A processing method suitable for large ship propeller blade root

Abstract

This invention discloses a machining method suitable for the root of large marine propeller blades. First, a workpiece coordinate system and transformation relationship are established on a machine tool using a standard ball and calibration pin to fit a positioning reference. The number of blades is identified based on the propeller point cloud, and the optimal imaging path is planned. A high-resolution vision sensor is integrated into the machine tool spindle, and raw 3D point cloud data of various parts of the blade root are collected along this path. The raw point cloud is transformed to the workpiece coordinate system, and data is directly stitched based on the machine tool's absolute coordinates to obtain a complete 3D point cloud of the blade root. After registering this point cloud with a theoretical model, the machining allowance is calculated and a machining trajectory is generated, ultimately completing the blade root machining. This invention reuses a high-precision machine tool for measurement, utilizing its micron-level positioning accuracy to achieve blind-zone-free high-resolution scanning of the narrow area of ​​the blade root; through a direct stitching framework based on machine tool coordinates, the accumulation of matching errors is avoided, achieving sub-millimeter-level stitching accuracy, and the method is computationally efficient and robust.

Description

Technical Field

[0001] This invention relates to the field of precision manufacturing and special processing technology for large and complex curved surface parts, and in particular to a processing method applicable to the root of large marine propeller blades. Background Technology

[0002] The blade root region of large propellers is located in the overlapping and obstructed area of ​​adjacent blades, resulting in narrow physical space and obstructed line of sight, leading to severely limited accessibility for inspection and systematic blind spots in geometric and surface quality data. Fixed equipment (such as coordinate measuring machines and laser scanners) and traditional manual or simple tools cannot effectively reach the back of the blade root and its complex surface, making it impossible to obtain the complete three-dimensional geometry of this region. Current methods mainly rely on manual visual inspection, tactile perception, or the use of simple measuring tools for local, contact-based qualitative or rough quantitative assessments, which have the following drawbacks:

[0003] (1) The test results are greatly affected by the experience and condition of the personnel, and the repeatability is poor, which cannot meet the requirements of high-precision processing and testing;

[0004] (2) The data is difficult to quantify and standardize, which brings difficulties to subsequent analysis. In order to obtain relatively complete blade root data, sometimes even the extreme method of "segmented disassembly" is required, or a lot of time is spent on manual attempts to locate and measure locally. The complete inspection of a large propeller can take 1-2 weeks, of which the inspection of overlapping areas (including blade roots) accounts for most of the time;

[0005] (3) The detection data of the leaf root area is incomplete and non-digital (mostly paper records or qualitative descriptions), which makes it impossible to make accurate comparison with the CAD design model, to establish accurate processing error feedback, and to form complete data that can be used for performance analysis, life prediction and maintenance decision-making.

[0006] In summary, insufficient accessibility of leaf roots leads to data gaps, outdated detection methods result in low data quality, and low efficiency makes it difficult to acquire data frequently and systematically. These three factors together contribute to the low level of digitalization in leaf root processing and control, hindering the realization of intelligent manufacturing and predictive maintenance. Summary of the Invention

[0007] This invention provides a processing method for the root of large marine propeller blades to overcome the above-mentioned technical problems.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows:

[0009] A processing method for the root of large marine propeller blades, comprising the following steps:

[0010] S1. Install the propeller and standard ball on the B-axis rotary table of the machine tool, and install the high-precision calibration needle on the A spindle. Use a fitting algorithm to determine the positioning reference of the propeller based on the positions of the standard ball and the high-precision calibration needle, and construct the workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system.

[0011] S2. Collect propeller point cloud data, and identify the number of blades based on the propeller point cloud data;

[0012] S3. Create a single-blade photography template, and obtain the optimal photography path based on the single-blade photography template and the number of blades; integrate a high-resolution vision sensor on the A and B axes of the machine tool, and enable the high-resolution vision sensor to collect the original three-dimensional point cloud data of each part of the blade root according to the optimal photography path.

[0013] S4. Based on the positioning reference, convert the original three-dimensional point cloud data of each part of the blade root into point cloud data in the machine tool coordinate system, and determine the point cloud data of each part of the blade root in the workpiece coordinate system based on the conversion relationship.

[0014] S5. The point cloud data of each part of the blade root in the workpiece coordinate system are stitched together to obtain the complete three-dimensional point cloud data of the blade root.

[0015] S6. Obtain the propeller theoretical model and register the complete three-dimensional point cloud data of the blade root with the propeller theoretical model.

[0016] S7. Calculate the processing allowance of the leaf root based on the registration result, generate the processing trajectory of the leaf root based on the processing allowance of the leaf root, and complete the processing of the leaf root region according to the processing trajectory.

[0017] Furthermore, in S1, the propeller positioning reference is determined based on the positions of a standard sphere and a high-precision calibration needle using a fitting algorithm, including:

[0018] Set spindle A to the 0-degree position, and control the B-axis rotary table to rotate at the set angle. Perform the following steps at each angle:

[0019] S11. Obtain the point cloud data of the standard sphere in the camera coordinate system at various rotation angles;

[0020] S12. Move spindle A so that the tip of the high-precision calibration needle contacts the preset contact point on the standard ball, and obtain the machine coordinates of the contact point in the machine coordinate system, that is, the true coordinates of the contact point;

[0021] S13. Add the actual machine coordinates of the contact point to the radius of the standard sphere to obtain the actual machine coordinates of the center of the standard sphere in the machine coordinate system at each rotation angle. The machine tool coordinate system includes the X, Y, Z, A-axis, and B-axis positions; k represents the rotation angle.

[0022] Place the A-axis at 0 degrees and at a set angle, move the A-spindle until the high-precision calibration pin touches the contact point of the standard ball, and record the machine tool coordinates at each angle.

[0023] The point cloud data of the standard sphere in the camera coordinate system is filtered and denoised to obtain standard point cloud data;

[0024] The standard point cloud data is fitted using the least squares method to determine the coordinates of the sphere center in the camera coordinate system, the rotation center OB of the B-axis rotary table, and the rotation center OA of the A-axis. The steps are as follows:

[0025] The standard point cloud data at each rotation angle is fitted using the least squares method to obtain the coordinates of the center of the standard sphere in the camera coordinate system at each rotation angle. ;

[0026] The rotation center OB of the B-axis rotary table is obtained by fitting the coordinates of the sphere center using the least squares method.

[0027] Connect the machine tool coordinates acquired at the 0-degree position and the set angle position of the A-axis, draw the perpendicular bisector of the line segment, and the intersection of the perpendicular bisector with the rotation axis of the A-axis is the rotation center OA of the A-spindle;

[0028] based on and Construct a hand-eye matrix, i.e., a positioning reference, to perform the transformation between the camera coordinate system and the machine tool coordinate system based on the hand-eye matrix. The steps are as follows:

[0029] Using the SVD decomposition algorithm By decomposing the matrix, the optimal rotation matrix and relative translation can be obtained. ;

[0030] Add the coordinates of OB and OA to the relative translation to obtain the translation vector;

[0031] Combining the rotation matrix and translation vector, we form the final 4×4 homogeneous transformation matrix, namely the hand-eye matrix, as follows:

[0032] ;

[0033] Where R represents the rotation matrix and t represents the relative translation.

[0034] Furthermore, the transformation relationship between the workpiece coordinate system and the machine tool coordinate system is established, including:

[0035] Establish the workpiece coordinate system with the rotation center OB of axis B as the origin of the workpiece coordinate system;

[0036] The transformation relationship between the workpiece coordinate system and the machine tool coordinate system is established as follows:

[0037] ;

[0038] in, Let the coordinates of the point be in the machine tool coordinate system. Let be the coordinates of the point in the workpiece coordinate system. The coordinates of the origin of the workpiece coordinate system in the machine tool coordinate system.

[0039] Furthermore, in S2, a wide-view camera is used to collect propeller point cloud data, and the number of blades is identified based on the propeller point cloud data. The specific steps include:

[0040] S21. Perform voxel grid downsampling processing on the propeller point cloud data:

[0041] The three-dimensional space of the propeller point cloud data is divided into several uniform grids according to a preset voxel grid size. The centroid of the point set formed by the points in the propeller point cloud data in each grid is taken as the centroid of the current grid. All points in each grid are replaced with the centroid to obtain downsampled point cloud data.

[0042] S22. An adaptive ROI extraction algorithm is used to extract the effective point cloud region that is only related to the blade from the downsampled point cloud data.

[0043] S23, Leaf point cloud segmentation:

[0044] The region growing segmentation algorithm is used to divide the effective point cloud region into two independent leaf point cloud sets by dividing points with the same normal direction and curvature into the same sub-region.

