A polishing track determination method, device and equipment

Abstract

The application provides a polishing track determination method, device and equipment. The method provided by the application comprises the following steps: collecting surface contour points of a blade to be polished, and obtaining an ideal contour track for characterizing a surface contour of a standard blade based on a three-dimensional model of the standard blade; determining data characteristics of an actual contour track to obtain an initial denoised contour track; for the initial denoised contour track, two track points with a distance less than a preset distance are determined as a pair of points to be processed; the semantic similarity of the pair of points to be processed is determined, and whether the two track points constituting the pair of points to be processed are repeated points is determined; if yes, a target track point in the pair of points to be processed is reserved to obtain a target denoised contour track; the target denoised contour track and the ideal contour track are registered to determine a polishing track. The polishing track determination method, device and equipment provided by the application are used to accurately determine a polishing track of an engine blade.

Description

Technical Field

[0001] This application relates to the field of engine blade processing technology, and in particular to a method, apparatus and equipment for determining grinding trajectory. Background Technology

[0002] Blades are critical components of engines, accounting for approximately one-third of the total number of blades, and their machining quality directly affects the overall performance of the engine. Blades with high geometric precision and excellent surface quality play a vital role in improving the overall performance and reliability of aero-engines. With the continuous development of engine technology, the design and machining precision requirements for blades are constantly increasing, especially the machining quality of blade edges, which has become a key factor affecting engine performance. Therefore, researching and optimizing blade edge machining technology to improve its machining precision and quality is particularly important.

[0003] Currently, research on the edge machining of precision-forged blades mainly focuses on mathematical modeling of grinding forces based on grinding machines, precise control of grinding allowances, and optimization of machining parameters to achieve precision machining. However, in terms of precisely controlling grinding allowances, how to effectively adapt to the fluctuations in profile and position distribution caused by the forging process, and accurately determine the grinding trajectory to improve the machining accuracy and quality of the blades, has become a current research hotspot. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus and equipment for determining the grinding trajectory, so as to accurately determine the grinding trajectory and improve the processing accuracy and processing quality.

[0005] Specifically, this application is implemented through the following technical solution:

[0006] The first aspect of this application provides a method for determining a grinding trajectory, the method comprising:

[0007] Collect surface contour points of the blade to be polished, combine the collected surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on a three-dimensional model of the standard blade.

[0008] The data characteristics of the actual contour trajectory are determined, the target denoising algorithm and the parameter values ​​of the target denoising algorithm are determined based on the data characteristics, and the actual contour trajectory is denoised based on the target denoising algorithm to obtain an initial denoised contour trajectory.

[0009] For the initial denoised contour trajectory, two trajectory points with a distance less than a preset distance are identified as a pair of points to be processed;

[0010] Determine the semantic similarity between the two trajectory points that constitute the pair of points to be processed, and determine whether the two trajectory points that constitute the pair of points to be processed are duplicate points based on the semantic similarity.

[0011] When the two trajectory points constituting the point pair to be processed are repeated points, the target trajectory point in the point pair to be processed is retained according to the semantic coherence and semantic heavyness of any trajectory point in the point pair to be processed with the adjacent points before and after it, so as to obtain the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic heavyness with the adjacent points before and after it.

[0012] The target denoised contour trajectory and the ideal contour trajectory are registered, and the grinding trajectory is determined based on the registration result.

[0013] A second aspect of this application provides a grinding trajectory determination device, the device comprising a data acquisition module, a noise reduction module, a processing module, and a determination module; wherein:

[0014] The acquisition module is used to acquire surface contour points of the blade to be polished, combine the acquired surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on the three-dimensional model of the standard blade.

[0015] The denoising module is used to determine the data characteristics of the actual contour trajectory, determine the target denoising algorithm and the parameter values ​​of the target denoising algorithm based on the data characteristics, and denoise the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory.

[0016] The processing module is used to identify two trajectory points whose distance is less than a preset distance as a pair of points to be processed for the initial denoised contour trajectory.

[0017] The determining module is used to determine the semantic similarity between the two trajectory points constituting the pair of points to be processed, and to determine whether the two trajectory points constituting the pair of points to be processed are duplicate points based on the semantic similarity.

[0018] The processing module is further configured to retain the target trajectory point in the point pair to be processed when the two trajectory points constituting the point pair to be processed are repeated points, based on the semantic coherence and semantic heavyness of any trajectory point in the point pair to be processed with the adjacent points before and after it, so as to obtain the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic heavyness with the adjacent points before and after it.

[0019] The processing module is further configured to register the target denoised contour trajectory and the ideal contour trajectory, and determine the grinding trajectory based on the registration result.

[0020] A third aspect of this application provides a grinding trajectory determination device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods provided in the first aspect of this application.

[0021] The grinding trajectory determination method, apparatus, and equipment provided in this application first collect surface contour points of the blade to be ground, combine the collected surface contour points into an actual contour trajectory to characterize the surface contour of the blade, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on a three-dimensional model of the standard blade; then, determine the data characteristics of the actual contour trajectory, determine the target denoising algorithm and its parameter values ​​based on the data characteristics, and denoise the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory; then, for the initial denoised contour trajectory, determine two trajectory points with a distance less than a preset distance as a pair of points to be processed; then... Next, the semantic similarity of the two trajectory points constituting the point pair to be processed is determined, and it is determined whether the two trajectory points constituting the point pair to be processed are duplicate points based on the semantic similarity. Then, when the two trajectory points constituting the point pair to be processed are duplicate points, the target trajectory point in the point pair to be processed is retained according to the semantic coherence and semantic repetition of any trajectory point in the point pair to be processed with its adjacent points before and after it, thus obtaining the target denoised contour trajectory. Among them, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic repetition with its adjacent points before and after it. Finally, the target denoised contour trajectory and the ideal contour trajectory are registered, and the polishing trajectory is determined based on the registration result. In this way, firstly, selecting a suitable target denoising algorithm and its parameter values ​​based on data characteristics ensures that the denoising algorithm is more closely suited to the blade scenario, guaranteeing reliability and accuracy compared to directly using traditional denoising algorithms. Secondly, after obtaining the initial denoised contour trajectory, internal repetition points and redundant noise points are further removed from the initial denoised contour trajectory. Thus, through multi-level denoising processing, the denoising effect can be effectively improved, making the final grinding trajectory more accurate. Thirdly, when determining repetition points, based on the semantic coherence and semantic heavyness of the trajectory points with their adjacent points, trajectory points with stronger semantic coherence and lower semantic heavyness are retained, thereby further removing redundant data and obtaining a more accurate target denoised contour trajectory. In this way, through precise data acquisition, effective denoising processing, accurate repetition point judgment, and precise registration process, the accurate determination of the grinding trajectory is achieved, ensuring the accuracy of the blade grinding trajectory determination. Attached Figure Description

[0022] Figure 1 A flowchart of Embodiment 1 of the grinding trajectory determination method provided in this application;

[0023] Figure 2 A comparative illustration of the noise reduction process before and after the noise reduction is provided in an exemplary embodiment of this application;

[0024] Figure 3 A flowchart of Embodiment 2 of the grinding trajectory determination method provided in this application;

[0025] Figure 4 A flowchart of Embodiment 3 of the grinding trajectory determination method provided in this application;

[0026] Figure 5 This is a hardware structure diagram of the grinding trajectory determination equipment in which the grinding trajectory determination device of this application is located;

[0027] Figure 6 This is a schematic diagram of the structure of Embodiment 1 of the grinding trajectory determination device provided in this application. Detailed Implementation

[0028] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0029] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0030] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0031] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0032] Figure 1 The flowchart below shows an embodiment of the grinding trajectory determination method provided in this application. Please refer to... Figure 1The method provided in this embodiment may include:

[0033] S101. Collect the surface contour points of the blade to be polished, combine the collected surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain the ideal contour trajectory to characterize the surface contour of the standard blade based on the three-dimensional model of the standard blade.

