Method for selecting points for blade profile of precision forging blade

Abstract

The present invention discloses a method for selecting points for the blade profile contour of a precision forged blade, which belongs to the technical field of blade contour detection, and includes: performing hierarchical filtering on the forging point cloud data; obtaining the structural features and corresponding coordinate data of the forging, and obtaining the leading edge curvature set of the forging under the coordinate data corresponding to each structural feature; inputting the leading edge curvature set into a weight determination model, wherein the weight determination model divides the forging into several regions based on the coordinate data set and the leading edge curvature set, and sets the identification threshold and verification mechanism of the intake edge and exhaust edge under each region; based on the identification threshold and verification mechanism and according to the leading edge curvature set, each region is adaptively segmented and selected points to correct the deviation of the intake edge and exhaust edge during the fitting process. The present application adopts hierarchical filtering and noise separation technology, and performs dual registration through curvature constraints and bidirectional geometric constraints, combined with the coupled deformation model of material-process-geometric features.

Description

Technical Field

[0001] The present invention relates to the technical field of blade profile detection, in particular to a method for taking points for the blade profile of a precision forged blade. Background Art

[0002] In the inspection of precision forged blade profiles, the profile measurement of the double-notch area on the inlet / exhaust edges faces a series of technical bottlenecks, which seriously affect the accuracy and efficiency of inspection.

[0003] When inspecting the contour of the double-notch area on the intake / exhaust edge, existing methods (such as discrete sampling on a three-dimensional coordinate machine) will generate coupling errors when processing the point cloud data in this area. This coupling error is caused by the complex double-notch structure of the intake / exhaust edge and the mutual influence between the individual notches. When inspecting both edges simultaneously, this coupling error is further amplified, with the maximum deviation reaching 0.1mm. This means that in actual inspection, the measurement result may deviate from the true value by 0.1mm. For components such as precision-forged aircraft engine blades, such deviations are unacceptable. Summary of the Invention

[0004] In order to solve the problems of the prior art, the present invention provides a method for obtaining the profile points of a precision forged blade, comprising the following steps:

[0005] Loading forging point cloud data, performing layered filtering on the forging point cloud data to obtain preprocessed point cloud data;

[0006] Acquire structural features and corresponding coordinate data of the forging according to the preprocessed point cloud data, and acquire a leading edge curvature set of the forging under the coordinate data corresponding to each structural feature according to the structural features of the forging;

[0007] The leading edge curvature set is input into a weight determination model, wherein the weight determination model divides the forging into a plurality of regions based on the coordinate data set and the leading edge curvature set, and sets an identification threshold and a verification mechanism for the intake edge and the exhaust edge in each region;

[0008] Based on the identification threshold and the verification mechanism and according to the leading edge curvature set, each region is adaptively segmented and points are taken to correct the deviation between the intake side and the exhaust side during the fitting process.

[0009] Furthermore, the layered filtering process includes:

[0010] Noise separation is achieved by setting curvature constraints and bidirectional geometric constraints; a coupled deformation model of material-process-geometric features is established, and the linear and nonlinear components of the deformation tensor are determined based on the coupled deformation model. A physical constraint framework supporting dual registration is constructed, and the deformation tensor component coefficients are introduced under the physical constraint framework to calculate the residual field. Entropy optimization compensation is performed based on the obtained residual field as the direct input of entropy optimization, and based on the compensation data obtained from the entropy optimization, the component coefficients of the deformation tensor are updated through inversion calculation to achieve adaptive correction of the coupled deformation model.

[0011] Furthermore, the curvature constraint condition includes: establishing dynamic structure points by morphological filtering to eliminate isolated noise points smaller than the structure points;

[0012] The bidirectional geometric constraint includes: using the DBSCAN clustering algorithm to positively eliminate small clusters and reversely retaining feature-related points to obtain constraint results, using the obtained constraint results as input for feature-sensitive area judgment, and reversely guiding the size selection of structural points of morphological filtering based on the real-time recognition results of the feature-sensitive areas.

