Part relative position detection method, system and medium under three-axis movement

Abstract

The application discloses a part relative position detection method and system under three-axis movement and a medium, relates to the technical field of position detection, and comprises the following steps: scanning a target part to obtain three-dimensional point cloud data, performing region segmentation and identification, and performing clustering analysis to obtain point cloud clusters of a plurality of feature points for global fitting to determine a plurality of feature space coordinates; a part local coordinate system is constructed, the plurality of feature space coordinates are mapped to the part local coordinate system for position calculation to obtain a plurality of feature relative position relationship graphs; deviation marks are added to the plurality of feature relative position relationship graphs, a plurality of out-of-tolerance items are obtained, and the target part is bound and stored to construct a position detection anomaly database, and position deviation training is performed to construct a position deviation prediction channel to feed back and optimize a detection path. The application solves the technical problems of insufficient part relative position detection precision and low efficiency in the prior art, and achieves the technical effects of improving part relative position detection precision and detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of position detection technology, specifically to a method, system, and medium for detecting the relative position of parts under three-axis movement. Background Technology

[0002] During the processing and assembly of parts, the relative position needs to be detected. Traditional detection methods rely on manual measurement or data collection under a single coordinate system. The utilization rate of point cloud data is limited, feature extraction is incomplete, coordinate mapping and position calculation processes are complex, which can easily cause measurement errors. The detection steps are lengthy and the data processing efficiency is low, which cannot meet the requirements of high precision and high efficiency detection. Summary of the Invention

[0003] This application provides a method, system, and medium for detecting the relative position of parts under three-axis movement, which is intended to address the technical problems of insufficient accuracy and low efficiency in the detection of the relative position of parts in the prior art.

[0004] In view of the above problems, this application provides a method, system and medium for detecting the relative position of parts under three-axis movement.

[0005] A first aspect of this application provides a method for detecting the relative position of a part under three-axis motion, the method comprising:

[0006] The process involves importing a 3D model of the target part to plan the detection path, constructing a three-axis motion control command to drive a three-coordinate measuring machine to move the probe head and scan the target part, obtaining 3D point cloud data. Based on the 3D point cloud data, the 3D model is segmented and identified, and the identification results are clustered to obtain point cloud clusters of multiple feature points. These clusters are then globally fitted to determine multiple feature spatial coordinates. A local coordinate system for the part is constructed, and the multiple feature spatial coordinates are mapped to this local coordinate system for position calculation, resulting in a relative positional relationship diagram of multiple features. The relative positional relationship diagrams of multiple features are traversed to mark deviations, and multiple out-of-tolerance items are bound and stored with the target part to construct a position detection anomaly database. Position deviation training is performed based on the position detection anomaly database, and a position deviation prediction channel is constructed to provide feedback optimization for the detection path.

[0007] A second aspect of this application provides a system for detecting the relative position of a part under three-axis movement, the system comprising:

[0008] The scanning module is used to import the 3D model of the target part, plan the detection path, construct three-axis motion control commands to drive the three-coordinate device to move the probe head to scan the target part, and obtain 3D point cloud data. The fitting module is used to perform region segmentation and recognition on the 3D model based on the 3D point cloud data, perform cluster analysis on the recognition results, obtain point cloud clusters of multiple feature points, perform global fitting, and determine multiple feature space coordinates. The calculation module is used to construct a local coordinate system of the part, map the multiple feature space coordinates to the local coordinate system of the part, perform position calculation, and obtain multiple feature relative position relationship diagrams. The binding and storage module is used to traverse the multiple feature relative position relationship diagrams, mark deviations, obtain multiple out-of-tolerance items, bind and store them with the target part, and construct a position detection anomaly database. The feedback optimization module is used to train the position deviation based on the position detection anomaly database, construct a position deviation prediction channel, and perform feedback optimization on the detection path.

[0009] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for detecting the relative position of a part under three-axis movement provided in this application.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] This application imports a 3D model of the target part to plan the detection path, constructs a three-axis motion control command to drive a three-coordinate device to move the probe head to scan the target part, and obtains 3D point cloud data; based on the 3D point cloud data, it performs region segmentation and recognition on the 3D model, performs cluster analysis on the recognition results to obtain point cloud clusters of multiple feature points, performs global fitting, and determines multiple feature space coordinates; it constructs a local coordinate system for the part, maps the multiple feature space coordinates to the local coordinate system for position calculation, and obtains multiple feature relative position relationship diagrams; it traverses the multiple feature relative position relationship diagrams to mark deviations, obtains multiple out-of-tolerance items, binds and stores them with the target part, and constructs a position detection anomaly database; it performs position deviation training based on the position detection anomaly database, constructs a position deviation prediction channel, and performs feedback optimization on the detection path. This invention solves the technical problems of insufficient accuracy and low efficiency in the prior art for relative position detection of parts, and achieves the technical effect of improving the accuracy and efficiency of relative position detection of parts through 3D point cloud acquisition, feature recognition, and deviation prediction optimization. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic flowchart of a method for detecting the relative position of a part under three-axis movement provided in an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of the structure of a part relative position detection system under three-axis movement provided in an embodiment of this application.

[0015] Figure labeling: Scanning module 11, Fitting module 12, Calculation module 13, Binding and storage module 14, Feedback optimization module 15. Detailed Implementation

[0016] This application provides a method, system, and medium for detecting the relative position of parts under three-axis movement. It addresses the technical problems of insufficient accuracy and low efficiency in the detection of the relative position of parts in the prior art. By optimizing the acquisition of three-dimensional point clouds, feature recognition, and deviation prediction, it achieves the technical effect of improving the accuracy and efficiency of the detection of the relative position of parts.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0019] Example 1, as Figure 1 As shown, this application provides a method for detecting the relative position of a part under three-axis movement, the method comprising:

[0020] Step S100: Import the 3D model of the target part, plan the detection path, construct a three-axis motion control command to drive the three-coordinate device to move the probe head to scan the target part and obtain 3D point cloud data.

[0021] In this embodiment, pre-prepared 3D model data is first imported. This 3D model is generated and stored in the form of a CAD file and contains the geometric feature information and dimensional data of the target part. Then, a pre-planned detection path is invoked based on this 3D model. This detection path includes a scanning trajectory covering the key feature areas of the part. Subsequently, a CNC interpolation algorithm is used to convert the pre-planned detection path into three-axis motion parameters, forming three-axis movement control commands. These three-axis movement control commands contain displacement data and velocity information in the X, Y, and Z directions.

