Aviation blade workpiece machining and positioning method based on three-dimensional point cloud

Abstract

The invention discloses a three-dimensional point cloud-based aviation blade workpiece processing and positioning method, which comprises the following steps of: respectively acquiring three-dimensional point cloud information of an aviation blade workpiece and a standard workpiece model, and preprocessing the three-dimensional point cloud information; segmenting the preprocessed three-dimensional point cloud information into a blade profile point cloud and a stem end point cloud, and extracting key points of the blade profile point cloud based on boundary features to obtain a point cloud picture; after the point cloud atlas is subjected to coarse registration, fine registration is carried out by adopting a Transform-based dynamic ICP method, and pose data of the aviation blade workpiece are obtained; and on the basis of the pose data, a machining path of a cutter in the aviation blade workpiece machining process is obtained and used for subsequent aviation blade workpiece machining. By dynamically optimizing the nearest neighbor distance threshold value, the normal vector included angle threshold value and the maximum number of iterations, adaptive threshold value adjustment is achieved, global context information between point clouds is fully captured, and therefore the robustness, stability and precision of the registration process are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of aviation blade processing, and more particularly relates to a three-dimensional point cloud-based aviation blade workpiece processing positioning method. BACKGROUND

[0002] In the field of modern aviation manufacturing, aviation blades, as the core components of aviation engines, directly affect the performance and service life of the engines. Aviation blades have complex geometrical shapes and high-precision processing requirements. Traditional processing positioning methods usually rely on special fixtures or manual alignment. However, these methods have many limitations, such as long processing preparation time, poor flexibility, error accumulation, and other issues, making it difficult to meet the demands of modern manufacturing for high efficiency and high precision.

[0003] In recent years, with the rapid development of laser scanning and vision technology, three-dimensional point cloud-based processing positioning methods have gradually attracted attention. Three-dimensional point cloud data is obtained through laser scanners or structured light cameras, which can comprehensively and accurately reflect the surface shape of the blade, providing a new solution for the positioning of complex surfaces. However, existing three-dimensional point cloud-based aviation blade processing positioning methods still have the following problems:

[0004] ① Low data acquisition and processing efficiency: three-dimensional point cloud data is large, and usually requires a lot of time for data preprocessing, such as denoising, registration, and feature extraction, affecting the overall processing efficiency.

[0005] ② Positioning accuracy is significantly affected by the environment: In the actual processing environment, changes in light, reflectivity of the workpiece surface, and stability of the scanning equipment will negatively affect the quality of the point cloud data and the positioning accuracy.

[0006] ③ Lack of stability: The robustness and accuracy of existing methods in complex surface feature extraction and matching are still insufficient, especially when dealing with highly reflective or less textured aviation blades, errors are prone to occur. SUMMARY

[0007] The purpose of the present application is to overcome the shortcomings of the prior art and provide a three-dimensional point cloud-based aviation blade workpiece processing positioning method and system, aiming to significantly improve the workpiece positioning accuracy and processing efficiency in the blade processing process through advanced three-dimensional point cloud scanning technology and optimized point cloud processing algorithms.

[0008] In a first aspect, the present application provides a three-dimensional point cloud-based aviation blade workpiece processing positioning method, comprising the following steps:

[0009] Obtain three-dimensional point cloud information of the aviation blade workpiece and the standard workpiece model, respectively;

[0010] Preprocess the three-dimensional point cloud information;

[0011] The pre-processed three-dimensional point cloud information is segmented into blade surface point cloud and handle end point cloud, and key points are extracted from the blade surface point cloud based on boundary features to obtain point cloud graphs of the aero blade workpiece and the standard workpiece model;

[0012] The point cloud graphs of the aero blade workpiece and the standard workpiece model are coarsely registered;

[0013] A dynamic ICP method based on Transformer is used for fine registration to obtain the pose data of the aero blade workpiece;

[0014] Based on the pose data, the machining path of the cutter in the machining process of the aero blade workpiece is obtained, which is used for subsequent machining of the aero blade workpiece.

[0015] As a preferred, the preprocessing includes denoising and voxelization downsampling. Specifically, to ensure the high precision and reliability of the point cloud data, Noise2Noise self-supervised denoising technology is used to remove the noise generated in the scanning process, thereby ensuring the accuracy of the point cloud data and avoiding the interference of noise in subsequent processing, ensuring data stability. After obtaining the preliminary point cloud data, the point cloud data is preprocessed by voxelization downsampling technology, aiming to reduce the amount of point cloud data and improve the calculation efficiency. Specifically, the voxelization technology divides the space into multiple small units (voxels), simplifies the points in each voxel, greatly reduces the complexity of the data, and provides optimized calculation conditions for subsequent processing steps, ensuring that the algorithm can run efficiently.