[0034] Specifically, the blade to be ground refers to an aero-engine blade that requires grinding. In practice, the surface profile points of the blade are obtained with high precision using a contact method. For example, in one embodiment, the probe of a Hexagon coordinate measuring machine directly contacts the blade surface to obtain the surface profile points.

[0035] Furthermore, surface profile points are points on the surface of the blade to be ground that can represent its shape and size. In practice, the profile of the blade to be ground is acquired by a measuring device, resulting in a combination of surface profile points containing all the surface profile points of the blade to be ground.

[0036] Furthermore, a standard blade is an aero-engine blade with an ideal shape and size. The surface profile of the standard blade can be obtained through its 3D model. In practice, the ideal trajectory profile of the standard blade is read from a pre-established 3D model. For example, in one embodiment, the standard blade is determined according to flight standards, thereby obtaining its 3D model, and the ideal trajectory profile is obtained based on the 3D model.

[0037] It is understandable that the actual contour trajectory can characterize the true surface shape and size of the blade to be polished; the ideal contour trajectory can characterize the surface shape and size of the standard blade. When polishing the blade to be polished, it is necessary to polish the blade to be polished into the shape of the standard blade.

[0038] In practice, the blades to be ground are inspected using a contact measurement method, which has higher accuracy and repeatability, effectively avoids problems such as surface reflection and noise interference, and ensures the stability and reliability of the data.

[0039] In practice, based on a Hexagon coordinate measuring machine, a high-precision probe is used to directly contact the blade surface, which can accurately obtain the surface profile of blades with complex geometries.

[0040] Optionally, in one possible implementation, acquiring the surface contour points of the blade to be ground includes:

[0041] (1) Measure the edge curvature of the blade to be polished, and draw the edge curve of the blade to be polished based on the measured edge curvature;

[0042] (2) Determine the maximum height difference between the edge curve and the standard curve corresponding to the standard blade;

[0043] (3) Determine the width range of the edge scan based on the maximum height difference;

[0044] (4) For the blade to be polished, scan the blade edge within the width range to obtain surface contour points.

[0045] In practice, the probe of a measuring device (such as a coordinate measuring machine) directly contacts the edge of the blade to be ground in order to obtain accurate curvature data of the edge.

[0046] Furthermore, using drawing software or calculation tools, the edge curves are drawn, which visually demonstrate the shape and curvature changes of the edge of the blade to be ground.

[0047] Furthermore, the edge curve of the blade to be polished is compared with the standard curve of a standard blade to find the maximum vertical distance difference between the two, i.e., the maximum height difference. Based on the maximum height difference, a suitable scanning width range is set to include all the points that need to be scanned on the edge of the blade to be polished, while also taking into account the balance between scanning efficiency and accuracy.

[0048] In a specific implementation, for example, in one embodiment, the sum of the maximum height difference (3mm) and the preset safety threshold (2mm) can be determined as the scanning width range, i.e., 5mm.

[0049] Furthermore, after determining the scanning width range, the blade edge within that range is scanned, and the three-dimensional coordinate information of each point on the edge is continuously recorded. These points are the surface contour points. That is, when acquiring blade data, only a 5mm range of the edge is scanned, for a total of 12 cross-sections.

[0050] Understandably, by accurately collecting the surface contour points of the blade to be ground and comparing and registering them with the ideal contour trajectory of the standard blade, it is possible to ensure that the grinding trajectory is highly consistent with the shape of the standard blade, thereby significantly improving the processing accuracy. At the same time, by combining the maximum height difference, the scanning width range can be accurately obtained, so that the data obtained from the scan can completely reflect the edge data of the blade and ensure the accuracy of the surface contour points.

[0051] It should be noted that traditional point cloud acquisition often uses a fixed scanning range, which may lead to data redundancy in irrelevant areas or insufficient data in critical areas. This method dynamically adjusts the scanning range by measuring edge curvature and calculating the maximum height difference, ensuring high data density in key areas while avoiding redundant acquisition in non-critical areas, thereby improving acquisition efficiency and accuracy.

[0052] S102. Determine the data characteristics of the actual contour trajectory, determine the target denoising algorithm and the parameter values ​​of the target denoising algorithm based on the data characteristics, and denoise the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory.

[0053] It should be noted that during the process of collecting surface contour points of the blade to be ground, due to measurement limitations, environmental interference, and the influence of the surface characteristics of the blade, the collected surface contour points may contain noise or outliers, so noise reduction processing is required.

[0054] In specific implementations, for example, in one possible approach, the target denoising algorithm can be determined based on the correspondence between data characteristics and denoising algorithms, and the actual contour trajectory can be denoised based on the target denoising algorithm. For example, in one embodiment, the data characteristics are: a lot of discrete information, high data repetition rate, many scattered noise points, and many redundant noise points, so the target denoising algorithm is determined to be a Gaussian filtering denoising algorithm.

[0055] Among these, discrete information refers to the distribution and spacing of data points in the actual contour trajectory, reflecting the degree of data dispersion and density. Data repetition rate refers to the frequency of identical or very close data points appearing in the actual contour trajectory, reflecting the degree of data redundancy. Scattered noise points refer to randomly distributed abnormal data points generated during data acquisition due to various reasons such as measurement errors and equipment vibrations; the distribution of scattered noise points differs significantly from that of surrounding data points. Redundant noise points refer to duplicate or invalid data points generated during data acquisition due to repeated measurements, data recording errors, or other reasons; redundant noise points contribute little or no contribution to the overall description of the data.

[0056] Furthermore, in one possible implementation, after determining the target denoising algorithm, all data in the actual contour trajectory can be clustered to obtain the cluster center and the cluster radius, and the parameter values ​​of the target denoising algorithm can be determined based on the cluster center and the radius.

[0057] In specific implementation, for example, after determining that the target denoising algorithm is Gaussian filtering, clustering can be performed on all data in the actual contour trajectory based on clustering algorithms such as K-means or DBSCAN. Among the multiple cluster centers obtained by clustering, since the cluster centers can better reflect the overall distribution of data points, a representative cluster center (such as the cluster center with the most data points) can be selected as the mean of Gaussian filtering. Since the standard deviation describes the dispersion of data points, the standard deviation determines the size and smoothness of the filter kernel. The cluster radius can be selected as a reference value for the standard deviation, or the standard deviation can be obtained by making appropriate adjustments based on the cluster radius (such as multiplying by an empirical coefficient).

[0058] Furthermore, after determining the target denoising algorithm and its parameter values, the target denoising algorithm is used to denoise the actual contour trajectory to obtain the initial denoised contour trajectory.

[0059] It is understandable that when the characteristics of the data change, other denoising algorithms such as median filtering, mean filtering, bilateral filtering, nonlocal mean filtering, wavelet transform denoising, sparse representation denoising, and Bayesian denoising can be selected as the target denoising algorithm. For example, in one embodiment, when the image contains complex noise and texture, sparse representation denoising is used as the target denoising algorithm. In another embodiment, when the image contains randomly occurring black and white dots, median filtering is used as the target denoising algorithm.

[0060] Figure 2 A comparative diagram of noise reduction before and after is provided in an exemplary embodiment of this application. Please refer to... Figure 2 , Figure 2 Figure A in the diagram shows the actual contour trajectory before denoising. Figure 2 Image B in the diagram shows the initial denoised contour trajectory after denoising. Please refer to... Figure 2 Before denoising, the blade outline had abnormal protrusions. After denoising, these protrusions were removed, resulting in a smooth blade surface trajectory. Thus, Gaussian filtering was chosen based on the characteristics of the noisy data. Through the linear smoothing properties of Gaussian filtering, noise can be effectively removed while preserving the detailed features of the point cloud data, thereby improving the quality of the point cloud data.