[0022] Then, the three-axis motion control command drives the three-coordinate device to move the probe head to scan the target part. The three-coordinate device moves the probe head to scan the surface of the part point by point at a fixed sampling frequency, collects spatial coordinate data in real time, and records and organizes the scanning results to finally obtain three-dimensional point cloud data.

[0023] Step S200: Based on the three-dimensional point cloud data, perform region segmentation and recognition on the three-dimensional model, perform cluster analysis on the recognition results, obtain point cloud clusters of multiple feature points, perform global fitting, and determine multiple feature space coordinates.

[0024] In this embodiment, when performing region segmentation and recognition on a 3D model based on 3D point cloud data, the 3D point cloud data is first mapped onto the 3D model to calculate normal vectors, constructing an initial point cloud normal vector field. An initial reference direction is then defined based on the initial point cloud normal vector field to segment the 3D model. Next, unsegmented point clouds are randomly selected as seed points based on the segmentation results. Neighboring points of the seed points are traversed and matched with the initial point cloud normal vector field. The normal vectors of the neighboring points are extracted, and the angle between the neighboring point normal vector and the seed point normal vector is calculated. When the angle is less than a preset angle threshold, the neighboring points are merged into the seed point, and the initial point cloud normal vector field is updated to obtain the point cloud normal vector field. Subsequently, based on the point cloud normal vector field and the segmentation results, the 3D model undergoes repeated region growth and merging until all unsegmented point clouds are segmented. Finally, multiple segmented regions are determined and geometric recognition is performed to obtain the recognition result.

[0025] Next, cluster analysis is performed on the recognition results to obtain point cloud clusters of multiple feature points for global fitting. In this process, the recognition results are first clustered using Euclidean distance to obtain point cloud clusters of multiple feature points. These clusters are then geometrically decomposed into aperture feature point cloud clusters, planar edge feature point cloud clusters, and curved surface feature point cloud clusters. Subsequently, for each type of point cloud cluster, cylindrical surface fitting, edge line fitting, and curved surface fitting are performed to obtain the corresponding aperture geometric feature parameters, planar edge geometric feature parameters, and curved surface geometric feature parameters. Finally, a global coordinate system is constructed based on the 3D model, and the aforementioned geometric feature parameters are transformed into this global coordinate system for identification, ultimately determining the coordinates of multiple feature spaces.

[0026] Furthermore, in the method provided in the application embodiments, the region segmentation and recognition of the 3D model based on the 3D point cloud data further includes:

[0027] The 3D point cloud data is mapped to a 3D model for normal vector calculation to construct an initial point cloud normal vector field, which corresponds to the 3D point cloud data. The 3D model is segmented according to the initial reference direction defined by the initial point cloud normal vector field. Unsegmented point clouds are randomly selected as seed points based on the segmentation results. Neighboring points of the seed points are traversed and matched with the initial point cloud normal vector field to extract neighboring point normal vectors. The angle between the neighboring point normal vector and the seed point normal vector is calculated. When the angle is less than a preset angle threshold, the neighboring points are merged into the seed point. The initial point cloud normal vector field is updated based on the neighboring point normal vector and the seed point normal vector to obtain the point cloud normal vector field. Based on the point cloud normal vector field and the segmentation results, the 3D model is repeatedly grown and merged into regions until all unsegmented point clouds are segmented. Multiple segmented regions are identified for geometric recognition to obtain the recognition results.

[0028] In this embodiment of the application, when mapping three-dimensional point cloud data to a three-dimensional model for normal vector calculation, each point in the three-dimensional point cloud data is a point with three-dimensional spatial coordinates. The neighborhood point set of each point is obtained by searching the neighborhood of a fixed radius. The neighborhood point set is fitted with a least squares plane and the plane normal vector is taken as the normal vector of the point. After calculating point by point, the initial point cloud normal vector field is obtained. The initial point cloud normal vector field maintains a one-to-one correspondence with each point in the three-dimensional point cloud data.

[0029] Next, when segmenting the 3D model according to the initial reference direction defined by the initial point cloud normal vector field, all normal vectors in the initial point cloud normal vector field are statistically analyzed and clustered by direction. The direction with the highest frequency is extracted as the initial reference direction. For each point in the 3D point cloud data, the angle between the normal vector of that point and the initial reference direction is calculated. The angle is obtained by dividing the vector dot product by the product of the magnitudes and taking the inverse cosine. When the angle is close to zero, the normal vector of that point is considered to be consistent with the initial reference direction, and the point is added to the set of segmented points. When the angle deviates significantly, the normal vector of that point is considered to be inconsistent with the initial reference direction, and the point is added to the set of unsegmented points. In this way, the point cloud data is divided into a set of segmented points and a set of unsegmented points. The segmented point set and the unsegmented point set are combined to form the segmentation result. Then, based on the segmentation result, an unsegmented point cloud is randomly selected as a seed point. A point is randomly selected from the unsegmented point set and defined as the seed point, which serves as the starting point for subsequent region growth.

[0030] Subsequently, when matching the neighboring points of the seed point with the initial point cloud normal vector field, a fixed k-nearest neighbor search is used to extract the normal vectors of the neighboring points. The angle between the normal vector of the neighboring point and the normal vector of the seed point is calculated. The angle is obtained by vector dot product and inverse cosine function. When the angle is less than a preset angle threshold, the neighboring point is merged into the region where the seed point is located.

[0031] After merging neighboring points, the initial point cloud normal vector field is updated based on the normal vectors of the neighboring points and the normal vector of the seed point. The update method is to calculate the average normal vector and replace the normal vector of the corresponding point to obtain a new point cloud normal vector field.

[0032] Finally, based on the updated point cloud normal vector field and the segmentation results, the region is repeatedly grown and incorporated. This process involves repeatedly performing neighborhood search, angle data calculation, threshold judgment, neighbor point incorporation, and average normal vector update until the set of unsegmented points is empty. Through this process, multiple segmented regions are ultimately obtained, and geometric recognition is performed on these regions. Plane fitting, cylindrical fitting, or surface fitting methods are used to identify the geometric features of the segmented regions, resulting in a complete recognition outcome.