[0016] As a preferred, the segmentation uses a region growing-based point cloud segmentation algorithm to accurately separate the blade surface and handle end point cloud regions, providing clear boundary information and clear structure for subsequent feature extraction and point cloud matching.

[0017] After completing the point cloud segmentation, the separated blade surface and handle end point cloud are further subjected to boundary feature extraction, and a key point extraction algorithm is used to accurately obtain representative key points, ensuring that errors are minimized during the matching process.

[0018] As a preferred, the coarse registration uses SAC-IA algorithm.

[0019] As a preferred, the fine registration includes the following steps:

[0020] Obtain the matching point set of the coarse registration result to generate an error feature matrix;

[0021] The error feature matrix is converted into a sequence input into a Transformer model to generate dynamically optimized ICP parameters; the Transformer model includes an input embedding layer, a multi-head self-attention layer, and an output layer; the ICP parameters include a nearest neighbor distance threshold, a normal vector angle threshold, and a maximum number of iterations;

[0022] The point cloud is precisely registered using the dynamically optimized ICP parameters.

[0023] After precisely registering the point cloud, the position and attitude of the blade workpiece in three-dimensional space can be accurately calculated. This position information provides an important basis for tool path planning, ensuring that the tool and workpiece can be precisely connected, avoiding errors and deviations in the machining process. In the tool path planning process, the system considers factors such as machining accuracy, machining sequence, and machining safety to optimize the path, thereby improving machining efficiency and ensuring high quality and high precision in the machining process, while effectively avoiding possible machining deviations and safety hazards.

[0024] In a second aspect, the present application provides an aviation blade workpiece machining positioning system based on three-dimensional point cloud, which mainly includes the following core modules:

[0025] An image acquisition module acquires three-dimensional point cloud information of the aviation blade workpiece;

[0026] A standard model generation module is used to generate a standard workpiece model and its three-dimensional point cloud information;

[0027] A point cloud processing module pre-processes the three-dimensional point cloud information, extracts boundary features and key points, and obtains point cloud graphs of the aviation blade workpiece and the standard workpiece model;

[0028] A point cloud coarse registration module coarsely registers the point cloud graphs of the aviation blade workpiece and the standard workpiece model;

[0029] A point cloud precise registration module uses a dynamic ICP method based on Transformer to precisely register the coarse registration result to obtain pose data of the aviation blade workpiece;

[0030] A tool path planning module obtains a machining path of the tool according to the pose data.

[0031] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the aviation blade workpiece machining positioning method based on three-dimensional point cloud as described in the first aspect.

[0032] In a fourth aspect, the present application provides a computer device, comprising:

[0033] A memory for storing instructions.

[0034] A processor for executing the instructions to cause the computer device to perform the three-dimensional point cloud-based aero blade workpiece machining positioning method as in the first aspect.

[0035] Compared with the prior art, the present application has the beneficial effects that:

[0036] 1. The present application adopts voxelization downsampling algorithm to significantly reduce the amount of data while effectively preserving the key features in the point cloud; through the segmentation algorithm based on region growing, the boundary features of the blade surface and the handle end are accurately separated, laying a foundation for the subsequent key point extraction.

[0037] 2. After preliminary implementation of coarse registration using the SAC-IA algorithm, the present application further introduces an ICP fine registration algorithm based on Transformer, which realizes adaptive threshold adjustment by dynamically optimizing the nearest neighbor distance threshold, the normal vector angle threshold and the maximum iteration number, fully captures the global context information between point clouds, and significantly improves the robustness, stability and accuracy of the registration process. Especially when dealing with complex shape and irregular surface blade workpieces, this method can effectively suppress the influence of noise and outliers, ensuring that the registration accuracy and convergence speed reach a higher level.

[0038] 3. Through accurate pose calculation and optimized tool path planning, the present application can effectively improve the accuracy and efficiency of the machining process, ensure accurate docking of the tool and the workpiece, and provide good machining allowance guarantee in complex machining environment, significantly improving the machining quality and safety. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The flowchart of the present application.

[0040] Figure 2 The specific flowchart of step six of the present application.

