Irregular part machine tool spindle self-adaptive positioning method and system

Abstract

The invention provides a self-adaptive positioning method and system for an irregular part machine tool spindle, and relates to the technical field of machine tool spindle positioning control. The method comprises the following steps: acquiring an actual measurement local feature set and a theoretical local feature set according to an actual measurement point cloud and a theoretical model point cloud of an irregular part; matching the actually measured local feature set with the theoretical local feature set to obtain a potential corresponding relation set; obtaining a reliable corresponding relation set according to the potential corresponding relation set; calculating an initial pose of the preprocessed point cloud relative to the theoretical model point cloud according to the reliable corresponding relation set; according to the initial pose, obtaining a final pose of the preprocessed point cloud relative to the theoretical model point cloud; and controlling the machine tool spindle to perform self-adaptive positioning according to the final pose. According to the method, the limitation of a traditional alignment method based on specific reference features is overcome, and accurate alignment of the actual poses of the parts and the theoretical model is achieved under the condition that the production takt requirement for rapid positioning is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine tool spindle positioning control, in particular to an irregular part machine tool spindle adaptive positioning method and system. BACKGROUND

[0002] In the field of large-scale complex equipment manufacturing, such as aerospace, energy equipment or heavy machinery industry, the precision machining of irregular parts is a crucial link. These parts, such as large structural parts, complex internal cavity castings or welded parts, are usually obtained by preliminary forming processes such as casting, forging or welding. However, due to the inherent characteristics of these processes, the geometry of the blank parts often deviates significantly from the ideal state of the theoretical design, with various and complex forms, including but not limited to overall distortion, local concave or convex, key size exceeding the allowable tolerance, and surface profile inconsistency and fluctuation. These geometric irregularities cause complex non-rigid deviations between the actual shape of the part and the theoretical CAD model.

[0003] In order to ensure that the subsequent cutting machining on the machine tool can meet the expected precision and safety requirements, it is necessary to first accurately determine the actual spatial pose of these irregular parts in the machine tool coordinate system. Traditional positioning methods often rely on specific geometric reference features on the part, such as planes, holes, cylinders, etc., and obtain the measured positions of these features through contact or non-contact measuring devices, and then match and align the measured features with the corresponding features in the theoretical model. However, for large-scale complex irregular parts, these theoretical reference features may also have geometric deviations on the actual blank parts, for example, a theoretical plane may become a slightly curved surface, and a theoretical positioning hole may not be round or may be offset. More challenging is that due to the complexity of the part structure and the field of view limitation of the measuring device, even if a non-contact scanning device (such as a laser scanner, a structured light camera) is used for multi-angle scanning, the point cloud data obtained may have local data blind spots and incompleteness, resulting in the reference features used for positioning being unable to be completely scanned or effectively extracted, thus making the traditional alignment method based on these biased or incomplete features unreliable, easily leading to positioning failure or positioning error exceeding the allowable range, and further affecting the subsequent machining quality, which may cause overcut or residual allowance.

[0004] Modern manufacturing processes have higher requirements for production efficiency, especially for the machining of large parts, which requires rapid completion of part pose determination within a limited machine tool non-cutting time. Usually, this process needs to be completed within a few minutes. Currently, the pose determination of irregular parts on the machine tool usually relies on the non-contact measurement system integrated on the machine tool, which obtains the surface point cloud data of the part by scanning the part fixed on the worktable. Due to the large volume and complex geometry of the part (including deep cavities, bosses, stiffeners, etc.), in order to obtain sufficient surface information for reliable pose determination, the scanning operation needs to cover most or even all of the accessible surface of the part. This full-coverage scanning process will generate a large-scale raw point cloud dataset, which has a data volume far exceeding that of conventional parts, and may contain millions or even tens of millions of (X, Y, Z) coordinate points. For example, a large structural part with a length of several meters, using a high-resolution scanner for full-surface scanning, will easily generate point cloud data in the order of tens of millions of points.

[0005] Processing such a large-scale, partially incomplete, and containing diverse geometric irregular information raw point cloud data, and performing effective feature extraction and alignment calculation with the theoretical model, is a computationally intensive task. Traditional point cloud processing algorithms (such as voxel downsampling, statistical filtering) take a long time to calculate when dealing with such a large amount of data. Common point cloud registration algorithms (such as Iterative Closest Point ICP and its variants) are prone to slow convergence or local optimal solution when facing incomplete data, large initial position deviation, and complex local deformation of the part, resulting in inaccurate alignment results, further increasing the calculation time or requiring manual intervention.

[0006] Therefore, in the automated machining scene of large and complex irregular parts, how to effectively process and utilize the large-scale, incomplete, and containing diverse geometric irregular information point cloud data from in-machine scanning within the limited positioning time, overcome the limitations of traditional alignment methods based on specific reference features, and achieve rapid, high-precision, and high-robustness determination of the actual pose of the part, is a key challenge to realize efficient and reliable automated machining. The existing positioning data processing methods are difficult to meet the multiple constraints of processing large-scale complex data, dealing with diverse geometric irregularities, handling data incompleteness, and meeting real-time requirements, limiting the efficiency and reliability of automated precision machining of irregular parts. SUMMARY

[0007] The purpose of the present application is to provide an irregular part machine tool spindle adaptive positioning method and system, which overcomes the limitations of traditional alignment methods based on specific reference features, and realizes accurate alignment of the actual pose of the part with the theoretical model under the requirement of fast positioning production rhythm.

[0008] In a first aspect, the present application provides an irregular part machine tool spindle adaptive positioning method, comprising the following steps:

[0009] S1. obtaining a measured point cloud and a theoretical model point cloud of an irregular part;

[0010] S2. preprocessing the measured point cloud to obtain a preprocessed point cloud;

[0011] S3. extracting a first local geometric feature according to the preprocessed point cloud to obtain a measured local feature set, and extracting a second local geometric feature according to the theoretical model point cloud to obtain a theoretical local feature set;

[0012] S4. matching between the measured local feature set and the theoretical local feature set to obtain a set of potential corresponding relationships;

[0013] S5. screening corresponding relationships that meet a preset spatial consistency constraint condition from the set of potential corresponding relationships to obtain a set of reliable corresponding relationships;

[0014] S6. calculating an initial pose of the preprocessed point cloud relative to the theoretical model point cloud according to the set of reliable corresponding relationships;

[0015] S7. performing iterative alignment of the preprocessed point cloud and the theoretical model point cloud according to the initial pose to obtain a final pose of the preprocessed point cloud relative to the theoretical model point cloud;

[0016] S8. controlling a machine tool spindle to perform adaptive positioning according to the final pose.

[0017] The irregular part machine tool spindle adaptive positioning method provided by the application is characterized in that local geometric features that are relatively insensitive to part irregularity and data incompleteness are used for matching and alignment of the measured point cloud and the theoretical model. The method extracts and describes local geometric features on the measured point cloud and the theoretical model, constructs feature corresponding relationships, and calculates the actual pose of the part based on these corresponding relationships. The entire process considers the processing efficiency of large-scale data and the timeliness of in-machine positioning.

[0018] In a second aspect, the application provides an irregular part machine tool spindle adaptive positioning system, comprising:

[0019] An acquisition module is configured to obtain a measured point cloud and a theoretical model point cloud of an irregular part;

[0020] A preprocessing module is configured to preprocess the measured point cloud to obtain a preprocessed point cloud;

[0021] An extraction module is configured to extract a first local geometric feature according to the preprocessed point cloud to obtain a measured local feature set, and extract a second local geometric feature according to the theoretical model point cloud to obtain a theoretical local feature set;

[0022] The matching module is configured to match the measured local feature set and the theoretical local feature set to obtain a set of potential corresponding relations;

[0023] The screening module is configured to screen corresponding relations that meet a preset spatial consistency constraint condition from the set of potential corresponding relations to obtain a set of reliable corresponding relations;

[0024] The calculation module is configured to calculate an initial pose of the preprocessed point cloud relative to the theoretical model point cloud according to the set of reliable corresponding relations.

