Intelligent positioning system and method for numerical control machining of mold accessories

Abstract

The invention discloses an intelligent positioning system and method for numerical control machining of a mold accessory, and the method comprises the steps: collecting a plurality of groups of three-dimensional feature points of the same mold accessory in different clamping states, and constructing unified feature point representation; uniformizing the feature points and establishing a cross-clamping uniform shape space; mapping the spatial relationship under different clamping states into continuous pose samples, and forming a sliding window according to a time sequence; performing group consistency calculation on poses in the window under a uniform space structure to obtain stable consensus poses; and dynamically updating a mapping relation between a machine tool coordinate system and a workpiece coordinate system based on the consensus pose, and outputting updated positioning parameters to a numerical control system to realize automatic positioning and coordinate self-correction under the condition of multiple times of clamping.

Description

Technical Field

[0001] This invention relates to the fields of CNC machining and industrial measurement technology, and in particular to an intelligent positioning system and method for CNC machining of mold parts. Background Technology

[0002] As mold manufacturing evolves towards higher precision, greater consistency, and automation, CNC machine tools typically require multiple clamping and process switching during mold component machining to meet the requirements of machining complex curved surfaces, precision holes, and assembly datums. To ensure machining accuracy, it is usually necessary to re-establish the spatial relationship between the machine tool coordinate system and the workpiece coordinate system after each clamping. Existing technologies largely rely on manual alignment, single-measurement positioning, or coordinate setting methods based on fixed datums, combined with probes or measurement programs to complete workpiece positioning and coordinate correction.

[0003] Existing CNC machining positioning technologies still have significant shortcomings in scenarios involving multiple clamping and repetitive positioning. On the one hand, traditional positioning methods typically treat each clamping as an independent process, lacking consistent modeling of the spatial relationships of the same workpiece under different clamping states. This makes them prone to coordinate drift accumulation due to measurement noise, clamping errors, or differences in point selection, affecting overall machining accuracy. On the other hand, existing technologies often simply average pose results in Euclidean space or directly select single measurement results, failing to fully consider the geometric constraints of rigid body poses under spatial group structures. This results in positioning results that are sensitive to abnormal measurements and lack stability, making it difficult to achieve automatic correction and reliable updates during continuous machining. These deficiencies are particularly prominent in the multiple clamping machining of high-precision mold parts, hindering the intelligent and consistent control of the CNC machining process.

[0004] Therefore, how to provide an intelligent positioning system and method for CNC machining of mold parts is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention proposes an intelligent positioning system and method for CNC machining of mold parts. By unifying the three-dimensional feature points collected from the same mold part under different clamping states, a unified shape space across clamping is constructed. Based on this, the spatial relationship between clamping is mapped into pose samples. A stable consensus pose is obtained by using a consistent pose solution mechanism on a group manifold. It has the advantages of high positioning stability, strong resistance to measurement errors, and adaptability to multiple clamping machining.

[0006] An intelligent positioning system and method for CNC machining of mold parts according to an embodiment of the present invention includes the following steps: The multi-clamping feature point acquisition module collects the three-dimensional feature point set of the same mold part under different clamping states, and establishes a corresponding clamping instance identifier and acquisition time sequence identifier for each set of three-dimensional feature points; The point set unification processing module receives a set of three-dimensional feature points with clamping instance identifiers, performs point order unification and rigid body constraint preprocessing on the three-dimensional feature point set, and generates a standard feature point set. The shape space unified modeling module receives a standard feature point set, constructs a unified shape space across clamping instances based on generalized Procrustes alignment, and outputs the rigid body transformation results of each clamping instance relative to the unified shape space. The pose sample mapping module maps the rigid body transformation results to pose samples defined on the SE(3) Lie group, and organizes the pose samples into a sliding window pose set based on the acquisition timing identifier. The consensus pose solving module performs Karcher mean calculation on the sliding window pose set on the SE(3) group manifold using geodesic distance as the metric to generate the corresponding consensus pose. The coordinate mapping update module updates the mapping relationship between the machine tool coordinate system and the workpiece coordinate system based on consensus pose, and outputs the updated positioning parameters to the CNC system.

[0007] Optionally, modules can be integrated using the following methods: Under different clamping conditions, the same mold component is measured multiple times to collect multiple sets of three-dimensional feature points, and clamping instance identifier and collection time sequence identifier corresponding to the clamping condition are established for each set of three-dimensional feature points. Perform point order unification and rigid body constraint preprocessing on the 3D feature point set with clamping instance identifiers to generate a standard feature point set for cross-clamping comparison; Based on the generalized Procrustes alignment, the standard feature point set is iteratively processed to construct a unified shape space across clamping instances, and the rigid body transformation results of each clamping instance relative to the unified shape space are obtained. The rigid body transformation results are mapped to pose samples defined on the SE(3) Lie group, and the pose samples are organized into a sliding window pose set according to the acquisition timing identifier. On the SE(3) group manifold, the geodesic distance between pose samples is used as a metric to perform Karcher mean calculation on the sliding window pose set to obtain the consensus pose corresponding to the current window; The mapping relationship between the machine tool coordinate system and the workpiece coordinate system is updated based on consensus pose, and the updated positioning parameters are output.

[0008] Optionally, establishing a clamping instance identifier and acquisition timing identifier corresponding to the clamping state includes: After the mold parts are clamped once, read the fixture number, fixture installation posture parameters, positioning element number and its position parameters in the machine tool coordinate system, and obtain the current clamping cycle count value. The fixture number, fixture installation posture parameters, positioning element number and its position parameters are combined with the clamping cycle count to form a clamping configuration data group, and the clamping configuration data group is subjected to consistent sorting and standardization processing. Based on the standardized clamping configuration data set, a unique clamping instance identifier corresponding to the current clamping configuration is generated, and a binding relationship is established between the clamping instance identifier and the current clamping configuration. Each time a 3D feature point acquisition operation is triggered, the system timestamp corresponding to the acquisition trigger time is read, and the program segment number and measurement instruction number currently being executed in the CNC system are read simultaneously. The collection sequence number is initialized during the first collection, and then incremented after each subsequent collection to form a collection sequence number sequence that is consistent with the collection order. The system timestamp, program segment number, measurement command number, and acquisition sequence number are combined to generate an acquisition timing identifier.

[0009] Optionally, generating a standard feature point set includes: Receive a set of three-dimensional feature points carrying a clamping instance identifier and a collection timing identifier, read the three-dimensional coordinates and feature point type identifier of each feature point, count the number of points of each type according to the feature point type identifier, compare the statistical results with the target point count table of the corresponding mold part in the configuration file item by item, and data batches that fail the comparison will not enter the point sequence unification process. In the batches of data that have been compared, a set of reference points of the same type is selected according to the priority of the feature point type identifier, and the first reference point is selected from the set of reference points of the same type. The coordinate point with the smallest lexicographical order; by Select a second reference point as the starting point To and The point with the largest Euclidean distance; select a third reference point from the remaining points. In order to make The point with the largest modulus; based on Establish a rigid body reference coordinate system and calculate the unit basis vectors; For each feature point in the three-dimensional feature point set Perform rigid body constraint preprocessing and calculate standardized coordinates. ; In a rigid body reference coordinate system, point order unification is performed, and all feature points are sorted in ascending order by feature point type identifier. of Component ascending order—plane polar angle Ascending order — plane radius Sort the data according to the "ascending order" rule and assign point numbers accordingly; Point sequence number, corresponding standardized coordinates The clamping instance identifier and the acquisition time sequence identifier are combined to generate a standard feature point set.

