A method and system for processing coordinate measuring machine data for automotive parts

Abstract

This invention belongs to the field of three-dimensional geometric measurement technology, specifically relating to a method and system for processing coordinate measuring machine (CMM) data for automotive parts. The method includes: acquiring actual point clouds and standard point clouds; filtering actual and standard feature point sets by calculating the maximum angular gap between data points; constructing noise-resistant geometric descriptors based on tangential weights for the feature points in both sets; matching the two sets of feature points based on the descriptors to generate an initial matching set; solving for the globally optimal transformation from the initial matching set using a random sampling consensus algorithm; applying the transformation to the actual point cloud and performing fine registration using an iterative nearest-point algorithm; and finally, performing deviation analysis. This invention solves the technical problem of traditional registration relying on initial alignment and easily getting trapped in local optima, improving the robustness and accuracy of registration.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional geometric measurement technology. More specifically, this invention relates to a method and system for processing coordinate measuring machine (CMM) data for automotive parts. Background Technology

[0002] In the automotive parts manufacturing industry, precise quality control is the cornerstone of ensuring the performance, safety, and assembly consistency of the whole vehicle. The coordinate measuring machine is the core equipment for performing this task, used to accurately measure the geometric dimensions and tolerances of parts to ensure that they strictly comply with design specifications.

[0003] The standard workflow for coordinate measuring machine (CMM) inspection relies on precisely aligning the actual point cloud data, which represents the actual shape of the part, collected by the measuring machine, with the CAD model, which represents the design datum, in three-dimensional space. This process is called registration, and the accuracy of registration is the basis for all subsequent deviation analysis, profile evaluation, and quality judgment.

[0004] Currently, mainstream registration algorithms in the industry, such as the Iterative Closest Point (TLP) algorithm, suffer from a fundamental flaw in their mathematical principles. As a local optimization algorithm, its registration results heavily rely on a good initial alignment position. This flaw is particularly pronounced when dealing with complex automotive parts, such as components with multiple holes or symmetrical structures. These parts typically contain a large number of similar or repetitive geometric features. If the initial pose deviation between the actual point cloud and the CAD model is too large, the TLP algorithm is prone to mismatching these similar features during iterative search, thus converging to an incorrect local optimum instead of a correct global optimum. This registration failure renders subsequent deviation analysis results completely invalid, making it impossible to accurately assess the manufacturing quality of the parts.

[0005] To address the initial dependency problem of the iterative nearest point algorithm, some industry research has shifted towards feature-based registration methods. However, when scanning automotive parts, coordinate measuring machines inevitably introduce measurement noise at edges or curvature changes. In actual point clouds, outlier noise points are often present. Existing feature extraction and description algorithms are highly susceptible to interference from this noise data when calculating features or constructing descriptors. For example, traditional weighting algorithms may assign excessive influence to these noise points, thereby polluting the descriptors and causing matching failures. Therefore, automatically achieving high-precision and robust global registration between automotive part point clouds and CAD models without manual intervention for initial alignment and resisting interference from actual measurement noise is a pressing technical challenge in the field of coordinate measuring machine inspection. Summary of the Invention

[0006] To address the technical problems mentioned above in the processing of coordinate measuring machine (CMM) data for automotive parts, such as the high dependence of existing registration algorithms on initial pose, easy convergence to local optima, and poor robustness under measurement noise, this invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a three-coordinate measuring machine (CCM) inspection data processing method for automotive parts, comprising: acquiring an actual point cloud and a standard point cloud of the automotive parts; for each data point in the actual point cloud and the standard point cloud, obtaining the maximum angular gap of the data point based on the projection angle distribution of its neighboring data points on the tangent plane of the data point; based on the maximum angular gap of each data point, selecting an actual feature point set and a standard feature point set from the actual point cloud and the standard point cloud respectively; for each feature point in the actual feature point set and the standard feature point set respectively, determining the maximum angular gap of the data point based on the distance between the feature point and its neighboring data points and the distance between the neighboring data points and the feature point; and for each feature point in the actual feature point set and the standard feature point set, determining the maximum angular gap of the data point based on the distance between the feature point and its neighboring data points and the distance between the neighboring data points and the feature point. The tangential consistency of the feature point normal vector is used to obtain the tangential weight between the feature point and its neighboring data points in their respective point clouds. Based on the tangential weight, the geometric descriptor of the feature point is obtained. For each actual feature point in the actual feature point set, the standard feature point with the closest geometric descriptor is found in the standard feature point set to generate an initial matching set composed of feature point pairs. A global optimal transformation is calculated from the initial matching set using a random sampling consistency algorithm. The global optimal transformation is applied to the actual point cloud, and the iterative nearest point algorithm is executed for fine registration to obtain the final transformation matrix. The three-coordinate detection process is completed based on the transformation matrix.