[0045] S24. Use an edge detection algorithm to extract the edge point cloud of the two segmented blades to obtain the edge point cloud data of the two blades. The edge point cloud data includes the areas of surface discontinuity and sharp curvature changes in the point cloud, as well as the physical boundaries, edges or surface intersections of the blades corresponding to these areas.

[0046] S25. Point cloud registration calculation:

[0047] Coarse registration: Calculate the centroid coordinates of the point clouds at the edges of the two blades respectively, obtain the initial rigid body transformation matrix by centroid alignment, and perform preliminary alignment of the point clouds at the edges of the two blades based on the initial rigid body transformation matrix;

[0048] Fine registration: The ICP algorithm is used to minimize the distance between corresponding points on the edge point clouds of two blades, thereby iteratively optimizing the initial rigid body transformation matrix to obtain the optimal transformation matrix, and further aligning the edge point clouds of two blades based on the optimal transformation matrix.

[0049] S26. Determining the number of blades:

[0050] Extract the rotation matrix from the optimal transformation matrix, and convert the rotation matrix into rotation angle values ​​using a quaternion-to-Euler angle conversion algorithm;

[0051] The number of blades is determined based on the rotation angle value.

[0052] Furthermore, in S3, the specific steps for creating a single-leaf photographing template and obtaining the optimal photographing path based on the single-leaf photographing template and the number of leaves include:

[0053] S31. Create a single-leaf photograph template:

[0054] The movement path of the large-view camera when collecting propeller point cloud data is obtained, and a single blade image template is obtained based on the movement path. The image template includes the spatial movement trajectory, attitude and shooting parameters of each point.

[0055] S32. Determine the reference blade:

[0056] Set the physical zero point of the turntable to 0 degrees in the workpiece coordinate system, and calculate the clockwise angle difference between the direction angle of each blade and the physical zero point of the turntable.

[0057] Compare the magnitudes of all clockwise angle differences and select the blade with the smallest clockwise angle difference as the reference blade closest to the physical zero point of the turntable;

[0058] S33. Generate the image path for the reference blade:

[0059] The transformation matrix between the reference blade and the photographic template is calculated using coarse and fine registration methods, and the rotation angle between the reference blade and the photographic template is obtained by converting quaternions to Euler angles.

[0060] The angle between the reference blade and the image template is converted into a transformation matrix. The image template is then transformed based on the transformation matrix to obtain the image path of the reference blade.

[0061] S34, Multi-leaf photography path extension:

[0062] The imaging path of the Nth blade is obtained by rotating and transforming the imaging path of the reference blade according to the rotation angle. This allows us to obtain the original photographic paths of all the blades. The calculation process for the rotation angle is as follows:

[0063] Rotation angle = rotation angle between the reference blade and the photo template + (N-1) × rotation angle value determined during the blade number recognition process;

[0064] S35. Optimize the original image path for all leaves:

[0065] The original photography paths of all leaves are integrated into a single point set. The problem of the order of visiting photography points is modeled as a traveling salesman problem, i.e., any two photography points are defined... S i and S j The cost of moving between them is the Euclidean distance between them:

[0066] d ij =|| S i - S j ||;

[0067] Starting from any point, a nearest neighbor greedy algorithm is used to select the next point with the closest Euclidean distance as the next destination, thus obtaining an initial shooting path;

[0068] The initial photography path is iteratively improved using the 2-opt local optimization method: try swapping any two segments in the path, calculate the change in the total path length before and after the swap, and only perform the swap if it can reduce the total distance. By iterating repeatedly, the optimal photography path is found by finding the order of accessing the photography point indexes that minimizes the sum of Euclidean distances.

[0069] The high-resolution vision sensor is controlled to scan the leaf root from multiple angles according to the optimal imaging path, thereby acquiring the original three-dimensional point cloud data of each part of the leaf root.

[0070] Furthermore, in S5, the specific steps for stitching together the point cloud data of each part of the blade root in the workpiece coordinate system to obtain the complete three-dimensional point cloud data of the blade root include:

[0071] S51. Perform coarse stitching of the point cloud data of each part of the blade root in the workpiece coordinate system directly according to the workpiece coordinate system.

[0072] S52. Based on the ICP algorithm, the coarse stitching results are finely stitched to obtain the complete three-dimensional point cloud data of the leaf roots.

[0073] Furthermore, in S6, the specific steps for obtaining the propeller theoretical model and registering the complete three-dimensional point cloud data of the blade root with the propeller theoretical model include:

[0074] S61. Perform point cloudification on the propeller theoretical model: Construct a CAD model of the blade root, and perform point cloudification on the CAD model according to a preset sampling density to convert it into theoretical point cloud data.

[0075] S62. Use the ISS feature point extraction algorithm to extract the ISS key points of the actual point cloud, namely the complete 3D point cloud data of the leaf root propeller and the theoretical point cloud data.

[0076] Perform FPFH feature matching on the extracted ISS key points to establish initial corresponding point pairs;

[0077] S63. Based on the SVD algorithm, perform best-fit registration on the initial corresponding point pairs to achieve registration of the actual collected point cloud data and the theoretical point cloud data, including:

[0078] Let the actual point cloud in the initial corresponding point pair be... The theoretical point cloud corresponds to the point as The centroids of the actual point cloud and the theoretical corresponding point cloud are calculated as follows:

[0079] ;

[0080] ;

[0081] The covariance matrix calculated based on the centroids of the two point clouds is as follows:

[0082] ;

[0083] The covariance matrix is ​​decomposed by SVD as follows:

[0084] ;

[0085] Obtain the optimal rotation matrix Translation vector for:

[0086] ;

[0087] ;

[0088] The optimal rotation matrix and translation vector are transformed and registered in the actual acquired point cloud to obtain:

[0089] .

[0090] Furthermore, the specific steps for calculating the machining allowance of the leaf root based on the registration results include:

[0091] S71. The normal vector of the theoretical point cloud is determined using principal component analysis, and the normal vector is corrected, including:

[0092] Extract any point from the theoretical point cloud Local neighborhood point cloud;

[0093] Calculate the covariance matrix of the local neighborhood point cloud;

[0094] Eigenvalue decomposition of the covariance matrix yields an orthogonal matrix. ;in Describes the main direction of the distribution of the neighborhood point cloud, i.e. , Description and The orthogonal second largest distribution direction, i.e. , Description and , All orthogonal directions, i.e. ;

[0095] Determine the normal vector:

[0096] For the upper surface of the propeller: normal vector ;

[0097] For the lower surface of the propeller: normal vector ;

[0098] Setting principal direction constraints to correct the normal vector includes:

[0099] Set the main direction of the leaf root region ;

[0100] Calculate the normal vector The angle between the normal vector and the principal direction of the leaf root region. If the angle is less than or equal to a set angle threshold, it means the normal vector direction conforms to the principal direction constraint of the leaf root region. This normal vector is retained and used as the corrected normal vector. If the angle is greater than the set angle threshold, then... Projected onto the main direction of the leaf root region Then, normalization is performed to obtain the corrected normal vector;

[0101] S72, with points Starting from the modified normal vector Search for the actual corresponding point:

[0102] For points on the upper surface of the leaf root, along Corrected normal vector A ray is emitted in the positive direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of the values ​​of all candidate points and the ray is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ;

[0103] For points on the lower surface of the leaf root, along Its normal vector after correction A ray is emitted in the negative direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of all candidate points and their values ​​is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ;

[0104] S73. Calculate the machining allowance at a single point:

[0105] For point pairs Calculate its machining allowance for:

[0106] ;

[0107] in, express and The straight-line distance between them;

[0108] S74. Calculate the processing allowance for the entire point cloud:

[0109] Iterate through all points in the theoretical point cloud model, execute steps S72 and S73, and generate a set of machining allowance values ​​that correspond one-to-one with the theoretical points. ;

[0110] S75. Construct a machining allowance distribution diagram:

[0111] Each theoretical point The coordinates and other values By associating these elements, a three-dimensional margin distribution mapping relationship is formed at the leaf root, thereby obtaining the processing margin at each point.

[0112] Furthermore, the specific steps for generating the leaf root machining trajectory based on the machining allowance of the leaf root include:

[0113] S101. Obtain local processing direction parameters based on the point cloud of the theoretical model, including:

[0114] S1011. Based on the radial and circumferential features of the propeller, initialize the global path direction to the circumferential direction and the scanning direction to the radial direction;

[0115] S1012. Optimize the global path direction and scanning direction using the principal curvature analysis method, so that the global path direction is along the local minimum curvature direction and the scanning direction is along the local maximum curvature direction. The specific steps are as follows:

[0116] Extracting the local point cloud neighborhood:

[0117] For each point to be optimized along the global path, a local point cloud neighborhood is extracted around it. The radius of the local point cloud neighborhood is set according to the point cloud density and the leaf feature size.