[0010] Optionally, the rigid body transformation results include: Receive standard feature point sets, and construct a point set sequence by combining the standard feature point sets formed by the same mold part under different clamping instances according to the clamping instance identifier, while keeping the point sequence number consistent within the point set sequence; Input the point set sequence into the generalized Procrustes alignment process, set the initial template of the uniform shape point set to any standard feature point set in the point set sequence or the initial mean point set obtained from the point set sequence, call the generalized Procrustes alignment process to output the uniform shape point set and the rigid body transformation result corresponding one-to-one with each clamping instance identifier in the point set sequence. For each clamping instance identifier in the point set sequence, the standard feature point set corresponding to the clamping instance is rotated and translated according to the rigid body transformation result to obtain the alignment point set, and the point sequence number is retained in the alignment point set so that the alignment point set and the uniform shape point set form a one-to-one correspondence in the point sequence number dimension. Define the uniform shape point set as a uniform shape space representation across clamping instances, and define the rigid body transformation result as the pose mapping parameters from clamping instances to the uniform shape space.

[0011] Optional, the generalized Procrustes alignment process includes: Let the number of point sets in the point set sequence be . The number of feature points in each point set is , No. The set of points is denoted as ,in This is the clamping instance number. For point sequence numbering, The point sequence number is 3D coordinate vector; For each candidate point set in the point set sequence Calculate the dispersion score and select the candidate point set with the smallest dispersion score as the initial template point set. The dispersion score is calculated using the following formula: ; in, For Dispersion score when used as a candidate template The index of the set of points to be aligned. To be Mapped to via a rigid body alignment transformation The next Alignment point coordinates, It is the Euclidean norm; In a rigid body alignment transformation calculation, the point sequence number is used as the basis for... and Pair up points one by one and calculate the geometric centers of the two point sets respectively. and The two point sets are centered separately; Calculate the correlation matrix based on the centralized point-by-point pairing results. And perform singular value decomposition on the correlation matrix to obtain an orthogonal matrix. ; Calculate the rotation matrix for ,when When the determinant is less than 0, The rotation matrix is ​​recalculated after taking the opposite of the third column; Calculate the translation vector for ; With the initial template point set Enter the iterative alignment process, let the first... The template point set for the round iteration is = ,in This represents the iteration round number. Let (i) be the three-dimensional coordinate vector of point number (i) in the template point set; For each iteration For each clamping instance point set Repeat the rigid body alignment transformation calculation to obtain the result relative to... rotation matrix With translation vector And perform rotation and translation on all points in the point set to obtain the aligned point set. ; Calculate the alignment residual for each clamping instance point set in the current round. Record the correspondence between residual values ​​and clamping instance identifiers; The entire set of alignment points will be sorted according to the clamping instance identifier. Divided into preset quantities There are non-overlapping groups, each containing the same number of alignment point sets; For each group and the sequence number of each point Calculate the group mean point ; Sequential numbering of the same point All group mean points Calculate the median values ​​of the coordinate components, respectively for , , The median value of the three coordinate components is taken and used to form the coordinates of the new template point. ; All point sequence numbers correspond to Combine to obtain the updated template point set ; Calculate the change in template point sets between two adjacent rounds. And take the maximum value as ; when The iteration terminates when the value is not greater than a preset threshold or the number of iterations reaches a preset upper limit. When the iteration terminates, output the final template point set. As a unified shape space representation across clamping instances; output the rigid body transformation result corresponding to each clamping instance identifier, the rigid body transformation result includes the rotation matrix and translation vector obtained from the last alignment in each iteration.

[0012] Optionally, generating the sliding window pose set includes: Receive the rigid body transformation results corresponding to each clamping instance identifier; Perform singular value decomposition on the rotation matrix, extract the two orthogonal matrices from the decomposition result, and multiply them to generate a new rotation matrix; When the determinant of the generated rotation matrix is ​​negative, take the opposite of the third column of an orthogonal matrix in the singular value decomposition result and multiply them again to obtain a rotation matrix with a positive determinant. Constructing the pose matrix involves creating a 4x4 matrix, writing the orthogonalized rotation matrix into the upper left 3x3 region of the 4x4 matrix, writing the 3D translation vector into the upper right 3x1 region of the 4x4 matrix, and writing the last four elements of the 4x4 matrix as "0, 0, 0, 1" in sequence. Perform SE(3) structure verification on the constructed 4x4 matrix; The pose samples that pass the SE(3) structure verification are written into the pose sample library along with the corresponding clamping instance identifier and acquisition timing identifier to form pose record entries. The pose record entries are sorted according to the acquisition sequence number in the acquisition time identifier to generate a pose sample sequence arranged in the acquisition order; When multiple pose record entries with the same acquisition sequence number exist, only the pose record entry with the newest system timestamp is retained; Set the sliding window length to a fixed value, and select consecutive pose record entries from the pose sample sequence in sequence to form a sliding window pose set; After a new pose record entry is written into the pose sample sequence, the new pose record entry is added to the end of the sliding window pose set, and the earliest written pose record entry is removed from the sliding window pose set to obtain the updated sliding window pose set.

[0013] Optional, the Karcher mean calculation includes: Receive the pose set of the sliding window, and read the poses within the window sorted by acquisition sequence number. There are several pose samples, each pose sample is a 4x4 homogeneous transformation matrix, and the upper left 3x3 is the rotation submatrix and the upper right 3x1 is the translation subvector. Using the pose sample with the center number collected in the sliding window as the reference sample, the relative poses of the other pose samples in the window are calculated respectively. Perform a logarithmic mapping once for each relative pose to obtain a six-dimensional increment vector. Take the median value of each component of all six-dimensional increment vectors in the window to form an initial increment vector. Perform an exponential mapping on the initial increment vector to generate the increment pose, and then multiply it with the reference sample to obtain the initial consensus pose value. Using the initial consensus pose value as the starting point for iteration, we enter the Karcher mean iteration, setting the maximum number of iterations and the stopping threshold. In each iteration, geodesic increment calculation is performed for each pose sample within the window: Using the current consensus pose as a reference, invert the consensus pose and multiply it with the pose sample to obtain the relative pose matrix; Extract the rotation submatrix and translation subvector from the relative pose matrix, and calculate the rotation logarithm of the rotation submatrix: Calculate the rotation angle, which is determined by the trace of the rotation submatrix; When the angle approaches zero, set the rotation increment to the zero vector; When the angle is not close to zero, the unit rotation axis is calculated based on the antisymmetric part of the rotation submatrix, and the rotation axis is multiplied by the angle to obtain the three-dimensional rotation increment vector; Generate the corresponding antisymmetric matrix based on the three-dimensional rotation increment vector; The 3x3 matrix required to calculate the logarithm of translation The It is generated by a three-term expansion of the antisymmetric matrix of the rotation increment; right After inverting the vector, multiply it by the translation sub-vector of the relative pose to obtain the three-dimensional translation increment vector; The six-dimensional geodesic increment vector of the pose sample is obtained by concatenating the three-dimensional rotation increment vector and the three-dimensional translation increment vector in a fixed order. Based on the acquisition sequence number, the pose samples within the sliding window are divided into time-ordered categories. Continuous segmentation, in which And each segment contains the same number of pose samples; Within each segment, the arithmetic mean of all six-dimensional geodesic increment vectors of the segment is taken to obtain the segmented average increment vector; The arithmetic mean of all segmented average increment vectors is then calculated to obtain the global update increment vector for the current round. Perform an exponential mapping on the global update increment vector to generate the increment pose, and then multiply it by the current consensus pose to obtain the next consensus pose. Calculate the stopping condition: terminate the iteration when the magnitude of the global update increment vector is not greater than the stopping threshold or the maximum number of iterations is reached. When the stopping condition is met, the current round consensus pose is output as the Karcher mean consensus pose corresponding to the current sliding window, and the consensus pose is associated with the start and end collection sequence number of the sliding window and then stored.