[0033] Furthermore, in the method provided in the application embodiments, the recognition results are subjected to cluster analysis to obtain point cloud clusters of multiple feature points, which are then globally fitted to determine the coordinates of multiple feature spaces. The method further includes:

[0034] Feature point clustering is performed using Euclidean distance traversal recognition results to obtain point cloud clusters of multiple feature points. These point cloud clusters are then geometrically decomposed to obtain hole feature point cloud clusters, planar edge feature point cloud clusters, and curved surface feature point cloud clusters. Cylindrical surface fitting is performed based on the hole feature point cloud clusters to obtain hole geometric feature parameters. Edge line fitting is performed based on the planar edge feature point cloud clusters to obtain planar edge geometric feature parameters. Surface fitting is performed based on the curved surface feature point cloud clusters to obtain curved surface geometric feature parameters. A global coordinate system is constructed based on a 3D model, and the hole geometric feature parameters, planar edge geometric feature parameters, and curved surface geometric feature parameters are transformed into the global coordinate system for identification, thus determining the spatial coordinates of the multiple feature points.

[0035] In this embodiment of the application, when processing the recognition results, Euclidean distance is first used to traverse all points. The spatial distance relationship between points is determined by setting a distance threshold. When the distance between an unassigned point and a clustered point is less than the threshold, the point is assigned to the same set. This process is repeated until all points are assigned, thereby obtaining point cloud clusters of multiple feature points. Each point cloud cluster represents a set of points that are close to each other in space and have similar features.

[0036] Next, the point cloud clusters with multiple feature points are geometrically decomposed. The point cloud clusters are fitted using the random sampling consistency method. By iteratively calculating the number of interior points under different geometric forms, the geometric form with the most interior points is selected as the classification criterion for the point cloud clusters. Finally, the point cloud clusters are divided into hole feature point cloud clusters, planar edge feature point cloud clusters, and curved surface feature point cloud clusters.

[0037] Subsequently, for the point cloud cluster of hole features, the least squares cylindrical surface fitting method is used to determine the hole's geometric feature parameters. Principal component analysis is used to obtain the axis vector, the average radius is calculated on the plane perpendicular to the axis, and the center coordinates are obtained from the centroid of the point cloud cluster. At the same time, the cylinder diameter is determined. Through this process, the hole's geometric feature parameters, including the center coordinates, axis vector, and diameter, are obtained.

[0038] For planar edge feature point cloud clusters, a least-squares line fitting method is used to determine the geometric feature parameters of the planar edges. First, principal component analysis is used to extract the first principal component as the direction vector of the edge line. The point cloud is projected onto this direction to form a one-dimensional coordinate sequence. The minimum and maximum points in the projection sequence are taken as the coordinates of key corner points. Then, the line direction is combined with the centroid coordinates to construct the line equation. Through this process, the geometric feature parameters of the planar edges are obtained, which include edge equation parameters and key corner point coordinates.

[0039] For a cluster of feature points on a curved surface, a B-spline surface least squares fitting method is used to determine the surface's geometric feature parameters. First, node vectors and a control mesh structure are defined based on the point cloud range, and a B-spline basis function matrix is ​​established. The point cloud coordinates are then substituted into an overdetermined system of equations, which are solved using the least squares method to obtain the control point coordinates. The principal curvature and Gaussian curvature of each point are calculated by fitting the surface. This process yields the surface's geometric feature parameters, which include the control point coordinates and curvature information.

[0040] After obtaining the geometric feature parameters of the hole, the planar edge, and the curved surface, a global coordinate system is constructed based on the 3D model. A homogeneous coordinate transformation method is then used to unify the parameters. Positional parameters, including the center coordinates, key corner coordinates, and control point coordinates, are transformed using rotation matrices and translation vectors. Orientational parameters, including the axis vector and line direction, are unified using rotation matrices. This process ultimately yields multiple feature space coordinates, which are then identified within the global coordinate system.

[0041] Step S300: Construct a local coordinate system for the part, map the spatial coordinates of the multiple features to the local coordinate system for position calculation, and obtain a relative position relationship diagram of the multiple features.

[0042] In this embodiment, when constructing the local coordinate system of a part, datum features are first extracted from the 3D model through datum analysis, and then mapped to the global coordinate system for identification to determine the datum plane, datum axis, and datum point. Then, the datum plane is used as the coordinate plane, the datum axis as the coordinate axis, and the datum point as the coordinate origin to complete the establishment of the local coordinate system of the part. Subsequently, a coordinate transformation matrix is ​​constructed, and multiple feature space coordinates are mapped to the local coordinate system of the part. The relative positional relationship diagram of multiple features is obtained through position calculation.

[0043] Furthermore, in the method provided in the application embodiments, constructing a local coordinate system for the part, mapping the multiple feature space coordinates to the local coordinate system for position calculation, and obtaining a relative position relationship diagram of multiple features, further includes:

[0044] Based on the 3D model, benchmark analysis is performed to obtain benchmark features. The benchmark features are then mapped to the global coordinate system for identification, and benchmark planes, benchmark axes, and benchmark points are determined. The benchmark plane is used as the coordinate plane, the benchmark axis as the coordinate axis, and the benchmark point as the coordinate origin to construct a local coordinate system for the part. A coordinate transformation matrix is ​​constructed, and the coordinates of the multiple feature spaces are mapped to the local coordinate system of the part according to the coordinate transformation matrix to calculate the position, thereby obtaining a relative position relationship diagram of the multiple features.

[0045] In this embodiment, when obtaining benchmark features based on a 3D model through benchmark analysis, the 3D point cloud data is first processed using a least-squares plane fitting method. The plane equation is calculated using the 3D point cloud data to obtain the normal vector, thus determining the benchmark plane. Next, the 3D point cloud data is processed using a least-squares line fitting method. The line equation is calculated using the 3D point cloud data to obtain the direction vector, thus determining the benchmark axis. Subsequently, the 3D point cloud data is processed using a centroid calculation method to obtain the 3D coordinate position, thus determining the benchmark point. The benchmark plane, benchmark axis, and benchmark point are then mapped to a global coordinate system for identification, obtaining the parameter information of the benchmark features in the global coordinate system.

[0046] Next, when constructing the local coordinate system of the part using the reference plane as the coordinate plane, the reference axes as the coordinate axes, and the reference point as the origin, the direction cosine method is used to establish a local three-dimensional orthogonal basis. Specifically, first, the local x-axis direction is aligned with the reference axis direction, then the local z-axis direction is aligned with the reference plane normal vector. Subsequently, the local y-axis direction is obtained through vector cross product, completing the three-axis definition of the part's local coordinate system. The reference point is then used as the origin to form the part's local coordinate system.

[0047] Subsequently, when constructing the coordinate transformation matrix, coordinate transformation analysis is first performed based on the coordinate data of the reference features in the global coordinate system to obtain the rotation matrix parameters and translation vector parameters. From these, the direction cosine and coordinate offset are calculated respectively, forming the rotation transformation matrix and translation transformation vector. These two are then combined to generate the initial coordinate transformation matrix, and its correctness is verified to obtain the final coordinate transformation matrix.