[0025] The iteration module is configured to perform iterative alignment on the preprocessed point cloud and the theoretical model point cloud according to the initial pose to obtain a final pose of the preprocessed point cloud relative to the theoretical model point cloud.

[0026] The control module is configured to control the machine tool spindle to perform adaptive positioning according to the final pose.

[0027] As can be seen from the above, the irregular part machine tool spindle adaptive positioning method provided by the present application significantly reduces the computational complexity of processing large-scale point cloud data through data downsampling (preprocessing), fast feature matching based on spatial indexing and robust screening algorithm, meets the production rhythm requirements of machine tool rapid positioning, and can effectively identify and eliminate false matching pairs, so that the pose calculation is not significantly affected by local incomplete areas and various geometric irregularities (such as local deformation and outliers), and the reliability of the positioning process is improved. Combined with iterative optimization for fine alignment, the accuracy problem caused by deviation of the reference feature or ICP falling into local optimum in the traditional method can be overcome, high-precision determination of the actual pose of the part is realized, and the method does not depend on specific geometric reference features preset on the part, but uses the local geometric characteristics universally existing on the part surface for matching, so it has stronger adaptability to the geometric shape and irregularity of the part, and is especially suitable for complex irregular parts with unreliable or difficult-to-extract traditional reference features.

[0028] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or can be learned by practice of the present application as hereinafter described. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 A flowchart of the irregular part machine tool spindle adaptive positioning method provided by the embodiment of the present application.

[0030] Figure 2 A structural schematic diagram of the irregular part machine tool spindle adaptive positioning system provided by the embodiment of the present application.

[0031] Label explanation:

[0032] 100, acquisition module; 200, preprocessing module; 300, extraction module; 400, matching module; 500, screening module; 600, calculation module; 700, iteration module; 800, control module. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0034] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0035] With reference to the accompanying drawings, Figure 1 the present application provides an irregular part machine tool spindle adaptive positioning method, comprising the following steps:

[0036] S1. acquiring a measured point cloud and a theoretical model point cloud of an irregular part;

[0037] S2. preprocessing the measured point cloud to obtain a preprocessed point cloud; the preprocessing includes data volume reduction, abnormal point removal, and surface direction information calculation;

[0038] S3. extracting a first local geometric feature according to the preprocessed point cloud to obtain a measured local feature set, and extracting a second local geometric feature according to the theoretical model point cloud to obtain a theoretical local feature set;

[0039] S4. matching between the measured local feature set and the theoretical local feature set to obtain a set of potential corresponding relationships;

[0040] S5. screening out corresponding relationships that meet a preset spatial consistency constraint condition according to the set of potential corresponding relationships to obtain a set of reliable corresponding relationships;

[0041] S6. calculating an initial pose of the preprocessed point cloud relative to the theoretical model point cloud according to the set of reliable corresponding relationships;

[0042] S7. According to the initial pose, iteratively aligning the pre-processed point cloud and the theoretical model point cloud to obtain the final pose of the pre-processed point cloud relative to the theoretical model point cloud;

[0043] S8. According to the final pose, controlling the main shaft of the machine tool to perform adaptive positioning.

[0044] The measured point cloud of an irregular part refers to a set of three-dimensional coordinate data obtained by scanning the surface of the irregular part with a measuring device. It can be achieved by using non-contact scanning devices such as laser scanners and structured light cameras. The main purpose is to obtain the actual geometric information of the part. The theoretical model point cloud refers to a set of point cloud data generated based on the theoretical design model of the irregular part. It can be achieved by using CAD model conversion software such as converting CAD models in STEP or IGES format to point cloud format. The main purpose is to provide ideal geometric reference for the part. Preprocessing of the measured point cloud refers to a series of operations on the original measured point cloud before subsequent processing. Preprocessing includes data reduction, outlier removal, and surface direction information calculation. The main purpose is to improve the quality and processing efficiency of point cloud data. Data reduction refers to reducing the number of point cloud data points. It can be achieved by using voxel downsampling or random sampling. The main purpose is to reduce the computational complexity. Outlier removal refers to identifying and removing points in the point cloud that do not conform to the distribution of surrounding points. It can be achieved by using statistical filtering or methods based on neighborhood analysis. The main purpose is to remove noise points. Surface direction information calculation refers to calculating the surface normal vector at each point in the point cloud. It can be achieved by using methods based on fitting a plane to the neighborhood points or principal component analysis. The main purpose is to provide local geometric information for subsequent feature extraction. Extracting local geometric features refers to calculating geometric descriptors of local regions in the point cloud. It can be achieved by using methods based on the shape distribution of point neighborhood or local coordinate system. The main purpose is to describe the local geometric characteristics of the point cloud. The measured local feature set refers to a set of local geometric features extracted from the preprocessed point cloud. The main purpose is to represent the local shape of the measured part. The theoretical local feature set refers to a set of local geometric features extracted from the theoretical model point cloud. The main purpose is to represent the local shape of the theoretical model. Matching between the measured local feature set and the theoretical local feature set refers to finding the correspondence between the measured local features and the theoretical local features. It can be achieved by using methods based on similarity comparison of feature descriptors. The main purpose is to preliminarily establish the correspondence between the measured point cloud and the theoretical model point cloud. The set of potential correspondence relationships refers to a set of preliminary correspondence relationships obtained by feature matching, which may contain incorrect matches. The main purpose is to provide a basis for subsequent screening. The spatial consistency constraint condition refers to a criterion for judging whether the potential correspondence relationship conforms to the overall spatial transformation rule. It can be achieved by using methods based on the relative position or distance invariance of corresponding point pairs. The main purpose is to eliminate incorrect matches. The reliable correspondence relationship set refers to a set of correspondence relationships retained after spatial consistency constraint screening, with a lower proportion of incorrect matches. The main purpose is to provide reliable input for subsequent pose calculation.The initial pose calculation refers to calculating a preliminary transformation matrix of the measured point cloud relative to the theoretical model point cloud according to a reliable correspondence set. It can be realized by using a method of solving a rigid body transformation based on corresponding point pairs, for example, using the RANSAC algorithm, which is mainly used to provide a good starting point for iterative alignment. Iterative alignment refers to gradually adjusting the pose of the measured point cloud to make it more accurately coincide with the theoretical model point cloud through an iterative optimization algorithm starting from the initial pose. It can be realized by using an iterative closest point (ICP) algorithm or its variants, which is mainly used to obtain a high-precision final alignment pose. The final pose refers to the accurate transformation matrix of the measured point cloud relative to the theoretical model point cloud obtained at the end of the iterative alignment process, which is mainly used to reflect the actual spatial position and attitude of the part. Controlling the machine tool spindle according to the final pose is to apply the calculated final pose information to the machine tool control system to adjust the motion trajectory or coordinate system of the machine tool spindle to match the actual spatial position and attitude of the part, which is mainly used to realize accurate machining of irregular parts.

[0045] The core working principle of the method is: by extracting and matching local geometric features (such as curvature, normal vector distribution, etc.) that are relatively insensitive to the overall irregularity of the part and the local data incompleteness, reliable local correspondence is established between the measured point cloud and the theoretical model. Using these reliable local correspondences, rather than relying on global reference features that may be biased or cannot be extracted, the actual pose of the part is calculated. The entire process ensures processing efficiency and robustness under large-scale, incomplete, and irregular data through data preprocessing, efficient feature matching, and robust screening technology.