[0014] The beneficial effects of this invention are: This invention collects three-dimensional feature points of the same mold component under multiple clamping conditions, and constructs a unified shape space across clamping conditions using a uniformization process and a generalized Procrustes alignment method. This allows the geometric information of the workpiece under different clamping conditions to be strictly aligned and compared in a unified space, thereby effectively reducing the impact of clamping errors, measurement noise, and differences in point selection on the positioning results, and significantly improving the overall positioning consistency and stability during multiple clamping processes.

[0015] This invention maps the spatial relationship corresponding to the clamping instance to a pose sample defined on the rigid body pose group, and calculates the consensus pose based on the geodesic distance using the Karcher mean within the sliding window. This avoids the destruction of rigid body geometric constraints by the traditional Euclidean averaging method, making the pose fusion result more robust to abnormal measurements, thereby realizing reliable and adaptive updating of the relationship between the machine tool coordinate system and the workpiece coordinate system during the machining process. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart below shows a CNC machining intelligent positioning system and method for mold parts proposed in this invention. Figure 2 This is a schematic diagram of the process based on generalized Procrustes proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1 - Figure 2A CNC machining intelligent positioning system and method for mold parts, comprising the following steps: The multi-clamping feature point acquisition module collects multiple sets of three-dimensional feature points of the same mold part under different clamping states, and establishes a corresponding clamping instance identifier and acquisition time sequence identifier for each set of three-dimensional feature points. The point set unification processing module receives a set of three-dimensional feature points with clamping instance identifiers, performs point order unification and rigid body constraint preprocessing on the three-dimensional feature point set, and generates a comparable standard feature point set. The shape space unified modeling module receives a standard feature point set, constructs a unified shape space across clamping instances based on generalized Procrustes alignment, and outputs the rigid body transformation results of each clamping instance relative to the unified shape space. The pose sample mapping module maps the rigid body transformation results to pose samples defined on the SE(3) Lie group, and organizes the pose samples into a sliding window pose set based on the acquisition timing identifier. The consensus pose solving module performs Karcher mean calculation on the sliding window pose set on the SE(3) group manifold using geodesic distance as the metric to generate the corresponding consensus pose. The coordinate mapping update module updates the mapping relationship between the machine tool coordinate system and the workpiece coordinate system based on consensus pose, and outputs the updated positioning parameters to the CNC system.

[0019] In this embodiment, the modules are interconnected using the following method: Under different clamping conditions, the same mold component is measured multiple times to collect multiple sets of three-dimensional feature points, and clamping instance identifier and collection time sequence identifier corresponding to the clamping condition are established for each set of three-dimensional feature points. Perform point order unification and rigid body constraint preprocessing on the 3D feature point set with clamping instance identifiers to generate a standard feature point set for cross-clamping comparison; Based on the generalized Procrustes alignment, the standard feature point set is iteratively processed to construct a unified shape space across clamping instances, and the rigid body transformation results of each clamping instance relative to the unified shape space are obtained. The rigid body transformation results are mapped to pose samples defined on the SE(3) Lie group, and the pose samples are organized into a sliding window pose set according to the acquisition timing identifier. On the SE(3) group manifold, the geodesic distance between pose samples is used as a metric to perform Karcher mean calculation on the sliding window pose set to obtain the consensus pose corresponding to the current window; The mapping relationship between the machine tool coordinate system and the workpiece coordinate system is updated based on consensus pose, and the updated positioning parameters are output.

[0020] In this embodiment, establishing a clamping instance identifier and a data acquisition timing identifier corresponding to the clamping state includes: After the mold parts are clamped once, read the fixture number, fixture installation posture parameters, positioning element number and its position parameters in the machine tool coordinate system, and obtain the current clamping cycle count value. The fixture number, fixture installation posture parameters, positioning element number and its position parameters are combined with the clamping cycle count to form a clamping configuration data group, and the clamping configuration data group is subjected to consistent sorting and standardization processing. Based on the standardized clamping configuration data set, a unique clamping instance identifier corresponding to the current clamping configuration is generated, and a binding relationship is established between the clamping instance identifier and the current clamping configuration. Each time a 3D feature point acquisition operation is triggered, the system timestamp corresponding to the acquisition trigger time is read, and the program segment number and measurement instruction number currently being executed in the CNC system are read simultaneously. The collection sequence number is initialized during the first collection, and then incremented after each subsequent collection to form a collection sequence number sequence that is consistent with the collection order. The system timestamp, program segment number, measurement instruction number and acquisition sequence number are combined to generate an acquisition timing identifier. The acquisition timing identifier is then stored after establishing a one-to-one correspondence with the corresponding 3D feature point set and clamping instance identifier.