[0118] Calculate the local surface normal vector and principal curvature of the local point cloud neighborhood, where:

[0119] Local surface normal vector: The covariance matrix of the local neighborhood point cloud is calculated by principal component analysis, and the eigenvector corresponding to the smallest eigenvalue is the local surface normal vector;

[0120] Calculate the principal curvature:

[0121] A quadratic surface is fitted to the local point cloud neighborhood to obtain the fitted surface equation;

[0122] The two principal curvatures of the point to be optimized are calculated based on the fitted surface equation. and ,and , respectively corresponding to the direction of maximum curvature and the direction of minimum curvature ;

[0123] Orientation projection and alignment:

[0124] Will and Fusion with global path direction and scan direction, including:

[0125] Project the global circumferential direction and the global radial direction onto the tangent planes respectively, calculate the dot product of the two tangent planes after projection, and use the two dot product results as the first direction similarity and the second direction similarity respectively;

[0126] Determine if the absolute value of the first-direction similarity is greater than the set similarity threshold. If it is, it means that the current circumferential direction is consistent with the direction of minimum curvature, and the global path direction remains the circumferential direction; otherwise, it means that the curvature feature of the current region is significant, and the global path direction is corrected to the direction of minimum curvature. The system checks if the absolute value of the second-direction similarity is greater than a set similarity threshold. If it is, it indicates that the current radial direction is consistent with the direction of maximum curvature, and the scanning direction remains radial. Otherwise, it indicates that the curvature features of the current region are significant, and the scanning direction is corrected to the direction of maximum curvature. ;

[0127] S102. Divide the processing allowance of the leaf root into equally spaced sections along the optimized global path direction, sort and resample the point cloud slices of each section at equal intervals, and generate a two-dimensional path point sequence, including:

[0128] Along the optimized global path direction, the machining allowance of the leaf root is divided into equal-spaced cross sections according to the preset path spacing, forming a series of parallel cross-sectional planes;

[0129] For each cross-sectional plane, calculate the distance between the projected values ​​of all points and the current cross-sectional reference value. Filter out all Points smaller than the cross-sectional tolerance range form the original cross-sectional point set;

[0130] The original cross-sectional point set of each cross-sectional plane is sorted from smallest to largest along the optimized scanning direction, and then the sorted original cross-sectional point set is resampled at equal intervals according to the preset point spacing to generate a two-dimensional path point sequence.

[0131] S103. Combine the two-dimensional path point sequence with the normal coordinates of the corresponding cross section to form a three-dimensional trajectory point, insert safe transition points, and perform trajectory smoothing to generate a three-dimensional smooth trajectory point sequence, including:

[0132] Each two-dimensional path point sequence is combined with the normal coordinate value of the corresponding cross section to synthesize a three-dimensional spatial trajectory point;

[0133] A safe transition point is inserted between the start and end points of two adjacent two-dimensional path point sequences to integrate the three-dimensional path point sequences of all sections, resulting in an integrated three-dimensional path point sequence; wherein, the safe transition point is formed by raising the midpoint of the line connecting the start and end points to a preset safe height along the modified normal vector direction;

[0134] The integrated 3D path point sequence is smoothed by moving average or spline curve to obtain the final leaf root processing trajectory.

[0135] Beneficial Effects: This invention achieves the positioning of the entire propeller by locating the propeller itself, and establishes a workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system. Based on this, a high-resolution vision sensor is integrated into the machine tool, enabling the high-resolution vision sensor to collect the original three-dimensional point cloud data of each part of the blade root according to the optimal imaging path. This invention directly reuses high-precision machining equipment as a high-precision measuring device. Through the inherent micron-level positioning accuracy of the machine tool, it achieves blind-spot-free, high-resolution scanning of the narrow area of ​​the blade root. A direct stitching framework based on the absolute coordinates of the machine tool is proposed to stitch together the point cloud data of each part of the blade root in the workpiece coordinate system, thereby obtaining the complete three-dimensional point cloud data of the blade root. This fundamentally avoids the problem of matching error accumulation, achieves sub-millimeter-level stitching accuracy, and has high computational efficiency and strong robustness. Attached Figure Description

[0136] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0137] Figure 1 This is a flowchart of a processing method applicable to the root of large marine propellers in this invention. Detailed Implementation

[0138] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0139] This embodiment provides a processing method suitable for the root of large marine propellers, such as... Figure 1 As shown, the specific steps include:

[0140] S1. Install the propeller and standard ball on the B-axis rotary table of the machine tool, and install the high-precision calibration needle on the A spindle. Use a fitting algorithm to determine the positioning reference of the propeller based on the positions of the standard ball and the high-precision calibration needle, and construct the workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system.

[0141] S2. Collect propeller point cloud data, and identify the number of blades based on the propeller point cloud data;

[0142] S3. Create a single-blade photography template, and obtain the optimal photography path based on the single-blade photography template and the number of blades; integrate a high-resolution vision sensor on the A and B axes of the machine tool, and enable the high-resolution vision sensor to collect the original three-dimensional point cloud data of each part of the blade root according to the optimal photography path.

[0143] S4. Based on the positioning reference, convert the original three-dimensional point cloud data of each part of the blade root into point cloud data in the machine tool coordinate system, and determine the point cloud data of each part of the blade root in the workpiece coordinate system based on the conversion relationship.

[0144] S5. The point cloud data of each part of the blade root in the workpiece coordinate system are stitched together to obtain the complete three-dimensional point cloud data of the blade root.

[0145] Specifically, this embodiment, through positioning and coordinate system transformation operations, ensures that the stitching accuracy directly inherits the machine tool's positioning accuracy, far exceeding the stitching accuracy based on software feature matching (which is susceptible to noise and feature repetition). It solves the stitching problem in scenarios lacking stable features, such as smooth metal surfaces, highly reflective surfaces, and large-curvature continuous surfaces with weak textures. It eliminates the time-consuming feature matching and iterative optimization calculation process, resulting in extremely fast stitching speed and deterministic results, free from the influence of random initial values. This method highly relies on the motion accuracy of each axis of the machine tool, the encoder resolution, and the rigid connection and precise synchronization between the sensor and the machine tool's motion. Through the above mechanism, the complex point cloud stitching problem is transformed into a high-precision machine tool spatial positioning and kinematic calculation problem, thereby achieving stable, accurate, and efficient markerless stitching.

[0146] S6. Obtain the propeller theoretical model and register the complete three-dimensional point cloud data of the blade root with the propeller theoretical model.

[0147] S7. Calculate the processing allowance of the leaf root based on the registration result, generate the processing trajectory of the leaf root based on the processing allowance of the leaf root, and complete the processing of the leaf root region according to the processing trajectory.

[0148] In a specific embodiment, the propeller is mounted on the machine tool rotary table, and the positioning reference of the propeller is determined by a fitting algorithm. The scheme for constructing the workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system is as follows:

[0149] I. Installing the workpiece and calibration device

[0150] The propeller is manually clamped onto the B-axis rotary table, and the line laser scanner is installed above the A-axis spindle. A standard ball with a known radius is fixed on the B-axis rotary table to ensure that the ball's position is stable as the rotary table rotates. A high-precision calibration needle is installed on the A-axis spindle, and its tip position has been calibrated by the machine tool probe.

[0151] II. The propeller positioning reference is determined by a fitting algorithm to position the propeller. The specific steps are as follows:

[0152] Step 1: Data Acquisition and Processing

[0153] 1. Control the B-axis rotary table to rotate according to the set rotation angle. At each rotation angle, scan the standard sphere on the B-axis rotary table with a line laser scanner to obtain the point cloud data Ccam(k) of the standard sphere in the camera coordinate system at each angle; and record the current machine tool coordinates as the theoretical coordinates M(k); k represents the rotation angle.

[0154] With the A-axis at 0 degrees, at each set rotation angle, move the A-spindle so that the tip of the high-precision calibration needle contacts a preset contact point on the standard ball. Obtain the machine coordinates of the contact point in the machine coordinate system, i.e., the true coordinates of the contact point, Creal(k). Add the true machine coordinates of the contact point to the radius of the standard ball to obtain the true machine coordinates of the center of the standard ball in the machine coordinate system at each rotation angle. The machine coordinates include the X, Y, Z, A-axis, and B-axis positions.

[0155] In this scheme, the B-axis rotary table is controlled to rotate sequentially to 0°, 90°, 180°, and 270°. Under each rotation, the A-axis drives the line laser scanner to scan the standard sphere at a constant speed to obtain the point cloud data of the standard sphere in the camera coordinate system.