[0048] After constructing the local coordinate system of the part and obtaining the coordinate transformation matrix, multiple feature space coordinates are mapped from the global coordinate system to the local coordinate system according to the coordinate transformation matrix. For position-type feature space coordinates, the global coordinates are subtracted from the translation transformation vector and multiplied by the transpose of the rotation transformation matrix to obtain the coordinate values ​​in the local coordinate system. For direction-type feature space coordinates, the global direction vector is multiplied by the transpose of the rotation transformation matrix to obtain the direction parameters in the local coordinate system. Scalar-type feature space coordinates remain unchanged, thus obtaining the complete feature geometric parameters in the local coordinate system. After obtaining the mapped feature space coordinates, position calculations are performed based on the local coordinate system. The Euclidean distance formula is used to calculate the distance between feature space coordinates, the direction vector dot product formula is used to calculate the angle between feature space coordinates, the parallelism and orthogonality criteria are used to calculate the parallelism and perpendicularity between feature space coordinates respectively, and the coaxiality criterion is used to calculate the coaxiality between feature space coordinates, thus obtaining the geometric relationships between multiple feature space coordinates.

[0049] After completing the position calculation, the mapping results of the hole geometric feature parameters, planar edge geometric feature parameters, and curved surface geometric feature parameters in the part's local coordinate system are associated with the calculated geometric relationships, organized and visualized as a diagram of the relative position relationships of multiple features. This diagram uses feature space coordinates as nodes and calculated indicators such as relative distance, relative angle, coaxiality, parallelism, and perpendicularity as edge attributes, intuitively expressing the spatial distribution and geometric constraints of multiple features in the part's local coordinate system.

[0050] Furthermore, in the method provided in the application embodiments, the process of constructing the coordinate transformation matrix further includes:

[0051] Coordinate data of the reference feature in the global coordinate system are extracted and subjected to coordinate transformation analysis to obtain coordinate transformation requirement parameters, which include rotation matrix parameters and translation vector parameters. The coordinate axes of the local coordinate system are mapped to the global coordinate system according to the rotation matrix parameters to calculate the direction cosine and construct a rotation transformation matrix. The origin of the local coordinate system is mapped to the global coordinate system according to the translation vector parameters to calculate the offset and construct a translation transformation vector. The rotation transformation matrix and the translation transformation vector are combined to construct an initial coordinate transformation matrix. The initial coordinate transformation matrix is ​​verified, and if the verification is successful, the final coordinate transformation matrix is ​​constructed.

[0052] In this embodiment, the coordinate data of the reference plane, reference axis, and reference point in the global coordinate system are first extracted, and the normal vector of the reference plane, the direction vector of the reference axis, and the spatial coordinates of the reference point are analyzed. The normal vector components of the reference plane in the global coordinate system and the direction vector components of the reference axis in the global coordinate system are calculated using vector algebra methods, and the three-dimensional coordinate position of the reference point in the global coordinate system is read. This process yields the coordinate transformation requirements parameters describing the mapping relationship between the local and global coordinate systems. These parameters include rotation matrix parameters and translation vector parameters.

[0053] Next, the coordinate axes of the local coordinate system are mapped to the global coordinate system according to the rotation matrix parameters. The direction cosines are calculated using the direction cosine method, which describes the directional relationship between the local coordinate system axes and the three axes of the global coordinate system. Taking the local X-axis as an example, the cosine of the angle is obtained by calculating the dot product of the local X-axis and the global X-axis and dividing by the magnitude. This dot product operation with the global Y-axis and global Z-axis is repeated to obtain the three direction cosines of the local X-axis in the global coordinate system. These three cosine values ​​constitute the first column of the rotation transformation matrix. Similarly, the above calculation process is repeated for the local Y-axis and Z-axis to obtain the second and third columns. This process ultimately yields a rotation transformation matrix that satisfies the orthogonality normalization condition.

[0054] Subsequently, the origin of the local coordinate system is mapped to the global coordinate system according to the translation vector parameters. By using the coordinate difference method, the three-dimensional coordinates of the reference point in the global coordinate system are used as a reference, and the three components of the three-dimensional coordinates are extracted and used as the components of the translation vector in the X, Y, and Z directions, thus forming the translation transformation vector.

[0055] Next, the rotation transformation matrix and translation transformation vector are combined to form a 4×4 initial coordinate transformation matrix using the homogeneous coordinate method. The rotation transformation matrix is ​​filled into the top left 3×3 sub-block of the 4×4 matrix, and the translation transformation vector is filled into the top right 3×1 sub-block. The bottom row of the matrix is ​​filled with 0, 0, 0 and 1, forming an initial homogeneous coordinate transformation matrix that can handle both position and orientation simultaneously.

[0056] After constructing the initial coordinate transformation matrix, it is then verified. This process involves first randomly selecting multiple feature points as verification points based on the part's local coordinate system. These verification points are then mapped to the global coordinate system to obtain multiple initial coordinates. The initial coordinate transformation matrix is ​​then used to perform coordinate transformation calculations to obtain multiple coordinate values, which are compared and verified with theoretical coordinate values. Error distribution parameters are determined through error analysis. Finally, the initial coordinate transformation matrix is ​​iteratively corrected based on the error distribution parameters to construct the final coordinate transformation matrix.

[0057] Furthermore, in the method provided in the application embodiment, verifying the initial coordinate transformation matrix and constructing the coordinate transformation matrix when the verification passes further includes:

[0058] Multiple feature points are randomly selected as verification points based on the local coordinate system; the multiple verification points are mapped to the global coordinate system to obtain multiple initial coordinates; the multiple initial coordinates are transformed according to the initial coordinate transformation matrix to obtain multiple coordinate values; the multiple coordinate values ​​are compared and verified with the theoretical coordinate values, and error analysis is performed based on the verification results to determine the error distribution parameters; the initial coordinate transformation matrix is ​​iteratively corrected according to the error distribution parameters to construct the coordinate transformation matrix.

[0059] In this embodiment of the application, a number of feature points are randomly selected from the local coordinate system of the part as multiple verification points using a uniform random sampling method. At the same time, the coordinates of these verification points in the local coordinate system of the part are recorded one by one. These coordinates are the theoretical coordinate values.