[0046] The core innovation of the present application is that by combining local geometric feature-based matching with spatial consistency constraint-based screening, and calculating the initial pose on this basis, and then performing iterative alignment, it can quickly, accurately, and robustly determine the pose of irregular parts when processing large-scale, locally incomplete, and containing diverse geometric irregular information point cloud data, overcoming the limitations of traditional alignment methods based on specific reference features.

[0047] Specifically, the method of the present application is carried out according to the following flow: first, the measured point cloud reflecting the actual shape of the part and the theoretical model point cloud representing the ideal shape are obtained as the basis data for subsequent processing. Then, the measured point cloud is pre-processed, including data reduction, abnormal point removal and surface direction information calculation, to obtain point cloud data with higher quality and easier to process. Then, local geometric features are extracted from the pre-processed point cloud and the theoretical model point cloud, respectively, to obtain the measured local feature set and the theoretical local feature set, which describe the local geometric characteristics of the point cloud. Subsequently, matching is performed between the measured local feature set and the theoretical local feature set, and a set of potential correspondence relationships between the measured point cloud and the theoretical model point cloud is initially established. Since the set of potential correspondence relationships may contain false matches, it is necessary to screen the correspondence relationships that meet the pre-set spatial consistency constraint conditions from the set of potential correspondence relationships to obtain a set of reliable correspondence relationships, thereby eliminating false matches and improving the accuracy of the correspondence relationships. Using the set of reliable correspondence relationships obtained by screening, the initial pose of the pre-processed point cloud relative to the theoretical model point cloud is calculated, providing a good starting point for subsequent fine alignment. Starting from the initial pose, the pre-processed point cloud and the theoretical model point cloud are iteratively aligned, and the pose is gradually optimized until the convergence condition is met, to obtain the final pose of the pre-processed point cloud relative to the theoretical model point cloud. Finally, the final pose calculated is used to control the spindle of the machine tool for adaptive positioning, and the actual spatial pose information of the part is applied to the machine tool machining process. The entire process combines local feature matching, robust screening and iterative optimization to achieve accurate determination of the pose of irregular parts.

[0048] As a preferred embodiment, the scheme of the present application is implemented as follows: a three-dimensional scanner installed on a machine tool is used to scan an irregular part fixed on a workbench to obtain measured point cloud data, and theoretical model point cloud data of the part is obtained from a computer-aided design system. The measured point cloud data obtained is input into a computer connected to the machine tool control system, and a point cloud preprocessing program is executed by the computer. The program performs voxel downsampling on the measured point cloud to reduce the data volume, removes abnormal points by executing a statistical filtering algorithm, and calculates the surface normal vector of each point. The preprocessed point cloud data is input into a feature extraction module, which calculates the local feature descriptor at the key points in the measured point cloud to form a measured local feature set. At the same time, a similar feature extraction process is performed on the theoretical model point cloud to obtain a theoretical local feature set. The measured local feature set and the theoretical local feature set are input into a feature matching module, which generates a set of potential corresponding relationships by comparing the similarity of the feature descriptors. The set of potential corresponding relationships is input into a screening module, which uses a method based on random sample consensus (RANSAC) to iteratively select a subset of potential corresponding relationships, calculate candidate rigid body transformations, and screen out a set of reliable corresponding relationships that meet the spatial consistency constraint condition according to the transformation results and local consistency evaluation. Using the set of reliable corresponding relationships, the initial pose of the measured point cloud relative to the theoretical model point cloud is calculated. The initial pose is input to execute an iterative alignment algorithm, such as the point-to-point or point-to-plane ICP algorithm, to calculate the optimal rigid body transformation that minimizes the distance between the two point clouds through iterative optimization, and obtain the final pose. The calculated final pose (including the rotation matrix and the displacement vector) is sent to the machine tool numerical control system, and the numerical control system adjusts the coordinate system or motion instructions of the machine tool spindle according to the pose to realize adaptive machining of the irregular part.

[0049] Through the above scheme, the present application can effectively process large-scale, locally incomplete and containing diverse geometric irregular information point cloud data, overcome the limitations of traditional alignment methods based on specific reference features, realize fast, high-precision and high-robustness determination of the pose of irregular parts, and improve the efficiency and reliability of automatic machining of irregular parts.

[0050] In some embodiments, the specific steps in step S4 include:

[0051] S41. For each measured local feature in the measured local feature set, search and determine a plurality of candidate theoretical local features in the theoretical local feature set;

[0052] S42. For each measured local feature and its corresponding plurality of candidate theoretical local features, calculate the feature similarity between the measured local feature and each candidate theoretical local feature;

[0053] S43. Obtain all pre-processed point clouds within the measured local feature neighborhood as a measured neighborhood point set, and all theoretical model point clouds within each candidate theoretical local feature neighborhood as a theoretical neighborhood point set;

[0054] S44. Based on the measured neighborhood point set and the theoretical neighborhood point set, evaluate the local geometric consistency of each candidate theoretical local feature as a corresponding relationship of the measured local feature;

[0055] S45. According to the evaluation results of the feature similarity and the local geometric consistency, determine one or more potential corresponding relationships for each measured local feature, and obtain a potential corresponding relationship set by collecting the potential corresponding relationships determined for each measured local feature.

[0056] Searching and determining a plurality of candidate theoretical local features refers to, for each local feature extracted from the measured point cloud, finding several feature points in the set of local features extracted from the theoretical model point cloud that are close to it in the feature space. This search can be implemented by constructing an index structure of the set of theoretical local features, for example using a k-d tree or octree, and then performing a nearest neighbor search based on the distance between feature descriptors (such as the Euclidean distance), selecting the closest plurality of theoretical local features. This approach expands the range of potential matching objects, increasing the probability of finding the correct correspondence, especially when the feature descriptors are not accurate due to noise or deformation. Calculating feature similarity refers to quantifying the matching degree between the measured local feature and each candidate theoretical local feature. This is usually achieved by comparing their local feature descriptors, such as calculating the distance or angle between descriptor vectors. Feature similarity provides a preliminary matching basis based on local surface properties. Obtaining all preprocessed point clouds within the neighborhood of the measured local feature as the measured neighborhood point set, and all theoretical model point clouds within the neighborhood of each candidate theoretical local feature as the theoretical neighborhood point set refers to extracting the original point cloud data around the local feature point within a predetermined neighborhood range (such as a fixed radius spherical region or a region composed of k nearest neighbors). These neighborhood point sets contain more abundant geometric information around the local feature point and are the basis for local geometric consistency evaluation. Evaluating the local geometric consistency of each candidate theoretical local feature as the correspondence of the measured local feature refers to comparing the geometric structure similarity between the measured neighborhood point set and the theoretical neighborhood point set. This evaluation does not rely on a single feature descriptor, but examines whether the shape, curvature distribution, or point pair relationship of the local region as a whole matches. Local geometric consistency provides a verification of the overall structure of the local region, effectively distinguishing points with similar geometric structures but different feature descriptors, or points with similar feature descriptors but mismatched geometric structures. According to the evaluation results of feature similarity and local geometric consistency, determining one or more potential corresponding relationships for each measured local feature refers to comprehensively utilizing the similarity of feature descriptors and the geometric consistency of local neighborhood point clouds to determine whether a candidate theoretical local feature constitutes an effective potential corresponding relationship with the current measured local feature. This can be achieved by setting thresholds, weighted combination scores, or using more complex decision logic. By collecting the potential corresponding relationships determined for each measured local feature, a potential corresponding relationship set is obtained, which refers to collecting all the corresponding relationships between the measured local features and their determined one or more candidate theoretical local features, forming a set containing all possible matching pairs, which will be used for subsequent global consistency screening.