[0013] Furthermore, the coupled deformation model is established according to the following method:

[0014] According to the constitutive relationship between the precision forging process parameters and the material, a four-dimensional deformation tensor model including strain components and displacement components is constructed. The four-dimensional deformation tensor model is expressed as:

[0015] ;

[0016] The tensile or compressive strain of the material in the x-axis direction is recorded as , the tensile or compressive strain of the material in the y-axis direction is recorded as , the linear strain component of the material in the thickness direction of the z axis is recorded as , the shear slip deformation of the material in the xy plane is recorded as , the shear strain component of the material in the xy plane is recorded as , satisfying the symmetry condition , the shear slip deformation of the material in the xz plane is recorded as , the shear strain component of the material in the xz plane is recorded as , satisfying the symmetry condition , the shear slip deformation of the material in the yz plane is recorded as , the shear strain component of the material in the yz plane is recorded as , satisfying the symmetry condition , the rigid translation of the material in the x-axis direction is recorded as , the rigid translation of the material in the y-axis direction is recorded as , the rigid translation of the material in the z-axis direction is recorded as ;

[0017] The linear component of the deformation tensor is the initial alignment of the measured point cloud and the CAD model using the improved ICP algorithm, and the corresponding point weights are dynamically calculated based on the deformation tensor, where the tensile or compressive strain of the material in the x-axis direction is used. , tensile or compressive strain in the y-axis direction , linear strain component in the thickness direction of the z axis , shear slip deformation of the material in the xy plane , shear slip deformation of the material in the xz plane , shear slip deformation of the material in the yz plane As the initial transformation matrix of the improved ICP algorithm;

[0018] The nonlinear component of the deformation tensor is a nonrigid deformation field established based on the thin plate spline function. The nonlinear displacement component in the deformation tensor is 、 and Initial displacement constraints as control points of the thin plate spline function;

[0019] The component coefficient of the deformation tensor is an empirical coefficient of the material constitutive relationship, and the empirical value of the empirical coefficient is 0.05-0.1.

[0020] Furthermore, interpolation calculation is performed based on the point cloud measured by the laser scanner and the reference data of the CAD model obtained by the three-coordinate measuring machine to obtain the residual field calculation result.

[0021] Furthermore, a simulated annealing algorithm is used to search for a minimum entropy compensation solution from the residual field calculation result as a direct input of entropy optimization to perform entropy optimization compensation, and compensation data is output through the entropy optimization process.

[0022] Furthermore, the local curvature calculated by the DBSCAN clustering algorithm Mapping to the feature sensitive area protection module;

[0023] Among them, >0.1mm -1 The protection band mechanism is automatically activated in the curvature area without compensation. -1 > >0.05mm -1 The sudden curvature region triggers the curvature compensation filter.

[0024] Furthermore, the size selection of the structure points includes: using a structure element with a 0.1mm sphere core in the feature protection zone and a structure element with a 0.3mm sphere core in the non-feature area to improve the denoising efficiency.

[0025] The beneficial effects of this application are:

[0026] 1. Utilizing layered filtering and noise separation technology, dual registration is performed using curvature constraints (dynamic point selection using morphological filtering) and bidirectional geometric constraints (DBSCAN clustering algorithm to eliminate noise and retain feature points), combined with a coupled deformation model (four-dimensional deformation tensor) based on material, process, and geometric characteristics. The noise filtering rate is ≥95%, and after residual field optimization and compensation, the coupling error is reduced to ≤0.02mm, significantly improving point cloud data quality.

[0027] 2. Utilizing regional decoupling and weight allocation technology, a dual-domain decoupling strategy (weighting 0.55 for the intake side and 0.45 for the exhaust side) is employed. Independent coordinate systems are created based on a curvature-weighted model to suppress multi-notch coupling interference. The error rate for both intake and exhaust side inspections is ≤1.2%, and the time required for simultaneous inspection is reduced to less than 1.5 times that of single-side inspection.

[0028] 3. Using adaptive segmentation and dynamic interruption mechanism, according to the local curvature ( >0.1mm -1 Trigger protection band, 0.1mm -1 > >0.05mm -1 Enable compensation filtering) and error accumulation threshold (≤0.03mm), dynamically adjust the fitting interval. The curvature gradient continuity meets C 2 Standard, spatial residual ≤ 0.02mm, to avoid fitting distortion caused by sudden changes in curvature.