[0156] 2. Filter and denoise the scanned point cloud data; a fixed rod is fixed on the B-axis rotary table under the standard sphere. Filter and denoise the point cloud data, mainly removing the point cloud data of the fixed rod, to ensure the coordinate accuracy of subsequent point cloud fitting.

[0157] 3. Place the A-axis at the 0-degree position, move the A spindle until the high-precision calibration pin touches the contact point of the standard ball, and record the current machine tool coordinates;

[0158] Rotate the A-axis to the set angle, and move the A-spindle again until the high-precision calibration pin touches the contact point of the standard ball. Record the current machine tool coordinates to solve for the rotation center of the A-axis; in this scheme, the A-spindle is rotated 90 degrees.

[0159] Step 2: Determine the coordinates of the sphere center in the camera coordinate system, the rotation center OB of the B-axis rotary table, the rotation center OA of the A-axis, and the camera-machine tool translation relationship between the camera coordinate system and the machine tool coordinate system based on the collected data.

[0160] 1. Fit the standard sphere point cloud data at each rotation angle using the least squares method to obtain the coordinates of the standard sphere's center in the camera coordinate system at each rotation angle. ;

[0161] 2. Fit the coordinates of the sphere center using the least squares method to obtain the rotation center OB of the B-axis rotary table;

[0162] 3. Connect the machine tool coordinates acquired at the 0-degree position and the set angle position of the A-axis, draw the perpendicular bisector of the line segment, and the intersection of the perpendicular bisector with the rotation axis of the A-axis is the rotation center OA of the A-spindle;

[0163] The specific coordinates of OA are obtained through geometric principles and axis constraints. This is a conventional geometric solution, and those skilled in the art can know how to construct equations for calculation. Therefore, the specific solution process will not be explained in detail.

[0164] Step 3: Construct a hand-eye matrix, i.e., a positioning reference, based on the data obtained in Step 1 and Step 2, to perform the transformation between the camera coordinate system and the machine tool coordinate system according to the hand-eye matrix:

[0165] 1. Using the SVD decomposition algorithm to... By decomposing the matrix, the optimal rotation matrix and relative translation can be obtained. ;

[0166] 2. Add the coordinates of OB and OA to the relative translation to obtain the translation vector;

[0167] Combining the rotation matrix and translation vector, we form the final 4×4 homogeneous transformation matrix, namely the hand-eye matrix, as follows:

[0168] ;

[0169] Where R represents the rotation matrix and t represents the relative translation;

[0170] Step 4: Construct the workpiece coordinate system:

[0171] Establish the workpiece coordinate system with the rotation center OB of axis B as the origin of the workpiece coordinate system;

[0172] Step 5: Establish the transformation relationship between the workpiece coordinate system and the machine tool coordinate system as follows:

[0173] ;

[0174] in, Let the coordinates of the point be in the machine tool coordinate system. Let be the coordinates of the point in the workpiece coordinate system. The coordinates of the workpiece coordinate system origin in the machine tool coordinate system;

[0175] In a specific embodiment, in S2, a wide-view camera is used to collect propeller point cloud data, and the number of blades is identified based on the propeller point cloud data. The specific steps include:

[0176] S21. Perform voxel grid downsampling processing on the propeller point cloud data:

[0177] The three-dimensional space of the propeller point cloud data is divided into several uniform grids according to a preset voxel grid size. The centroid of the point set formed by the points in the propeller point cloud data in each grid is taken as the centroid of the current grid. All points in each grid are replaced with the centroid to obtain downsampled point cloud data.

[0178] Specifically, by downsampling, the amount of point cloud data can be significantly reduced while keeping the propeller's geometric features unchanged, thereby improving the speed of subsequent data processing.

[0179] S22. An adaptive ROI (Region of Interest) extraction algorithm is used to extract the effective point cloud region related only to the blade from the downsampled point cloud data;

[0180] Specifically, the adaptive ROI extraction algorithm can remove interference point clouds (such as environmental debris reflection points) in the vertical direction of the camera and the point cloud data of the turntable itself, retaining only the effective point cloud area related to the blade, thus avoiding irrelevant data from interfering with subsequent segmentation and registration.

[0181] S23, Leaf point cloud segmentation:

[0182] The region growing segmentation algorithm is used to divide the effective point cloud region into two independent leaf point cloud sets by dividing points with the same normal direction and curvature into the same sub-region.

[0183] Specifically, this embodiment uses the local geometric features (normal direction and curvature) of the point cloud as the judgment basis. Starting from the preset seed point, it iteratively calculates and gradually merges adjacent points with the same normal direction and curvature into the same region. Finally, the effective point cloud is divided into two independent leaf point cloud sets, realizing the separation of the leaf point cloud from other region point clouds.

[0184] S24. Use an edge detection algorithm to extract the edge point cloud of the two segmented blades to obtain the edge point cloud data of the two blades. The edge point cloud data includes the areas of surface discontinuity and sharp curvature changes in the point cloud, as well as the physical boundaries, edges or surface intersections of the blades corresponding to these areas.

[0185] S25. Point cloud registration calculation:

[0186] Coarse registration: Calculate the centroid coordinates of the point clouds at the edges of the two blades respectively, obtain the initial rigid body transformation matrix by aligning the centroids, and perform preliminary alignment of the point clouds at the edges of the two blades based on the initial rigid body transformation matrix to provide an initial estimate for fine registration;

[0187] Fine registration: The ICP (Iterative Closest Point) algorithm is used to minimize the distance between corresponding points on the edge point clouds of two blades, thereby iteratively optimizing the initial rigid body transformation matrix to obtain the optimal transformation matrix, and further aligning the edge point clouds of two blades based on the optimal transformation matrix.

[0188] S26. Determining the number of blades:

[0189] Extract the rotation matrix from the optimal transformation matrix, and convert the rotation matrix into rotation angle values ​​using a quaternion-to-Euler angle conversion algorithm;

[0190] The number of blades is determined based on the rotation angle value.

[0191] Specifically, if the rotation angle is around 90 degrees, the propeller is determined to be a four-bladed propeller; if the rotation angle is around 72 degrees, the propeller is determined to be a five-bladed propeller. The identification results are stored as key parameters for subsequent image path planning and processing trajectory generation.

[0192] In a specific embodiment, S3, the specific steps of creating a single-leaf photographing template and obtaining the optimal photographing path based on the single-leaf photographing template and the number of leaves include:

[0193] S31. Create a single-leaf photograph template:

[0194] The movement path of the large-view camera when collecting propeller point cloud data is obtained, and a single blade image template is obtained based on the movement path. The image template includes the spatial movement trajectory (coordinate sequence), attitude and shooting parameters of each point.

[0195] S32. Determine the reference blade:

[0196] Set the physical zero point of the turntable to 0 degrees in the workpiece coordinate system, and calculate the clockwise angle difference between the direction angle of each blade and the physical zero point of the turntable.

[0197] Compare the magnitudes of all clockwise angle differences, select the blade with the smallest clockwise angle difference as the reference blade closest to the physical zero point of the turntable, and use the reference blade as the starting blade for subsequent path extension;

[0198] S33. Generate the image path for the reference blade:

[0199] The transformation matrix between the reference blade and the photographic template is calculated using coarse registration (centroid method) and fine registration (ICP algorithm), and the rotation angle between the reference blade and the photographic template is obtained by converting quaternions to Euler angles.

[0200] The angle between the reference blade and the image template is converted into a transformation matrix. The image template is then transformed based on the transformation matrix to obtain the image path of the reference blade.

[0201] Specifically, for the reverse side of the reference blade, the XY coordinates of the shooting point are kept completely consistent. By rotating the Z axis by 180 degrees, the shooting path of the reverse side of the reference blade can be obtained, ensuring that both the front and back sides of the blade can be completely scanned.

[0202] In this embodiment, all transformation matrices contain only rotational transformations, and the translation component is set to zero. Since the blades rotate around the center of the turntable and are placed horizontally on the turntable, their vertical (Z-axis) position remains unchanged, ensuring the accuracy of path expansion.

[0203] S34, Multi-leaf photography path extension:

[0204] The imaging path of the Nth blade is obtained by rotating and transforming the imaging path of the reference blade according to the rotation angle. This allows us to obtain the original photographic paths of all the blades. The calculation process for the rotation angle is as follows:

[0205] Rotation angle = rotation angle between the reference blade and the photo template + (N-1) × rotation angle value determined during the blade number recognition process, such as the rotation angle between adjacent blades of a four-bladed propeller is 90 degrees, and that of a five-bladed propeller is 72 degrees.