[0057] The reason why the matching method is used in the present application is that in the machine scanning scene of a large irregular part, due to local deformation, machining allowance, scanning noise or data incompleteness, the local feature descriptors of the measured point cloud may be different from the feature descriptors of the corresponding positions of the theoretical model. Simply relying on feature similarity for matching is easy to produce false matching or missing matching. It is because on the basis of preliminarily screening a plurality of candidate theoretical local features based on feature similarity, the local geometric consistency between the measured local feature neighborhood point set and the candidate theoretical local feature neighborhood point set is further introduced, so that the matching process can consider the local surface attribute and the overall geometric structure of the local region at the same time. By comprehensively utilizing the feature similarity and the local geometric consistency, the potential corresponding relationship can be judged more robustly, the influence of local deformation and machining allowance on the feature descriptor can be effectively coped with, the number of false matching can be reduced, and the accuracy and reliability of the potential corresponding relationship set can be improved. This method, combined with the previous steps of obtaining the measured point cloud and the theoretical model point cloud, preprocessing the measured point cloud and extracting local geometric features, provides high-quality input for subsequent global consistency screening and initial pose calculation based on the potential corresponding relationship set, thereby improving the robustness and precision of the entire irregular part pose determination method.

[0058] As a specific implementation, when matching, for each measured local feature in the measured local feature set, k nearest neighbor theoretical local features in the theoretical local feature set can be searched based on the Euclidean distance of the feature descriptors as candidate theoretical local features. Then, the feature similarity between the measured local feature and the k candidate theoretical local features can be calculated, for example, the normalized cross-correlation coefficient of the descriptor vector can be used as the similarity measure. At the same time, for each candidate theoretical local feature, all point clouds in a predetermined radius neighborhood in the theoretical model point cloud are extracted as the theoretical neighborhood point set, and all point clouds in the same radius neighborhood in the preprocessed point cloud corresponding to the measured local feature are extracted as the measured neighborhood point set. Then, based on the measured neighborhood point set and the theoretical neighborhood point set, the local geometric consistency is evaluated, for example, the Hausdorff distance or the chamfer distance between the two neighborhood point sets can be calculated, or the principal direction, curvature distribution and other statistical features of the two neighborhood point sets are compared. Finally, the feature similarity score and the local geometric consistency evaluation score are combined, for example, by weighted summation or by setting a double threshold, to screen the candidate theoretical local features that meet the conditions, and form potential corresponding relationships with the corresponding measured local features. All potential corresponding relationships determined by the measured local features are collected to form a potential corresponding relationship set.

[0059] By the matching method, the similarity of local features and the geometric consistency of local neighborhood point clouds are combined to jointly determine the potential correspondence, which can effectively deal with the feature descriptor difference problem caused by local deformation and machining allowance in the scene of in-machine scanning of large irregular parts. Thus, the accuracy and robustness of matching between the measured local feature set and the theoretical local feature set are improved, providing more reliable input for subsequent global consistency screening and pose calculation, thereby improving the overall precision and reliability of the adaptive positioning of the spindle of the irregular part machine tool.

[0060] In some embodiments, the specific steps in step S44 include:

[0061] S441. Calculate the relative geometric relationship between the point pairs inside the measured neighborhood point set to obtain a measured neighborhood geometric relationship set;

[0062] S442. Calculate the relative geometric relationship between the point pairs inside the theoretical neighborhood point set to obtain a theoretical neighborhood geometric relationship set;

[0063] S443. Calculate the neighborhood similarity between the measured neighborhood point set and the theoretical neighborhood point set by comparing the measured neighborhood geometric relationship set and the theoretical neighborhood geometric relationship set;

[0064] S444. According to the neighborhood similarity, evaluate the local geometric consistency of each candidate theoretical local feature as the corresponding relationship of the measured local feature.

[0065] The relative geometric relationship refers to the geometric measurement between point pairs, which can be realized by the Euclidean distance between point pairs, the angle between the connecting line of point pairs and the surface normal, or the edge length ratio or angle of a triangle formed by three points. The measured neighborhood geometric relationship set and the theoretical neighborhood geometric relationship set refer to a set of one or more sets of numerical values obtained by calculating the relative geometric relationship between the point pairs inside the neighborhood point set. Comparing the measured neighborhood geometric relationship set and the theoretical neighborhood geometric relationship set refers to comparing and analyzing the relative geometric relationships in the two sets, which can be realized by calculating the difference statistics (such as mean square error, correlation coefficient) of the corresponding relationships (such as distance to distance, angle to angle) in the two sets, or constructing a histogram of the two sets and comparing the similarity of the histograms (such as Bhattacharyya distance, KL divergence). The neighborhood similarity refers to a numerical value that quantitatively represents the similarity degree of the shape of the measured neighborhood point set and the theoretical neighborhood point set, which is usually between 0 and 1, and the larger the value, the higher the similarity. Evaluating the local geometric consistency refers to determining the matching degree of the measured neighborhood and the theoretical neighborhood in the local geometric shape according to the calculated neighborhood similarity, which can directly use the calculated neighborhood similarity as the evaluation score of the local geometric consistency, or map the similarity to the consistency score through a function.

[0066] The application obtains a measured neighborhood geometric relation set by calculating relative geometric relations between point pairs inside a measured neighborhood point set, and obtains a theoretical neighborhood geometric relation set by calculating relative geometric relations between point pairs inside a theoretical neighborhood point set. These relative geometric relations have invariance or weak invariance relative to overall pose changes, and can capture the inherent local shape features of the measured neighborhood and the theoretical neighborhood, thereby quantitatively representing the geometric information thereof. By comparing the two geometric relation sets, the similarity of the measured neighborhood and the theoretical neighborhood in shape can be more robustly measured. This comparison based on relative relations can effectively deal with the interference of local deformation of irregular parts and data incompleteness on direct shape comparison, thereby calculating a similarity value that more reflects the matching degree of the real shape. According to the neighborhood similarity, the local geometric consistency of each candidate theoretical local feature as the corresponding relation of the measured local feature is evaluated. The higher the neighborhood similarity, the more matched the shape of the measured neighborhood and the theoretical neighborhood, and the higher the local geometric consistency evaluated. This evaluation based on the quantified similarity provides a more accurate and reliable basis for subsequently determining potential corresponding relations in combination with feature similarity. In the entire method flow, accurate local geometric consistency evaluation improves the quality of the potential corresponding relation set, and further improves the accuracy and robustness of subsequent screening of reliable corresponding relations, calculation of an initial pose, and final iterative alignment. In particular, when processing large and complex irregular parts, this scheme can effectively deal with the actual local non-rigid deformation, and improve the accuracy and reliability of pose determination.

[0067] In one embodiment, when calculating the relative geometric relations between point pairs inside the measured neighborhood point set, the Euclidean distances between each point in the measured neighborhood point set and all other points in the neighborhood can be calculated, and these distance values can be collected to form the measured neighborhood geometric relation set. Similarly, when calculating the relative geometric relations between point pairs inside the theoretical neighborhood point set, the Euclidean distances between each point in the theoretical neighborhood point set and all other points in the neighborhood can be calculated, and these distance values can be collected to form the theoretical neighborhood geometric relation set. When comparing the measured neighborhood geometric relation set and the theoretical neighborhood geometric relation set, the distance values in the two sets can be sorted respectively, and then the sum of the squares of the difference values of the distances at corresponding positions can be calculated, and the reciprocal thereof can be taken as the neighborhood similarity. For example, if the distance values of the two sets are {d s1 , d s2 ,..., d sn} and {d t1 , d t2 ,..., d tn} (sorted), the similarity can be calculated as When evaluating the local geometric consistency according to the neighborhood similarity, the calculated neighborhood similarity value can be directly taken as the evaluation score of the local geometric consistency.