[0061] S103. For the initial denoised contour trajectory, two trajectory points with a distance less than a preset distance are determined as a pair of points to be processed.

[0062] Specifically, the exact value of the preset distance is set according to actual needs, and this embodiment does not limit it.

[0063] In practice, after the denoising process is completed, any two trajectory points in the initial denoised contour trajectory are traversed, and two trajectory points with a distance less than a preset distance are found. These two trajectory points are then identified as the point pair to be processed.

[0064] S104. Determine the semantic similarity between the two trajectory points constituting the point pair to be processed, and determine whether the two trajectory points constituting the point pair to be processed are duplicate points based on the semantic similarity.

[0065] Specifically, semantic similarity refers to whether two trajectory points are similar in shape, size, position, etc., and whether they represent the same feature on the blade surface. One possible implementation is to obtain the feature vectors of the two trajectory points, and then determine their semantic similarity based on these vectors. If the semantic similarity between the two trajectory points is higher than a set similarity threshold, then the two trajectory points are considered to be duplicates.

[0066] In a specific implementation, for example, in one possible approach, features can be extracted from each trajectory point in the pair of points to be processed, obtaining features such as the trajectory point's 3D coordinates, normal vector, curvature, and distribution of neighboring points. Then, a semantic model is established to understand the geometric features and semantic information of the blade surface. The extracted features are input into the semantic model, which calculates the similarity score between two trajectory points based on these features. Two trajectory points with similarity scores higher than a preset value are then identified as duplicate trajectory points. For example, in one embodiment, the similarity score is a value between 0 and 1; two trajectory points with similarity scores above 0.8 are identified as duplicate trajectory points.

[0067] S105. When the two trajectory points constituting the point pair to be processed are duplicate points, the target trajectory point in the point pair to be processed is retained according to the semantic coherence and semantic heavyness of any trajectory point in the point pair to be processed with its adjacent points before and after it, so as to obtain the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic heavyness with its adjacent points before and after it.

[0068] Specifically, if two trajectory points are detected as duplicate points, in this step, the target trajectory points to be retained are selected based on semantic coherence and semantic heavyness, and finally the target denoised contour trajectory is constructed.

[0069] That is, after determining that the point pair to be processed is a duplicate point pair, the adjacent points are combined to determine which trajectory point to keep. In this step, the data with higher semantic coherence and lower semantic heavyness with the adjacent point is kept, while the other trajectory point in the point pair to be processed is deleted to remove redundancy.

[0070] Semantic coherence measures the semantic consistency between a trajectory point and its preceding and following adjacent points, i.e., the reasonableness of the point within the entire trajectory. In practice, semantic coherence can be determined based on spatial consistency, i.e., determining whether the spatial distribution of the trajectory point and its adjacent points is continuous, for example, by calculating its spatial similarity with preceding and following trajectory points.

[0071] In a specific implementation, one possible approach is to put any one trajectory point in the pair of points to be processed and its adjacent trajectory points into the semantic model to obtain numerical values ​​representing semantic coherence and semantic severity. Then, put the other trajectory point in the pair of points to be processed and its adjacent trajectory points into the semantic model to obtain numerical values ​​representing semantic coherence and semantic severity. The target trajectory point with a large semantic coherence value and a low semantic severity value is retained, while the other trajectory point is deleted.

[0072] It should be noted that trajectory points and their adjacent trajectory points can be input into the semantic model. The semantic model calculates the corresponding semantic vectors, which represent these trajectory points. Then, the semantic coherence score of these semantic vectors is calculated, and trajectory points with high semantic coherence scores are identified as coherent trajectory points. In practice, the semantic consistency between semantic vectors can be measured by calculating the cosine similarity or Euclidean distance between them, thus obtaining the semantic coherence score.

[0073] Understandably, by identifying the point pairs to be processed and performing semantic judgment, redundant data and noise points in the actual contour trajectory can be effectively removed, improving the accuracy and reliability of the data. At the same time, combining the blade attributes for denoising can better preserve the real features of the blade surface, providing higher quality data support for subsequent point cloud registration and polishing trajectory determination.

[0074] It should be noted that this application designs a two-stage denoising approach. In the first stage of denoising, the target denoising algorithm and its parameters are determined based on the data characteristics. Compared with using a fixed denoising algorithm and fixed parameters, this approach can adapt to different data distributions and avoid the problems of overly smooth transitions or insufficient denoising caused by using fixed empirical parameters. This ensures that the denoising process can effectively denoise different blade surface morphologies. Furthermore, based on this, the second stage of denoising mainly addresses the redundancy problems caused by data duplication and local outliers. In this application, adjacent point pair matching combined with semantic similarity analysis is used to remove heavy points, ensuring that the removed points do not affect the true blade morphology features. This allows the retained point cloud data to better represent the blade surface morphology, which is beneficial for accurately determining the grinding trajectory.

[0075] S106. Register the target denoised contour trajectory and the ideal contour trajectory, and determine the grinding trajectory based on the registration result.

[0076] In practice, the specific implementation process of this step may include:

[0077] S1061. Using the target denoised trajectory as the initial point cloud, for each actual contour trajectory point in the initial point cloud, determine the corresponding point of the actual contour trajectory point based on the distance between each ideal contour trajectory point in the ideal contour trajectory and the actual contour trajectory point, and combine the actual contour trajectory point and the corresponding point of the actual contour trajectory point into a neighboring point pair.

[0078] Specifically, the ideal contour trajectory point closest to an actual contour trajectory point is determined as the corresponding point of that actual contour trajectory point. In practice, for an actual contour trajectory point, the Euclidean distance between the actual contour trajectory point and any ideal contour trajectory point can be calculated, and then the ideal contour trajectory point closest to it is determined as the corresponding point of the actual contour trajectory point.

[0079] For example, in one embodiment, the ideal contour trajectory points are a, b, c, d, e, f, and g. The actual contour trajectory point A is closest to the ideal contour trajectory point a in Euclidean distance. Therefore, the ideal contour trajectory point a is determined as the corresponding point of the actual contour trajectory point A.

[0080] Furthermore, in conjunction with the above embodiments, the actual contour trajectory point A and the ideal contour trajectory point a are determined as a neighboring point pair.

[0081] The following specific embodiments will be given to describe in detail the steps of determining the corresponding point of the actual contour trajectory point based on the distance between each ideal contour trajectory point and the actual contour trajectory point. These steps will not be repeated here.

[0082] S1062. Construct a registration evaluation function, and based on multiple sets of neighboring point pairs, solve for the rotation and translation matrices that minimize the registration evaluation function.

[0083] The rotation matrix and translation matrix are used to register the actual contour trajectory points in the target denoised contour trajectory to the ideal contour trajectory; the registration evaluation function is used to evaluate the alignment accuracy of the target denoised contour trajectory registered to the ideal contour trajectory.

[0084] Specifically, the registration evaluation function is used to assess the alignment between the denoised target contour trajectory and the ideal contour trajectory.

[0085] In a specific implementation, in one embodiment, the target denoised contour trajectory is denoted as the initial point cloud, and the initial point cloud is denoted as P. Furthermore, the ideal contour trajectory is denoted as the target point cloud, and the target point cloud is denoted as X. At this point, the constructed registration evaluation function is:

[0086] ;

[0087] in, This represents the number of actual contour trajectory points contained in the initial point cloud P. For rotation matrix, It is a translation matrix. Let i be the ideal trajectory point of the i-th point in the target point cloud. Let be the i-th actual contour trajectory point in the initial point cloud, where the i-th ideal trajectory point and the i-th actual contour trajectory point are a neighboring point pair.