[0206] S35. Optimize the original image path for all leaves:

[0207] The photo points along the original photo paths of all leaves are integrated into a single point set. The problem of the order of photo point visits is modeled as a Traveling Salesman Problem (TSP), that is, any two photo points are defined... S i and S j The cost of moving between them is the Euclidean distance between them:

[0208] d ij =|| S i - S j ||;

[0209] Starting from any point, a nearest neighbor greedy algorithm is used to select the next point with the closest Euclidean distance as the next destination, thus obtaining an initial shooting path;

[0210] The initial photography path is iteratively improved using the 2-opt local optimization method: try swapping any two segments in the path, calculate the change in the total path length before and after the swap, and only perform the swap if it can reduce the total distance. By iterating repeatedly, the optimal photography path is found by finding the order of accessing the photography point indexes that minimizes the sum of Euclidean distances.

[0211] A high-resolution vision sensor (such as a line laser scanner) is controlled to scan the leaf root from multiple angles according to the optimal imaging path, thereby acquiring the original three-dimensional point cloud data of each part of the leaf root.

[0212] Specifically, since the blade root region is typically a deep cavity structure, ordinary three-axis scanning (moving only XYZ) is easily obstructed by the blades, resulting in blind spots. In this embodiment, a high-resolution vision sensor is directly integrated into the spindle of a high-precision machine tool, reusing the processing equipment as a detection device. Utilizing the rotation of the A-axis and B-axis, the scanning head can flexibly change its orientation, thereby bypassing obstructions and entering the physical blind spots caused by blade root overlap and obstruction, directly illuminating the inner wall of the blade root. Specifically, the A-axis (usually the main rotation axis) is responsible for a wide range of orientation adjustments. For example, the scanning head can be positioned to probe the deep cavity from above the blade, or the pitch angle of the scanning head can be adjusted so that its beam is perpendicular to the inner wall of the blade root. The B-axis (usually the auxiliary rotation axis) is responsible for fine-tuning the orientation of the scanning head, working in conjunction with the A-axis to achieve omnidirectional coverage of the scanning head within the deep cavity, ensuring no blind spots. The A-axis and B-axis move in real time in coordination with the propeller's CAD model (containing the precise geometry of the blade root). For example, when the scanning head moves along the circumference of the leaf root, the A-axis and B-axis rotate synchronously, always keeping the scanning beam perpendicular to the surface normal of the current scanning point, thus ensuring acquisition accuracy.

[0213] Specifically, in this embodiment, in order to ensure the accuracy of subsequent stitching, the point cloud data of two adjacent scans are set to have at least 30% overlap.

[0214] In a specific embodiment, S4, the specific steps of converting the original three-dimensional point cloud data of each part of the blade root into point cloud data in the machine tool coordinate system based on the positioning reference, and determining the point cloud data in the workpiece coordinate system based on the conversion relationship, include:

[0215] S41. Perform point cloud preprocessing and noise reduction on the original three-dimensional point cloud data of each part of the leaf root:

[0216] A statistical outlier removal method is adopted to eliminate noise points; a voxel grid filter is used to uniformly downsample the point cloud, which reduces computational complexity while preserving geometric features.

[0217] Statistical outlier removal is a denoising algorithm based on statistical analysis of the local neighborhood of point clouds. The neighborhood distances of normal points follow a normal distribution, while the neighborhood distances of outliers deviate significantly from this distribution. The algorithm removes the detected outliers from the original point cloud. Voxel mesh downsampling divides the three-dimensional space into uniform grids (voxels) and replaces all points within each voxel with a representative point (usually the centroid), thereby significantly reducing the amount of point cloud data while preserving geometric features. Statistical outlier removal methods and voxel mesh filters are common techniques in point cloud processing. Those skilled in the art know how to use these two algorithms to process point cloud data, so they will not be elaborated upon further.

[0218] S42. Convert the preprocessed point cloud data from the camera coordinate system to the machine tool coordinate system using a hand-eye matrix. Then, based on the conversion relationship between the machine tool coordinate system and the workpiece coordinate system, convert the point cloud data in the machine tool coordinate system to the point cloud data in the workpiece coordinate system.

[0219] In a specific embodiment, S5 involves stitching together the point cloud data of each part of the blade root in the workpiece coordinate system to obtain the complete three-dimensional point cloud data of the blade root. The specific steps include:

[0220] S51. Perform coarse stitching of the point cloud data of each part of the blade root in the workpiece coordinate system directly according to the workpiece coordinate system.

[0221] S52. Based on the ICP algorithm, the coarse stitching results are finely stitched to obtain the complete three-dimensional point cloud data of the leaf roots.

[0222] Specifically, due to mechanical and cumulative errors, coarsely stitching the point cloud data of each part of the blade root in the workpiece coordinate system directly according to the workpiece coordinate system may result in misalignment or ghosting of the point cloud. This embodiment uses the ICP algorithm to perform fine stitching on the coarse stitching result (using the coarse stitching result as an excellent initial value for the ICP algorithm). Utilizing the 30% overlap area between point cloud data from two adjacent scans, the ICP algorithm automatically finds the optimal matching position between adjacent point clouds. Through iterative calculation, it minimizes the error in the overlap area, thereby obtaining the complete three-dimensional point cloud data of the blade root.

[0223] Specifically, this embodiment abandons the traditional ICP approach that relies on the features of the point cloud itself for matching, and establishes a direct geometric transformation stitching method based on the high-precision encoder readings of the machine tool motion system. This fundamentally avoids the problems of matching failure or error accumulation in traditional ICP on weak textures at the blade root and highly reflective metal surfaces. This operation transforms the stitching calculation into a simple matrix operation, eliminating the need for iterative optimization and significantly improving the efficiency of real-time or near real-time field data processing.

[0224] In a specific embodiment, S6 involves obtaining the propeller theoretical model and registering the complete three-dimensional point cloud data of the blade root with the propeller theoretical model. The specific steps include:

[0225] S61. Perform point cloud processing on the propeller theoretical model: Construct a CAD model of the blade root, convert it into theoretical point cloud data according to the preset sampling density, and retain the complete surface features;

[0226] Specifically, in this embodiment, the STEPCAFControl_Reader and XCAFDoc_ShapeTool functions of Opencascade are used to convert the CAD model into a point cloud to obtain a theoretical point cloud.

[0227] S62,

[0228] The ISS feature point extraction algorithm was used to extract the ISS key points of the actual collected point cloud, namely the complete 3D point cloud data of the leaf root and the theoretical point cloud data.

[0229] The extracted ISS keypoints are subjected to FPFH (Fast Point Feature Histogram) feature matching to establish initial corresponding point pairs;

[0230] Specifically, the ISS feature point extraction algorithm identifies feature points with significant geometric changes by analyzing the distribution of eigenvalues ​​of the covariance matrix of each point's neighborhood. The algorithm assumes that the local surface at a feature point exhibits significant changes in all three principal directions, manifested by relatively large eigenvalues ​​in all three directions; the algorithm will extract points that meet these requirements.

[0231] Specifically, IPFH is a local feature descriptor used to describe the geometric features of a point's neighborhood in a 3D point cloud. For each query point, it is necessary to calculate its relationships with all neighboring points. When ISS and FPFH are used simultaneously, FPFH will only perform feature matching on the feature points acquired by ISS, greatly improving the registration speed and quality.

[0232] ISS and FPFH are both commonly used techniques in the field of technical expertise, so they will not be described in detail.

[0233] S63. Based on the SVD algorithm, perform best-fit registration on the initial corresponding point pairs to achieve registration of the actual collected point cloud data and the theoretical point cloud data, including:

[0234] Let the actual point cloud in the initial corresponding point pair be... The theoretical point cloud corresponds to the point as The centroids of the actual point cloud and the theoretical corresponding point cloud are calculated as follows:

[0235] ;

[0236] ;

[0237] The covariance matrix calculated based on the centroids of the two point clouds is as follows:

[0238] ;

[0239] The covariance matrix is ​​decomposed by SVD as follows:

[0240] ;

[0241] Obtain the optimal rotation matrix Translation vector for:

[0242] ;

[0243] ;

[0244] The optimal rotation matrix and translation vector are transformed and registered in the actual acquired point cloud to obtain:

[0245] .

[0246] In a specific embodiment, the specific steps for calculating the machining allowance of the leaf root based on the registration result include:

[0247] S71. The normal vector of the theoretical point cloud is determined using principal component analysis, and the normal vector is corrected, including:

[0248] Extract any point from the theoretical point cloud Local neighborhood point cloud;

[0249] Calculate the covariance matrix of the local neighborhood point cloud;

[0250] Eigenvalue decomposition of the covariance matrix yields an orthogonal matrix. ;in Describes the main direction of the distribution of the neighborhood point cloud, i.e. , Description and The orthogonal second largest distribution direction, i.e. , Description and , All orthogonal directions, i.e. ;

[0251] Determine the normal vector:

[0252] For the upper surface of the propeller (pressure surface / suction surface): normal vector (Pointing outwards from the workpiece, in the direction of material surplus);

[0253] For the lower surface of the propeller (non-working surface): normal vector (Pointing to the inside of the workpiece).