[0068] By calculating the relative geometric relationship between the point pairs inside the measured neighborhood point set and obtaining the measured neighborhood geometric relationship set, and calculating the relative geometric relationship between the point pairs inside the theoretical neighborhood point set and obtaining the theoretical neighborhood geometric relationship set, the application can capture the inherent local shape features of the measured neighborhood and the theoretical neighborhood. By comparing the two geometric relationship sets and calculating the neighborhood similarity, the similarity of the shape between the measured neighborhood and the theoretical neighborhood can be more robustly measured, effectively dealing with the local deformation of irregular parts and data incompleteness. According to the neighborhood similarity, the local geometric consistency of each candidate theoretical local feature as the corresponding relationship of the measured local feature is evaluated, which improves the accuracy of the local geometric consistency evaluation, provides more accurate and reliable basis for subsequent determination of potential corresponding relationships, and further improves the accuracy and robustness of the in-machine positioning of irregular parts.

[0069] In some embodiments, the specific steps in step S5 include:

[0070] S51. Selecting a subset from the set of potential corresponding relationships;

[0071] S52. According to the selected subset, calculating a candidate global rigid body transformation;

[0072] S53. According to the candidate global rigid body transformation, calculating the distance between the position of the measured point after the global rigid body transformation and the position of the corresponding theoretical point in each corresponding relationship in the set of potential corresponding relationships, to obtain the global residual of each corresponding relationship;

[0073] S54. According to the global residual of each corresponding relationship and the evaluation result of the local geometric consistency, determining whether each corresponding relationship meets the spatial consistency constraint condition by judging whether each corresponding relationship meets the preset global residual threshold condition and the preset local geometric consistency threshold condition at the same time;

[0074] S55. Screening the corresponding relationships that meet the spatial consistency constraint condition to obtain the set of reliable corresponding relationships.

[0075] The set of potential corresponding relationships refers to the sum of the corresponding relationships obtained after matching the set of measured local features and the set of theoretical local features, which includes correct corresponding relationships and incorrect corresponding relationships.

[0076] The subset refers to part of the corresponding relationships selected from the set of potential corresponding relationships, which is used to estimate the spatial transformation relationship between the measured point cloud and the theoretical model point cloud.

[0077] The global rigid body transformation candidate refers to a transformation calculated according to the selected subset, which describes the position and attitude relationship between the measured point cloud and the theoretical model point cloud, and the transformation can be expressed in the form of a matrix.

[0078] The global residual refers to the spatial distance between the transformed position of each corresponding relationship in the potential corresponding relationship set after applying the global rigid body transformation candidate and the position of the corresponding theoretical point, which reflects the fitting degree of the corresponding relationship under the overall transformation.

[0079] The local geometric consistency evaluation result refers to an evaluation value calculated based on the geometric similarity between the measured local feature neighborhood point set and the corresponding theoretical local feature neighborhood point set in the matching process, which reflects the geometric fitting degree of the local area where the corresponding relationship is located.

[0080] The preset global residual threshold condition refers to a numerical or logical judgment standard used to judge whether the global residual of the corresponding relationship meets the requirements, for example, the global residual is less than a certain preset distance value.

[0081] The preset local geometric consistency threshold condition refers to a numerical or logical judgment standard used to judge whether the local geometric consistency evaluation result of the corresponding relationship meets the requirements, for example, the local geometric consistency evaluation result is greater than a certain preset similarity value.

[0082] The spatial consistency constraint condition refers to a comprehensive judgment criterion used to filter potential corresponding relationships, which requires that the corresponding relationship meets the preset global residual threshold condition and the preset local geometric consistency threshold condition at the same time.

[0083] The reliable corresponding relationship set refers to the corresponding relationship set retained after being filtered by the spatial consistency constraint condition, and the corresponding relationship in this set is accurate.

[0084] The scheme describes how to filter out reliable corresponding relationships from the potential corresponding relationship set to solve the technical problem that in the scene of in-machine scanning and machining, due to the actual local non-rigid deformation of the part, even if the corresponding relationship is accurate, after applying the overall rigid body transformation, there may be a residual error between the measured point and the theoretical point that exceeds the preset threshold, affecting the filtering accuracy based on the strict rigid body transformation. The scheme judges the reliability of the corresponding relationship by combining global consistency information (global residual) and local consistency information (local geometric consistency evaluation result).

[0085] Step S51 selects a subset from the potential corresponding relationship set, which is to estimate the transformation relationship from a subset composed of accurate corresponding relationships in the set containing false matches. The subset-based estimation method should deal with noise and false corresponding relationships in the potential corresponding relationship set.

[0086] Step S52 calculates a candidate global rigid body transformation according to the selected subset. The global rigid body transformation describes the transformation of the measured point cloud relative to the theoretical model point cloud in terms of position and orientation. Through the estimated global transformation, the measured point cloud can be aligned to the coordinate system of the theoretical model point cloud.

[0087] Step S53 calculates the global residual of each correspondence in the potential correspondence set according to the candidate global rigid body transformation. The global residual is the distance between the position of the measured point after the global rigid body transformation and the position of the corresponding theoretical point. The global residual reflects the overall deviation of the correspondence under the current global transformation. A smaller residual indicates that the correspondence conforms to the current global transformation model.

[0088] Step S54 determines whether each correspondence meets the spatial consistency constraint condition by judging whether each correspondence meets the preset global residual threshold condition and the preset local geometric consistency threshold condition according to the global residual of each correspondence and the evaluation result of the local geometric consistency. It combines the global residual and the evaluation result of the local geometric consistency. A correspondence that meets both the global consistency requirement (the global residual is less than the preset threshold) and the local geometric similarity requirement (the evaluation result of the local geometric consistency meets the preset threshold) is determined to be reliable. This double judgment mechanism identifies and excludes false matches, especially those correspondences that are locally similar but globally inconsistent, and those correspondences that are accurately matched locally but have a global residual exceeding the preset range due to local deformation of the part or incomplete data, thereby improving the screening accuracy.

[0089] Step S55 screens out the correspondences that meet the spatial consistency constraint condition to obtain a reliable correspondence set. These screened correspondences are accurate.

[0090] The present scheme is combined with the potential correspondence set and the local geometric consistency evaluation result provided by step S4. Step S4 preliminarily identifies the potential correspondence and provides the local similarity information through local feature matching and local geometric consistency evaluation. However, the matching result of S4 contains false matching, and the local evaluation cannot fully reflect the global spatial relationship. Step S5 uses the information provided by S4 to form a comprehensive and robust screening criterion by introducing global rigid body transformation and global residual calculation, and combining global consistency judgment with the local consistency evaluation result provided by S4. This combination makes the screening process consider the position relationship of the correspondence in the whole space and the geometric consistency of the local area where the correspondence is located, so as to accurately identify the true correspondence relationship in the presence of local deformation of parts, incomplete data or false matching. This combination of global and local information screening method overcomes the limitations of relying on a single information source for screening, and obtains pure and reliable correspondence set.

[0091] Specifically, the global rigid body transformation includes a rotation matrix and a displacement vector.