[0008] This invention filters the actual feature point set and the standard feature point set by calculating the maximum angular gap of the data points, and then constructs a noise-resistant geometric descriptor based on tangential weights for these feature points. This method does not depend on the initial position of the automotive parts on the coordinate measuring machine, so that the subsequent random sampling consensus algorithm can robustly estimate the global optimal transformation from the initial matching set containing erroneous matches. This provides a good initial value for the iterative nearest point algorithm, thereby avoiding registration getting trapped in local optima and improving the accuracy and reliability of coordinate measuring machine detection data processing.

[0009] Preferably, obtaining the maximum angular gap of a data point based on the projection angle distribution of its neighboring data points on the tangent plane of the data point includes: obtaining the tangent plane basis vector at the data point through principal component analysis; projecting all neighboring data points of the data point onto the tangent plane to obtain projection vectors; calculating the angle of each projection vector relative to the tangent plane basis vector and storing it in an angle set; sorting the angle set in ascending order; calculating the angle difference between two adjacent angles after sorting, and the first-to-last wraparound angle difference between the last angle and the first angle; and taking the maximum value among all angle differences as the maximum angular gap of the data point.

[0010] This invention utilizes the asymmetry of the neighborhood distribution of data points at the edges or corners of automotive parts, while the neighborhood distribution of planar data points is symmetrical, by projecting neighborhood data points onto the tangent plane at the data points and calculating the maximum angle difference in the 360-degree distribution of the projection angle. This makes the maximum angle gap a reliable basis for distinguishing structural feature points from non-feature points, thus ensuring the quality of the selected feature point set.

[0011] Preferably, the projection vector satisfies the expression: ;in, For data points The projection vector on the tangent plane; From the center point Pointing to neighboring data points ; and These are the first and second basis vectors of the tangent plane, respectively. This represents the dot product operation of vectors.

[0012] Preferably, the tangential weights satisfy the expression: ;in, Representing neighborhood points relative center point Tangential weights; The central feature point; for A neighboring data point within a neighborhood; express and The Euclidean distance between them; central feature point The normal vector; From point to , unit vector; Represents the dot product operation of vectors; It represents the absolute value.

[0013] The tangential weight calculation method proposed in this invention utilizes the dot product of the normal vector of the central feature point and the unit vector pointing from the central feature point to the neighboring data points during calculation. By subtracting the absolute value of this dot product from 1, the normal component is suppressed. This reduces the weight of outlier noise points that deviate from the true surface generated in coordinate measuring machine measurements, while retaining the weight of neighboring data points located on the same surface. This provides a basis for constructing noise-resistant geometric descriptors.

[0014] Preferably, the geometry descriptor satisfies the expression: ;in, central feature point The geometry descriptor; and They are and Simplified histogram vector; for ; For neighborhood data points Compared to Tangential weights.

[0015] This invention constructs a final geometric descriptor by superimposing a simplified histogram of the central feature point with a weighted average of the simplified histograms of its neighboring data points. The weighted average uses the aforementioned calculated tangential weights, ensuring that when aggregating neighborhood information, only those neighboring data points on the same surface contribute the main information, while the information of measurement noise points is filtered out. This allows the constructed geometric descriptor to more accurately reflect the true geometric shape of the local surface of automotive parts.