[0254] Setting principal direction constraints to correct the normal vector includes:

[0255] Set the main direction of the leaf root region ,like The direction may be specified according to process requirements;

[0256] Calculate the normal vector The angle between the normal vector and the principal direction of the leaf root region is considered. If the angle is less than or equal to a set angle threshold, it indicates that the normal vector direction conforms to the constraint of the principal direction of the leaf root region. This normal vector is retained and used as the corrected normal vector. If the angle is greater than the set angle threshold, it indicates that the normal vector direction error is too large, and it is... Projected onto the main direction of the leaf root region Then, normalization is performed to obtain the corrected normal vector;

[0257] S72, with points Starting from the modified normal vector Search for the actual corresponding point:

[0258] For points on the upper surface of the leaf root, along Corrected normal vector A ray is emitted in the positive direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of the values ​​of all candidate points and the ray is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ;

[0259] For points on the lower surface of the leaf root, along Corrected normal vector A ray is emitted in the negative direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of all candidate points and their values ​​is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ;

[0260] S73. Calculate the machining allowance at a single point:

[0261] For the matched point pairs Calculate its machining allowance for:

[0262] ;

[0263] in, express and The straight-line distance between them;

[0264] S74. Calculate the processing allowance for the entire point cloud:

[0265] Iterate through all points in the theoretical point cloud model, execute steps S72 and S73, and generate a set of machining allowance values ​​that correspond one-to-one with the theoretical points. ;

[0266] S75. Construct a machining allowance distribution diagram:

[0267] Each theoretical point The coordinates and other values By associating these elements, a three-dimensional allowance distribution mapping relationship is formed for the blades, thereby obtaining the machining allowance for each point.

[0268] In a specific embodiment, the specific steps for generating the leaf root machining trajectory based on the machining allowance of the leaf root include:

[0269] Receiving theoretical point clouds and the set of machining allowance values ​​corresponding one-to-one with the theoretical points. ;

[0270] Specifically, the theoretical model point cloud is a set of theoretically designed points that perfectly match the actual propeller's spatial pose, satisfying the "horizontal placement constraint" (the bottom surface of the propeller hub is horizontal, and the axis is parallel to the Z-axis). This serves as the core geometric data for subsequent path planning, coordinate system establishment, and thickness calculation. The machining allowance distribution set includes data from the theoretical point cloud. Each point in The one-to-one corresponding normal machining allowance value provides data support for the accuracy compensation of subsequent machining trajectories.

[0271] S101. Obtain local processing direction parameters based on the point cloud of the theoretical model, including:

[0272] S1011. Based on the radial and circumferential features of the propeller, initialize the global path direction to the circumferential direction and the scanning direction to the radial direction;

[0273] S1012. Optimize the global path direction and scanning direction using the principal curvature analysis method, ensuring that the global path direction follows the local minimum curvature direction and the scanning direction follows the local maximum curvature direction. This ensures that the machining trajectory is evenly covered in the curvature abrupt change region, avoiding sparse or overly dense trajectories. The specific steps are as follows:

[0274] Extracting the local point cloud neighborhood:

[0275] For each point to be optimized along the global path, a local point cloud neighborhood is extracted around it. The radius of the local point cloud neighborhood is set according to the point cloud density and the leaf feature size.

[0276] Calculate the local surface normal vector and principal curvature of the local point cloud neighborhood, where:

[0277] Local surface normal vector: The covariance matrix of the local neighborhood point cloud is calculated by principal component analysis (PCA), and the eigenvector corresponding to the smallest eigenvalue is the local surface normal vector;

[0278] Calculate the principal curvature:

[0279] By fitting a quadratic surface (such as a plane, cylinder, or parabola) to the local point cloud neighborhood, the equation of the fitted surface is obtained.

[0280] The two principal curvatures of the point to be optimized are calculated based on the fitted surface equation. and ,and , respectively corresponding to the direction of maximum curvature and the direction of minimum curvature ;

[0281] Orientation projection and alignment:

[0282] Will and Fusion with global path direction and scan direction:

[0283] Project the global circumferential direction and the global radial direction onto the tangent planes respectively, calculate the dot product of the two tangent planes after projection, and use the two dot product results as the first direction similarity and the second direction similarity respectively;

[0284] Determine if the absolute value of the first-direction similarity is greater than the set similarity threshold. If it is, it means that the current circumferential direction is consistent with the direction of minimum curvature, and the global path direction remains the circumferential direction; otherwise, it means that the curvature feature of the current region is significant, and the global path direction is corrected to the direction of minimum curvature. The system checks if the absolute value of the second-direction similarity is greater than a set similarity threshold. If it is, it indicates that the current radial direction is consistent with the direction of maximum curvature, and the scanning direction remains radial. Otherwise, it indicates that the curvature features of the current region are significant, and the scanning direction is corrected to the direction of maximum curvature. ;

[0285] S102. Divide the processing allowance of the leaf root into equally spaced sections along the optimized global path direction, sort and resample the point cloud slices of each section at equal intervals, and generate a two-dimensional path point sequence, including:

[0286] Along the optimized global path direction, the machining allowance of the leaf root is divided into equal-spaced cross sections according to the preset path spacing, forming a series of parallel cross-sectional planes;

[0287] For each cross-sectional plane, calculate the distance between the projected values ​​of all points and the current cross-sectional reference value. Filter out all Points smaller than the cross-sectional tolerance range form the original cross-sectional point set;

[0288] Point cloud slicing based on directional vector projection is a basic operation in 3D data processing and a common technique used by those skilled in the art. Therefore, the specific calculation process will not be described in detail.

[0289] The original cross-sectional point sets of each cross-sectional plane are sorted from smallest to largest along the optimized scanning direction. Then, the sorted original cross-sectional point sets are resampled at equal intervals according to the preset point spacing to generate a two-dimensional path point sequence. This not only preserves the key shape features of the cross-section, but also avoids trajectory jitter caused by uneven density of the original point cloud.

[0290] S103. Combine the two-dimensional path point sequence with the normal coordinates of the corresponding cross section to form a three-dimensional trajectory point, insert safe transition points, and perform trajectory smoothing to generate a three-dimensional smooth trajectory point sequence, including:

[0291] Each two-dimensional path point sequence is combined with the normal coordinate value of the corresponding cross section to synthesize a three-dimensional spatial trajectory point, ensuring the spatial accuracy of the trajectory point;

[0292] A safe transition point is inserted between the start and end points of two adjacent two-dimensional path point sequences to integrate the three-dimensional path point sequences of all sections, resulting in an integrated three-dimensional path point sequence. The safe transition point is formed by raising the midpoint of the line connecting the start and end points to a preset safe height along the corrected normal vector direction. Inserting a safe transition point can prevent interference between the tool and the workpiece and ensure safe and smooth movement between paths.

[0293] The integrated 3D path point sequence is smoothed by moving average or spline curve to obtain the final leaf root processing trajectory; this eliminates the slight jitter introduced by discrete sampling, ensuring smooth robot movement during processing and improving the quality of the processed surface.

[0294] In this embodiment, based on the processing requirements of the leaf root region, roughing, semi-finishing, and finishing are performed, wherein:

[0295] Roughing (milling): Removes most of the machining allowance, uses larger cutting parameters, quickly approaches the theoretical shape, and at the same time leaves a uniform allowance for semi-finishing.

[0296] Semi-finishing (milling): Further optimize the workpiece shape, remove surface defects caused by roughing, and leave a small amount of finishing allowance;

[0297] Precision machining (grinding): For key areas such as the blade root fillet transition area, a fine grinding process is used to ensure that the surface roughness and dimensional accuracy meet the design requirements.

[0298] Specifically, in practice, the processing includes:

[0299] Macro motion control (machine tool): Based on the three-dimensional machining trajectory, control the machine tool to quickly and roughly position the tool to a safe position near the blade root, and provide a stable machining platform to ensure the basic accuracy of the tool movement;

[0300] Micro-motion control (FTS): Under conditions where the machine tool is almost stationary or feeds at low speed, the FTS (Flexible Transmission System) is controlled to execute the machining trajectory of fine parts at extremely high frequency and precision, so as to achieve high-precision forming.

[0301] Specifically, this embodiment also includes closed-loop detection and data archiving:

[0302] Real-time inspection during processing: After each section of the blade root fillet is processed to a preset length (e.g., 5mm), the area is immediately scanned to measure dimensional errors and assess contour accuracy and surface roughness. If the inspection results are unqualified, the processing parameters are immediately adjusted based on the error data, and compensation and fine-tuning are performed in that local area to avoid error accumulation.