[0060] Next, when mapping multiple verification points to the global coordinate system, a model mapping method is used. This method leverages the correspondence between the local and global coordinate systems established in the 3D model to map the coordinates of the verification points in the part's local coordinate system to the global coordinate system proportionally and oriented. Through this process, multiple initial coordinates corresponding to the multiple verification points are obtained.

[0061] Subsequently, when performing coordinate transformation calculations on multiple initial coordinates according to the initial coordinate transformation matrix, the homogeneous coordinate matrix multiplication method is used to expand each initial coordinate into a homogeneous coordinate form, and then matrix operations are performed with the initial coordinate transformation matrix. Through this process, multiple coordinate values ​​in the local coordinate system of the part are obtained.

[0062] After obtaining multiple coordinate values, these coordinate values ​​are compared and verified with the theoretical coordinate values. During the comparison process, residual analysis is used to calculate the residual vector for each verification point. The residual vector is obtained by subtracting the theoretical coordinate value from the coordinate value. By statistically analyzing the residuals of all verification points, indicators such as root mean square error, mean absolute error, and maximum absolute error are obtained. Simultaneously, the residual distribution is fitted to extract error distribution parameters, including the mean deviation vector and the residual covariance matrix.

[0063] Next, the initial coordinate transformation matrix is ​​iteratively corrected according to the error distribution parameters. During this correction process, the least squares rigid body registration method is used to update the rotation transformation matrix and translation transformation vector. Specifically, the theoretical coordinate values ​​and the coordinate values ​​obtained from the initial coordinates through the initial coordinate transformation matrix are taken as corresponding point pairs. The centroids of the two sets of points are calculated and decentralized. A cross-covariance matrix is ​​constructed and singular value decomposition is performed to extract the updated rotation transformation matrix, ensuring that it satisfies the conditions of orthogonality and a determinant of 1. Then, the updated translation vector is obtained based on the centroid offset. Subsequently, the updated rotation transformation matrix and translation vector are combined to replace and update the initial coordinate transformation matrix until the root mean square error converges to a preset threshold or the number of iterations reaches the upper limit. After the correction is completed, the final matrix obtained is the coordinate transformation matrix.

[0064] Step S400: Traverse the relative position relationship diagram of the multiple features to mark the deviation, obtain multiple out-of-tolerance items and bind them to the target part for storage, and build a position detection anomaly database.

[0065] In this embodiment, during the position detection process, firstly, deviation analysis is performed by traversing multiple feature relative position relationship graphs, and feature relative position relationships exceeding the tolerance range are extracted as out-of-tolerance items. Then, a part identification code is assigned to the target part based on the part's multi-dimensional information, and the corresponding 3D point cloud data fragments are extracted by further traversing the feature relative position relationship graphs for multiple out-of-tolerance items. Next, multiple out-of-tolerance items and part identification codes are associated and bound according to multiple 3D point cloud data fragments to form a bound data list. Finally, this bound data list is stored in a null-structured database, thereby constructing a position detection anomaly database.

[0066] Furthermore, in the method provided in the application embodiment, the deviation marking is performed by traversing the multiple feature relative position relationship diagrams, multiple out-of-tolerance items are obtained and bound to the target part for storage, and a position detection anomaly database is constructed. The method also includes:

[0067] Deviation analysis is performed by traversing the multiple feature relative position relationship diagrams, and feature relative position relationships that exceed the tolerance range are extracted as out-of-tolerance items, thus identifying multiple out-of-tolerance items; multi-dimensional analysis is performed on the target part, and part identification codes are determined and assigned to the target part based on the multi-dimensional information of the part; multiple three-dimensional point cloud data fragments are extracted by traversing the multiple feature relative position relationship diagrams based on the multiple out-of-tolerance items; the multiple out-of-tolerance items and the part identification codes are associated and bound according to the multiple three-dimensional point cloud data fragments to construct a bound data list; a null value structured database is constructed, and the bound data list is stored in the database to construct the position detection anomaly database.

[0068] In this embodiment, when performing deviation analysis by traversing multiple feature relative position relationship diagrams, the geometric parameters such as relative distance, relative angle, coaxiality, parallelism, and perpendicularity of each feature pair are first compared with the theoretical design parameters. A threshold comparison method is used to determine whether the deviation value is within the tolerance zone. When the deviation value is greater than the upper limit of the tolerance zone or less than the lower limit of the tolerance zone, the relative position relationship of the feature is marked as an out-of-tolerance item, and finally multiple out-of-tolerance items are obtained.

[0069] When performing multidimensional analysis on a target part to assign a part identification code, multidimensional information such as geometric information, material information, and processing information of the part is extracted. This information is then combined using a unique coding method to generate a part identification code, which is then assigned to the target part.

[0070] Next, when extracting multiple 3D point cloud data segments by traversing multiple feature relative position relationship maps based on multiple out-of-tolerance terms, a neighborhood search method is used to set a fixed search radius, with the feature space coordinates corresponding to each out-of-tolerance term as the center, to perform neighborhood retrieval on the 3D point cloud data, and to crop and extract the local point cloud regions related to the out-of-tolerance terms, thereby obtaining multiple 3D point cloud data segments containing deviation features.

[0071] Subsequently, multiple tolerance items are associated and bound to part identification codes according to multiple 3D point cloud data fragments. Using the part identification code as an index, each tolerance item is bound and stored with its corresponding 3D point cloud data fragment, forming a bound data list. Simultaneously, detection path parameters and detection environment parameters are written into the bound data list. The detection path parameters are derived from the control command data when the probe performs the scanning task, including the coordinate sequence of scanning trajectory points, movement speed, and acceleration. The detection environment parameters are derived from the data collected by external environmental sensors of the detection equipment, including temperature, humidity, and vibration amplitude. Through this process, a bound data list containing tolerance items, tolerance types, tolerance values, part identification codes, 3D point cloud data fragments, detection path parameters, and detection environment parameters is obtained.

[0072] Finally, when constructing the null-value structured database, relational database design methods were used to establish the table structure and field constraints. The bound data list was written into the database tables, and indexes were created for the fields to ensure query efficiency and data consistency. Through this process, the location detection anomaly database was completed.

[0073] Step S500: Perform position deviation training based on the position detection anomaly database, and construct a position deviation prediction channel to optimize the detection path.

[0074] In this embodiment, when training for position deviation based on the position detection anomaly database, historical deviation data is first extracted from the database. This historical deviation data includes historical deviation types, historical deviation values, historical probe path parameters, and historical probe environment parameters. Then, the historical probe path parameters and historical probe environment parameters are used as input features, and the historical deviation types and historical deviation values ​​are used as prediction targets for training, thereby constructing a position deviation prediction channel. In a new detection task, the probe path is synchronized to the position deviation prediction channel for prediction, obtaining the position deviation prediction result. Finally, the position deviation prediction result is fed back to the probe path for optimization and adjustment, obtaining an optimized detection path, which is used to drive the coordinate measuring machine to move the probe head to detect the relative position of the target part.