[0021] In this embodiment, generating a standard feature point set for cross-clamp comparison includes: Receive a set of three-dimensional feature points carrying a clamping instance identifier and a collection timing identifier, read the three-dimensional coordinates and feature point type identifier of each feature point, count the number of points of each type according to the feature point type identifier, compare the statistical results with the target point count table of the corresponding mold part in the configuration file item by item, and data batches that fail the comparison will not enter the point sequence unification process. In the batches of data that have been compared, a set of reference points of the same type is selected according to the priority of the feature point type identifier, and the first reference point is selected from the set of reference points of the same type. The lexicographically smallest coordinate point; with Select a second reference point as the starting point To and The point with the largest Euclidean distance; select a third reference point from the remaining points. In order to make The point with the largest modulus; based on Establish a rigid body reference coordinate system and calculate the unit basis vectors: ; ; ; in, The selected three-dimensional reference point coordinate vector, It is the Euclidean norm. For vector cross product, is the unit axis vector of the rigid body reference coordinate system; For each feature point in the three-dimensional feature point set Perform rigid body constraint preprocessing and calculate the normalized coordinates using the following formula. : ; in, For the first The three-dimensional coordinate vector of each feature point For the first The three-dimensional coordinate vectors of each feature point in the rigid body reference coordinate system. This is a transpose operation; In a rigid body reference coordinate system, point order unification is performed, and all feature points are sorted in ascending order by feature point type identifier. of Component ascending order—plane polar angle Ascending order — plane radius Sort the data according to the "ascending order" rule and assign point numbers accordingly; Point sequence number, corresponding standardized coordinates The combination of clamping instance identifier and acquisition time sequence identifier generates a standard feature point set for cross-clamping comparison. In this embodiment, the rigid body transformation result is constructed as follows: Receive the generated standard feature point set, and form a point set sequence by combining multiple standard feature point sets formed by the same mold part under different clamping instances according to the clamping instance identifier, and keep the point sequence number consistent within the point set sequence, so that there is a one-to-one correspondence between any two standard feature point sets within the point set sequence based on the point sequence number. Input the point set sequence into the generalized Procrustes alignment process, set the initial template of the uniform shape point set to any standard feature point set in the point set sequence or the initial mean point set obtained from the point set sequence, call the generalized Procrustes alignment process to output the uniform shape point set and the rigid body transformation result corresponding one-to-one with each clamping instance identifier in the point set sequence. For each clamping instance identifier in the point set sequence, the standard feature point set corresponding to the clamping instance is rotated and translated according to the rigid body transformation result to obtain the alignment point set, and the point sequence number is retained in the alignment point set so that the alignment point set and the uniform shape point set form a one-to-one correspondence in the point sequence number dimension. The unified shape point set is defined as a unified shape space representation across clamping instances, the rigid body transformation result is defined as the pose mapping parameter from the clamping instance to the unified shape space, and the unified shape point set, clamping instance identifier, rigid body transformation result and alignment point set are associated and stored so that subsequent steps can directly retrieve the corresponding pose mapping parameter based on the clamping instance identifier.

[0022] In this embodiment, the generalized Procrustes alignment process includes: Receive a point set sequence, which contains multiple standard feature point sets corresponding to each clamping instance identifier. Each standard feature point set contains three-dimensional normalized coordinates with consistent point sequence numbers. Let the number of point sets in the point set sequence be... The number of feature points in each point set is , No. The set of points is denoted as ,in This is the clamping instance number. For point sequence numbering, The point sequence number is 3D coordinate vector; For each candidate point set in the point set sequence Calculate the dispersion score and select the candidate point set with the smallest dispersion score as the initial template point set. The dispersion score is calculated using the following formula: ; in, For Dispersion score when used as a candidate template The index of the set of points to be aligned. To be Mapped to via the rigid body alignment transformation described above The next Alignment point coordinates, It is the Euclidean norm; In a rigid body alignment transformation calculation, the point sequence number is used as the basis for... and Pair up points one by one and calculate the geometric centers of the two point sets respectively. and The geometric center is calculated by considering all points within the point set. Calculate the arithmetic mean of the coordinates of the points; The two point sets are centered by translating each point set by its respective geometric center; a 3x3 correlation matrix is ​​then calculated based on the point-by-point pairing results after centeredness. The correlation matrix is ​​calculated by summing the outermost coordinate vectors of the aligned points and the template points for all points in sequence numbering. Performing singular value decomposition on the correlation matrix yields an orthogonal matrix. Calculate the rotation matrix for ,when When the determinant is less than 0, Recalculate the rotation matrix after taking the opposite of the third column; calculate the translation vector. for ; With the initial template point set Enter the iterative alignment process, let the first... The template point set for the round iteration is = ,in This represents the iteration round number. Let (i) be the three-dimensional coordinate vector of point number (i) in the template point set; For each iteration For each clamping instance point set Repeat the rigid body alignment transformation calculation to obtain the result relative to... rotation matrix With translation vector And perform rotation and translation on all points in the point set to obtain the aligned point set. The alignment point coordinates are calculated as "for the first "First multiply each point by the rotation matrix and then add the translation vector"; Calculate the alignment residual for each clamping instance point set in the current round. The alignment residual is calculated as follows: number all points sequentially. Calculate separately and The square of the Euclidean norm of the difference vector, and for arrive The residual values ​​are accumulated to obtain the residual values; the correspondence between the residual values ​​and the clamping instance identifiers is recorded. The entire set of alignment points will be sorted according to the clamping instance identifier. Divided into preset quantities There are non-overlapping groups, each containing the same number of alignment point sets; For each group and the sequence number of each point Calculate the group mean point The group mean is calculated by sequentially numbering all aligned points within the group. Calculate the arithmetic mean of the three-dimensional coordinates at the location; Sequential numbering of the same point All group mean points Calculate the median values ​​of the coordinate components, respectively for , , The median value of the three coordinate components is taken and used to form the coordinates of the new template point. ; All point sequence numbers corresponding to Combine to obtain the updated template point set ; Calculate the change in template point sets between two adjacent rounds. The change is calculated by numbering all points sequentially. calculate The Euclidean norm of the given information is determined, and the maximum value is taken as the standard value. ;when The iteration terminates when the value is not greater than a preset threshold or the number of iterations reaches a preset upper limit. When the iteration terminates, output the final template point set. As a unified shape space representation of cross-clamping instances; Output the rigid body transformation result corresponding to each clamping instance identifier. The rigid body transformation result includes the rotation matrix and translation vector obtained from the last alignment in each iteration. Output the alignment residual record corresponding to each clamping instance identifier.

[0023] In this embodiment, generating the sliding window pose set includes: Receive the rigid body transformation results corresponding to each clamping instance identifier. The rigid body transformation results include at least a 3x3 rotation matrix and a three-dimensional translation vector. Simultaneously read the acquisition timing identifier corresponding to the rigid body transformation results. The rotation matrix is ​​orthogonalized by performing singular value decomposition on it, extracting two orthogonal matrices from the decomposition results and multiplying them to generate a new rotation matrix. When the determinant of the generated rotation matrix is ​​negative, the third column of one of the orthogonal matrices in the singular value decomposition results is taken as the opposite and then multiplied again to obtain a rotation matrix with a positive determinant. Constructing the pose matrix involves creating a 4x4 matrix, writing the orthogonalized rotation matrix into the upper left 3x3 region of the 4x4 matrix, writing the 3D translation vector into the upper right 3x1 region of the 4x4 matrix, and writing the last four elements of the 4x4 matrix as "0, 0, 0, 1" in sequence. Perform SE(3) structure verification on the constructed 4x4 matrix, specifically including: Check if the last row of the four-by-four matrix is ​​"0, 0, 0, 1"; Check the result of multiplying the top left 3x3 rotation submatrix by its transpose with the 3x3 identity matrix element by element, and the maximum absolute difference is not greater than a preset threshold. Check if the determinant of the rotated submatrix is ​​positive and if the absolute value of the difference between it and 1 is not greater than a preset threshold. If all the above checks pass, the four-by-four matrix is ​​determined to be a pose sample defined on the SE(3) Lie group; The pose samples that pass the SE(3) structure verification are written into the pose sample library along with the corresponding clamping instance identifier and acquisition timing identifier to form pose record entries. The pose record entries are sorted according to the acquisition sequence number in the acquisition time identifier to generate a pose sample sequence arranged in the acquisition order; When multiple pose record entries with the same acquisition sequence number exist, only the pose record entry with the newest system timestamp is retained; Set the sliding window length to a fixed value, and select consecutive pose record entries from the pose sample sequence in sequence to form a sliding window pose set; After a new pose record entry is written into the pose sample sequence, the new pose record entry is added to the end of the sliding window pose set, and the earliest written pose record entry is removed from the sliding window pose set to obtain the updated sliding window pose set.