[0303] Post-processing full-area inspection: After processing, the propeller blades are continuously scanned in three dimensions to generate a final three-dimensional quality map, which is then compared with the theoretical model to verify whether the processing accuracy meets the requirements.

[0304] Data archiving: All data during the processing (including point cloud data, registration results, processing allowance, trajectory parameters, inspection reports, etc.) are automatically archived and associated with the digital twin model of the workpiece to achieve full lifecycle traceability;

[0305] Handling of non-conformities: If the overall inspection results do not meet the design requirements, return to the machining allowance calculation step, regenerate the machining allowance distribution and machining trajectory, and perform secondary machining until the inspection is qualified.

[0306] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A processing method for the root of a large marine propeller blade, characterized in that, The specific steps include: S1. Install the propeller and standard ball on the B-axis rotary table of the machine tool, and install the high-precision calibration needle on the A spindle. Use a fitting algorithm to determine the positioning reference of the propeller based on the positions of the standard ball and the high-precision calibration needle, and construct the workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system. S2. Collect propeller point cloud data, and identify the number of blades based on the propeller point cloud data; S3. Create a single-blade photography template, and obtain the optimal photography path based on the single-blade photography template and the number of blades; integrate a high-resolution vision sensor on the A and B axes of the machine tool, and enable the high-resolution vision sensor to collect the original three-dimensional point cloud data of each part of the blade root according to the optimal photography path. S4. Based on the positioning reference, convert the original three-dimensional point cloud data of each part of the blade root into point cloud data in the machine tool coordinate system, and determine the point cloud data of each part of the blade root in the workpiece coordinate system based on the conversion relationship. S5. The point cloud data of each part of the blade root in the workpiece coordinate system are stitched together to obtain the complete three-dimensional point cloud data of the blade root. S6. Obtain the propeller theoretical model and register the complete three-dimensional point cloud data of the blade root with the propeller theoretical model. S7. Calculate the processing allowance of the leaf root based on the registration result, generate the processing trajectory of the leaf root based on the processing allowance of the leaf root, and complete the processing of the leaf root region according to the processing trajectory. The specific steps for calculating the machining allowance of the leaf root based on the registration results include: S71. The normal vector of the theoretical point cloud is determined using principal component analysis, and the normal vector is corrected, including: Extract any point from the theoretical point cloud Local neighborhood point cloud; Calculate the covariance matrix of the local neighborhood point cloud; Eigenvalue decomposition of the covariance matrix yields an orthogonal matrix. ;in Describes the main direction of the distribution of the neighborhood point cloud, i.e. , Description and The orthogonal second largest distribution direction, i.e. , Description and , All orthogonal directions, i.e. ; Determine the normal vector: For the upper surface of the propeller: normal vector ; For the lower surface of the propeller: normal vector ; Setting principal direction constraints to correct the normal vector includes: Set the main direction of the leaf root region ; Calculate the normal vector The angle between the normal vector and the principal direction of the leaf root region. If the angle is less than or equal to a set angle threshold, it means the normal vector direction conforms to the principal direction constraint of the leaf root region. This normal vector is retained and used as the corrected normal vector. If the angle is greater than the set angle threshold, then... Projected onto the main direction of the leaf root region Then, normalization is performed to obtain the corrected normal vector; S72, with points Starting from the modified normal vector Search for the actual corresponding point: For points on the upper surface of the leaf root, along Corrected normal vector A ray is emitted in the positive direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of the values ​​of all candidate points and the ray is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ; For points on the lower surface of the leaf root, along Its normal vector after correction A ray is emitted in the negative direction. All points within a set threshold range near this ray in the actual point cloud are searched as candidate points, and the sum of all candidate points and their values ​​is calculated. Find the point closest to the theoretical point along the ray direction as the distance to the theoretical point. Corresponding actual measurement points ; S73. Calculate the machining allowance at a single point: For point pairs Calculate its machining allowance for: ； in, express and The straight-line distance between them; S74. Calculate the processing allowance for the entire point cloud: Iterate through all points in the theoretical point cloud model, execute steps S72 and S73, and generate a set of machining allowance values ​​that correspond one-to-one with the theoretical points. ; S75. Construct a machining allowance distribution diagram: Each theoretical point The coordinates and other values By associating these elements, a three-dimensional margin distribution mapping relationship is formed at the leaf root, thereby obtaining the processing margin at each point.

2. The processing method for the root of large marine propeller blades according to claim 1, characterized in that, In S1, the propeller positioning reference is determined based on the positions of a standard sphere and a high-precision calibration needle using a fitting algorithm, including: Set spindle A to the 0-degree position, and control the B-axis rotary table to rotate at the set angle. Perform the following steps at each angle: S11. Obtain the point cloud data of the standard sphere in the camera coordinate system at various rotation angles; S12. Move spindle A so that the tip of the high-precision calibration needle contacts the preset contact point on the standard ball, and obtain the machine coordinates of the contact point in the machine coordinate system, that is, the true coordinates of the contact point; S13. Add the actual machine coordinates of the contact point to the radius of the standard sphere to obtain the actual machine coordinates of the center of the standard sphere in the machine coordinate system at each rotation angle. The machine tool coordinate system includes the X, Y, Z, A-axis, and B-axis positions; k represents the rotation angle. Place the A-axis at 0 degrees and at a set angle, move the A-spindle until the high-precision calibration pin touches the contact point of the standard ball, and record the machine tool coordinates at each angle. The point cloud data of the standard sphere in the camera coordinate system is filtered and denoised to obtain standard point cloud data; The standard point cloud data is fitted using the least squares method to determine the coordinates of the sphere center in the camera coordinate system, the rotation center OB of the B-axis rotary table, and the rotation center OA of the A-axis. The steps are as follows: The standard point cloud data at each rotation angle is fitted using the least squares method to obtain the coordinates of the center of the standard sphere in the camera coordinate system at each rotation angle. ; The rotation center OB of the B-axis rotary table is obtained by fitting the coordinates of the sphere center using the least squares method. Connect the machine tool coordinates acquired at the 0-degree position and the set angle position of the A-axis, draw the perpendicular bisector of the line segment, and the intersection of the perpendicular bisector with the rotation axis of the A-axis is the rotation center OA of the A-spindle; based on and Construct a hand-eye matrix, i.e., a positioning reference, to perform the transformation between the camera coordinate system and the machine tool coordinate system based on the hand-eye matrix. The steps are as follows: Using the SVD decomposition algorithm By decomposing the matrix, the optimal rotation matrix and relative translation can be obtained. ; Add the coordinates of OB and OA to the relative translation to obtain the translation vector; Combining the rotation matrix and translation vector, we form the final 4×4 homogeneous transformation matrix, namely the hand-eye matrix, as follows: ； Where R represents the rotation matrix and t represents the relative translation.

3. The processing method for the root of large marine propeller blades according to claim 2, characterized in that, Establish the workpiece coordinate system and the transformation relationship between the workpiece coordinate system and the machine tool coordinate system, including: Establish the workpiece coordinate system with the rotation center OB of axis B as the origin of the workpiece coordinate system; The transformation relationship between the workpiece coordinate system and the machine tool coordinate system is established as follows: ； in, Let the coordinates of the point be in the machine tool coordinate system. Let be the coordinates of the point in the workpiece coordinate system. The coordinates of the origin of the workpiece coordinate system in the machine tool coordinate system.

4. The processing method for the root of large marine propeller blades according to claim 3, characterized in that, In S2, a wide-view camera is used to collect propeller point cloud data, and the number of blades is identified based on the propeller point cloud data. The specific steps include: S21. Perform voxel grid downsampling processing on the propeller point cloud data: The three-dimensional space of the propeller point cloud data is divided into several uniform grids according to a preset voxel grid size. The centroid of the point set formed by the points in the propeller point cloud data in each grid is taken as the centroid of the current grid. All points in each grid are replaced with the centroid to obtain downsampled point cloud data. S22. An adaptive ROI extraction algorithm is used to extract the effective point cloud region that is only related to the blade from the downsampled point cloud data. S23, Leaf point cloud segmentation: The region growing segmentation algorithm is used to divide the effective point cloud region into two independent leaf point cloud sets by dividing points with the same normal direction and curvature into the same sub-region. S24. Use an edge detection algorithm to extract the edge point cloud of the two segmented blades to obtain the edge point cloud data of the two blades. The edge point cloud data includes the areas of surface discontinuity and sharp curvature changes in the point cloud, as well as the physical boundaries, edges or surface intersections of the blades corresponding to these areas. S25. Point cloud registration calculation: Coarse registration: Calculate the centroid coordinates of the point clouds at the edges of the two blades respectively, obtain the initial rigid body transformation matrix by centroid alignment, and perform preliminary alignment of the point clouds at the edges of the two blades based on the initial rigid body transformation matrix; Fine registration: The ICP algorithm is used to minimize the distance between corresponding points on the edge point clouds of two blades, thereby iteratively optimizing the initial rigid body transformation matrix to obtain the optimal transformation matrix, and further aligning the edge point clouds of two blades based on the optimal transformation matrix. S26. Determining the number of blades: Extract the rotation matrix from the optimal transformation matrix, and convert the rotation matrix into rotation angle values ​​using a quaternion-to-Euler angle conversion algorithm; The number of blades is determined based on the rotation angle value.