[0092] The global rigid body transformation refers to a mathematical representation of the transformation of an object from one position and posture to another position and posture in a three-dimensional space, which maintains the relative distance between the internal points of the object unchanged, i.e. no deformation, which can be represented in various mathematical forms, such as quaternion plus displacement vector, Euler angle plus displacement vector, etc. The rotation matrix refers to a 3x3 matrix used to describe the rotation operation in a three-dimensional space, which can be calculated based on, for example, axis angle, Euler angle or quaternion. The displacement vector refers to a 3-dimensional vector used to describe the translation operation in a three-dimensional space, which can be obtained by, for example, directly measuring or calculating the coordinate difference between two points.

[0093] In some embodiments, the specific steps in step S54 include:

[0094] S541. For each correspondence in the potential correspondence set, obtain a set of measured neighborhood points of the measured local feature and a set of corresponding theoretical neighborhood points of the corresponding theoretical local feature;

[0095] S542. According to the set of measured neighborhood points and the set of theoretical neighborhood points, evaluate the data integrity of the set of measured neighborhood points and the set of theoretical neighborhood points to obtain a neighborhood data integrity evaluation result corresponding to each correspondence;

[0096] S543. According to the neighborhood data integrity evaluation result, calculate a reliability factor of the local geometric consistency evaluation result corresponding to each correspondence;

[0097] S544. According to the reliability factor, the local geometric consistency evaluation result corresponding to each correspondence is corrected to obtain a corrected local geometric consistency evaluation score;

[0098] S545. The global residual of each correspondence is compared with a preset global residual threshold, and the corrected local geometric consistency evaluation score is compared with a preset local geometric consistency threshold;

[0099] S546. According to the comparison result, it is judged whether each correspondence satisfies the preset global residual threshold condition and the preset local geometric consistency threshold condition at the same time, so as to determine whether each correspondence meets the spatial consistency constraint condition.

[0100] The set of potential correspondences refers to the set of possible correspondences obtained after matching the set of measured local features and the set of theoretical local features; the set of measured neighborhood points of the measured local feature refers to the measured point cloud data within a certain range around the measured local feature; the set of theoretical neighborhood points of the theoretical local feature refers to the theoretical point cloud data within a certain range around the theoretical local feature; the neighborhood data integrity evaluation result refers to the degree of data missing or incompleteness of the set of measured neighborhood points or the set of theoretical neighborhood points, which can be evaluated by using point cloud density, proportion of point number to theoretical point number, surface coverage rate and the like; the reliability factor refers to the weight or coefficient reflecting the credibility of the local geometric consistency evaluation result, which can be calculated according to the neighborhood data integrity evaluation result, for example, the higher the integrity, the closer the factor to 1; the corrected local geometric consistency evaluation score refers to the adjusted score obtained by multiplying the original local geometric consistency evaluation result or combining the reliability factor through a functional relationship; the global residual refers to the distance between the measured point after global rigid body transformation and the corresponding theoretical point; the preset global residual threshold refers to the upper limit of the distance for judging global consistency; the preset local geometric consistency threshold refers to the lower limit or upper limit of the score for judging local consistency; the spatial consistency constraint condition refers to the condition of simultaneously satisfying the global residual less than or equal to the preset global residual threshold and the corrected local geometric consistency evaluation score greater than or equal to the preset local geometric consistency threshold or less than or equal to the preset local geometric consistency threshold.

[0101] The application provides a specific method for judging whether a potential correspondence relationship meets a spatial consistency constraint condition when screening the potential correspondence relationship. The core of the method is to consider the influence of local point cloud data integrity on the reliability of local geometric consistency evaluation results, and to correct the evaluation results accordingly, thereby improving the accuracy and robustness of screening. Specifically, first, for each correspondence relationship in the set of potential correspondence relationships, the measured neighborhood point set of the corresponding measured local feature and the theoretical neighborhood point set of the corresponding theoretical local feature are obtained, which are the basis for evaluating local data integrity and local geometric consistency. Then, according to the measured neighborhood point set and the theoretical neighborhood point set, the data integrity of the two point sets is evaluated, and the neighborhood data integrity evaluation result corresponding to each correspondence relationship is obtained. By evaluating the data integrity, the reliability of the point cloud data in the local area where the correspondence relationship is located can be quantified, and the area with missing or incomplete data can be identified. Then, according to the neighborhood data integrity evaluation result, the reliability factor of the local geometric consistency evaluation result corresponding to each correspondence relationship is calculated. The higher the data integrity, the more reliable the calculated local geometric consistency evaluation result, and the higher the corresponding reliability factor; on the contrary, the lower the data integrity, the lower the reliability factor. Subsequently, according to the calculated reliability factor, the local geometric consistency evaluation result corresponding to each correspondence relationship is corrected to obtain a corrected local geometric consistency evaluation score. By weighting or adjusting the original local geometric consistency evaluation result with the reliability factor, the local geometric consistency score of the area with poor data integrity is reduced, reflecting the unreliability of the evaluation result and avoiding misjudgment caused by incomplete local data. Further, the global residual error of each correspondence relationship is compared with a preset global residual error threshold, and the corrected local geometric consistency evaluation score is compared with a preset local geometric consistency threshold. Here, the local geometric consistency score corrected by data integrity is used to ensure the reliability of the judgment. Finally, according to the comparison result, it is judged whether each correspondence relationship meets the preset global residual error threshold condition and the preset local geometric consistency threshold condition, so as to determine whether each correspondence relationship meets the spatial consistency constraint condition. Only the correspondence relationship that meets both the global consistency and the locally consistent reliability is considered reliable, thereby effectively eliminating the false correspondence relationship caused by incomplete local data. The scheme further considers the influence of local point cloud data integrity on the reliability of local geometric consistency evaluation results and corrects the evaluation results based on the use of global residual error and local geometric consistency evaluation results for correspondence relationship screening. This improvement makes the screening process more accurate and robust when facing the possible local incompleteness of irregular part point cloud data, and can more reliably screen out reliable correspondence relationships for subsequent pose calculation from the set of potential correspondence relationships, thereby improving the accuracy and robustness of the entire positioning method.

[0102] In one embodiment, for each correspondence in the set of potential correspondences, the set of measured neighborhood points of the measured local feature and the set of corresponding theoretical neighborhood points of the corresponding theoretical local feature can be obtained by setting a spherical or cylindrical search radius around the measured local feature and the corresponding theoretical local feature, and collecting all points within the radius. Based on the set of measured neighborhood points and the set of corresponding theoretical neighborhood points, evaluating the data completeness of the set of measured neighborhood points and the set of corresponding theoretical neighborhood points can calculate the point density of the set of neighborhood points, or compare the point density of the set of neighborhood points with the point density of the equivalent region in the theoretical point cloud to obtain a completeness score, for example, the proportion of the actual number of points to the theoretical number of points. Based on the evaluation result of the neighborhood data completeness, calculating the reliability factor of the local geometric consistency evaluation result corresponding to each correspondence can be a function, for example, the completeness score is directly used as the reliability factor, or the completeness score is mapped to the reliability factor using a Sigmoid function. Based on the reliability factor, modifying the local geometric consistency evaluation result corresponding to each correspondence to obtain the modified local geometric consistency evaluation score can multiply the original score by the reliability factor, or use a weighted average or the like. Comparing the global residual of each correspondence with the preset global residual threshold, and comparing the modified local geometric consistency evaluation score with the preset local geometric consistency threshold, wherein the global residual threshold and the local geometric consistency threshold can be determined according to experience or experiment. Based on the comparison result, it is judged whether each correspondence satisfies the preset global residual threshold condition and the preset local geometric consistency threshold condition at the same time, so as to determine whether each correspondence meets the spatial consistency constraint condition, and the judgment condition can be: the global residual is less than or equal to the global residual threshold and the modified local geometric consistency evaluation score is greater than or equal to the preset local geometric consistency threshold, assuming that a high score represents high consistency.