[0016] Preferably, the step of using the random sampling consensus algorithm to calculate a globally optimal transformation from the initial matching set includes: repeatedly performing the following operations within a preset number of iterations: randomly extracting a preset number of correspondences from the initial matching set, calculating a temporary rigid transformation based on the extracted correspondences, applying the temporary rigid transformation to the actual feature point set, then traversing the initial matching set to count the number of interior points whose distance between the transformed feature points and their corresponding feature points is less than a preset distance threshold; after all iterations are completed, the temporary rigid transformation that obtains the most interior points among all calculated temporary rigid transformations is taken as the globally optimal transformation.

[0017] This invention employs a random sampling consensus algorithm to handle the initial matching set containing a large number of incorrect matches due to the symmetrical or repetitive structure of automotive parts. By iteratively testing small subsets of matches and counting the number of interior points that support the transformation of that subset, the global optimal transformation can be found probabilistically by identifying the transformation model supported by the most correct matches. This ensures that the correct transformation matrix can be estimated even when the initial matching quality is low.

[0018] Preferably, applying the global optimal transformation to the actual point cloud includes: applying the rotation matrix and translation vector of the global optimal transformation to all data points in the actual point cloud to obtain an aligned point cloud.

[0019] Preferably, the step of selecting the actual feature point set and the standard feature point set from the actual point cloud and the standard point cloud respectively based on the maximum angular gap of each data point includes: presetting an angular gap threshold; in response to the maximum angular gap of each data point in the actual point cloud being greater than the angular gap threshold, storing the data point in the actual feature point set; in response to the maximum angular gap of each data point in the standard point cloud being greater than the angular gap threshold, storing the data point in the standard feature point set.

[0020] Preferably, the normal vector is obtained by performing principal component analysis on the neighborhood of the central feature point.

[0021] Secondly, the present invention provides a coordinate measuring machine (CMM) inspection data processing system for automotive parts, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned coordinate measuring machine (CMM) inspection data processing method for automotive parts is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned three-coordinate inspection data processing method for automotive parts, and stored in the memory so that it can be loaded and executed by the processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows:

[0024] This invention filters the actual feature point set and the standard feature point set by calculating the maximum angular gap, matches a noise-resistant geometric descriptor based on tangential weights, and then uses a random sampling consensus algorithm to solve the global optimal transformation. It does not depend on the initial position of the automotive parts on the coordinate measuring machine, thus avoiding the problem of traditional iterative nearest point algorithms getting stuck in local optima due to poor initial pose.

[0025] This invention introduces tangential weights when constructing geometric descriptors. These weights, by analyzing the degree to which neighboring data points deviate from the tangential plane of the central feature point, can adaptively suppress the influence of outlier noise points generated during coordinate measuring machine (CMM) measurement. This ensures that the constructed geometric descriptor accurately reflects the local true geometric shape of automotive parts, thereby improving the accuracy of subsequent matching.

[0026] This invention uses the maximum angular gap as a screening criterion to objectively and consistently extract structural feature points with high geometric asymmetry located at edges or corners from actual point clouds and standard point clouds. This provides a stable data foundation for subsequent global optimal transformation solutions. Combined with the final iterative nearest point algorithm for fine registration, it ensures high accuracy of the entire coordinate measuring machine detection data processing flow during automated execution. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a three-coordinate measuring machine (CMM) inspection data processing method for automotive parts according to the present invention.

[0028] Figure 2 This is a schematic diagram illustrating an misaligned initial point cloud in this invention;

[0029] Figure 3This is a schematic diagram illustrating the maximum angular gap distribution of the actual point cloud in this invention;

[0030] Figure 4 This is a schematic diagram illustrating the maximum angular gap distribution of the standard point cloud in this invention;

[0031] Figure 5 This is a schematic diagram illustrating the feature point extraction results of an actual point cloud in this invention;

[0032] Figure 6 This is a schematic diagram illustrating the feature point extraction results of the standard point cloud in this invention;

[0033] Figure 7 This is a schematic diagram illustrating the RANSAC coarse registration in this invention;

[0034] Figure 8 This is a schematic illustration of the deviation chromatogram in the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] This invention discloses a method for processing coordinate measuring machine (CMM) data for automotive parts, referring to... Figure 1 This includes steps S1-S5:

[0038] S1: Obtain the actual point cloud and standard point cloud of automotive parts.