[0041] Figure 3 The structure diagram of the Transformer model of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples.

[0043] In view of the above defects or improvement needs of the prior art, the present application provides a three-dimensional point cloud-based aero blade workpiece machining positioning method, as shown in Figure 1 The method comprises the following steps:

[0044] Step 1: Use binocular structured light camera to shoot the precision-cast aviation blade workpiece, obtain high-precision three-dimensional point cloud information, and use it for high-precision three-dimensional reconstruction of the object surface structure. The CAD model is used as the standard for subsequent registration of the application. The STL format file of the CAD model is imported into Geomagic Studio 2014 software. Adjust the camera view angle in Geomagic Studio 2014, click the "Simulate Scan" command in the "Toolbar", set the resolution to 500x500, and simulate the camera scanning of the CAD model. The scanned model is converted into unordered points, and the point cloud data of the CAD model is obtained.

[0045] Step 2: Preprocess the collected point cloud data of the aviation blade workpiece and the CAD model. First, use the Noise2Noise self-supervised denoising algorithm for denoising. The application uses different noise versions of the same data , uses the deep learning point cloud network PointNet+ to extract features, and generates denoised point cloud .

[0046] Loss function: Calculate the error between the noisy point cloud to make the model learn the distribution characteristics of the noise.

[0047]

[0048] Denoised point cloud Use the voxel-based point cloud downsampling algorithm to extract information points sufficient to represent the characteristics of the object, replacing the original massive point cloud data to improve processing efficiency. This method divides the point cloud into a voxel grid and replaces all points in the voxel with the point closest to the center of gravity in each voxel. While reducing the number of point clouds, it can better preserve the shape characteristics of the point cloud, thereby improving the efficiency and accuracy of subsequent point cloud operations. The specific implementation process of this algorithm is as follows:

[0049] (1) According to the original point cloud data in , , axis direction to create a cuboid minimum bounding box, the side length of the minimum bounding box is , , , the calculation method is as follows:

[0050]

[0051]

[0052]

[0053] wherein , , , , , respectively are the maximum value and minimum value corresponding to the point cloud coordinate value of three-axis direction. , , The maximum value and minimum value corresponding to the point cloud coordinate value of three-axis direction.

[0054] (2) Determine the side length of the voxel grid , The larger the sampling point cloud resolution is lower, the point cloud is more sparse, The smaller is the opposite. By L, the rectangular inside is divided into grid, the number of the sampling cube is × × The calculation method is as follows:

[0055]

[0056]

[0057]

[0058] (3) Assuming that the local point cloud set in each voxel is According to the formula, the center of gravity of each voxelized grid is calculated, the point closest to the center of gravity under the voxelized grid is used to replace all points under the voxelized grid, and then the obtained point set is combined into a new point cloud, which is used as the result of voxel sampling.

[0059]

[0060] The application combines binocular structured light camera and Noise2Noise self-supervised denoising technology, can effectively remove noise in the acquisition process, ensure the accuracy of point cloud data, significantly improve the data quality, and provide a stable and reliable basis for subsequent point cloud processing, matching and registration.

[0061] Step 3: the point cloud data model after pretreatment is separated from the blade surface point cloud and the handle end point cloud by using a segmentation algorithm based on region growing to obtain the blade surface point cloud , so as to distinguish the blade surface from the handle end part which may have more interference or deformation, and improve the matching accuracy. In the segmentation process, the curvature characteristics of the point cloud are used as the judgment strategy, the adjacent points meeting the conditions are gradually added to the current region from the selected seed point, so as to form a complete region, and the blade surface and the handle end part are effectively distinguished by analyzing the curvature information, so as to realize accurate segmentation.

[0062] Step 4: for the blade surface point cloud obtained by segmentation Further keypoint extraction based on boundary features is performed to obtain a point cloud map. This method projects the neighborhood points of a point in the point cloud onto a tangent plane. The angle between the vector from the point to the projected point determines whether it is a boundary point; if it is, the angle between two projected vectors will be significantly larger than the angle between other vectors. Simultaneously, an angle distribution histogram is constructed to statistically analyze the angle distribution characteristics of the projected points in the local coordinate system, thereby further distinguishing between outer and inner boundary points. Outer boundary points exhibit large blank areas in their angle histograms, while the histograms of inner boundary points show a more uniform distribution.