[0103] By evaluating the neighborhood data completeness and calculating the reliability factor to modify the local geometric consistency evaluation result, the screening process can effectively identify and reduce the influence of unreliable evaluation caused by incomplete local data. This improves the accuracy of screening reliable correspondence sets from the set of potential correspondences, and enhances the robustness of the entire method, especially when processing large-scale irregular part point clouds containing local data loss or incompleteness.

[0104] In some embodiments, the specific steps in step S7 include:

[0105] S71. Obtain the preset process key area information; the process key area information is the area in the theoretical model point cloud that needs high-precision alignment;

[0106] S72. In each iteration process, find a set of corresponding point pairs in the preprocessed point cloud according to the current pose of the preprocessed point cloud and the theoretical model point cloud;

[0107] S73. For each of the corresponding point pairs in the set of corresponding point pairs, judging whether the theoretical model point in the corresponding point pair is located in a process critical region or not;

[0108] S74. According to the judging result, determining a weight for each of the corresponding point pairs; the corresponding point pairs located in the process critical region have higher weights than the corresponding point pairs not located in the process critical region;

[0109] S75. According to the set of corresponding point pairs and the weight determined for each of the corresponding point pairs, calculating an optimal rigid body transformation that minimizes the distance between the weighted corresponding point pairs;

[0110] S76. Applying the optimal rigid body transformation to the pre-processed point cloud to update the current pose of the pre-processed point cloud;

[0111] S77. Repeating steps S72 to S76 until a preset convergence condition is met or a preset maximum iteration number is reached to obtain a final pose.

[0112] The process critical region information refers to a specific geometric region in the theoretical model point cloud that is pre-designated as having a decisive influence on subsequent machining precision and needs to be aligned with high precision, which can be defined by an index list of points in the theoretical model point cloud, surface patch identifiers, or through geometric boundaries such as bounding boxes, spheres, or polygons. The weight refers to a numerical factor assigned to each corresponding point pair when calculating the optimal rigid body transformation, which is used to measure the importance of the point pair in the alignment calculation, which can be determined by a fixed numerical value, a discrete value based on region type, or a function value related to the properties of the point pair. The optimal rigid body transformation that minimizes the distance between the weighted corresponding point pairs refers to the combination of rotation and translation obtained by optimization algorithm, which can make the weighted distance square sum of all corresponding point pairs reach the minimum, which can be calculated by standard optimization techniques such as weighted least squares method. The preset convergence condition refers to the stopping criterion for judging whether the iteration process has reached sufficient alignment accuracy, which can be set by the amount of pose change calculated by two consecutive iterations being less than a threshold value, or the average weighted distance of the corresponding point pairs being less than a threshold value. The preset maximum iteration number refers to the upper limit of iterations set to prevent the algorithm from circulating indefinitely, which can be determined according to experience or calculation resource limitations.

[0113] The core of the scheme is to introduce a weighting mechanism based on the process key area to improve the alignment accuracy of the key area. First, the pre-set process key area information is obtained, which clearly defines which areas in the theoretical model point cloud are the key areas that need high-precision alignment in subsequent processing, providing a basis for subsequent differentiated processing. In each iteration process, according to the current pose of the pre-processed point cloud, find the corresponding point pair set in the pre-processed point cloud and the theoretical model point cloud, which is the standard operation of the iterative alignment algorithm, used to establish the correspondence between the two point clouds. For each corresponding point pair found, determine whether the theoretical model point in it falls within the pre-set process key area, which is the key step of associating the correspondence with the key area information. According to the judgment result, determine a weight for each corresponding point pair. The corresponding point pair located in the process key area is given a higher weight, while the corresponding point pair in the non-key area has a relatively low weight. This differentiated weight allocation reflects the different requirements for alignment accuracy in different areas, making the alignment error of the key area have a greater impact in subsequent calculations. According to the set of corresponding point pairs with weights, calculate the optimal rigid transformation that minimizes the sum of the squares of the distances between the weighted corresponding point pairs. Unlike traditional minimization of unweighted distances, weighted minimization allows high-weight key area point pairs to contribute more to the calculated optimal transformation, thereby driving the algorithm to preferentially reduce the alignment error of the key area. Apply the calculated optimal rigid transformation to the pre-processed point cloud to update its current pose, making the pre-processed point cloud further align to the theoretical model point cloud. Repeat the above steps until the pre-set convergence condition is met or the pre-set maximum number of iterations is reached. Through this weighted iteration process, the final pose can better ensure the alignment accuracy of the process key area, thereby providing a more accurate pose reference for subsequent machine tool spindle adaptive positioning. This weighted iterative alignment method, combined with the initial pose obtained based on local feature matching and spatial consistency screening as described above, can fully utilize the global coarse alignment information provided by the initial pose, and through the weighting of the key area in the iteration process, overcome the limitations of traditional methods in handling irregular part local deformation and data incompleteness, making the final alignment result in areas with high machining precision requirements have higher reliability.

[0114] In one embodiment, the process critical region information can be stored as a list containing the indices of all points in the theoretical model point cloud that belong to the critical region. In each iteration, after finding the set of corresponding point pairs, for each corresponding point pair in the set, check if the index of its theoretical model point exists in the critical region index list. If it exists, the corresponding point pair is determined to be located within the process critical region. The determination of the weight can be in a hierarchical manner, for example, the corresponding point pairs located within the process critical region are assigned a weight of 10, while the corresponding point pairs in the non-critical region are assigned a weight of 1. The calculation of the optimal rigid transformation that minimizes the distance between the weighted corresponding point pairs can be realized by a variant of the weighted Iterative Closest Point (ICP) algorithm, which multiplies the distance error of each point pair by its corresponding weight when calculating the transformation between the point pairs. The iteration process can set the convergence condition as the change in the rotation angle between two iterations being less than 0.01 degrees and the change in the translation vector being less than 0.05 millimeters, or the maximum number of iterations being 100.

[0115] By introducing a weighting mechanism based on the process critical region, the present scheme can prioritize the optimization of the alignment accuracy in the process critical region during the iterative alignment process. This enables the final obtained pre-processed point cloud to have a higher accuracy in the region that is crucial for subsequent machining relative to the theoretical model point cloud. This improved alignment result provides a more reliable input for the adaptive positioning of the machine tool spindle, which helps to ensure that the tool path can accurately refer to the theoretical model during the precision machining of irregular parts, thereby improving the machining quality and reducing the scrap rate or rework caused by alignment errors.

[0116] With reference to the accompanying drawings Figure 2 , the present application provides an irregular part machine tool spindle adaptive positioning system, comprising:

[0117] An acquisition module 100 is configured to acquire a measured point cloud of an irregular part and a theoretical model point cloud; and a preprocessing module 200 is configured to preprocess the measured point cloud to obtain a pre-processed point cloud.

[0118] An extraction module 300 is configured to extract a first local geometric feature from the pre-processed point cloud to obtain a measured local feature set, and extract a second local geometric feature from the theoretical model point cloud to obtain a theoretical local feature set.

[0119] A matching module 400 is configured to match the measured local feature set and the theoretical local feature set to obtain a set of potential corresponding relationships.

[0120] A screening module 500 is configured to screen corresponding relationships that meet a preset spatial consistency constraint condition from the set of potential corresponding relationships to obtain a set of reliable corresponding relationships.