[0039] It should be noted that the purpose of this invention is to obtain two core data sources for registration; in the quality inspection scenario of automotive parts, there are two types of data: one is the standard point cloud representing the design benchmark. It is obtained by sampling from the CAD model; the other is the actual point cloud representing the measurement results of the physical entity. The fundamental task of this invention is to calculate an accurate spatial transformation, namely rotation and translation, such that... Able to Precise alignment in space.

[0040] Specifically, the actual three-dimensional surface of automotive parts is obtained through laser scanning using a coordinate measuring machine, generating actual point clouds. From the CAD model of the component, such as a STEP file, its surface is uniformly sampled to generate a standard point cloud. This standard point cloud is used as the registration target; after acquiring the data, the actual point cloud can be... and standard point cloud Downsampling is performed to reduce data redundancy and make the point cloud densities of the two roughly equivalent, thereby improving the computational speed and matching stability of subsequent steps.

[0041] For example, Figure 2 This is a schematic diagram of the unaligned initial point clouds. The blue point clouds represent the standard point clouds, corresponding to the CAD model; the red point clouds represent the actual point clouds, corresponding to coordinate measuring machine (CMM) data. This diagram shows the original states of the two point clouds before registration begins. Their positions and orientations in space—i.e., rotation and translation—are completely different, visually illustrating the initial alignment problem that the algorithm needs to solve.

[0042] S2: For each data point in the actual point cloud and the standard point cloud, obtain the maximum angular gap of the data point based on the projection angle distribution of its neighboring data points on the tangent plane of the data point; based on the maximum angular gap of each data point, select the actual feature point set and the standard feature point set from the actual point cloud and the standard point cloud respectively.

[0043] It should be noted that, in order to achieve global registration independent of initial positions, this invention abandons the strategy of attempting to match all points in the traditional iterative nearest-point algorithm. In complex automotive component point cloud data, most data points, such as those located on a plane, lack uniqueness or symmetry in their local geometric features, which can easily lead to ambiguity in registration. Conversely, data points located in regions with drastic geometric changes, such as edges or corners, exhibit highly asymmetrical local neighborhood distributions, which are key to achieving stable alignment. The purpose of this invention is to evaluate this local geometric asymmetry and filter out feature points with high information entropy.

[0044] Specifically, this invention calculates the maximum angular gap. This is used to filter feature points; the calculation is for... and Each individual data point To be carried out;

[0045] For each data point Determine its neighborhood This neighborhood Can be set to radius For all points within the range, calculate the neighborhood using principal component analysis. The covariance matrix is ​​used to obtain three eigenvectors, where the eigenvector corresponding to the smallest eigenvalue is defined as the normal vector. The eigenvectors corresponding to the two largest eigenvalues ​​are defined as follows: Tangent plane basis vectors at the location and ;

[0046] Furthermore, for the neighborhood All data points and Calculate its projection vector on the tangent plane. :

[0047]

[0048] in, For data points The projection vector on the tangent plane; From the center point Pointing to neighboring data points ; and These are the first and second basis vectors of the tangent plane, respectively. This represents the dot product operation of vectors.

[0049] Furthermore, each projection vector is calculated. Relative to basis vectors Angle ,angle The range of values ​​is [0, 360°); all angles Store in an angle set And sort the set of angles in ascending order; calculate the data points. Maximum angular gap :

[0050]

[0051] in:

[0052]

[0053]

[0054] In the formula, Representing data points Maximum angular gap; Center point; for The set of angles of all projection points in the neighborhood; It is a set Number of angles included; Angle set The Middle The sorted angles; Angle set The Middle The sorted angles; Angle set The Middle The sorted angles; For the first The angle difference between each sorting angle and its next angle; The difference in the first and last circumference angles between the last angle and the first angle; This indicates taking the maximum value among all angle differences; This represents the dot product operation of vectors.