[0088] It should be noted that, through registration, point cloud P is made to coincide with point cloud X through a rigid transformation (rotation and translation), thus obtaining X = R•P + And through continuous iteration, the mean square error is reduced. To reach the minimum.

[0089] Specifically, a rotation matrix is ​​a three-dimensional rotation matrix used to describe a rotational transformation in space. Understandably, during registration, a rotation matrix is ​​used to adjust the orientation of the initial point cloud to rotate and align it with the ideal contour trajectory. A translation matrix is ​​also a three-dimensional translation vector used to describe the distance moved in space along a certain direction. Understandably, during registration, a translation matrix is ​​used to adjust the position of the initial point cloud to move and align it with the ideal contour trajectory.

[0090] Furthermore, alignment accuracy refers to the degree or accuracy of the match between the initial point cloud and the ideal contour trajectory. It can be understood that higher alignment accuracy indicates a smaller deviation between the two. In practice, the registration evaluation function evaluates the alignment accuracy of the initial point cloud registered to the ideal contour trajectory by calculating the mean square error of the rotation and translation matrices. For example, in one embodiment, multiple combinations of rotation and translation matrices are calculated based on the registration evaluation function, and the combination that minimizes the value of the registration evaluation function is determined as the final rotation and translation matrices.

[0091] Specifically, the implementation process of this step may include:

[0092] (1) Transform the solution objective into a first objective and a second objective; wherein the first objective involves the optimization objective of the rotation matrix; the second objective involves the optimization objective of the rotation matrix and the translation matrix; (2) Solve the first objective using the singular value decomposition method to obtain the rotation matrix; (3) Substitute the obtained rotation matrix into the second objective to obtain the translation matrix.

[0093] Specifically, as introduced earlier, the objective is to calculate the rotation and translation matrices that minimize the value of the registration evaluation function; that is, the objective is:

[0094] ;

[0095] In the specific implementation, for the initial point cloud P: (p1, p2, ..., pN) P The target point cloud X corresponding to the ideal trajectory: (x1, x2, ..., xN) X First, calculate the centroids of these two point cloud sets based on the following formula:

[0096] ;

[0097] ;

[0098] in, Let P be the centroid of the initial point cloud. This represents the number of actual contour trajectory points contained in the initial point cloud P. Let i be the coordinates of the i-th actual contour trajectory point. The centroid of the ideal contour trajectory, Let i be the coordinates of the i-th ideal contour trajectory point. The number of points contained in the target point cloud.

[0099] Furthermore, the coordinates of the points in the two sets of point clouds with the centroid as the origin are obtained as follows:

[0100] , ;

[0101] Furthermore, ;

[0102] Right now, ;

[0103] Right now, ;

[0104] Right now, ;

[0105] Right now, ;

[0106] Furthermore, Based on this formula, the above formula is simplified to obtain: .

[0107] Furthermore, because: , ;

[0108] Therefore, the objective is transformed into:

[0109] ;

[0110] Therefore, based on the above transformation, for ease of calculation, the solution objective is divided into a first objective (involving the optimization of the optimal rotation matrix) and a second objective (involving the optimization of the optimal rotation matrix and the optimal translation matrix).

[0111] The primary goal is:

[0112] ;

[0113] The second objective is:

[0114] ;

[0115] Wherein, P is the initial point cloud, and the The initial point cloud contains the number of actual contour trajectory points. For the rotation matrix, the Let x be the translation matrix. i Let i be the coordinates of the i-th ideal contour trajectory point. Let μ be the coordinate value of the i-th actual contour trajectory point. p μ is the centroid of the initial point cloud. x Let be the centroid of the ideal contour trajectory.

[0116] Furthermore, in step (2), the singular value decomposition method can be used to solve for the first target, so as to obtain the rotation matrix:

[0117] Specifically, ;

[0118] ;

[0119] make ;

[0120] ;

[0121] Thus, the rotation matrix can be solved using the following method:

[0122] (1) Calculate the intermediate quantity H according to the first formula;

[0123] (2) Perform singular value decomposition on the intermediate quantity H according to the second formula to obtain the first matrix U and the second matrix V;

[0124] (3) Solve for the rotation matrix according to the third formula based on the first matrix and the second matrix;

[0125] The first formula is:

[0126] ;

[0127] The second formula is:

[0128] ;

[0129] The third formula is:

[0130] ;

[0131] Where, trace is the trace of the matrix, H is an intermediate quantity, U is the first matrix, V is the second matrix, and so on. Denotes the transpose of the second matrix, the Represents the transpose of the first matrix; the R * This is the rotation matrix obtained by solving.

[0132] In other words, the coordinates of points in the two point clouds P and X with the centroid as the origin are: and Substitution Let the rotation matrix be... To obtain the optimal solution, the calculation yields... .

[0133] Furthermore, simplifying the above formula yields... , ,make Calculations yielded At this point, the maximum value is calculated. Thus, the final rotation matrix is ​​obtained. optimal solution .

[0134] (2) Substitute the solved rotation matrix into the second target to solve for the translation matrix.

[0135] In specific implementation, let , And set a translation matrix. The optimal solution Substituting into the equation yields... Combined with rotation matrix Substituting it into the second objective, we get Determine the final translation matrix. optimal solution .

[0136] The method provided in this embodiment decouples the optimization of rotation and translation by transforming the solution objective into a first objective and a second objective. The rotation matrix is ​​solved independently first, and then the result is substituted into the calculation of the translation matrix. This effectively reduces the complexity of the optimization problem, thereby improving computational efficiency. Furthermore, this method avoids the complex nonlinear optimization involved in jointly solving the rotation and translation matrices. Further, the point cloud is centered using centroids during the calculation. The calculation of centroids eliminates the influence of translation on the data, making the solution of the rotation matrix more independent. This approach enhances the algorithm's robustness to noise and effectively adapts to deviations caused by acquisition errors in actual point cloud data.

[0137] S1063. For each actual contour trajectory point in the target denoised contour trajectory, process the target denoised contour trajectory based on the rotation matrix and translation matrix to obtain the registration point of the actual contour trajectory point.

[0138] In practice, each actual contour trajectory point in the target denoised contour trajectory is rotated according to the rotation matrix, and then translated according to the translation matrix. The position after rotation and translation is determined as the registration point of the actual contour trajectory point.

[0139] In practice, the registration points can be calculated based on the following formula:

[0140] ;

[0141] in, The coordinates of the registration points for the actual contour trajectory points. To obtain the rotation matrix, This is the translation matrix obtained by solving. Let i be the coordinates of the i-th actual contour trajectory point. Let i be the registration point for the i-th actual contour trajectory point, where i ranges from 1 to N. P .

[0142] S1064. When the distance from the registration point of all actual contour trajectory points to the corresponding point of the actual contour trajectory point is less than a preset threshold, the registration trajectory formed by the registration points of each actual contour trajectory point in the target denoised contour trajectory is determined as the polishing trajectory; otherwise, the registration trajectory is used as the initial point cloud, and the step of determining the corresponding point of each actual contour trajectory point in the initial point cloud based on the distance from each ideal contour trajectory point in the ideal contour trajectory to the actual contour trajectory point is executed again.

[0143] Specifically, the distance from the registration point of each actual contour trajectory point to the corresponding point of that actual contour trajectory point is calculated and compared with a preset threshold.

[0144] That is, calculation and The Euclidean distance is used to determine whether it is less than a preset threshold. When the distance from the registration point of all actual contour trajectory points to the corresponding point of that actual contour trajectory point is less than the preset threshold, the registration trajectory formed by the registration points of each actual contour trajectory point in the denoised contour trajectory is determined as the polishing trajectory. That is, the registration trajectory is determined as the polishing trajectory when the following conditions are met:

[0145] ;

[0146] in, Let x be the coordinates of the registration point for the i-th actual contour trajectory point. i Let i be the coordinates of the point corresponding to the i-th actual contour trajectory point, where i ranges from 1 to N. P ε is a preset threshold.