[0121] The computing module 600 is configured to calculate an initial pose of the preprocessed point cloud relative to the theoretical model point cloud according to the reliable correspondence set.

[0122] The iteration module 700 is configured to perform iterative alignment of the preprocessed point cloud and the theoretical model point cloud according to the initial pose, to obtain a final pose of the preprocessed point cloud relative to the theoretical model point cloud.

[0123] The control module 800 is configured to control the self-adaptive positioning of the machine tool spindle according to the final pose.

[0124] In this article, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions.

[0125] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for adaptive positioning of a spindle of a machine tool for irregular parts, characterized in that: The following steps are involved: S1. Obtain the measured point cloud and theoretical model point cloud of the irregular part; S2. preprocessing the measured point cloud to obtain a preprocessed point cloud; S3. Extracting the first local geometric features based on the preprocessed point cloud to obtain a measured local feature set, and extracting the second local geometric features based on the theoretical model point cloud to obtain a theoretical local feature set; S4. matching the measured local feature set with the theoretical local feature set to obtain a potential correspondence relationship set; S5. Based on the potential correspondence set, select correspondences that meet the preset spatial consistency constraints to obtain a reliable correspondence set; S6. Calculate the initial pose of the preprocessed point cloud relative to the theoretical model point cloud based on the reliable correspondence set; S7. Iteratively aligning the preprocessed point cloud with the theoretical model point cloud based on the initial pose to obtain a final pose of the preprocessed point cloud relative to the theoretical model point cloud; S8. Control the machine tool spindle to perform adaptive positioning according to the final posture.

2. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 1, characterized in that: Preprocessing includes data reduction, outlier removal, and surface orientation information calculation.

3. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 1, characterized in that: The specific steps in step S4 include: S41. For each measured local feature in the measured local feature set, search and determine multiple candidate theoretical local features in the theoretical local feature set; S42. For each measured local feature and its corresponding multiple candidate theoretical local features, calculate the feature similarity between the measured local feature and each candidate theoretical local feature; S43. Obtain all pre-processed point clouds within the measured local feature neighborhood as a measured neighborhood point set, and all theoretical model point clouds within each candidate theoretical local feature neighborhood as a theoretical neighborhood point set; S44. Based on the measured neighborhood point set and the theoretical neighborhood point set, evaluating the local geometric consistency of each candidate theoretical local feature as a corresponding relationship of the measured local feature; S45. Determine one or more potential correspondences for each measured local feature based on the evaluation results of feature similarity and local geometric consistency, and obtain a potential correspondence set by aggregating the potential correspondences determined for each measured local feature.

4. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 3, characterized in that: The specific steps in step S44 include: S441. Calculate the relative geometric relationship between the internal point pairs of the measured neighborhood point set to obtain the measured neighborhood geometric relationship set; S442. Calculate the relative geometric relationship between the internal point pairs of the theoretical neighborhood point set to obtain a theoretical neighborhood geometric relationship set; S443. By comparing the measured neighborhood geometric relationship set and the theoretical neighborhood geometric relationship set, the neighborhood similarity between the measured neighborhood point set and the theoretical neighborhood point set is calculated; S444. Evaluate the local geometric consistency of the correspondence between each candidate theoretical local feature and the measured local feature based on the neighborhood similarity.

5. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 4, characterized in that: The specific steps in step S5 include: S51. Select a subset from the set of potential correspondences; S52. Calculate a candidate global rigid body transformation based on the selected subset; S53. Based on the candidate global rigid body transformation, calculate the distance between the position of the measured point after the global rigid body transformation and the position of the corresponding theoretical point in each corresponding relationship in the potential corresponding relationship set to obtain the global residual of each corresponding relationship; S54. Based on the global residual and local geometric consistency evaluation results of each correspondence, determine whether each correspondence satisfies the preset global residual threshold condition and the preset local geometric consistency threshold condition, and determine whether each correspondence satisfies the spatial consistency constraint condition; S55. Filter out the corresponding relationships that meet the spatial consistency constraint conditions to obtain a reliable corresponding relationship set.

6. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 5, characterized in that: The global rigid body transformation consists of a rotation matrix and a displacement vector.

7. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 5, characterized in that: The specific steps in step S54 include: S541. For each corresponding relationship in the potential corresponding relationship set, obtain a measured neighborhood point set of the measured local feature and a theoretical neighborhood point set of the corresponding theoretical local feature; S542. According to the measured neighborhood point set and the theoretical neighborhood point set, the data integrity of the measured neighborhood point set and the theoretical neighborhood point set is evaluated to obtain the neighborhood data integrity evaluation result corresponding to each corresponding relationship; S543. Calculate the reliability factor of the local geometric consistency evaluation result corresponding to each correspondence according to the neighborhood data integrity evaluation result; S544. According to the reliability factor, the local geometric consistency evaluation result corresponding to each correspondence is corrected to obtain a corrected local geometric consistency evaluation score; S545. Compare the global residual of each correspondence with a preset global residual threshold, and compare the corrected local geometric consistency evaluation score with a preset local geometric consistency threshold; S546. Based on the comparison result, determine whether each corresponding relationship satisfies both the preset global residual threshold condition and the preset local geometric consistency threshold condition, thereby determining whether each corresponding relationship meets the spatial consistency constraint condition.

8. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 1, characterized in that: The specific steps in step S7 include: S71. Get preset process key area information; S72. In each iteration, according to the current pose of the preprocessed point cloud, find a set of corresponding point pairs in the preprocessed point cloud and the theoretical model point cloud; S73. For each corresponding point pair in the corresponding point pair set, determine whether the theoretical model point in the corresponding point pair is located in the process critical area; S74. According to the judgment result, a weight is determined for each corresponding point pair; the corresponding point pairs located in the process critical area have a higher weight than the corresponding point pairs in the non-critical area; S75. Calculate the optimal rigid body transformation that minimizes the distance between the weighted corresponding point pairs based on the corresponding point pair set and the weight determined for each corresponding point pair; S76. Apply the optimal rigid body transformation to the preprocessed point cloud, and update the current pose of the preprocessed point cloud; S77. Repeat steps S72 to S76 until the preset convergence condition is met or the preset maximum number of iterations is reached to obtain the final pose.

9. The method for self-adaptive positioning of a spindle of a machine tool for irregular parts according to claim 8, characterized in that: The process critical area information is the area in the theoretical model point cloud that requires high-precision alignment.

10. An adaptive spindle positioning system for irregular parts machine tools, characterized in that: include: Acquisition module, used to obtain measured point clouds and theoretical model point clouds of irregular parts; A preprocessing module is used to preprocess the measured point cloud to obtain a preprocessed point cloud; An extraction module is used to extract a first local geometric feature based on the preprocessed point cloud to obtain a measured local feature set, and to extract a second local geometric feature based on the theoretical model point cloud to obtain a theoretical local feature set; A matching module is used to match the measured local feature set with the theoretical local feature set to obtain a potential correspondence relationship set; A screening module is used to screen out corresponding relationships that meet preset spatial consistency constraints based on the potential corresponding relationship set to obtain a reliable corresponding relationship set; A calculation module is used to calculate the initial pose of the preprocessed point cloud relative to the theoretical model point cloud based on the reliable correspondence set; An iterative module is used to iteratively align the preprocessed point cloud with the theoretical model point cloud based on the initial pose to obtain the final pose of the preprocessed point cloud relative to the theoretical model point cloud; The control module is used to control the machine tool spindle to perform adaptive positioning according to the final posture.

Images