[0055] In the formula, Describes neighborhood data points exist Orientation on the tangent plane; Described The spacing of neighboring data points in terms of angular distribution; This means The greatest discontinuity in the angular distribution of neighborhood data points; if the center point Located on a flat surface, its neighborhood The points inside will be evenly distributed. Around, gathering The included angles are evenly distributed, resulting in all angle differences. They are all relatively small, therefore The value is very small; conversely, if the center point Located at an edge, its neighborhood This will only be distributed on one side, for example, within a range of about 180°, which will result in a very large difference in the front and rear circumference angles. Its value is close to 180°, therefore The value is very large; if the center point Located at a corner, its neighborhood The asymmetry is stronger, for example, it is only distributed within a 90° range, which will result in a larger difference in the head-to-tail wraparound angle, which is close to 270°, therefore The value is very high.

[0056] Furthermore, set an angle gap threshold. ,For example To satisfy all The data points are extracted and used to form actual feature point sets. and standard feature point set .

[0057] For example, Figure 3 and Figure 4These are schematic diagrams showing the maximum angular gap distribution of an actual point cloud and a standard point cloud, respectively. The light blue bars represent the number of points with specific angular gap ranges; the red dashed lines represent preset angular gap thresholds, such as 90 degrees.

[0058] For example, Figure 5 and Figure 6 These are schematic diagrams showing the feature point extraction results of an actual point cloud and a standard point cloud, respectively. Gray and semi-transparent points are non-feature points; red and prominent points are feature points selected based on the maximum angular gap.

[0059] S3: For each feature point in the actual feature point set and the standard feature point set, based on the distance between the feature point and its neighboring data points and the tangential consistency between the neighboring data points and the feature point's normal vector, obtain the tangential weight between the feature point and its neighboring data points in their respective point clouds; obtain the geometric descriptor of the feature point based on the tangential weight.

[0060] It should be noted that after identifying the key feature points, the next step is to describe these points; the reason for this description is that knowing only one point... Coordinates are useless before registration because they are in and The coordinates in the original text are completely different; this invention must calculate a geometric descriptor, which is a mathematical summary that does not depend on coordinates and only describes the shape of the neighborhood around the point; theoretically, A corner point on and The corresponding corner points should have very similar geometric descriptors; however, the coordinate measuring machine scans... Outlier noise points may occur, which, although physically nonexistent, are very close to feature points in the data. Traditional algorithms incorrectly include these noise points when calculating the shape of the surrounding neighborhood, leading to distorted geometric descriptors. The core objective of this invention is to obtain a tangential weight that automatically reduces the influence of noise points that deviate from the true surface when calculating descriptors, focusing only on neighborhood points that are truly on the same surface.

[0061] Specifically, the present invention provides for each feature point, with For example, obtain an adaptive weighted geometric descriptor; for the central feature point and its neighborhood Each neighboring point in Calculate the tangential weights between the two. :

[0062]

[0063] In the formula, Representing neighborhood points relative center point Tangential weights; The central feature point; for Neighborhood A neighboring point; express and The Euclidean distance between them; Center point The normal vector; From point to , unit vector; Represents the dot product operation of vectors; It represents the absolute value.

[0064] In the formula, It is a distance-weighted term, which retains the characteristic that the closer the distance, the higher the correlation; It is the normal component term, which is calculated as follows: point to The vector in The size of the projection along the normal direction; if the neighboring points lie in On the tangent plane, that is and On the same surface, then vectors With normal Almost perpendicular, the product of these points is close to 0; if the neighboring points It is an outlier noise point, located in If the vector is directly above or below the surface, that is, deviating from the surface along the normal direction, then the vector... With normal Since they are almost parallel, the dot product value is close to 1; therefore, This constitutes the tangential consistency factor, when When located on the tangent plane, the factor is close to 1, and the weights are fully preserved. When the point is an outlier or noise point, the factor is close to 0, and the weight is suppressed to zero; the final It prioritizes retaining neighboring points that are on the same surface and close to each other, while penalizing or eliminating noise points that are off the surface.