[0075] Furthermore, in the method provided in the application embodiments, the method further includes: training on position deviation based on the position detection anomaly database, constructing a position deviation prediction channel to optimize the detection path; and performing feedback optimization.

[0076] Historical deviation data is extracted from the position detection anomaly database. This historical deviation data includes historical deviation types, historical deviation values, historical detection path parameters, and historical detection environment parameters. The historical detection path parameters and historical detection environment parameters are used as input features, and the historical deviation types and historical deviation values ​​are used as prediction targets for training to construct a position deviation prediction channel. The detection path is synchronized to the position deviation prediction channel for prediction to construct a position deviation prediction result. The position deviation prediction result is fed back to the detection path for optimization and adjustment, resulting in an optimized detection path that drives the coordinate measuring machine to move the probe head to detect the relative position of the target part.

[0077] In this embodiment of the application, historical out-of-tolerance data is first extracted from the location detection anomaly database. The historical out-of-tolerance data includes historical out-of-tolerance type, historical out-of-tolerance value, historical detection path parameters and historical detection environment parameters. At the same time, the numerical fields are standardized using the Z-score standardization method, and the timestamps are aligned to ensure that the historical detection path parameters and historical detection environment parameters correspond to the historical out-of-tolerance type and historical out-of-tolerance value on the same time scale.

[0078] Next, historical detection path parameters and historical detection environment parameters are used as input features, and historical deviation types and historical deviation values ​​are used as prediction targets. The gradient boosting tree method is used for modeling. During training, the training and validation sets are divided according to a fixed ratio. First, a deviation type classifier is trained to output the historical deviation type, and then a deviation value regressor is trained to output the historical deviation value. Key hyperparameters such as learning rate, maximum depth, and number of weak learners are selected based on the performance of the validation set. After training convergence, the classifier and regressor are encapsulated into a position deviation prediction channel using a unified interface.

[0079] Next, the detection path to be executed is synchronized to the position deviation prediction channel for prediction. The position deviation prediction channel receives the detection path parameters and the current detection environment parameters as input, and uses a batch forward inference method to calculate and generate the position deviation prediction results. The prediction results include the deviation type and deviation value corresponding to the trajectory segment or sampling point. For example, under the conditions of a scanning speed of 200 mm / s, a sampling frequency of 100 Hz, and an ambient temperature of 35℃, a certain trajectory segment is predicted to have a coaxiality deviation of 0.03 mm, which is marked in the output to clearly indicate the potentially high-risk trajectory segment.

[0080] Finally, the position deviation prediction results are fed back to the detection path for optimization. In this process, a Bayesian optimization method is employed to reduce the prediction deviation, while considering constraints such as the maximum speed, maximum acceleration, and travel boundaries of the coordinate measuring machine. Optimization variables include trajectory spacing, scanning speed, sampling frequency, and probe attitude angle. The prediction deviation is gradually reduced by iteratively adjusting the path parameters until the optimization objective converges or the maximum number of iterations is reached. The optimized detection path is then used to drive the coordinate measuring machine to move the probe head to detect the relative position of the target part, thereby improving detection accuracy and efficiency.

[0081] In summary, the embodiments of this application have at least the following technical effects:

[0082] This application imports a 3D model of the target part to plan the detection path, constructs a three-axis motion control command to drive a three-coordinate device to move the probe head to scan the target part, and obtains 3D point cloud data; based on the 3D point cloud data, it performs region segmentation and recognition on the 3D model, performs cluster analysis on the recognition results to obtain point cloud clusters of multiple feature points, performs global fitting, and determines multiple feature space coordinates; it constructs a local coordinate system for the part, maps the multiple feature space coordinates to the local coordinate system for position calculation, and obtains multiple feature relative position relationship diagrams; it traverses the multiple feature relative position relationship diagrams to mark deviations, obtains multiple out-of-tolerance items, binds and stores them with the target part, and constructs a position detection anomaly database; it performs position deviation training based on the position detection anomaly database, constructs a position deviation prediction channel, and performs feedback optimization on the detection path. This invention solves the technical problems of insufficient accuracy and low efficiency in the prior art for relative position detection of parts, and achieves the technical effect of improving the accuracy and efficiency of relative position detection of parts through 3D point cloud acquisition, feature recognition, and deviation prediction optimization.

[0083] Example 2, based on the same inventive concept as the part relative position detection method under three-axis movement in the previous examples, such as... Figure 2 As shown, this application provides a part relative position detection system under three-axis movement. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0084] The scanning module 11 is used to import the 3D model of the target part, plan the detection path, construct a three-axis motion control command to drive the three-coordinate device to move the probe head to scan the target part, and obtain 3D point cloud data; the fitting module 12 is used to perform region segmentation and recognition on the 3D model based on the 3D point cloud data, perform cluster analysis on the recognition results, obtain point cloud clusters of multiple feature points for global fitting, and determine multiple feature space coordinates; the calculation module 13 is used to construct a local coordinate system of the part, map the multiple feature space coordinates to the local coordinate system of the part for position calculation, and obtain multiple feature relative position relationship diagrams; the binding and storage module 14 is used to traverse the multiple feature relative position relationship diagrams to mark deviations, obtain multiple out-of-tolerance items and bind and store them with the target part, and construct a position detection anomaly database; the feedback optimization module 15 is used to perform position deviation training based on the position detection anomaly database, construct a position deviation prediction channel to perform feedback optimization on the detection path.

[0085] Furthermore, the system is also used to implement the following functions:

[0086] The 3D point cloud data is mapped to a 3D model for normal vector calculation to construct an initial point cloud normal vector field, which corresponds to the 3D point cloud data. The 3D model is segmented according to the initial reference direction defined by the initial point cloud normal vector field. Unsegmented point clouds are randomly selected as seed points based on the segmentation results. Neighboring points of the seed points are traversed and matched with the initial point cloud normal vector field to extract neighboring point normal vectors. The angle between the neighboring point normal vector and the seed point normal vector is calculated. When the angle is less than a preset angle threshold, the neighboring points are merged into the seed point. The initial point cloud normal vector field is updated based on the neighboring point normal vector and the seed point normal vector to obtain the point cloud normal vector field. Based on the point cloud normal vector field and the segmentation results, the 3D model is repeatedly grown and merged into regions until all unsegmented point clouds are segmented. Multiple segmented regions are identified for geometric recognition to obtain the recognition results.