[0029] 4. Utilizing nonlinear deformation compensation and entropy optimization, a nonrigid deformation field is constructed using thin plate spline (TPS) algorithms. Simulated annealing is then used to search for a minimum entropy compensation solution, optimizing both the linear component (improved ICP algorithm) and the nonlinear component (TPS displacement field) of the deformation tensor. This improves nonlinear displacement compensation accuracy by 40%, and entropy optimization improves the uniformity of residual distribution, with an RMS value of ≤0.02mm. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of the profile comparison provided by the present invention;

[0031] Figure 2 The present invention provides a flow chart of the method. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] The present invention provides a method for taking points for the blade profile of a precision forged blade, comprising the following steps: loading point cloud data of a forging, performing layered filtering processing on the point cloud data of the forging to obtain pre-processed point cloud data; obtaining structural features and corresponding coordinate data of the forging according to the pre-processed point cloud data, and obtaining a leading edge curvature set of the forging under the coordinate data corresponding to each structural feature according to the structural features of the forging; inputting the leading edge curvature set into a weight determination model, the weight determination model divides the forging into several areas based on the coordinate data set and the leading edge curvature set, and sets an identification threshold and a verification mechanism for the intake edge and the exhaust edge under each area; based on the identification threshold and the verification mechanism and according to the leading edge curvature set, adaptively taking points in each area in sections to correct deviations of the intake edge and the exhaust edge that occur during the fitting process.

[0034] In some embodiments, forging point cloud data is obtained using a laser scanner or other device to capture 3D point cloud data of precision-forged blades, serving as raw data for subsequent processing. This data provides discrete coordinate information on the blade surface, laying the foundation for profile measurement. Layered filtering applies multi-level noise suppression to the raw point cloud data, including curvature constraints, geometric constraints, and physical deformation compensation. This filtering removes measurement noise (such as equipment errors and surface burrs) while preserving true geometric features. A noise filtering rate of ≥95% is achieved.

[0035] In some embodiments, structural features can be understood as key blade structures (such as the intake and exhaust edges and double notches). Point cloud analysis is used to extract these key blade structures (such as the intake and exhaust edges and double notches) and their spatial coordinates (x, y, z). The spatial coordinates (x, y, z) are used to determine the positioning detection area, providing a benchmark for curvature calculation and region division.

[0036] In some embodiments, the curvature value of each point on the leading edge (inlet edge / exhaust edge) of the blade is calculated to form a curvature distribution set ( The curvature distribution set reflects the changes in the blade surface shape and is used for area division and point selection strategy adjustment (such as increasing the number of points in high curvature areas).

[0037] In some embodiments, the weight determination model is based on coordinate and curvature data, dividing the blade into multiple detection areas (such as the high curvature area on the intake side and the flat area on the exhaust side), and setting detection weights for each area (such as 0.55 for the intake side and 0.45 for the exhaust side). Its function is to decouple the double-notch coupling interference and independently optimize the detection parameters of each area (such as recognition threshold and verification mechanism).

[0038] In some embodiments, a dual-domain decoupling strategy is used to overcome multi-notch coupling interference, and regional decoupling is achieved by establishing independent coordinate systems on the intake and exhaust sides. The weight of the intake side is set to 0.55, and the weight of the exhaust side is set to 0.45. The weight distribution is based on the importance of the intake side and the exhaust side in the entire blade structure, as well as their own geometric characteristics. Specifically, regional decoupling is achieved by establishing independent coordinate systems for the intake / exhaust sides. When determining the weight distribution of the independent coordinate systems, the weight of the intake side is set to 0.55, and the weight of the exhaust side is set to 0.45, based on the characteristics of the intake / exhaust sides. This weight distribution is based on a comprehensive consideration of multiple factors, such as the importance of the intake / exhaust sides in the entire blade structure and their own geometric characteristics.

[0039] The weights are obtained based on a mathematical model, which is expressed as:

[0040] ;

[0041] ;

[0042] in, Indicates the intake weight, represents the exhaust weight, and Represent the local curvature of the intake and exhaust respectively, and the final weight distribution result is =0.55 ， =0.45.