[0065] Furthermore, using this tangential weight To calculate the center point Geometric descriptor, center point The geometry descriptor satisfies the expression:

[0066]

[0067] In the formula, central feature point The geometry descriptor; and They are and The simplified histogram vector is obtained based on the Simplified Point Feature Histogram (SPFH) calculation method in the existing technology; for The neighborhood; For neighborhood data points Compared to Tangential weights.

[0068] Further explanation is needed regarding the specific calculation method for the simplified histogram vector: For any central feature point and any neighboring data point within its neighborhood, a local reference coordinate system is established attached to the surface of the central feature point. This coordinate system uses the normal vector of the central feature point as the first axis, the cross product of the line vector connecting the central feature point to the neighboring data point and the normal vector as the second axis, and the cross product of the first and second axes as the third axis. Based on this local reference coordinate system, three angular feature values ​​describing the relative attitude of the two points are calculated: the first feature value is obtained by calculating the dot product of the connecting vector and the normal vector of the central feature point, reflecting the angle of the neighboring point's deviation from the tangent plane of the central point; the second feature value is obtained by calculating the angle of the neighboring data point's deviation from the tangent plane of the central point; the third feature value is obtained by calculating the angle of the neighboring data point's deviation from the tangent plane of the central point. The dot product of the point's normal vector and the second axis is obtained, reflecting the degree of distortion of the surface normal. The third eigenvalue is obtained by calculating the projection angle of the normal vector of the neighboring data points onto the plane formed by the first and third axes, reflecting the rotation direction of the normal. The simplified histogram vector of the central feature point itself is obtained by comprehensively calculating its relative relationship with all points in its neighborhood. The process is as follows: traverse all neighboring data points in the neighborhood of the central feature point, calculate the above three angular eigenvalues ​​for each neighboring data point, and map these eigenvalues ​​to a preset statistical interval for counting and accumulation. The statistical results of the three independent dimensions after accumulation are spliced ​​together to form a simplified histogram vector that reflects the complete local geometry around the central feature point.

[0069] S4: For each actual feature point in the actual feature point set, find the standard feature point in the standard feature point set whose geometric descriptor is closest to it, and generate an initial matching set consisting of feature point pairs.

[0070] It should be noted that the present invention has already provided... and Each feature point in the dataset has a reliable geometric descriptor calculated; the purpose of this invention is to find a match between two sets, but automotive parts, such as gearbox housings, may have many symmetrical or repeating structures, such as multiple identical screw holes; this leads to a Feature points in, may be in If several candidate points with very similar geometric descriptors are found, fuzzy matching will occur, resulting in a large number of incorrect matches in the initial matching set. Therefore, this invention adopts an algorithm that can tolerate a large number of incorrect matches, namely the Random Sample Consensus (RANSAC) algorithm, to robustly find the only correct global optimal transformation from these mixed matches.

[0071] Specifically, this invention employs a random sample consensus algorithm to accomplish this task; in the descriptor space, for Each geometry descriptor in ,exist Find the geometric descriptor that is closest to it in Euclidean distance. This generates an initial matching set. This set inevitably contains a large number of false matches caused by geometric similarity.

[0072] Furthermore, iterative random sample consistency estimation is performed; in each iteration, from the initial matching set... In, random sampling A correspondence, for example ; Assuming this The correspondence is correct. By minimizing the sum of squared distances between corresponding points, a temporary rigid transformation can be calculated. The transformation includes a temporary rotation matrix. and temporary translation vector The temporary rigid transformation Applied to All feature points in the initial matching set are traversed. All correspondences Check whether they satisfy consistency, that is, the transformed points Corresponding point Is the distance between them less than a preset distance threshold? All correspondences satisfying this condition are called interior points, i.e., correspondences that satisfy the consistency of the current temporary rigid transformation. Record the number of interior points in this iteration; after repeating this process a large number of times, such as 1000 times, the transformation calculated in the iteration with the highest number of interior points will be obtained. ,Right now and This serves as the globally optimal transformation for solving the problem.