[0147] It should be noted that the specific value of the preset threshold is determined based on the actual situation, and this embodiment does not limit it. In specific implementation, the preset threshold can be set to 0.001 based on expert experience.

[0148] Furthermore, referring to the preceding description, it can be understood that when all distance values ​​are determined to be less than the preset threshold, the registration trajectory at this time is determined to meet the requirements, and the registration trajectory is determined as the grinding trajectory. In other words, the blade to be ground can be ground according to the movement process of the registration trajectory to obtain a blade that is the same as the ideal blade.

[0149] Furthermore, when at least one distance value is greater than or equal to a preset threshold, the registered trajectory is used as the initial point cloud, and the step of determining the corresponding point of each actual contour trajectory point in the initial point cloud based on the distance between each ideal contour trajectory point in the ideal contour trajectory and the actual contour trajectory point is executed again.

[0150] It should be noted that, firstly, by denoising the actual contour trajectory, useless noise points can be removed, improving the quality of the denoised contour trajectory and making the subsequent registration and grinding trajectory determination processes more accurate, thus avoiding unnecessary interference factors. Secondly, by registering the actual contour trajectory with the ideal contour trajectory of the standard blade using rotation and translation matrices for precise alignment, the actual surface contour can be accurately aligned with the standard ideal contour, ensuring high precision in the final generated grinding trajectory and reducing errors. Thirdly, this method uses a registration evaluation function to dynamically evaluate the registration effect, ensuring registration accuracy, and gradually optimizes the registration trajectory through a threshold judgment mechanism. In each registration process, if the registration accuracy does not meet the requirements, automatic re-registration is performed, ensuring the accuracy and reliability of the grinding trajectory and effectively... The method improves the quality of blade surface treatment, ensuring the consistency of surface morphology during grinding and avoiding local over-grinding or under-grinding. Fourthly, through iterative calculation and gradual optimization of registration accuracy, the grinding trajectory more closely matches the actual blade surface contour, avoiding potential deviations from the initial point cloud and effectively improving the accuracy of the final grinding trajectory. Fifthly, based on point-to-point matching between the ideal contour trajectory and the actual trajectory, personalized adjustments can be made for different blade surface morphologies, adapting to blades of different shapes and sizes, exhibiting strong adaptability and flexibility. Sixthly, this method can automatically handle the denoising, point cloud registration, and trajectory generation processes, reducing the need for manual intervention. Users only need to provide the surface contour data of the blade to be ground, and the grinding trajectory can be automatically generated, improving work efficiency and automation levels.

[0151] The method provided in this embodiment first collects surface contour points of the blade to be ground, combines the collected surface contour points into an actual contour trajectory to characterize the surface contour of the blade, and obtains an ideal contour trajectory to characterize the surface contour of the standard blade based on a 3D model of the standard blade; then, it determines the data characteristics of the actual contour trajectory, determines the target denoising algorithm and its parameter values ​​based on the data characteristics, and denoises the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory; then, for the denoised contour trajectory, it determines two trajectory points with a distance less than a preset distance as a pair of points to be processed; then it determines... The semantic similarity of the point pairs to be processed is used to determine whether the two trajectory points constituting the point pairs are duplicate points. Then, when the two trajectory points constituting the point pairs are duplicate points, the target trajectory point in the point pairs is retained according to the semantic coherence and semantic heavyness of any trajectory point in the point pairs with its adjacent points before and after it, thus obtaining the target denoised contour trajectory. The target trajectory point is the trajectory point in the point pairs with stronger semantic coherence and lower semantic heavyness with its adjacent points before and after it. Finally, the target denoised contour trajectory and the ideal contour trajectory are registered, and the polishing trajectory is determined based on the registration result. In this way, firstly, selecting a suitable target denoising algorithm and its parameter values ​​based on data characteristics ensures that the denoising algorithm is more closely suited to the blade scenario, guaranteeing reliability and accuracy compared to directly using traditional denoising algorithms. Secondly, after obtaining the initial denoised contour trajectory, internal repetition points and redundant noise points are further removed from the initial denoised contour trajectory. Thus, through multi-level denoising processing, the denoising effect can be effectively improved, making the final grinding trajectory more accurate. Thirdly, when determining repetition points, based on the semantic coherence and semantic heavyness of the trajectory points with their adjacent points, trajectory points with stronger semantic coherence and lower semantic heavyness are retained, thereby further removing redundant data and obtaining a more accurate target denoised contour trajectory. In this way, through precise data acquisition, effective denoising processing, accurate repetition point judgment, and precise registration process, the accurate determination of the grinding trajectory is achieved, ensuring the accuracy of the blade grinding trajectory determination.

[0152] Figure 3 The flowchart for Embodiment 2 of the grinding trajectory determination method provided in this application is shown below. Please refer to... Figure 3 The method provided in this embodiment, based on the above embodiments, includes determining the corresponding point of the actual contour trajectory point according to the distance between each ideal contour trajectory point and the actual contour trajectory point, comprising:

[0153] S301. Determine the minimum number of attributes that can distinguish different leaves, and determine the number of levels of the retrieval tree based on the minimum number of attributes.

[0154] Specifically, leaves have many similar properties, but also some unique properties. These unique properties can distinguish different leaves. The minimum number of these properties that can uniquely distinguish a leaf can be determined, that is, the minimum number of properties that can define or distinguish a leaf.

[0155] In a specific implementation, for example, in one embodiment, three attributes (such as length, width, and thickness) are sufficient to distinguish all leaves. These three attributes are the minimum number of attributes. Based on this number, the number of layers in the retrieval tree can be set to 3.

[0156] Specifically, different blades have multiple attributes, such as geometric features (chord length, chord thickness, blade height, twist angle), material properties (material type, alloy composition), etc. In order to determine the minimum number of attributes required to distinguish blades, feature selection methods, such as principal component analysis and mutual information, can be used to determine k attributes (k being the minimum number of attributes) that can completely distinguish all blades.

[0157] S302. Based on the degree of influence of each attribute on the surface quality of the blade, sort the attributes corresponding to the minimum number of attributes to obtain the sorting result.

[0158] Specifically, different attributes have varying degrees of impact on blade surface quality. For example, in one embodiment, length has a relatively small impact on blade surface quality, while thickness has a relatively large impact.

[0159] Understandably, attributes are ranked based on their impact on blade surface quality so that more important attributes can be prioritized when constructing the retrieval tree.

[0160] In practice, the influence of each attribute on the blade surface quality can be determined by experimentally verifying the degree of influence of the minimum number of attributes, and a sequence table of the influence of each attribute on the blade surface quality can be constructed. For example, in one embodiment, the constructed sequence table of the influence of each attribute on the blade surface quality is as follows: material hardness > surface roughness > heat treatment state > curvature > thickness > edge sharpness > length > width.

[0161] As described above, it can be understood that the number of levels in the retrieval tree is determined by the minimum number of attributes k, with each level corresponding to one attribute: that is, if k = 4, then the retrieval tree has 4 levels, and each level is classified according to different attributes.

[0162] S303. Select the attribute with the highest influence in the sorting results as the root layer of the retrieval tree, and take the point with the largest attribute value corresponding to the ideal contour trajectory point under that attribute as the initial node.

[0163] In a specific implementation, for example, in one embodiment, thickness has the greatest impact on surface quality. The thickness is used as the root layer of the retrieval tree, and the point corresponding to the ideal trajectory point with the largest thickness value is selected as the initial node.