[0024] In this embodiment, the Karcher mean calculation includes: Receive the pose set of the sliding window, and read the poses within the window sorted by acquisition sequence number. There are several pose samples, each pose sample is a 4x4 homogeneous transformation matrix, and the upper left 3x3 is the rotation submatrix and the upper right 3x1 is the translation subvector. Using the pose sample with the center number collected in the sliding window as the reference sample, the relative pose of the other pose samples in the window is calculated. The calculation method is: the inverse of the reference sample is multiplied by the target sample. Perform a logarithmic mapping once for each relative pose to obtain a six-dimensional increment vector. Take the median value of each component of all six-dimensional increment vectors in the window to form an initial increment vector. Perform an exponential mapping on the initial increment vector to generate the increment pose, and then multiply it with the reference sample to obtain the initial consensus pose value. Using the initial consensus pose value as the starting point for iteration, we enter the Karcher mean iteration, setting the maximum number of iterations and the stopping threshold. In each iteration, geodesic increment calculation is performed for each pose sample within the window: Using the current consensus pose as a reference, invert the consensus pose and multiply it with the pose sample to obtain the relative pose matrix; Extract the rotation submatrix and translation subvector from the relative pose matrix, and calculate the rotation logarithm of the rotation submatrix. Specifically: Calculate the rotation angle, which is determined by the trace of the rotation submatrix; When the angle approaches zero, set the rotation increment to the zero vector; When the angle is not close to zero, the unit rotation axis is calculated based on the antisymmetric part of the rotation submatrix, and the rotation axis is multiplied by the angle to obtain the three-dimensional rotation increment vector; Generate the corresponding antisymmetric matrix based on the three-dimensional rotation increment vector; The 3x3 matrix required to calculate the logarithm of translation The The antisymmetric matrix with rotation increment is generated by a three-term expansion, specifically by combining the first-order and second-order terms of the identity matrix and the antisymmetric matrix in Rodrigues form. right After inverting the vector, multiply it by the translation sub-vector of the relative pose to obtain the three-dimensional translation increment vector; The six-dimensional geodesic increment vector of the pose sample is obtained by concatenating the three-dimensional rotation increment vector and the three-dimensional translation increment vector in a fixed order. Based on the acquisition sequence number, the pose samples within the sliding window are divided into at least two consecutive segments in chronological order, with each segment containing the same number of pose samples. Within each segment, the arithmetic mean of all six-dimensional geodesic increment vectors of the segment is taken to obtain the segmented average increment vector; The arithmetic mean of all segmented average increment vectors is then calculated to obtain the global update increment vector for the current round. The global update increment vector is subjected to an exponential mapping to generate the increment pose, which is then multiplied by the current consensus pose to obtain the next consensus pose. The exponential mapping specifically includes: The global update increment vector is split into rotation increment and translation increment; The exponential mapping of the rotation submatrix is ​​calculated based on the rotation increment, and the rotation matrix is ​​generated from the angle and the unit axis using the Rodrigues formula; Generate a matrix consistent with the aforementioned logarithmic mapping based on the rotation increment. and use Multiplying the translation increment by the left side yields the translation subvector; Write the rotation matrix and translation subvector into a 4x4 homogeneous transformation matrix and write the last row as "0, 0, 0, 1" as the incremental pose; Calculate the stopping condition: terminate the iteration when the magnitude of the global update increment vector is not greater than the stopping threshold or the maximum number of iterations is reached. When the stopping condition is met, the current round consensus pose is output as the Karcher mean consensus pose corresponding to the current sliding window, and the consensus pose is associated with the start and end collection sequence number of the sliding window and then stored.

[0025] In this embodiment, the rotation angle is determined by the trace of the rotation matrix, the rotation logarithmic exponent adopts the Rodrigues form, and the logarithm / exponent of SE(3) is obtained through the matrix. The related translation part is derived from the Lie group / Lie algebra theory in mathematics and the original formula system of SO(3) / SE(3) exponential and logarithmic mappings in classical robotics / computer vision: first from the original relation Push Then by The rotation increment vector is obtained and an antisymmetric matrix is ​​constructed. Then, Rodrigues' original formula is used. The rotation increment is restored to the rotation matrix; in SE(3), this application restores the classic translation coupling relationship. , It is clearly defined as "constructed by combining the linear and quadratic terms of an antisymmetric matrix in Rodrigues form". Then use The feasible derivation process is as follows: "Left multiply the translation subvector to obtain the translation increment, and then left multiply the translation increment by (V) to restore the translation subvector." The modification of this application based on the original formula is reflected in the process-level construction path: the logarithmic / exponential mapping is deduced from abstract operations to "trace to find angle—antisymmetric to find axis—Rodrigues to generate rotation—..." A stepwise computational chain for "translation" is generated, and in the Karcher mean iteration, a six-dimensional geodesic increment for each sample is generated according to the chain, and then the pose is updated by backtracking. In terms of dimensions, the rotation matrix and trace are both dimensionless quantities, and the angle... In radians (considered dimensionless in engineering), Rodrigues and Both ends are dimensionless matrices, and the translation subvectors are... It is a unit of length. It is composed of dimensionless matrix terms and is dimensionless as a whole, therefore and Both ends maintain the dimension of length. Finally, the rotation block of the homogeneous transformation matrix is ​​dimensionless, the translation block is of length, and the structural constraints are consistent, which satisfies the principle that the dimensions of both sides of the formula are consistent and conforms to the norm of rigid body pose.