[0043] Through such a mathematical model, the relationship between the intake / exhaust edges in the independent coordinate system can be described more accurately, thereby effectively achieving regional decoupling and reducing the impact of coupling errors on profile detection.

[0044] In some embodiments, the identification threshold is used as a criterion for setting the curvature or coordinates of the intake / exhaust edge feature points, such as >0.05mm -1 Determined as a gap edge. Cross-region residual comparison (such as the deviation between the intake edge fitting result and the theoretical position of the exhaust edge) verifies detection accuracy and ensures feature point recognition accuracy, avoiding deviations in contour calculation caused by misjudgment.

[0045] In some embodiments, adaptive segmented point selection is used. Specifically, the point density and fitting interval are dynamically adjusted according to the regional curvature characteristics (such as dense point selection in high curvature areas and sparse point selection in flat areas), the fitting deviation is corrected, and the fitting accuracy of complex structures (double gaps) is improved. The spatial residual is ≤ 0.02 mm, meeting the C 2 Curvature continuity requirement.

[0046] In some embodiments, the layered filtering processing includes: achieving noise separation by setting curvature constraints and bidirectional geometric constraints; establishing a coupled deformation model of material-process-geometric features, determining the linear component and nonlinear component of the deformation tensor based on the coupled deformation model, and constructing a physical constraint framework that supports dual registration. Under the physical constraint framework, the deformation tensor component coefficients are introduced to perform residual field calculation, and entropy optimization compensation is performed based on the obtained residual field as the direct input of entropy optimization. Based on the compensation data obtained by entropy optimization, the component coefficients of the deformation tensor are updated through inversion calculation to achieve adaptive correction of the coupled deformation model.

[0047] In some embodiments, the curvature constraint can be understood as dynamically adjusting the size of the structure points through morphological filtering algorithms (such as opening and closing operations) to filter out isolated noise points smaller than the structural elements. For example, a 0.3mm sphere kernel is used for fast denoising in non-feature areas, and a 0.1mm sphere kernel is used to preserve details in the feature protection band. Specifically, in the feature protection band (high curvature area, >0.1 mm -1 : Use the smallest structural element (0.1 mm sphere kernel) to avoid detail loss. Non-feature areas (flat areas, <0.05 mm -1 ): Use larger structural elements (0.3 mm spherical core) to improve denoising efficiency. Figure 1 Middle, showing high curvature protection band ( >0.1 mm -1 , disable compensation), curvature mutation area (0.1mm -1 > >0.05mm -1 , enable Gaussian filtering) and adaptively take the direction of point density change (from dense to sparse).

[0048] In some embodiments, bidirectional geometric constraints include forward culling, which uses the DBSCAN (Density-Based Spatial Clustering of Application with Noise) clustering algorithm to identify and remove small clusters (noise point clusters). Reverse retention, which retains point clusters associated with leaf features (such as notch edges) as criteria for feature-sensitive regions. Through forward culling and reverse retention, the morphological filter's structural point size is adjusted inversely based on the identified feature regions, achieving adaptive denoising.

[0049] In some embodiments, reverse retention involves calculating the Hausdorff distance between the point clusters obtained by DBSCAN clustering and the gap edge line of the CAD model. Point clusters with a distance < 0.1 mm are retained, and the rest are treated as noise. For example, the theoretical edge line of the intake edge gap is a NURBS (Non Uniform Rational B-spline) curve in CAD. If the maximum distance between a measured point cluster and this curve is < 0.1 mm, it is considered a feature-related point.

[0050] In some embodiments, the dual registration framework uses a linear component based on an improved ICP algorithm, using the strain component as the initial transformation matrix to achieve coarse alignment between the point cloud and the CAD model. It also uses a nonlinear component based on thin plate splines (TPS) to use the displacement component as a control point constraint to construct a non-rigid deformation field and compensate for local deformation.