[0164] S304. The sorting results are used as the classification basis for each layer, and for each layer, the ideal contour trajectory points are classified according to the classification basis of that layer to obtain a retrieval tree.

[0165] In practical implementation, each attribute in the sorting results can be determined as the classification criterion for the corresponding layer of the search tree, and the order of the sorting results can be determined as the order of the layer division. For example, if the sorting results are length, thickness, and width, the search tree consists of three layers: the first layer classifies by length, the second layer classifies by thickness, and the third layer classifies by width. Thus, when constructing the search tree, for the first layer, the point with the most prominent length value is selected as the initial node, and then the ideal trajectory points are divided into first-level layers according to length. Further, for the second layer, the ideal trajectory points are divided into second-level layers according to thickness, and further, for the third layer, the ideal trajectory points are divided into third-level layers according to width. After completing the division of all layers, the search tree is obtained.

[0166] It should be noted that, compared to random or balanced partitioning, this method has the following advantages:

[0167] (1) Clear hierarchy, reducing redundancy

[0168] Since attributes are filtered according to the minimum distinguishing criterion, each level of division is based on the most effective classification factor, minimizing the depth of the tree and improving retrieval efficiency.

[0169] (2) The semantics are reasonable and consistent with the factors affecting blade surface quality.

[0170] Blade retrieval is used for surface quality analysis. It uses the attribute of ranking the degree of influence of surface quality as the classification basis, so that the retrieval results are more in line with engineering needs.

[0171] (3) Classify by level to enhance interpretability

[0172] Because the tree is constructed based on the ordering of specific leaf attributes, the division of each layer has a clear physical or processing meaning, rather than random partitioning, which enhances interpretability.

[0173] (4) Reduce computational load and improve retrieval efficiency

[0174] By using the fewest distinguishing attributes, invalid or redundant calculations are avoided, reducing search complexity and making leaf retrieval more efficient.

[0175] In this application, the above method is used to construct a hierarchical and accurately classified retrieval tree based on the minimum distinguishable attributes of the blade and the degree of influence of each attribute on the surface processing quality of the blade, so that the retrieval process is efficient and conforms to the influencing factors of the surface quality of the blade.

[0176] S305. For each actual contour trajectory point in the initial point cloud, search for the point closest to the actual contour trajectory point from the retrieval tree, and determine the searched point as the corresponding point of the actual contour trajectory point.

[0177] In practice, for each actual contour trajectory point in the initial point cloud, the search is performed by traversing the pre-built search tree layer by layer until the point closest to it is found. The found point is the corresponding point of the actual contour trajectory point in the ideal contour trajectory.

[0178] The method provided in this embodiment significantly reduces the search space and ensures the reliability and accuracy of the search tree construction by determining the minimum number of attributes that can distinguish different leaves and constructing a search tree based on these attributes. Simultaneously, by sorting the attributes according to their influence on the leaf surface quality and prioritizing more important attributes when constructing the search tree, the structure of the search tree reflects the key features of the leaf surface, helping to find the corresponding point that best matches the actual contour trajectory point more quickly during the registration process, thereby improving registration accuracy. Furthermore, using the attribute with the highest influence as the root layer of the search tree ensures that the data is arranged sequentially from the most accurate, allowing for priority comparison of the most accurate data during retrieval, thus guaranteeing retrieval accuracy. Additionally, selecting the point with the largest attribute value as the initial node ensures optimal performance of the initial node, preventing confusion caused by poor performance during downward retrieval and ensuring retrieval accuracy.

[0179] Specifically, in one possible implementation, the process of implementing this step includes:

[0180] (1) Extract the feature information of the actual contour trajectory points and determine the retrieval level based on the feature information.

[0181] Specifically, feature information can include the curvature change of the blade edge, the direction of the normal vector of a point, etc. Similar points can be quickly located in the retrieval tree using feature information.

[0182] In practice, feature information can be extracted from the actual contour trajectory points collected by methods such as principal component analysis or local binary pattern analysis.

[0183] Furthermore, the extracted features are analyzed and quantified, and converted into numerical or vector forms that can be compared. These data are then compared to evaluate the importance of different features in distinguishing different leaves or leaf surface features. Based on the importance of the features and the actual application requirements, the hierarchical structure of the retrieval tree is determined.

[0184] In a specific implementation, for example, in one embodiment, the L0 layer of the retrieval tree corresponds to the overall shape of the contour, the L1 layer corresponds to the local curvature, the L2 layer corresponds to the processing error, the L3 layer corresponds to the position information, and the L4 layer corresponds to the texture / surface characteristics. Before querying, the features of the contour trajectory points are extracted, and the most prominent feature of the contour trajectory point is determined. When the local curvature value of the contour trajectory point is the largest (i.e., the local curvature is the most prominent), the L1 layer is determined as the retrieval level, that is, the retrieval starts from the L1 layer.

[0185] The following is another specific embodiment to illustrate in detail how to determine the retrieval level. Specifically, in one possible implementation, this step includes:

[0186] (1) Normalize the feature information to generate a feature vector;

[0187] (2) Calculate the hierarchical similarity score of the feature vector at each level;

[0188] (3) The level corresponding to the highest level similarity score is determined as the retrieval level.

[0189] Specifically, the similarity score is the degree of matching between the feature vector and the attributes at the retrieval tree level, and the similarity can be calculated based on a pre-trained similarity calculation model.

[0190] In specific implementation, for example, in one embodiment, the retrieval tree hierarchy is layer L0-L5. The feature information of the actual contour trajectory points includes geometric, topological, error, position, surface texture, and other features. After normalization, a feature vector X is obtained. A pre-trained similarity scoring model is used to calculate the similarity score between the feature vector X and each layer from L0 to L5. The similarity score between the feature vector X and layer L0 is 0.5, with layer L1 it is 0.6, with layer L2 it is 0.8, with layer L3 it is 0.7, with layer L4 it is 0.6, and with layer L5 it is 0.5. It is determined that the feature vector X has the highest similarity score with layer L2. Therefore, layer L2 is determined as the retrieval level, and the retrieval starts from layer L2.

[0191] (2) Starting from the retrieval level, candidate points are filtered layer by layer, and in the process of filtering candidate points layer by layer, the matching result of each layer is used as the input of the next layer, and candidate points are filtered layer by layer until the point closest to the actual contour trajectory point is found; wherein, in the process of filtering candidate points layer by layer, the similarity threshold gradually increases with the increase of the number of layers, and the number of candidate points filtered gradually decreases with the increase of the number of layers.

[0192] The similarity threshold for each layer is different. The larger the layer, that is, the later the layer, the larger the similarity threshold. Furthermore, the number of candidate points to be selected in each layer is also different. The larger the layer, that is, the later the layer, the fewer candidate points to be selected.

[0193] Specifically, as described above, since each layer of the retrieval tree is classified based on an attribute—that is, each layer corresponds to one attribute—and the influence of each attribute on the leaf surface quality gradually decreases, a gradually decreasing similarity threshold can be set for each layer. For example, in one embodiment, for the first layer, data with a similarity greater than 50% can be retrieved; for the second layer, data with a similarity greater than 60% can be retrieved; and for the third layer, data with a similarity greater than 70% can be retrieved.

[0194] In a specific implementation, for example, in one embodiment, the first layer of the retrieval tree corresponds to the shape attribute, and points with a similarity greater than 50% are retrieved to obtain 8 nodes; the second layer corresponds to the shading level attribute, and points with a similarity greater than 60% are selected in the second layer below these 8 nodes to obtain 5 nodes; further, the third layer corresponds to the edge sharpness attribute, and data points with a similarity greater than 70% are selected in the third layer below these 5 nodes to obtain 2 nodes; the fourth layer corresponds to the hardness attribute, and points with a similarity greater than 20% are selected in the fourth layer below these 2 nodes to obtain 1 node, which is the final result.