[0026] In this embodiment, generating the updated positioning parameters includes: Receive consensus pose, which is given in the form of a 4x4 homogeneous transformation matrix. The upper left 3x3 element in the matrix represents the rotation relationship from the machine coordinate system to the workpiece coordinate system, and the upper right 3x1 element represents the translation offset from the origin of the machine coordinate system to the origin of the workpiece coordinate system. Read the coordinate mapping parameters between the currently effective machine tool coordinate system and the workpiece coordinate system from the CNC system. The coordinate mapping parameters are stored in the form of the position offset and attitude parameters of the zero point of the workpiece coordinate system in the machine tool coordinate system. Then, restore the parameters to the corresponding homogeneous coordinate transformation matrix as the mapping matrix before the update. According to the coordinate transformation order from the machine tool coordinate system to the workpiece coordinate system, the mapping matrix before the update and the homogeneous transformation matrix corresponding to the consensus pose are combined to obtain the updated mapping matrix that reflects the latest relative relationship between the machine tool coordinate system and the workpiece coordinate system in the current machining stage. Perform a structural consistency check on the updated mapping matrix. The check includes: whether the last row of the matrix is ​​fixed as "0, 0, 0, 1", whether the rotation submatrix satisfies the constraint that the pairwise column vectors are orthogonal and have a magnitude of 1, and whether the values ​​of the translation subvectors in each coordinate axis direction are within the allowable range of the current axis travel of the machine tool. When the structural consistency verification is passed, the three-axis offset of the workpiece coordinate system origin in the machine tool coordinate system is directly read from the updated mapping matrix as the workpiece zero-point offset parameter, and the attitude angle parameter corresponding to the rotation sub-matrix is ​​read as the workpiece coordinate system attitude parameter. Write the workpiece zero-point offset parameters and workpiece coordinate system attitude parameters into the workpiece coordinate system parameter register area specified by the CNC system, so that the updated mapping relationship can be used as a valid coordinate reference in the subsequent CNC machining program. The starting acquisition number, ending acquisition number, and update timestamp of the sliding window corresponding to this mapping update are recorded synchronously, and the recorded data are correlated with the positioning parameters written into the CNC system to complete the update and output of the mapping relationship between the machine tool coordinate system and the workpiece coordinate system.

[0027] Example: To fully demonstrate the practical effectiveness of this invention in solving the problem of inconsistent positioning during multiple clamping operations of mold components, this embodiment uses a typical CNC machining process of a precision mold core part as an application scenario. This type of part typically requires multi-face machining, precision hole machining, and complex curved surface cavity machining during processing, inevitably necessitating multiple changes in clamping posture. Since each clamping introduces new clamping errors and measurement deviations, traditional machining methods often rely on manual experience to repeatedly correct the machine tool coordinate system, making it difficult to maintain a stable and consistent spatial positioning relationship between multiple clamping operations. This is precisely the core technical problem that this invention addresses and attempts to solve.

[0028] In the application scenario described above, the intelligent positioning system and method for CNC machining of mold parts of the present invention is deployed as an independent positioning calculation unit between the machine tool measurement and CNC system. Structurally, the system includes at least a multi-clamping feature point acquisition module, a point set consistency processing module, a unified shape space modeling module, a pose sample mapping module, a group-consistent pose solving module, and a coordinate mapping update module. These modules are sequentially connected through data interfaces to form a complete positioning calculation link.

[0029] In actual machining, once the mold core completes one clamping and enters the measurement state, the multi-clamping feature point acquisition module is triggered first. This module controls the machine tool measuring device to perform contact or non-contact measurements at several pre-configured measurement positions on the workpiece surface, acquiring the corresponding three-dimensional spatial coordinate data. Each set of measurement data is automatically appended with a clamping instance identifier upon generation. This identifier is generated by combining the current fixture configuration, clamping cycle, and related parameters to clearly distinguish the data source under different clamping states. Simultaneously, the system also generates an acquisition timing identifier for each measurement, enabling subsequent processing to perceive the sequential relationship of the data.

[0030] When the 3D feature point data enters the point set consistency processing module, the system first verifies the number and type of feature points. Only when the preset configuration requirements are met does it proceed to the next processing step. In the verified dataset, the system automatically selects reference points based on feature point type priority and constructs a rigid body reference coordinate system through geometric relationships. Under this reference coordinate system, the system performs a standardized transformation on the original feature point coordinates under rigid body constraints. Simultaneously, it sorts all feature points according to unified rules and assigns them fixed point sequence numbers, thereby obtaining a structurally consistent standard feature point set that can be used for cross-clamp comparison. This process ensures that data collected under different clamping states maintains mathematical consistency, laying the foundation for subsequent unified modeling.

[0031] The unified shape space modeling module receives standard feature point sets from different clamping instances and establishes a one-to-one correspondence between the points based on their sequence numbers. Internally, the system calls the generalized Procrustes alignment algorithm to iteratively align multiple standard feature point sets. During the alignment process, the system calculates the rigid body transformation relationships between each point set, gradually eliminating the influence of clamping posture differences on the overall geometry. Ultimately, it obtains a unified shape space representing the true geometry of the mold component and outputs the rigid body transformation result of each clamping instance relative to this unified shape space.

[0032] After obtaining the rigid body transformation results, the pose sample mapping module maps each rigid body transformation result to a pose sample defined in the rigid body pose space, and organizes these pose samples sequentially according to the acquisition time sequence identifier to form a continuously updated sliding window pose set. The introduction of the sliding window enables the system to integrate multiple clamping measurement results during pose fusion, rather than relying on a single measurement, thereby improving the stability of the overall positioning results.

[0033] In the group-consistent pose solving module, the system calculates geodesic distances based on the geometric relationships between pose samples within a sliding window, and performs Karcher mean calculation under the pose group structure. Through this calculation method, the system obtains a consensus pose that represents the overall spatial relationships of the current processing stage. Compared with the traditional Euclidean averaging method, this consensus pose strictly satisfies rigid body motion constraints in its mathematical structure, exhibits stronger suppression of abnormal pose samples, and effectively avoids the excessive influence of single measurement errors on the final positioning result.

[0034] After obtaining the consensus pose, the coordinate mapping update module combines the consensus pose with the currently effective coordinate mapping relationship of the machine tool to generate an updated mapping relationship between the machine tool coordinate system and the workpiece coordinate system. After verifying the structural consistency of the update result, the system writes the new workpiece coordinate system parameters into the CNC system, enabling subsequent machining programs to be automatically executed based on the updated coordinate reference, thereby achieving continuous maintenance of the workpiece positioning relationship under multiple clamping conditions.

[0035] To verify the effectiveness of the above-mentioned technical features in actual processing, a comparative test was conducted on the processing results using the traditional positioning method and the positioning method of this invention. The test object was a mold core part with the same structure and the same process route. After all multiple clamping and processing operations were completed, the key dimensions and spatial positions were uniformly inspected, and the following data were obtained.

[0036] Table 1 Comparison of positioning effects under multiple clamping machining conditions ; As shown in Table 1, this invention achieves simultaneous improvements in three main areas: accuracy consistency, efficiency, and quality risk control. Regarding positioning accuracy, the maximum spatial deviation of key holes decreased from 0.083 mm to 0.029 mm, a reduction of approximately 65.1%, and the average spatial deviation decreased from 0.055 mm to 0.018 mm, a reduction of approximately 67.3%. This indicates that by using unified shape space modeling across clamping and group-consistent pose fusion, the worst-case error and overall error level caused by multiple clamping operations were significantly reduced. Regarding datum consistency, the relative deviation of the datum surface under different clamping conditions decreased from 0.068 mm. The error was reduced to 0.024 mm, a decrease of approximately 64.7%. Simultaneously, the overall consistency pass rate increased from 87% to 98%, an increase of 11 percentage points, indicating that the system not only reduced the error amplitude but also improved the probability of stable compliance. Regarding efficiency and rework control, the auxiliary time related to single-piece positioning was shortened from 39.6 minutes to 18.2 minutes, a reduction of 21.4 minutes, or approximately 54.0%. The rework rate decreased from 12% to 3%, a reduction of 75%. This demonstrates that automatic coordinate mapping updates reduced manual alignment and repetitive measurements, and robust suppression of abnormal pose samples reduced subsequent quality risks caused by unstable positioning.