[0051] In some embodiments, the residual field refers to the position deviation field of each point calculated by interpolation comparison between the laser point cloud and the three-coordinate CAD model. Entropy optimization uses a simulated annealing algorithm to search for the minimum entropy compensation scheme, optimize the deformation tensor coefficients, and invert and correct the coupling model to improve the uniformity of the residual distribution, with an RMS (root mean square) value of ≤0.02mm. In this embodiment, the entropy value reflects the disorder of the residual distribution. The smaller the entropy value, the more concentrated the residual is in the small value interval. The simulated annealing algorithm searches for the minimum entropy value globally, forcing the compensation scheme to prioritize eliminating large deviation points, thereby making the residual distribution more uniform, and the RMS value drops from 0.05mm to below 0.02mm.

[0052] In some embodiments, the curvature constraint conditions include establishing dynamic structure points using morphological filtering to eliminate isolated noise points whose size is smaller than the structure points; the bidirectional geometric constraint includes: using the DBSCAN clustering algorithm to positively eliminate small clusters and reversely retain feature-related points to obtain constraint results, using the obtained constraint results as input for feature-sensitive area judgment, and reversely guiding the size selection of the structure points of the morphological filter based on the real-time recognition results of the feature-sensitive areas.

[0053] In some embodiments, the dynamic point selection of morphological filtering is to dynamically switch the size of the structural element according to the curvature change of the feature sensitive area (such as the edge of the gap), such as within the feature protection band ( >0.1mm -1 ): Use a 0.1mm spherical kernel to avoid filtering out key details. Non-feature areas: Use a 0.3mm spherical kernel to effectively remove large areas of noise.

[0054] In some embodiments, the DBSCAN clustering algorithm uses forward elimination and reverse retention techniques. Forward elimination involves setting a clustering threshold (e.g., a neighborhood radius of 0.2 mm, a minimum number of points of 5) to eliminate small clusters with fewer points than the threshold (considered noise). Reverse retention involves retaining large clusters associated with features such as leaf edges and notches as initial criteria for determining feature-sensitive areas.

[0055] In some embodiments, the feature sensitive area criterion can be obtained as follows: input: DBSCAN clustering result (feature point distribution) and curvature calculation value ( Output: Real-time marking of high curvature areas ( >0.1mm -1 ) and curvature mutation area (0.1mm -1 > >0.05mm -1 ), which guides the adjustment of filtering parameters.

[0056] Furthermore, the coupled deformation model is established according to the following method: based on the precision forging process parameters and the material constitutive relationship, a four-dimensional deformation tensor model including strain components and displacement components is constructed. The four-dimensional deformation tensor model is expressed as:

[0057] ;

[0058] The tensile or compressive strain of the material in the x-axis direction is recorded as , the tensile or compressive strain of the material in the y-axis direction is recorded as , the linear strain component of the material in the thickness direction of the z axis is recorded as , the shear slip deformation of the material in the xy plane is recorded as , the shear strain component of the material in the xy plane is recorded as , satisfying the symmetry condition , the shear slip deformation of the material in the xz plane is recorded as , the shear strain component of the material in the xz plane is recorded as , satisfying the symmetry condition , the shear slip deformation of the material in the yz plane is recorded as , the shear strain component of the material in the yz plane is recorded as , satisfying the symmetry condition , the rigid translation of the material in the x-axis direction is recorded as , the rigid translation of the material in the y-axis direction is recorded as , the rigid translation of the material in the z-axis direction is recorded as ;

[0059] The linear component of the deformation tensor is the initial alignment of the measured point cloud and the CAD model using the improved ICP algorithm, and the corresponding point weights are dynamically calculated based on the deformation tensor, where the tensile or compressive strain of the material in the x-axis direction is used. , tensile or compressive strain in the y-axis direction , linear strain component in the thickness direction of the z axis , shear slip deformation of the material in the xy plane , shear slip deformation of the material in the xz plane , shear slip deformation of the material in the yz plane As the initial transformation matrix of the improved ICP algorithm;

[0060] The nonlinear component of the deformation tensor is a non-rigid deformation field established based on the thin plate spline function TPS. The nonlinear displacement component in the deformation tensor is 、 and As the initial displacement constraint of the TPS control point;

[0061] The component coefficients of the deformation tensor are empirical coefficients of the material constitutive relation, with empirical values ​​ranging from 0.05 to 0.1. In some embodiments, the empirical value range is 0.05-0.1, with the specific value depending on the material: for titanium alloy blades, the coefficient is 0.08 (plastic deformation dominates, requiring enhanced nonlinear compensation); for nickel-based high-temperature alloy blades, the coefficient is 0.06 (elastic deformation dominates, with a higher proportion of linear components).