5. The processing method for the root of large marine propeller blades according to claim 4, characterized in that, In S3, the specific steps for creating a single-leaf photographing template and obtaining the optimal photographing path based on the single-leaf photographing template and the number of leaves include: S31. Create a single-leaf photograph template: The movement path of the large-view camera when collecting propeller point cloud data is obtained, and a single blade image template is obtained based on the movement path. The image template includes the spatial movement trajectory, attitude and shooting parameters of each point. S32. Determine the reference blade: Set the physical zero point of the turntable to 0 degrees in the workpiece coordinate system, and calculate the clockwise angle difference between the direction angle of each blade and the physical zero point of the turntable. Compare the magnitudes of all clockwise angle differences and select the blade with the smallest clockwise angle difference as the reference blade closest to the physical zero point of the turntable; S33. Generate the image path for the reference blade: The transformation matrix between the reference blade and the photographic template is calculated using coarse and fine registration methods, and the rotation angle between the reference blade and the photographic template is obtained by converting quaternions to Euler angles. The angle between the reference blade and the image template is converted into a transformation matrix. The image template is then transformed based on the transformation matrix to obtain the image path of the reference blade. S34, Multi-leaf photography path extension: The imaging path of the Nth blade is obtained by rotating and transforming the imaging path of the reference blade according to the rotation angle. This allows us to obtain the original photographic paths of all the blades. The calculation process for the rotation angle is as follows: Rotation angle = rotation angle between the reference blade and the photo template + (N-1) × rotation angle value determined during the blade number recognition process; S35. Optimize the original image path for all leaves: The original photography paths of all leaves are integrated into a single point set. The problem of the order of visiting photography points is modeled as a traveling salesman problem, i.e., any two photography points are defined... S i and S j The cost of moving between them is the Euclidean distance between them: d ij =|| S i - S j ||； Starting from any point, a nearest neighbor greedy algorithm is used to select the next point with the closest Euclidean distance as the next destination, thus obtaining an initial shooting path; The initial photography path is iteratively improved using the 2-opt local optimization method: try swapping any two segments in the path, calculate the change in the total path length before and after the swap, and only perform the swap if it can reduce the total distance. By iterating repeatedly, the optimal photography path is found by finding the order of accessing the photography point indexes that minimizes the sum of Euclidean distances. The high-resolution vision sensor is controlled to scan the leaf root from multiple angles according to the optimal imaging path, thereby acquiring the original three-dimensional point cloud data of each part of the leaf root.

6. The processing method for the root of a large marine propeller blade according to claim 5, characterized in that, In S5, the specific steps for stitching together the point cloud data of each part of the blade root in the workpiece coordinate system to obtain the complete three-dimensional point cloud data of the blade root include: S51. Perform coarse stitching of the point cloud data of each part of the blade root in the workpiece coordinate system directly according to the workpiece coordinate system. S52. Based on the ICP algorithm, the coarse stitching results are finely stitched to obtain the complete three-dimensional point cloud data of the leaf roots.

7. The processing method for the root of large marine propeller blades according to claim 6, characterized in that, In S6, the specific steps for obtaining the propeller theoretical model and registering the complete three-dimensional point cloud data of the blade root with the propeller theoretical model include: S61. Perform point cloudification on the propeller theoretical model: Construct a CAD model of the blade root, and perform point cloudification on the CAD model according to a preset sampling density to convert it into theoretical point cloud data. S62. Use the ISS feature point extraction algorithm to extract the ISS key points of the actual point cloud, namely the complete 3D point cloud data of the leaf root propeller and the theoretical point cloud data. Perform FPFH feature matching on the extracted ISS key points to establish initial corresponding point pairs; S63. Based on the SVD algorithm, perform best-fit registration on the initial corresponding point pairs to achieve registration of the actual collected point cloud data and the theoretical point cloud data, including: Let the actual point cloud in the initial corresponding point pair be... The theoretical point cloud corresponds to the point as The centroids of the actual point cloud and the theoretical corresponding point cloud are calculated as follows: ； ； The covariance matrix calculated based on the centroids of the two point clouds is as follows: ； The covariance matrix is ​​decomposed by SVD as follows: ； Obtain the optimal rotation matrix Translation vector for: ； ； The optimal rotation matrix and translation vector are transformed and registered in the actual acquired point cloud to obtain: 。 8. The processing method for the root of a large marine propeller blade according to claim 7, characterized in that, The specific steps for generating the leaf root machining trajectory based on the machining allowance of the leaf root include: S101. Obtain local processing direction parameters based on the point cloud of the theoretical model, including: S1011. Based on the radial and circumferential features of the propeller, initialize the global path direction to the circumferential direction and the scanning direction to the radial direction; S1012. Optimize the global path direction and scanning direction using the principal curvature analysis method, so that the global path direction is along the local minimum curvature direction and the scanning direction is along the local maximum curvature direction. The specific steps are as follows: Extracting the local point cloud neighborhood: For each point to be optimized along the global path, a local point cloud neighborhood is extracted around it. The radius of the local point cloud neighborhood is set according to the point cloud density and the leaf feature size. Calculate the local surface normal vector and principal curvature of the local point cloud neighborhood, where: Local surface normal vector: The covariance matrix of the local neighborhood point cloud is calculated by principal component analysis, and the eigenvector corresponding to the smallest eigenvalue is the local surface normal vector; Calculate the principal curvature: A quadratic surface is fitted to the local point cloud neighborhood to obtain the fitted surface equation; The two principal curvatures of the point to be optimized are calculated based on the fitted surface equation. and ,and , respectively corresponding to the direction of maximum curvature and the direction of minimum curvature ; Orientation projection and alignment: Will and Fusion with global path direction and scan direction, including: Project the global circumferential direction and the global radial direction onto the tangent planes respectively, calculate the dot product of the two tangent planes after projection, and use the two dot product results as the first direction similarity and the second direction similarity respectively; Determine if the absolute value of the first-direction similarity is greater than the set similarity threshold. If it is, it means that the current circumferential direction is consistent with the direction of minimum curvature, and the global path direction remains the circumferential direction; otherwise, it means that the curvature feature of the current region is significant, and the global path direction is corrected to the direction of minimum curvature. The system checks if the absolute value of the second-direction similarity is greater than a set similarity threshold. If it is, it indicates that the current radial direction is consistent with the direction of maximum curvature, and the scanning direction remains radial. Otherwise, it indicates that the curvature features of the current region are significant, and the scanning direction is corrected to the direction of maximum curvature. ; S102. Divide the processing allowance of the leaf root into equally spaced sections along the optimized global path direction, sort and resample the point cloud slices of each section at equal intervals, and generate a two-dimensional path point sequence, including: Along the optimized global path direction, the machining allowance of the leaf root is divided into equal-spaced cross sections according to the preset path spacing, forming a series of parallel cross-sectional planes; For each cross-sectional plane, calculate the distance between the projected values ​​of all points and the current cross-sectional reference value. Filter out all Points smaller than the cross-sectional tolerance range form the original cross-sectional point set; The original cross-sectional point set of each cross-sectional plane is sorted from smallest to largest along the optimized scanning direction, and then the sorted original cross-sectional point set is resampled at equal intervals according to the preset point spacing to generate a two-dimensional path point sequence. S103. Combine the two-dimensional path point sequence with the normal coordinates of the corresponding cross section to form a three-dimensional trajectory point, insert safe transition points, and perform trajectory smoothing to generate a three-dimensional smooth trajectory point sequence, including: Each two-dimensional path point sequence is combined with the normal coordinate value of the corresponding cross section to synthesize a three-dimensional spatial trajectory point; A safe transition point is inserted between the start and end points of two adjacent two-dimensional path point sequences to integrate the three-dimensional path point sequences of all sections, resulting in an integrated three-dimensional path point sequence; wherein, the safe transition point is formed by raising the midpoint of the line connecting the start and end points to a preset safe height along the modified normal vector direction; The integrated 3D path point sequence is smoothed by moving average or spline curve to obtain the final leaf root processing trajectory.

Images