[0037] The data results show that this invention effectively reduces spatial deviations caused by multiple clamping operations through multi-clamping feature point consistency processing, unified shape space modeling, and group-consistent pose solving based on Karcher's mean, ensuring that the workpiece maintains a highly consistent positioning relationship under different clamping states. Simultaneously, because the system can automatically update coordinate mapping parameters, it significantly reduces the time required for manual alignment and repeated measurements, lowering the risk of rework due to unstable positioning.

[0038] In practical use, operators do not need to understand complex mathematical calculations; they only need to complete the clamping and measurement operations according to the normal procedure. The system can then automatically complete the positioning calculations and parameter updates in the background. This implementation method, which encapsulates complex geometric modeling and group structure calculations within the system, ensures that the invention is technologically advanced while possessing good engineering feasibility.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A CNC machining intelligent positioning system for mold parts, characterized in that, include: The multi-clamping feature point acquisition module collects the three-dimensional feature point set of the same mold part under different clamping states, and establishes a corresponding clamping instance identifier and acquisition time sequence identifier for each set of three-dimensional feature points; The point set unification processing module receives a set of three-dimensional feature points with clamping instance identifiers, performs point order unification and rigid body constraint preprocessing on the three-dimensional feature point set, and generates a standard feature point set. The shape space unified modeling module receives a standard feature point set, constructs a unified shape space across clamping instances based on generalized Procrustes alignment, and outputs the rigid body transformation results of each clamping instance relative to the unified shape space. The pose sample mapping module maps the rigid body transformation results to pose samples defined on the SE(3) Lie group, and organizes the pose samples into a sliding window pose set based on the acquisition timing identifier. The consensus pose solving module performs Karcher mean calculation on the sliding window pose set on the SE(3) group manifold using geodesic distance as the metric to generate the corresponding consensus pose. The coordinate mapping update module updates the mapping relationship between the machine tool coordinate system and the workpiece coordinate system based on consensus pose, and outputs the updated positioning parameters to the CNC system.

2. A method for intelligent positioning in CNC machining of mold parts, characterized in that, The modules are connected in the following way: Under different clamping conditions, the same mold component is measured multiple times to collect multiple sets of three-dimensional feature points, and clamping instance identifier and collection time sequence identifier corresponding to the clamping condition are established for each set of three-dimensional feature points. Perform point order unification and rigid body constraint preprocessing on the 3D feature point set with clamping instance identifiers to generate a standard feature point set for cross-clamping comparison; Based on the generalized Procrustes alignment, the standard feature point set is iteratively processed to construct a unified shape space across clamping instances, and the rigid body transformation results of each clamping instance relative to the unified shape space are obtained. The rigid body transformation results are mapped to pose samples defined on the SE(3) Lie group, and the pose samples are organized into a sliding window pose set according to the acquisition timing identifier. On the SE(3) group manifold, the geodesic distance between pose samples is used as a metric to perform Karcher mean calculation on the sliding window pose set to obtain the consensus pose corresponding to the current window; The mapping relationship between the machine tool coordinate system and the workpiece coordinate system is updated based on consensus pose, and the updated positioning parameters are output.

3. The intelligent positioning method for CNC machining of mold parts according to claim 2, characterized in that, Establishing clamping instance identifiers and acquisition timing identifiers corresponding to the clamping status includes: After the mold parts are clamped once, read the fixture number, fixture installation posture parameters, positioning element number and its position parameters in the machine tool coordinate system, and obtain the current clamping cycle count value. The fixture number, fixture installation posture parameters, positioning element number and its position parameters are combined with the clamping cycle count to form a clamping configuration data group, and the clamping configuration data group is subjected to consistent sorting and standardization processing. Based on the standardized clamping configuration data set, a unique clamping instance identifier corresponding to the current clamping configuration is generated, and a binding relationship is established between the clamping instance identifier and the current clamping configuration. Each time a 3D feature point acquisition operation is triggered, the system timestamp corresponding to the acquisition trigger time is read, and the program segment number and measurement instruction number currently being executed in the CNC system are read simultaneously. The collection sequence number is initialized during the first collection, and then incremented after each subsequent collection to form a collection sequence number sequence that is consistent with the collection order. The system timestamp, program segment number, measurement command number, and acquisition sequence number are combined to generate an acquisition timing identifier.

4. The intelligent positioning method for CNC machining of mold parts according to claim 2, characterized in that, Generating a standard feature point set includes: Receive a set of three-dimensional feature points carrying a clamping instance identifier and a collection timing identifier, read the three-dimensional coordinates and feature point type identifier of each feature point, count the number of points of each type according to the feature point type identifier, compare the statistical results with the target point count table of the corresponding mold part in the configuration file item by item, and data batches that fail the comparison will not enter the point sequence unification process. In the batches of data that have been compared, a set of reference points of the same type is selected according to the priority of the feature point type identifier, and the first reference point is selected from the set of reference points of the same type. The coordinate point with the smallest lexicographical order; by Select a second reference point as the starting point To and The point with the largest Euclidean distance; select a third reference point from the remaining points. In order to make The point with the largest modulus; based on Establish a rigid body reference coordinate system and calculate the unit basis vectors; For each feature point in the three-dimensional feature point set Perform rigid body constraint preprocessing and calculate standardized coordinates. ; In a rigid body reference coordinate system, a point order unification is performed, sorting all feature points in ascending order based on their feature point type identifier. of Component ascending order—plane polar angle Ascending order — plane radius Sort the data according to the "ascending order" rule and assign point numbers accordingly; Point sequence number, corresponding standardized coordinates The clamping instance identifier and the acquisition time sequence identifier are combined to generate a standard feature point set.

5. The intelligent positioning method for CNC machining of mold parts according to claim 2, characterized in that, The generated rigid body transformation results include: Receive standard feature point sets, and construct a point set sequence by combining the standard feature point sets formed by the same mold part under different clamping instances according to the clamping instance identifier, while keeping the point sequence number consistent within the point set sequence; Input the point set sequence into the generalized Procrustes alignment process, set the initial template of the uniform shape point set to any standard feature point set in the point set sequence or the initial mean point set obtained from the point set sequence, call the generalized Procrustes alignment process to output the uniform shape point set and the rigid body transformation result corresponding one-to-one with each clamping instance identifier in the point set sequence. For each clamping instance identifier in the point set sequence, the standard feature point set corresponding to the clamping instance is rotated and translated according to the rigid body transformation result to obtain the alignment point set, and the point sequence number is retained in the alignment point set so that the alignment point set and the uniform shape point set form a one-to-one correspondence in the point sequence number dimension. Define the uniform shape point set as a uniform shape space representation across clamping instances, and define the rigid body transformation result as the pose mapping parameters from clamping instances to the uniform shape space.