[0062] It should be noted that 、 and It reflects local non-uniform deformation (such as slight warping at the edge of the notch) in addition to rigid translation. / / Reflects the tensile / compressive deformation in the x / y / z axis direction (such as thickness reduction during precision forging) ). / / This reflects in-plane shear slip deformation (e.g., plane distortion caused by material flow). It should also be noted that the integration of process parameters (e.g., forging pressure) and material constitutive properties (e.g., elastic modulus) quantifies the deformation difference between the blade from the CAD model to the actual forging, providing physical constraints for point cloud registration.

[0063] In some embodiments, interpolation calculations are performed based on the laser scanner's measured point cloud and the CAD model's reference data obtained by a coordinate measuring machine (CMM) to generate the residual field calculation results. The laser point cloud represents the measured coordinates of discrete points on the blade surface, which include machining errors and noise. The CAD reference data represents the theoretical model coordinates obtained by the CMM, serving as a reference for the ideal contour.

[0064] In some embodiments, interpolation calculations employ kriging or polynomial interpolation to interpolate the deviations between corresponding points in the measured point cloud and the CAD model into a continuous residual field, which intuitively reflects the global deformation distribution. This provides a quantitative basis for subsequent compensation, such as identifying areas of maximum deviation (up to 0.1 mm, requiring focused compensation).

[0065] In some embodiments, a simulated annealing algorithm is used to search for a minimum entropy compensation solution from the residual field calculation results, which serves as a direct input for entropy optimization and compensation. Compensation data is then generated through entropy optimization. The simulated annealing algorithm searches for a compensation solution in the residual field that minimizes the system entropy (i.e., the solution with the most uniform residual distribution). By simulating a physical annealing process, the temperature is gradually lowered (reducing search randomness), avoiding local optima and improving global compensation accuracy. Entropy optimization input: residual field calculation results (deviations at each point); entropy optimization output: compensation data (such as corrections to deformation tensor coefficients), which is used to update the coupled model, achieving a closed-loop optimization of "calculation-compensation-recalculation."

[0066] In some embodiments, the local curvature calculated by the DBSCAN clustering algorithm Mapped to the feature sensitive area protection module; >0.1mm -1 The protection band mechanism is automatically activated in the high curvature area without compensation. -1 > >0.05mm -1 The mutation area triggers the curvature compensation filter.

[0067] In some embodiments, the high curvature region ( >0.1mm -1): For example, at the tip of a notch, the guard band mechanism is enabled and compensation is disabled to avoid damaging the true geometric features. Mutation area (0.1mm -1 > >0.05mm -1 ): For example, in the gap transition area, curvature compensation filtering is triggered to smooth local noise but retain shape trends.

[0068] In some embodiments, according to the curvature With the coordinates (x, y, z), the blade is divided into 4 types of areas: the high curvature area on the inlet side ( >0.1mm -1 , weight 0.6), the intake side flat area ( <0.05mm -1 , weight 0.4), high curvature area of ​​exhaust edge ( >0.1mm -1 , weight 0.5) and exhaust edge flat area ( <0.05mm -1 , weight 0.3). Recognition threshold setting: Inlet edge feature points: >0.07mm -1 And the y coordinate > 0 (assuming the y axis is the blade span direction); exhaust edge feature point: >0.06mm -1 And the y coordinate is <0.

[0069] In some embodiments, segmentation strategy: high curvature region ( >0.1mm -1 ): Take 1 point every 0.2mm, the fitting interval length is ≤1mm, and the protection band mechanism is enabled (deformation compensation is prohibited); mutation area (0.05mm -1 < <0.1mm -1 ): Take 1 point every 0.5mm, the fitting interval length is ≤3mm, and apply curvature compensation filter (Gaussian filter) =0.1mm); flat area ( <0.05mm -1 ): Take 1 point every 1mm, the fitting interval length is ≤5mm, and linear interpolation is used. When fitting in segments, if the difference between the residuals of two adjacent segments is ≥0.03mm, a transition point is inserted at the junction to ensure that the curvature continuity meets C 2 Standard, C 2 The canonical is a fundamental concept in geometry that describes the continuity and smoothness of a curve or surface. Specifically, C 2 The criterion means that a geometric object has at least second-order continuous derivatives in its domain.

