate relative pose, calculate the consistency measure between the current feature correspondence and the candidate relative pose; A524. Determine whether the consistency measure of the current feature correspondence meets the determined consistency test threshold; If satisfied, the current feature correspondence is determined as a candidate relative pose. A53. After traversing all feature correspondences in the feature correspondence set, all feature correspondences that match the candidate relative pose are obtained.

8. The stamping die assembly control method according to claim 5, characterized in that, The first calculation module performs the following when determining the consistency check threshold for the current feature correspondence based on the acquired threshold adjustment rules, the acquired basic consistency threshold, and the reliability assessment results: Obtain the feature types involved in the current feature correspondence; Based on the feature type, determine the threshold adjustment sub-rule corresponding to the feature type from the obtained threshold adjustment rules; Based on the determined threshold adjustment sub-rules, the obtained basic consistency threshold, and the reliability assessment results, determine the consistency check threshold for the current feature correspondence.

Images