[0073] It should be added that the distance threshold The setting should be between 1.5 and 3 times the average resolution of the point cloud. If the value is too small, even correct matches may be misclassified as outliers. If the value is too large, it will introduce incorrect matching and affect the accuracy of the transformation matrix.

[0074] For example, Figure 7 This is a schematic diagram of RANSAC coarse registration. The blue point cloud represents the standard point cloud; the red point cloud represents the actual point cloud after coarse registration following the globally optimal transformation estimated by RANSAC. This diagram shows the result of global registration; the red point cloud's pose roughly overlaps with the blue point cloud, providing a good initial position for subsequent ICP fine registration.

[0075] S5: Calculate a globally optimal transformation from the initial matching set using the random sampling consensus algorithm; apply the globally optimal transformation to the actual point cloud and perform fine registration using the iterative nearest point algorithm to obtain the final transformation matrix; complete the three-coordinate detection process based on the transformation matrix.

[0076] It should be noted that the globally optimal transformation has already been obtained. This transformation is calculated based on a small number of feature points. While globally correct, ensuring the part isn't incorrectly matched to a different location, its accuracy may not be high enough to meet the precision tolerances required for coordinate measuring machine (CMM) inspection. Therefore, this invention utilizes this highly accurate... As the initial position, the traditional Iterative Closest Point (ICP) algorithm is called again, allowing ICP to use all the point cloud data for a final fine-tuning step.

[0077] Specifically, the solved global optimal transformation is applied to the original actual point cloud, that is, the rotation and translation operations included in the transformation are applied to all data points in the actual point cloud, thereby obtaining an aligned point cloud; further, this aligned point cloud is used as the basis for further analysis. and standard point cloud As input, execute a standard iterative nearest-point algorithm once; because Very close The iterative nearest point algorithm will no longer face the risk of local optima and will immediately converge to its precise final position.

[0078] The system outputs the finely registered transformation matrix and performs subsequent deviation calculations and tolerance analysis based on it. Once the system outputs the finely registered transformation matrix, it is immediately applied to the original, unaligned actual point cloud, ensuring perfect alignment with the standard point cloud in the same coordinate system. The system calculates the normal distance from each point in the aligned actual point cloud to the nearest CAD model surface. This process generates a global deviation chromatogram covering the entire component surface, using different colors to visually display the magnitude of the deviation. Based on the predefined geometric dimensions and tolerance specifications in the CAD model, the system automatically detects whether the measured values ​​of key features exceed the design's allowable error range, ultimately generating a detailed quality inspection report, such as the deviation chromatogram, thus completing the coordinate measuring machine (CMM) inspection process.

[0079] For example, Figure 8 This is a deviation chromatogram. The colors of the point cloud, from blue to red, represent the deviation between that point and its nearest point on the standard point cloud after fine registration; the color bars next to the graph show the mapping from color to specific deviation values. This graph represents the final result of the entire invention; blue areas indicate very small deviations, while red areas indicate larger deviations, serving as a key basis for analyzing the manufacturing quality of parts.

[0080] Thus, a method for processing coordinate measuring machine (CMM) data for automotive parts has been realized.

[0081] This invention also discloses a coordinate measuring machine (CMM) inspection data processing system for automotive parts, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a CMM inspection data processing method for automotive parts according to the present invention.