[0087] Furthermore, the system is also used to implement the following functions:

[0088] Feature point clustering is performed using Euclidean distance traversal recognition results to obtain point cloud clusters of multiple feature points. These point cloud clusters are then geometrically decomposed to obtain hole feature point cloud clusters, planar edge feature point cloud clusters, and curved surface feature point cloud clusters. Cylindrical surface fitting is performed based on the hole feature point cloud clusters to obtain hole geometric feature parameters. Edge line fitting is performed based on the planar edge feature point cloud clusters to obtain planar edge geometric feature parameters. Surface fitting is performed based on the curved surface feature point cloud clusters to obtain curved surface geometric feature parameters. A global coordinate system is constructed based on a 3D model, and the hole geometric feature parameters, planar edge geometric feature parameters, and curved surface geometric feature parameters are transformed into the global coordinate system for identification, thus determining the spatial coordinates of the multiple feature points.

[0089] Furthermore, the system is also used to implement the following functions:

[0090] Based on the 3D model, benchmark analysis is performed to obtain benchmark features. The benchmark features are then mapped to the global coordinate system for identification, and benchmark planes, benchmark axes, and benchmark points are determined. The benchmark plane is used as the coordinate plane, the benchmark axis as the coordinate axis, and the benchmark point as the coordinate origin to construct a local coordinate system for the part. A coordinate transformation matrix is ​​constructed, and the coordinates of the multiple feature spaces are mapped to the local coordinate system of the part according to the coordinate transformation matrix to calculate the position, thereby obtaining a relative position relationship diagram of the multiple features.

[0091] Furthermore, the system is also used to implement the following functions:

[0092] Coordinate data of the reference feature in the global coordinate system are extracted and subjected to coordinate transformation analysis to obtain coordinate transformation requirement parameters, which include rotation matrix parameters and translation vector parameters. The coordinate axes of the local coordinate system are mapped to the global coordinate system according to the rotation matrix parameters to calculate the direction cosine and construct a rotation transformation matrix. The origin of the local coordinate system is mapped to the global coordinate system according to the translation vector parameters to calculate the offset and construct a translation transformation vector. The rotation transformation matrix and the translation transformation vector are combined to construct an initial coordinate transformation matrix. The initial coordinate transformation matrix is ​​verified, and if the verification is successful, the final coordinate transformation matrix is ​​constructed.

[0093] Furthermore, the system is also used to implement the following functions:

[0094] Multiple feature points are randomly selected as verification points based on the local coordinate system; the multiple verification points are mapped to the global coordinate system to obtain multiple initial coordinates; the multiple initial coordinates are transformed according to the initial coordinate transformation matrix to obtain multiple coordinate values; the multiple coordinate values ​​are compared and verified with the theoretical coordinate values, and error analysis is performed based on the verification results to determine the error distribution parameters; the initial coordinate transformation matrix is ​​iteratively corrected according to the error distribution parameters to construct the coordinate transformation matrix.

[0095] Furthermore, the system is also used to implement the following functions:

[0096] Deviation analysis is performed by traversing the multiple feature relative position relationship diagrams, and feature relative position relationships that exceed the tolerance range are extracted as out-of-tolerance items, thus identifying multiple out-of-tolerance items; multi-dimensional analysis is performed on the target part, and part identification codes are determined and assigned to the target part based on the multi-dimensional information of the part; multiple three-dimensional point cloud data fragments are extracted by traversing the multiple feature relative position relationship diagrams based on the multiple out-of-tolerance items; the multiple out-of-tolerance items and the part identification codes are associated and bound according to the multiple three-dimensional point cloud data fragments to construct a bound data list; a null value structured database is constructed, and the bound data list is stored in the database to construct the position detection anomaly database.

[0097] Furthermore, the system is also used to implement the following functions:

[0098] Historical deviation data is extracted from the position detection anomaly database. This historical deviation data includes historical deviation types, historical deviation values, historical detection path parameters, and historical detection environment parameters. The historical detection path parameters and historical detection environment parameters are used as input features, and the historical deviation types and historical deviation values ​​are used as prediction targets for training to construct a position deviation prediction channel. The detection path is synchronized to the position deviation prediction channel for prediction to construct a position deviation prediction result. The position deviation prediction result is fed back to the detection path for optimization and adjustment, resulting in an optimized detection path that drives the coordinate measuring machine to move the probe head to detect the relative position of the target part.

[0099] In Embodiment 3, based on the same inventive concept as the part relative position detection method under three-axis movement in the foregoing embodiments, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of any of the methods described in Embodiment 1 above.

[0100] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0101] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0102] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for detecting the relative position of a part under three-axis movement, characterized in that, The method includes: Import the 3D model of the target part to plan the detection path, construct a three-axis motion control command to drive the three-coordinate device to move the probe head to scan the target part and obtain 3D point cloud data; Based on the three-dimensional point cloud data, the three-dimensional model is segmented and identified. The identification results are then subjected to cluster analysis to obtain point cloud clusters of multiple feature points. Global fitting is then performed to determine the spatial coordinates of multiple features. The process of performing cluster analysis on the recognition results to obtain point cloud clusters of multiple feature points, performing global fitting, and determining multiple feature space coordinates includes: The Euclidean distance traversal recognition results are used to cluster feature points to obtain point cloud clusters of multiple feature points; Geometrically decompose point cloud clusters with multiple feature points to obtain point cloud clusters with hole features, point cloud clusters with planar edge features, and point cloud clusters with curved surface features. Based on the hole feature point cloud cluster, a cylindrical surface is fitted to obtain the hole geometric feature parameters; Edge line fitting is performed based on the planar edge feature point cloud cluster to obtain the planar edge geometric feature parameters; Surface fitting is performed based on the surface feature point cloud cluster to obtain surface geometric feature parameters; A global coordinate system is constructed based on the 3D model. The geometric feature parameters of the hole, the geometric feature parameters of the planar edge, and the geometric feature parameters of the curved surface are transformed into the global coordinate system for identification, thereby determining the coordinates of the multiple feature spaces. A local coordinate system for the part is constructed, and the spatial coordinates of the multiple features are mapped to the local coordinate system for position calculation to obtain a relative position relationship diagram of multiple features, including: Based on the 3D model, benchmark analysis is performed to obtain benchmark features. The benchmark features are then mapped to the global coordinate system for identification, thereby determining the benchmark plane, benchmark axis, and benchmark point. A local coordinate system for the part is constructed by using the reference plane as the coordinate plane, the reference axis as the coordinate axis, and the reference point as the origin. Construct a coordinate transformation matrix, and map the spatial coordinates of the multiple features to the local coordinate system of the part according to the coordinate transformation matrix to perform position calculation, thereby obtaining a relative position relationship diagram of the multiple features; By traversing the relative position relationship diagrams of the multiple features, deviation marking is performed, and multiple out-of-tolerance items are obtained and bound to the target part for storage, thus constructing a position detection anomaly database; The location deviation is trained based on the location detection anomaly database, and a location deviation prediction channel is constructed to provide feedback optimization for the detection path.