6. The intelligent positioning method for CNC machining of mold parts according to claim 5, characterized in that, The generalized Procrustes alignment process includes: Let the number of point sets in the point set sequence be . The number of feature points in each point set is , No. The set of points is denoted as ,in This is the clamping instance number. For point sequence numbering, The point sequence number is 3D coordinate vector; For each candidate point set in the point set sequence Calculate the dispersion score and select the candidate point set with the smallest dispersion score as the initial template point set. The dispersion score is calculated using the following formula: ； in, For Dispersion score when used as a candidate template The index of the set of points to be aligned. To be Mapped to via a rigid body alignment transformation The next Alignment point coordinates, It is the Euclidean norm; In a rigid body alignment transformation calculation, the point sequence number is used as the basis for... and Pair up points one by one and calculate the geometric centers of the two point sets respectively. and The two point sets are centered separately; Calculate the correlation matrix based on the centralized point-by-point pairing results. And perform singular value decomposition on the correlation matrix to obtain an orthogonal matrix. ; Calculate the rotation matrix for ,when When the determinant is less than 0, The rotation matrix is ​​recalculated after taking the opposite of the third column; Calculate the translation vector for ; With the initial template point set Enter the iterative alignment process, let the first... The template point set for the round iteration is = ,in This represents the iteration round number. Let (i) be the three-dimensional coordinate vector of point number (i) in the template point set; For each iteration For each clamping instance point set Repeat the rigid body alignment transformation calculation to obtain the result relative to... rotation matrix With translation vector And perform rotation and translation on all points in the point set to obtain the aligned point set. ; Calculate the alignment residual for each clamping instance point set in the current round. Record the correspondence between residual values ​​and clamping instance identifiers; The entire set of alignment points will be sorted according to the clamping instance identifier. Divided into preset quantities There are non-overlapping groups, each containing the same number of alignment point sets; For each group and the sequence number of each point Calculate the group mean point ; Sequential numbering of the same point All group mean points Calculate the median values ​​of the coordinate components, respectively for , , The median value of the three coordinate components is taken and used to form the coordinates of the new template point. ; All point sequence numbers correspond to Combine to obtain the updated template point set ; Calculate the change in template point sets between two adjacent rounds. And take the maximum value as ; when The iteration terminates when the value is not greater than a preset threshold or the number of iterations reaches a preset upper limit. When the iteration terminates, output the final template point set. As a unified shape space representation across clamping instances; output the rigid body transformation result corresponding to each clamping instance identifier, the rigid body transformation result includes the rotation matrix and translation vector obtained from the last alignment in each iteration.

7. The intelligent positioning method for CNC machining of mold parts according to claim 2, characterized in that, The generated sliding window pose set includes: Receive the rigid body transformation results corresponding to each clamping instance identifier; Perform singular value decomposition on the rotation matrix, extract the two orthogonal matrices from the decomposition result, and multiply them to generate a new rotation matrix; When the determinant of the generated rotation matrix is ​​negative, take the opposite of the third column of an orthogonal matrix in the singular value decomposition result and multiply them again to obtain a rotation matrix with a positive determinant. Construct the pose matrix: Create a 4x4 matrix, write the orthogonalized rotation matrix into the upper left 3x3 region of the 4x4 matrix, write the 3D translation vector into the upper right 3x1 region of the 4x4 matrix, and write the last four elements of the 4x4 matrix as "0, 0, 0, 1" in sequence. Perform SE(3) structure verification on the constructed 4x4 matrix; The pose samples that pass the SE(3) structure verification are written into the pose sample library along with the corresponding clamping instance identifier and acquisition timing identifier to form pose record entries. The pose record entries are sorted according to the acquisition sequence number in the acquisition time identifier to generate a pose sample sequence arranged in the acquisition order; When multiple pose record entries with the same acquisition sequence number exist, only the pose record entry with the newest system timestamp is retained; Set the sliding window length to a fixed value, and select consecutive pose record entries from the pose sample sequence in sequence to form a sliding window pose set; After a new pose record entry is written into the pose sample sequence, the new pose record entry is added to the end of the sliding window pose set, and the earliest written pose record entry is removed from the sliding window pose set to obtain the updated sliding window pose set.

8. The intelligent positioning method for CNC machining of mold parts according to claim 2, characterized in that, Karcher mean calculation includes: Receive the pose set of the sliding window, and read the poses within the window sorted by acquisition sequence number. There are several pose samples, each pose sample is a 4x4 homogeneous transformation matrix, and the upper left 3x3 is the rotation submatrix and the upper right 3x1 is the translation subvector. Using the pose sample with the center number collected in the sliding window as the reference sample, the relative poses of the other pose samples in the window are calculated respectively. Perform a logarithmic mapping once for each relative pose to obtain a six-dimensional increment vector. Take the median value of each component of all six-dimensional increment vectors in the window to form an initial increment vector. Perform an exponential mapping on the initial increment vector to generate the increment pose, and then multiply it with the reference sample to obtain the initial consensus pose value. Using the initial consensus pose value as the starting point for iteration, we enter the Karcher mean iteration, setting the maximum number of iterations and the stopping threshold. In each iteration, geodesic increment calculation is performed for each pose sample within the window: Using the current consensus pose as a reference, invert the consensus pose and multiply it with the pose sample to obtain the relative pose matrix; Extract the rotation submatrix and translation subvector from the relative pose matrix, and calculate the rotation logarithm of the rotation submatrix: Calculate the rotation angle, which is determined by the trace of the rotation submatrix; When the angle approaches zero, set the rotation increment to the zero vector; When the angle is not close to zero, the unit rotation axis is calculated based on the antisymmetric part of the rotation submatrix, and the rotation axis is multiplied by the angle to obtain the three-dimensional rotation increment vector; Generate the corresponding antisymmetric matrix based on the three-dimensional rotation increment vector; The 3x3 matrix required to calculate the logarithm of translation The It is generated by a three-term expansion of the antisymmetric matrix of the rotation increment; right After inverting the vector, multiply it by the translation sub-vector of the relative pose to obtain the three-dimensional translation increment vector; The six-dimensional geodesic increment vector of the pose sample is obtained by concatenating the three-dimensional rotation increment vector and the three-dimensional translation increment vector in a fixed order. Based on the acquisition sequence number, the pose samples within the sliding window are divided into time-ordered categories. Continuous segmentation, in which And each segment contains the same number of pose samples; Within each segment, the arithmetic mean of all six-dimensional geodesic increment vectors of the segment is taken to obtain the segmented average increment vector; The arithmetic mean of all segmented average increment vectors is then calculated to obtain the global update increment vector for the current round. Perform an exponential mapping on the global update increment vector to generate the increment pose, and then multiply it by the current consensus pose to obtain the next consensus pose. Calculate the stopping condition: terminate the iteration when the magnitude of the global update increment vector is not greater than the stopping threshold or the maximum number of iterations is reached. When the stopping condition is met, the current round consensus pose is output as the Karcher mean consensus pose corresponding to the current sliding window, and the consensus pose is associated with the start and end collection sequence number of the sliding window and then stored.

Images