[0070] In some embodiments, the curvature threshold is based on: >0.1mm -1 : Corresponding to sharp edge structures such as notch tips, with a curvature radius of <10mm, belonging to key points of geometric features, compensation is prohibited to avoid distortion; 0.05mm -1 < <0.1mm -1 : Corresponding to the notch transition area, the curvature radius is 10-20mm, allowing Gaussian filtering to smooth noise while retaining the overall shape trend. The width of the protection band is automatically set to 1 / 10 of the curvature radius, such as =0.1mm -1 When , the width of the protection band = 10mm × 1 / 10 = 1mm.

[0071] In some embodiments, the guard band mechanism sets a "no modification" mark in the algorithm, and does not perform any deformation compensation on the point cloud coordinates in the high curvature area to ensure the accuracy of the detection benchmark.

[0072] In some embodiments, the size selection of the structure point includes: using the minimum structure element 0.1mm sphere core within the feature protection band, and using the 0.3mm sphere core in the non-feature area to improve the denoising efficiency. Among them, the feature protection band is a local area around the high curvature feature (such as the edge of the notch), and the width is determined by the curvature gradient (usually 0.1-0.3mm). When taking points, the minimum structure element (0.1mm sphere core) is used to ensure that the point cloud density of the feature edge is high enough to avoid loss of details. The non-feature area is the flat area on the blade surface (such as the middle of the blade body), and the curvature is <0.05mm -1 When selecting points, a 0.3mm sphere kernel is used to reduce redundant point clouds and improve denoising efficiency and calculation speed.

[0073] In some embodiments, the ICP (Iterative Closest Point) algorithm (linear component) is improved: the strain component is directly used ( , , , The dynamic weighting algorithm uses point cloud matching weights in high-strain regions (such as those near gaps) to prevent noise in low-strain regions from dominating the registration results.

[0074] In some embodiments, the TPS thin plate spline function (nonlinear component): , , As the initial displacement of the TPS control points, the model is forced to learn local deformations (such as blade twist). Through TPS interpolation, a continuously deforming surface is generated to compensate for complex deformations that cannot be described by linear models, such as edge wrinkles caused by material flow.

[0075] In some embodiments, the residual field can be defined as the coordinate deviation between the measured point cloud and the CAD model, reflecting the global deformation distribution, with a maximum deviation of up to 0.1 mm.

[0076] In some embodiments, the simulated annealing process includes: randomly generating deformation tensor coefficients, and using their modified values ​​as initial solutions; calculating the entropy value of the current solution, and accepting it if it is better than the historical optimal solution; lowering the search temperature according to an exponential cooling strategy (which can be set according to actual conditions) to reduce randomness; iterating until the entropy value converges, and outputting the optimal compensation solution (such as Correction factor 0.08).