[0063] The specific method is as follows: Assuming a boundary point cloud have First, follow the method described above to... Fit a spatial plane Find the normal vector of the plane. , will dot clouds All points according to Direction projection onto the plane In the middle, the projection points on the plane are used Indicates. For any point in Establish a local coordinate system according to the following principles: With the origin as the point, for Axial direction, a point in With center of gravity The direction of the connecting line is Axial direction, shaft and The cross product of the axes is axial direction, where:

[0064]

[0065]

[0066] In the local coordinate system, draw Except All other points A vector of points, assuming a point is... Then the vector is represented as Calculate the vector and The clockwise angle of the axis The set of all included angles is obtained as follows: The angle set is plotted as a histogram with intervals of 30 degrees, and if a point is an outer boundary point, there are many continuous intervals in the histogram of the point without points, and if it is an inner boundary point, the histogram distribution is more uniform. If it is found that there are less than 2 points in three or more consecutive intervals in the histogram, the point is considered an outer boundary point, and the remaining points are removed as inner boundary points.

[0067] Step 5: Based on the characteristics of the boundary points as key points, the SAC-IA algorithm is used to coarsely register the point cloud graph of the aviation blade and the point cloud graph of the CAD model; specifically:

[0068] The FPFH (Fast Persistent Feature Histograms) feature descriptor of the above aviation blade workpiece and CAD model point cloud graph is calculated respectively:

[0069]

[0070] Wherein represents the weight, generally and is the Euclidean distance. The present application finds the points with similar FPFH features in the sampling points and the point cloud to be registered by the SAC-IA algorithm. The algorithm steps are as follows:

[0071] (1) Take the point cloud graph of the aviation blade workpiece as the source point cloud, and select ( ≥3) sampling points from the source point cloud , while ensuring that the distance between the sampling points is greater than the pre-set threshold ;

[0072] (2) Take the point cloud graph of the CAD model as the target point cloud; for each sampling point, find the point with similar FPFH features in the target point cloud , and form a list, and randomly select a point from the list as the corresponding point of the sampling point; (3) Calculate the rigid transformation matrix between the corresponding points, and use the Huber penalty function to calculate the distance error sum after the corresponding point transformation, to judge the performance of the current registration transformation, denoted as . Finally, in order to minimize the error function, it is necessary to find a set of optimal transformations from all transformations, so as to perform initial registration, wherein:

[0073]

[0074] In the formula, is a pre-set value, is the distance difference of the th set of corresponding points after transformation.

[0075] Step 6: After the coarse registration of the SAC-IA algorithm is completed, to further improve the registration accuracy of the workpiece point cloud and the CAD model point cloud, the present application performs fine registration based on the dynamic optimization ICP of the Transformer model.

[0076] As shown in Figure 2 , the present application first extracts the matching point pair information in the coarse registration result through feature extraction and error modeling, calculates the error vector and constructs the error feature matrix, and introduces the error modulus to enhance the sensitivity of the model to the size of the point pair deviation. Then, the error feature matrix is converted into a sequence input and input into the Transformer model based on the multi-head self-attention mechanism (structure see Figure 3 ) to capture the global dependency and feature relationship of the point pair deviation. The model generates a dynamically optimized ICP parameter set, including the nearest neighbor distance threshold, the normal vector angle threshold, and the maximum number of iterations, through the synergistic effect of the embedding layer, the multi-head attention mechanism, and the output layer. These parameters dynamically adjust the convergence conditions of ICP through specific mapping functions, thereby optimizing the iteration process of the algorithm. In the optimized ICP process, the accuracy of the point cloud registration is further improved by minimizing the point-to-plane registration error objective function, and the robustness and efficiency of the registration process are ensured through the dynamically adjusted residual threshold and error reduction rate. The specific process is as follows:

[0077] (1) Feature extraction and error modeling

[0078] The coarse registration result output by the SAC-IA algorithm contains a matching point set , where represents a matching point pair in the two point clouds, and N is the number of matching point pairs. Define the error vector to represent the difference between the th matching point pair, and the error feature matrix is represented as:

[0079]

[0080] Each column vector reflects the spatial deviation of a point pair.

[0081] The error modulus is used to measure the deviation size, and is defined as:

[0082]

[0083] where , , represent the components of the error vector in the three-dimensional space coordinate axes.

[0084] (2) Transformer model input construction

[0085] Convert the error feature matrix into an error feature sequence , where each Concat , Concat represents concatenation. By adding the error modulus , the sensitivity of the model to the size of the deviation is enhanced, ensuring that the model can capture the global features of the point pair deviation.