2. The method for detecting the relative position of a part under three-axis movement as described in claim 1, characterized in that, The method for performing region segmentation and recognition of a 3D model based on the aforementioned 3D point cloud data includes: The three-dimensional point cloud data is mapped to a three-dimensional model to calculate the normal vectors, and an initial point cloud normal vector field is constructed. The initial point cloud normal vector field has a corresponding relationship with the three-dimensional point cloud data. The three-dimensional model is segmented according to the initial reference direction defined by the initial point cloud normal vector field, and the unsegmented point cloud is randomly selected as seed point based on the segmentation result. The neighboring points of the seed point are traversed and matched with the initial point cloud normal vector field to extract the normal vectors of the neighboring points and calculate the angle data between the normal vectors of the neighboring points and the normal vector of the seed point. When the included angle data is less than a preset angle threshold, the neighboring points are merged into the seed point, and the initial point cloud normal vector field is updated according to the normal vectors of the neighboring points and the normal vectors of the seed point to obtain the point cloud normal vector field. Based on the point cloud normal vector field and the segmentation results, the 3D model is repeatedly grown and incorporated into the region until all unsegmented point clouds are segmented. Multiple segmented regions are then identified for geometric recognition to obtain the recognition results.

3. The method for detecting the relative position of a part under three-axis movement as described in claim 1, characterized in that, The process of constructing the coordinate transformation matrix includes the following methods: The coordinate data of the reference feature in the global coordinate system are extracted and subjected to coordinate transformation analysis to obtain the coordinate transformation requirement parameters, which include rotation matrix parameters and translation vector parameters. The coordinate axes of the local coordinate system are mapped to the global coordinate system according to the rotation matrix parameters, and the direction cosines are calculated to construct the rotation transformation matrix. The local coordinate system origin is mapped to the global coordinate system according to the translation vector parameters to calculate the offset and construct the translation transformation vector. The rotation transformation matrix and the translation transformation vector are combined to construct an initial coordinate transformation matrix. The initial coordinate transformation matrix is ​​verified. When the verification is successful, the coordinate transformation matrix is ​​constructed.

4. The method for detecting the relative position of a part under three-axis movement as described in claim 3, characterized in that, Verify the initial coordinate transformation matrix. If the verification passes, construct the coordinate transformation matrix. The method includes: Multiple feature points are randomly selected as multiple verification points based on the local coordinate system; The multiple verification points are mapped to the global coordinate system to obtain multiple initial coordinates; The multiple initial coordinates are transformed and calculated according to the initial coordinate transformation matrix to obtain multiple coordinate values; The multiple coordinate values ​​are compared and verified with the theoretical coordinate values. Based on the verification results, error analysis is performed to determine the error distribution parameters. The coordinate transformation matrix is ​​constructed by iteratively correcting the initial coordinate transformation matrix according to the error distribution parameters.

5. The method for detecting the relative position of a part under three-axis movement as described in claim 1, characterized in that, The method involves iterating through the relative positional relationship diagrams of multiple features, marking deviations, obtaining multiple out-of-tolerance items, binding and storing them with the target part, and constructing a position detection anomaly database. The method includes: By traversing the multiple feature relative position relationship diagrams, deviation analysis is performed, and feature relative position relationships that exceed the tolerance range are extracted as out-of-tolerance items, thus identifying multiple out-of-tolerance items. Perform multidimensional analysis on the target part, and determine the part identification code to be assigned to the target part based on the multidimensional information of the part; Based on the multiple out-of-range terms, multiple 3D point cloud data fragments are extracted by traversing the multiple feature relative position relationship graphs; The multiple out-of-tolerance items and the part identification code are associated and bound according to the multiple three-dimensional point cloud data fragments to construct a bound data list; Construct a null-structured database, store the bound data list in the database, and build the location detection anomaly database.

6. The method for detecting the relative position of a part under three-axis movement as described in claim 1, characterized in that, The method includes training on position deviation based on the aforementioned position detection anomaly database, constructing a position deviation prediction channel to optimize the detection path, and the following steps: Historical out-of-tolerance data is extracted based on the location detection anomaly database. The historical out-of-tolerance data includes historical out-of-tolerance type, historical out-of-tolerance value, historical detection path parameters, and historical detection environment parameters. The historical detection path parameters and historical detection environment parameters are used as input features, and the historical deviation types and historical deviation values ​​are used as prediction targets for training to construct a position deviation prediction channel. The detection path is synchronized to the position deviation prediction channel for prediction, and a position deviation prediction result is constructed. The position deviation prediction result is fed back to the detection path for optimization and adjustment. The optimized detection path drives the coordinate measuring machine to move the probe head to detect the relative position of the target part.

7. A part relative position detection system under three-axis movement, characterized in that, The system is used to perform the part relative position detection method under three-axis movement as described in any one of claims 1-6, and the system includes: The scanning module is used to import the 3D model of the target part, plan the detection path, and construct three-axis motion control commands to drive the three-coordinate device to move the probe head to scan the target part and obtain 3D point cloud data. The fitting module is used to perform region segmentation and recognition on the 3D model based on the 3D point cloud data, perform cluster analysis on the recognition results, obtain point cloud clusters of multiple feature points for global fitting, and determine multiple feature space coordinates. The calculation module is used to construct a local coordinate system for the part, map the spatial coordinates of the multiple features to the local coordinate system for position calculation, and obtain a relative position relationship diagram of the multiple features. The binding and storage module is used to traverse the relative position relationship diagram of the multiple features, mark the deviation, obtain multiple out-of-tolerance items and bind and store them with the target part, and build a position detection anomaly database. The feedback optimization module is used to train the position deviation based on the position detection anomaly database and construct a position deviation prediction channel to perform feedback optimization on the detection path.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the method for detecting the relative position of a part under three-axis movement as described in any one of claims 1-6.

Images