[0082] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for processing coordinate measuring machine (CMM) data for automotive parts, characterized in that, include: Acquire actual and standard point clouds of automotive parts; For each data point in the actual point cloud and the standard point cloud, the maximum angular gap of the data point is obtained based on the projection angle distribution of its neighboring data points on the tangent plane of that data point. This includes: obtaining the tangent plane basis vector at the data point through principal component analysis; projecting all neighboring data points of the data point onto the tangent plane to obtain projection vectors; calculating the angle of each projection vector relative to the tangent plane basis vector and storing it in an angle set; sorting the angle set in ascending order; calculating the angle difference between two adjacent angles after sorting, and the first-to-last wraparound angle difference between the last angle and the first angle; using the maximum value among all angle differences as the maximum angular gap of the data point; the projection vector satisfies the expression: , For data points The projection vector on the tangent plane. From the center point Pointing to neighboring data points The vector, and These are the first and second basis vectors of the tangent plane, respectively. Represents the dot product operation of vectors; Based on the maximum angular gap of each data point, the actual feature point set and the standard feature point set are obtained by filtering from the actual point cloud and the standard point cloud, respectively. For each feature point in both the actual feature point set and the standard feature point set, based on the distance between the feature point and its neighboring data points and the tangential consistency between the neighboring data points and the feature point's normal vector, the tangential weight between the feature point and its neighboring data points in their respective point clouds is obtained; the geometric descriptor of the feature point is obtained based on the tangential weight. For each actual feature point in the actual feature point set, find the standard feature point in the standard feature point set whose geometric descriptor is closest to it, and generate an initial matching set consisting of feature point pairs; A global optimal transformation is calculated from the initial matching set using a random sampling consensus algorithm. The global optimal transformation is then applied to the actual point cloud, and an iterative nearest point algorithm is executed for fine registration to obtain the final transformation matrix. The three-coordinate detection process is then completed based on the transformation matrix.

2. The method for processing coordinate measuring machine (CMM) data for automotive parts according to claim 1, characterized in that, The tangential weights satisfy the expression: ； in, Representing neighborhood points relative center point Tangential weights; The central feature point; for A neighboring data point within a neighborhood; express and The Euclidean distance between them; central feature point The normal vector; From point to , unit vector; Represents the dot product operation of vectors; It represents the absolute value.

3. The method for processing coordinate measuring machine (CMM) data for automotive parts according to claim 2, characterized in that, The geometry descriptor satisfies the expression: ； in, central feature point The geometry descriptor; and They are and Simplified histogram vector; for ; For neighborhood data points Compared to Tangential weights.

4. The method for processing coordinate measuring machine data for automotive parts according to claim 1, characterized in that, The step of using a random sample consensus algorithm to calculate a globally optimal transformation from the initial matching set includes: Within a preset number of iterations, the following operations are repeated: a preset number of correspondences are randomly selected from the initial matching set; a temporary rigid transformation is calculated based on the selected correspondences; the temporary rigid transformation is applied to the actual feature point set; then the initial matching set is traversed to count the number of interior points whose distance between the transformed feature points and their corresponding feature points is less than a preset distance threshold; after all iterations are completed, the temporary rigid transformation that obtains the most interior points among all calculated temporary rigid transformations is taken as the global optimal transformation.

5. The method for processing coordinate measuring machine data for automotive parts according to claim 1, characterized in that, The application of the global optimal transformation to the actual point cloud includes: The rotation matrix and translation vector of the globally optimal transformation are applied to all data points in the actual point cloud to obtain an aligned point cloud.

6. The method for processing coordinate measuring machine data for automotive parts according to claim 1, characterized in that, The process of selecting actual feature point sets and standard feature point sets from actual point clouds and standard point clouds respectively, based on the maximum angular gap of each data point, includes: A preset angle gap threshold is set; if the maximum angle gap of each data point in the actual point cloud is greater than the angle gap threshold, the data point is stored in the actual feature point set; if the maximum angle gap of each data point in the standard point cloud is greater than the angle gap threshold, the data point is stored in the standard feature point set.

7. A method for processing coordinate measuring machine (CMM) data for automotive parts according to claim 2, characterized in that, The normal vector is obtained by performing principal component analysis on the neighborhood of the central feature point.

8. A coordinate measuring machine (CMM) inspection data processing system for automotive parts, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a coordinate measuring machine data processing method for automotive parts according to any one of claims 1-7.

Images