[0086] (3) Transformer model structure

[0087] Build a Transformer model based on the multi-head self-attention mechanism to learn the relationship between error features and generate dynamically optimized ICP parameters:

[0088] Input embedding layer: map the error feature sequence to a high-dimensional feature space to obtain .

[0089]

[0090] where and are embedding layer weights and biases.

[0091] Multi-head self-attention layer: extract the global dependency between point pairs, the key formula is:

[0092]

[0093] where:

[0094]

[0095] , , are projection matrices; is the feature dimension.

[0096] Output layer: map the extracted global features to a dynamically adjusted ICP threshold set

[0097]

[0098] where is the distance threshold of the nearest neighbor point pair, is the normal vector angle threshold, is the maximum number of iterations.

[0099] (4) Dynamically optimized ICP threshold value according to the dynamic threshold set output by the Transformer model , dynamically adjust the convergence condition of the ICP algorithm:

[0100] Distance threshold of the nearest neighbor point pair

[0101]

[0102] wherein is the Euclidean distance of the nearest neighbor point pair, , is a weight parameter.

[0103] Normal vector angle threshold

[0104]

[0105] wherein , is an adjustment parameter.

[0106] denotes the angle between the normal vectors of the point pair:

[0107]

[0108] wherein and are the normal vectors of points and , respectively.

[0109] Maximum number of iterations

[0110]

[0111] wherein is a minimum number of iterations set artificially. is a proportionality coefficient, is the number of points of the workpiece point cloud.

[0112] (5) Precise registration solution

[0113] Using the dynamically optimized ICP parameters, the point cloud precise registration is performed to minimize the point-to-plane error, and the objective function is:

[0114]

[0115] wherein: is a rotation matrix, satisfying , is an identity matrix; is the distance that each point needs to be translated along the space during the transformation; is the normal vector of the point.

[0116] Error convergence condition: ICP convergence is judged by residual threshold and error reduction rate to ensure that the iteration stopping condition meets the accuracy and robustness requirements.

[0117] Residual threshold

[0118]

[0119] Error reduction rate

[0120]

[0121] When or , stop iteration and complete fine registration using ICP parameters optimized based on Transformer. Among them is the residual convergence threshold set by man, is the residual reduction rate threshold set by man, is the current iteration round.

[0122] Step 7: After completing the fine registration based on Transformer dynamic ICP, the rotation matrix , translation vector and geometric information of the workpiece are obtained. These pose data can accurately describe the position and attitude of the workpiece in space, providing reliable basis for tool path planning. Based on the above pose data, the system plans and optimizes the tool trajectory, the specific steps are as follows:

[0123] (1) Coordinate system conversion

[0124] Convert the point cloud data in the workpiece coordinate system (obtained from steps 1 to 6) to the machine tool coordinate system through the rotation matrix and translation vector , so as to complete the accurate alignment of the actual position of the workpiece and the machining path. This process ensures that the position and attitude of the workpiece in the machine tool coordinate system are consistent with the planned path, providing accurate geometric basis for subsequent machining operations.

[0125]

[0126] Among them: : Actual point cloud coordinates of the workpiece in the machine tool coordinate system; : Point cloud coordinates defined in the CAD model; : Rotation matrix, representing the attitude change of the workpiece; ​: Fixed translation vector between workpiece coordinate system and machine tool coordinate system, indicating the displacement change of the workpiece.

[0127] (2) Start position determination of tool path

[0128] Based on the workpiece position after the pose transformation and the shape data thereof, the start point position of the tool path is accurately determined by comprehensively considering the machining process requirements and the tool feed direction, so as to ensure the rationality and efficiency of the machining process.

[0129]

[0130] wherein is a preset reference start point in the CAD model.

[0131] The above only describes the preferred embodiments of the present application, and it should be noted that for ordinary skilled persons in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered within the protection scope of the present application.

Claims

1. A method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds, characterized in that, The method includes the following steps: The three-dimensional point cloud information of the aircraft blade workpiece and the standard workpiece model were obtained respectively; Preprocessing of 3D point cloud information; The preprocessed 3D point cloud information is segmented into blade profile point cloud and stem tip point cloud. Key points of the blade profile point cloud are extracted based on boundary features to obtain point cloud images of aerospace blade workpieces and standard workpiece models. The point cloud images of the aircraft blade workpiece and the standard workpiece model are coarsely registered. A Transformer-based dynamic ICP method is used for fine registration to obtain the pose data of the aerospace blade workpiece. Based on the pose data, the machining path of the tool in the machining process of the aerospace blade workpiece is obtained, which is used for the subsequent machining of the aerospace blade workpiece.

2. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, The preprocessing includes denoising and voxelization downsampling.

3. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, A region-growing-based segmentation algorithm was used to segment the blade profile point cloud and the stem tip cloud.

4. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, Keypoint extraction based on boundary features includes the following steps: By projecting the neighboring points of a point in the point cloud onto the tangent plane, we can determine whether it is a boundary point based on the angle between the vectors from the point to the projection point. If it is a boundary point, there must be a point where the angle between two projection vectors is significantly larger than the angle between other vectors. Meanwhile, by constructing an angle distribution histogram, the angle distribution characteristics of the projection points in the local coordinate system are statistically analyzed, thereby further distinguishing between outer boundary points and inner boundary points, and then removing the inner boundary points.

5. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, The coarse registration specifically refers to: Calculate the FPFH feature descriptors of the point cloud maps of the aircraft blade workpiece and the standard workpiece model, respectively; Using the point cloud image of the aircraft blade workpiece as the source point cloud, select from the source point cloud Each sampling point is selected, and the distance between any two sampling points is ensured to be greater than a pre-set threshold. ≥3; The point cloud map of the standard workpiece model is used as the target point cloud; for each sampling point, points with similar FPFH features are found in the target point cloud and a list is formed, and a point is randomly selected from the list as the corresponding point of the sampling point. Calculate the rigid body transformation matrix between corresponding points, and use the Huber penalty function to calculate the distance error after the transformation of corresponding points in order to judge the performance of the current registration transformation; Repeat the above steps until the distance error is minimized. Record this registration transformation as the optimal transformation and apply it to the entire source point cloud to complete the coarse registration.

6. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, The Transformer-based dynamic ICP method includes the following steps: (1) Feature extraction and error modeling: The coarse registration result contains a set of matching points. ,in This represents a pair of points matched in two point clouds, where N is the number of matched point pairs; Define error vector Indicates the first The difference between each pair of matching points is the error feature matrix. Represented as: Each column vector reflects the spatial deviation of a point pair; the error feature matrix Transform into error feature sequence Each of them Concat Concat means concatenation. The error modulus is defined as: in, , , These represent the error vectors respectively. In three-dimensional space Components on the coordinate axes; (2) Construct the Transformer model: A Transformer model based on a multi-head self-attention mechanism is constructed to learn the relationships between error features and generate dynamically optimized ICP parameters. The Transformer model comprises a cascaded input layer, an input embedding layer, a multi-head self-attention layer, and an output layer. The input embedding layer is used to process the error feature sequence... Mapped to a high-dimensional feature space, the multi-head self-attention layer is used to extract global dependencies between point pairs, and the output layer is used to map the extracted global features to a dynamically adjusted ICP threshold set. (3) The convergence conditions of the ICP algorithm are dynamically adjusted according to the dynamically adjusted ICP threshold set to obtain dynamically optimized ICP parameters; (4) Use dynamically optimized ICP parameters to perform fine registration of point clouds, minimize the error between points and surfaces, and complete the fine registration.

7. The method for machining and positioning aerospace blade workpieces based on three-dimensional point clouds according to claim 1, characterized in that, The pose data of the aircraft blade workpiece includes the rotation matrix, translation vector, and geometric information of the aircraft blade workpiece. The registered point cloud data is converted to the machine tool coordinate system by using pose data, thereby achieving precise alignment between the actual position of the workpiece and the machining path. Then, taking into account the machining process requirements and the tool feed direction, the starting point of the tool path is determined.

8. A positioning system for machining aerospace blade workpieces based on three-dimensional point clouds, characterized in that, The system includes the following modules: The image acquisition module acquires the three-dimensional point cloud information of the aircraft blade workpiece; The standard model generation module is used to generate standard workpiece models and their 3D point cloud information; The point cloud processing module preprocesses the 3D point cloud information, extracts boundary features and key points, and obtains point cloud images of aircraft blade workpieces and standard workpiece models. The point cloud coarse registration module performs coarse registration of the point cloud images of the aerospace blade workpiece and the standard workpiece model; The point cloud fine registration module uses a Transformer-based dynamic ICP method to fine register the coarse registration results, thereby obtaining the pose data of the aerospace blade workpiece. The tool path planning module obtains the machining path of the tool based on the pose data.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images