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The method for selecting the point of blade profile of precision forging blade is characterized by: The steps include: Loading forging point cloud data, performing layered filtering on the forging point cloud data to obtain preprocessed point cloud data; Acquire structural features and corresponding coordinate data of the forging according to the preprocessed point cloud data, and acquire a leading edge curvature set of the forging under the coordinate data corresponding to each structural feature according to the structural features of the forging; The leading edge curvature set is input into a weight determination model, wherein the weight determination model divides the forging into a plurality of regions based on the coordinate data set and the leading edge curvature set, and sets an identification threshold and a verification mechanism for the intake edge and the exhaust edge in each region; Adaptively segmenting each region into points based on the identification threshold and verification mechanism and the leading edge curvature set to correct deviations between the intake and exhaust edges during the fitting process; The layered filtering process includes: Noise separation is achieved by setting curvature constraints and bidirectional geometric constraints; a coupled deformation model of material-process-geometric features is established, and the linear and nonlinear components of the deformation tensor are determined based on the coupled deformation model. A physical constraint framework supporting dual registration is constructed, and the deformation tensor component coefficients are introduced under the physical constraint framework to calculate the residual field. Entropy optimization compensation is performed based on the obtained residual field as the direct input of entropy optimization, and based on the compensation data obtained from the entropy optimization, the component coefficients of the deformation tensor are updated through inversion calculation to achieve adaptive correction of the coupled deformation model; The curvature constraint condition includes: establishing dynamic structure points by morphological filtering to eliminate isolated noise points with a size smaller than the structure points; The bidirectional geometric constraint includes: using the DBSCAN clustering algorithm to positively eliminate small clusters and reversely retaining feature-related points to obtain constraint results, using the obtained constraint results as input for feature-sensitive area judgment, and reversely guiding the size selection of structural points of morphological filtering based on the real-time recognition results of the feature-sensitive areas; The coupled deformation model is established according to the following method: According to the constitutive relationship between the precision forging process parameters and the material, a four-dimensional deformation tensor model including strain components and displacement components is constructed. The four-dimensional deformation tensor model is expressed as: The tensile or compressive strain of the material in the x-axis direction is recorded as ε x , the tensile or compressive strain of the material in the y-axis direction is recorded as ε y , the linear strain component of the material in the z-axis thickness direction is recorded as ε z , the shear slip deformation of the material in the xy plane is recorded as γ xy , the shear strain component of the material in the xy plane is recorded as γ yx , satisfying the symmetry condition γ xy =γ yx , the shear slip deformation of the material in the xz plane is recorded as γ xz , the shear strain component of the material in the xz plane is recorded as γ zx , satisfying the symmetry condition γ xz =γ zx , the shear slip deformation of the material in the yz plane is recorded as γ yz , the shear strain component of the material in the yz plane is recorded as γ zy , satisfying the symmetry condition γ yz =γ zy , the rigid translation of the material in the x-axis direction is recorded as δ x , the rigid translation of the material in the y-axis direction is recorded as δ y , the rigid translation of the material in the z-axis direction is recorded as δ z .

2. The method for obtaining the point of blade profile of a precision forged blade according to claim 1, wherein: The linear component of the deformation tensor is the initial alignment of the measured point cloud and the CAD model using the improved ICP algorithm, and the corresponding point weights are dynamically calculated based on the deformation tensor, where the tensile or compressive strain ε of the material in the x-axis direction is used. x , tensile or compressive strain ε in the y-axis direction y , linear strain component ε in the thickness direction of the z axis z , shear slip deformation of the material in the xy plane γ xy , shear slip deformation of the material in the xz plane γ xz , shear slip deformation of the material in the yz plane γ yz As the initial transformation matrix of the improved ICP algorithm; The nonlinear component of the deformation tensor is a nonrigid deformation field established based on the thin plate spline function. The nonlinear displacement component δ in the deformation tensor is x , δ y and δ z Initial displacement constraints as control points of the thin plate spline function; The component coefficient of the deformation tensor is an empirical coefficient of the material constitutive relationship, and the empirical value of the empirical coefficient is 0.05-0.

1.

3. The method for obtaining the point of blade profile of a precision forged blade according to claim 1, wherein: Interpolation calculation is performed based on the point cloud measured by the laser scanner and the benchmark data of the CAD model obtained by the three-coordinate measuring machine to obtain the residual field calculation result.

4. The method for obtaining the point of blade profile of a precision forged blade according to claim 3, wherein: A simulated annealing algorithm is used to search for a minimum entropy compensation solution from the residual field calculation result as a direct input of entropy optimization to perform entropy optimization compensation, and compensation data is output through the entropy optimization process.

5. The method for obtaining the point of blade profile of a precision forged blade according to claim 3, wherein: The local curvature κ calculated by the DBSCAN clustering algorithm is mapped to the feature sensitive area protection module; Among them, for κ>0.1mm -1 The protection band mechanism is automatically activated in the curvature area without compensation. -1 >κ>0.05mm -1 The sudden curvature region triggers the curvature compensation filter.

6. The method for obtaining the point of blade profile of a precision forged blade according to claim 1, wherein: The size selection of the structure points includes: using a structure element with a 0.1 mm sphere core in the feature protection zone and a structure element with a 0.3 mm sphere core in the non-feature area to improve the denoising efficiency.

Images