[0195] It should be noted that the method provided in this embodiment extracts the feature information of actual contour trajectory points and determines the retrieval level, directly starting the retrieval from the most relevant level, avoiding the inefficient operation of traversing the entire tree layer by layer from the root node. Furthermore, at each level, the selected candidate points serve as input for the next level. This retrieval process is a continuously converging process, from coarse matching to precise matching, avoiding interference from irrelevant data. Through layer-by-layer filtering, the final matched points maintain high similarity while avoiding excessive computational overhead caused by global search. Further, by setting different similarity thresholds for each level, with lower thresholds for layers with higher influence, early screening is more lenient, ensuring no loss of key matching information, while later matching is more stringent, improving the final matching accuracy. That is, root-level matching is coarse-grained, and final-level matching is fine-grained, ensuring that each level's matching task fulfills its purpose and reducing unnecessary computation. This strategy reduces computational load while maintaining the accuracy of matching results, avoiding the problem of prematurely discarding correct matching points due to strict global screening.

[0196] In summary, the method provided in this embodiment, through hierarchical filtering, progressive matching, and dynamic adjustment of similarity thresholds, significantly improves search efficiency while maintaining retrieval accuracy through the synergistic effect of these three factors. Compared to traditional traversal methods, this method is more targeted, avoids irrelevant calculations, makes the retrieval process faster and more accurate, and maintains good scalability in large-scale data environments.

[0197] Figure 4 For the flowchart of Embodiment 3 of the grinding trajectory determination method provided in this application, please refer to... Figure 4 Based on the above embodiments, the step of determining the corresponding point of each actual contour trajectory point in the initial point cloud according to the distance between each ideal contour trajectory point and the actual contour trajectory point includes:

[0198] S401. Extract key points from the initial point cloud and the ideal contour trajectory to obtain the initial key point set of the initial point cloud and the ideal key point set of the ideal contour trajectory.

[0199] Specifically, the key points are some distinctive features that are highly identifiable and few in number.

[0200] In practice, although denoising has been performed on each ideal contour trajectory point in the ideal contour trajectory and each actual contour trajectory point in the initial point cloud, the amount of denoised point cloud data is still large, and the direct search speed is slow, which will affect the overall efficiency. Therefore, key points can be extracted, and then the search can be performed based on the key points.

[0201] In practice, for the target point cloud (ideal contour trajectory, denoted as target point cloud), the normal vector of each point xi in the target point cloud under different neighborhoods can be calculated, and then the absolute value of the sine of the angle between the two normal vectors can be obtained. Data whose absolute sine value exceeds the first preset sine value is determined as the ideal key point set. Similarly, the normal vector of each point pi in the initial point cloud under different neighborhoods can be calculated, and then the absolute value of the sine of the angle between the two normal vectors can be selected. Data whose absolute sine value exceeds the second preset sine value is determined as the initial key point set.

[0202] In this way, the characteristics of different regions corresponding to different data are determined by the absolute value of the sine. When the absolute value of the sine is small, the data in this part is flat and the geometric features are not obvious; when the absolute value of the sine is large, the data in this part varies greatly and the geometric features are obvious, which are the key points.

[0203] S402. For each actual contour trajectory point in the initial key points, find the corresponding point of the actual contour trajectory point from the ideal key points.

[0204] In practice, each actual contour trajectory point in the initial keypoints is mapped to an ideal keypoint to determine the corresponding point of that actual contour trajectory point.

[0205] In a specific implementation, a retrieval tree can be constructed for the point set composed of ideal key points; further, for each initial key point, the point closest to the initial key point is searched from the retrieval tree, and the searched point is determined as the corresponding point of the initial key point.

[0206] The grinding trajectory determination method provided in this embodiment extracts key points from the initial point cloud and the ideal contour trajectory, respectively, to extract key point sets that can represent the main shape features of each. Further searching is performed based on the key points, which can reduce the number of data points that need to be registered. Compared with the traditional registration method based on the entire point cloud, the registration method based on key points can significantly reduce the amount of computation and improve the registration efficiency.

[0207] Furthermore, to verify the effectiveness of the above registration algorithm, some ideal point clouds X and target point clouds P to be registered were extracted from the three-dimensional model of the blade. The algorithm was developed in MATLAB software to simulate registration, and the point clouds X and P were found to overlap, proving that the algorithm is effective.

[0208] Furthermore, during the registration process, error is a crucial indicator for evaluating the algorithm. Following the requirements of the above calculation process, the error is calculated each time, and experiments can be conducted to confirm that the error gradually approaches zero.

[0209] In practical implementation, the process test was conducted using an A-type engine blade as the corresponding material. During blade data acquisition, only about 5mm of the edge was scanned, with a total of 12 cross sections scanned. The scanned data was imported into MATLAB software, and Gaussian filtering was used to remove noise. The filtered point cloud data was then registered using the ICP algorithm to obtain the geometric relationship and machining allowance between the theoretical model and the actual blade edge. Based on the RIG software platform, the machining trajectory was generated on the theoretical model. To verify the feasibility and accuracy of the proposed algorithm, it was validated using the FANUC LR-Mate-200iD robot grinding experimental platform. This platform consists of a robot body, a grinding wheel, and sensors. The robot's repeatability is ±0.02mm, the clamping weight is <4KG, and the radius of motion is ≤700mm. The roughing and finishing grinding wheels were nylon wheels with mesh counts of P80 and P240, respectively. After inspection with a coordinate measuring machine and a surface roughness tester, the polished blade profile was within the tolerance zone (±0.1 mm), and the blade roughness Ra was 0.3. Both the edge profile accuracy and surface roughness met the process requirements. Therefore, this application effectively solves the problem that a single grinding program cannot guarantee edge profile accuracy, addressing the high-precision requirements of aero-engine blade edge machining. It achieves efficient and high-precision machining of complex-shaped blade edges and has been verified by building a robotic grinding experimental platform. Experimental results show that this method can effectively adapt to fluctuations in blade edge profile, significantly improve machining accuracy and quality, and provide reliable technical support for high-quality manufacturing of aero-engine blades.

[0210] Corresponding to the aforementioned embodiment of a grinding trajectory determination method, this application also provides an embodiment of a grinding trajectory determination device.

[0211] An embodiment of the grinding trajectory determination device disclosed in this application can be applied to a grinding trajectory determination device. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the grinding trajectory determination device reading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, such as... Figure 5 The diagram shown is a hardware structure diagram of the grinding trajectory determination device in this application, except for... Figure 5 In addition to the processor, memory, network interface, and non-volatile memory shown, the grinding trajectory determination device in the embodiment may also include other hardware depending on the actual function of the grinding trajectory determination device, which will not be described in detail here.

[0212] Figure 6 This is a schematic diagram of the structure of Embodiment 1 of the grinding trajectory determination device provided in this application. Please refer to... Figure 6The apparatus provided in this embodiment includes a data acquisition module 610, a noise reduction module 620, a processing module 630, and a determination module 640; wherein,

[0213] The acquisition module 610 is used to acquire surface contour points of the blade to be polished, combine the acquired surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on the three-dimensional model of the standard blade.

[0214] The denoising module 620 is used to determine the data characteristics of the actual contour trajectory, determine the target denoising algorithm and the parameter values ​​of the target denoising algorithm based on the data characteristics, and denoise the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory.

[0215] The processing module 630 is used to determine two trajectory points with a distance less than a preset distance as a pair of points to be processed for the initial denoised contour trajectory;

[0216] The determining module 640 is used to determine the semantic similarity between the two trajectory points constituting the point pair to be processed, and to determine whether the two trajectory points constituting the point pair to be processed are duplicate points based on the semantic similarity.

[0217] The processing module 630 is further configured to retain the target trajectory point in the point pair to be processed when the two trajectory points constituting the point pair to be processed are repeated points, based on the semantic coherence and semantic repetition of any trajectory point in the point pair to be processed with the adjacent points before and after it, so as to obtain the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic repetition with the adjacent points before and after it.

[0218] The processing module 630 is further configured to register the target denoised contour trajectory and the ideal contour trajectory, and determine the grinding trajectory based on the registration result.

[0219] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0220] Please continue to refer to Figure 5 This application also provides a grinding trajectory determination device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods provided in the first aspect of this application.

[0221] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods provided in this application.

[0222] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0223] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0224] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for determining a grinding trajectory, characterized in that, The method includes: Collect surface contour points of the blade to be polished, combine the collected surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on a three-dimensional model of the standard blade. The data characteristics of the actual contour trajectory are determined, the target denoising algorithm and the parameter values ​​of the target denoising algorithm are determined based on the data characteristics, and the actual contour trajectory is denoised based on the target denoising algorithm to obtain an initial denoised contour trajectory. For the initial denoised contour trajectory, two trajectory points with a distance less than a preset distance are identified as a pair of points to be processed; Determine the semantic similarity between the two trajectory points that constitute the pair of points to be processed, and determine whether the two trajectory points that constitute the pair of points to be processed are duplicate points based on the semantic similarity. When the two trajectory points constituting the point pair to be processed are duplicate points, the target trajectory point in the point pair to be processed is retained according to the semantic coherence and semantic heavyness of any trajectory point in the point pair to be processed with its adjacent points before and after it, thus obtaining the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic heavyness with its adjacent points before and after it. The target denoised contour trajectory and the ideal contour trajectory are registered, and the grinding trajectory is determined based on the registration result.

2. The method according to claim 1, characterized in that, The step of registering the target denoised contour trajectory and the ideal contour trajectory, and determining the polishing trajectory based on the registration result, includes: Using the target denoised contour trajectory as the initial point cloud, for each actual contour trajectory point in the initial point cloud, the corresponding point of the actual contour trajectory point is determined according to the distance between each ideal contour trajectory point in the ideal contour trajectory and the actual contour trajectory point, and the actual contour trajectory point and the corresponding point of the actual contour trajectory point are combined into a neighboring point pair. Construct a registration evaluation function, and solve for the rotation and translation matrices that minimize the registration evaluation function based on multiple sets of neighboring point pairs; For each actual contour trajectory point in the target denoised contour trajectory, the target denoised contour trajectory is processed based on the rotation matrix and translation matrix to obtain the registration point of the actual contour trajectory point. When the distance from the registration point of all actual contour trajectory points to the corresponding point of the actual contour trajectory point is less than a preset threshold, the registration trajectory formed by the registration points of each actual contour trajectory point in the target denoised contour trajectory is determined as the polishing trajectory; otherwise, the registration trajectory is used as the initial point cloud, and the step of determining the corresponding point of each actual contour trajectory point in the initial point cloud based on the distance from each ideal contour trajectory point in the ideal contour trajectory to the actual contour trajectory point is executed again.

3. The method according to claim 2, characterized in that, The step of determining the corresponding point of the actual contour trajectory point based on the distance between each ideal contour trajectory point and the actual contour trajectory point includes: Determine the minimum number of attributes required to distinguish different leaves, and then determine the number of levels in the retrieval tree based on the minimum number of attributes. Based on the degree of influence of each attribute on the surface quality of the blade, the attributes corresponding to the minimum number of attributes are sorted to obtain the sorting result; The attribute with the highest influence in the sorting results is selected as the root of the retrieval tree, and the point with the largest attribute value corresponding to the ideal contour trajectory point under that attribute is selected as the initial node. The sorting results are used layer by layer as the classification criteria for each layer, and for each layer, the ideal contour trajectory points are classified according to the classification criteria of that layer to obtain a retrieval tree; For each actual contour trajectory point in the initial point cloud, the point closest to the actual contour trajectory point is searched from the retrieval tree, and the searched point is determined as the corresponding point of the actual contour trajectory point.

4. The method according to claim 3, characterized in that, The step of searching the retrieval tree for the point closest to the actual contour trajectory point includes: Extract the feature information of the actual contour trajectory points, and determine the retrieval level based on the feature information; Starting from the retrieval level, candidate points are filtered layer by layer. During the process of filtering candidate points layer by layer, the matching result of each layer is used as the input of the next layer. The process continues until the point closest to the actual contour trajectory point is found. In the process of filtering candidate points layer by layer, the similarity threshold gradually increases with the number of layers, and the number of candidate points filtered gradually decreases with the number of layers.

5. The method according to claim 4, characterized in that, The step of extracting feature information of the actual contour trajectory points and determining the retrieval level based on the feature information includes: The feature information is normalized to generate a feature vector; Calculate the hierarchical similarity score of the feature vector at each level; The level corresponding to the highest level similarity score is determined as the retrieval level.

6. The method according to claim 1, characterized in that, The process of collecting surface contour points of the blade to be ground includes: Measure the edge curvature of the blade to be polished, and plot the edge curve of the blade to be polished based on the measured edge curvature; Determine the maximum height difference between the edge curve and the standard curve corresponding to the standard blade; The width range of the edge scan is determined based on the maximum height difference; For the blade to be polished, the blade edge within the specified width range is scanned to obtain surface contour points.

7. The method according to claim 1, characterized in that, The data characteristics used to determine the actual contour trajectory include: The actual contour trajectory is clustered, and the cluster center and cluster radius are determined as the data characteristics of the actual contour trajectory.

8. The method according to claim 2, characterized in that, For each actual contour trajectory point in the initial point cloud, determining the corresponding point of the actual contour trajectory point based on the distances between each ideal contour trajectory point and the actual contour trajectory point includes: Key points are extracted from the initial point cloud and the ideal contour trajectory to obtain the initial key point set of the initial point cloud and the ideal key point set of the ideal contour trajectory. For each actual contour trajectory point in the initial key points, find the corresponding point of that actual contour trajectory point from the ideal key points.

9. A grinding trajectory determining device, characterized in that, The device includes an acquisition module, a noise reduction module, a processing module, and a determination module; The acquisition module is used to acquire surface contour points of the blade to be polished, combine the acquired surface contour points into an actual contour trajectory to characterize the surface contour of the blade to be polished, and obtain an ideal contour trajectory to characterize the surface contour of the standard blade based on the three-dimensional model of the standard blade. The denoising module is used to determine the data characteristics of the actual contour trajectory, determine the target denoising algorithm and the parameter values ​​of the target denoising algorithm based on the data characteristics, and denoise the actual contour trajectory based on the target denoising algorithm to obtain an initial denoised contour trajectory. The processing module is used to identify two trajectory points whose distance is less than a preset distance as a pair of points to be processed for the initial denoised contour trajectory. The determining module is used to determine the semantic similarity between the two trajectory points constituting the pair of points to be processed, and to determine whether the two trajectory points constituting the pair of points to be processed are duplicate points based on the semantic similarity. The processing module is further configured to retain the target trajectory point in the point pair to be processed when the two trajectory points constituting the point pair to be processed are repeated points, based on the semantic coherence and semantic heavyness of any trajectory point in the point pair to be processed with the adjacent points before and after it, so as to obtain the target denoised contour trajectory; wherein, the target trajectory point is the trajectory point in the point pair to be processed with stronger semantic coherence and lower semantic heavyness with the adjacent points before and after it. The processing module is further configured to register the target denoised contour trajectory and the ideal contour trajectory, and determine the grinding trajectory based on the registration result.

10. A grinding trajectory determination device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the method according to any one of claims 1-8.

Images