A method for detecting the profile of double-notch blades on the intake and exhaust edges of precision forged blades

Abstract

The present invention discloses a method for detecting the profile of a double-notch blade on the inlet and exhaust edges of a precision-forged blade, comprising the following steps: a) dividing the inlet / exhaust edge regions into independent coordinate systems based on a dual-domain decoupling strategy to suppress multi-notch coupling interference; b) performing multi-level noise suppression processing on noisy cloud data, including dynamic clustering analysis and deformation compensation mechanisms; c) employing a bidirectional collaborative iterative optimization algorithm to synchronously perform forward fitting of the inlet edge and reverse fitting of the exhaust edge, and achieving convergence control through cross-domain residual verification; d) dynamically adjusting the data fitting interval based on an adaptive segmentation strategy for feature-sensitive areas; and e) constructing an incremental data management module to achieve versioned storage and compression optimization of the iterative process. The dual-domain decoupling strategy is employed to overcome multi-notch coupling interference; the present invention achieves synchronous detection of double notches, significantly shortening detection time compared to traditional methods.

Description

Technical Field

[0001] The present invention relates to the technical field of blade profile detection, and in particular to a method for detecting the profile of a double-notch blade profile at the intake and exhaust edges of a precision forged blade. Background Art

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

[0003] First, there is the coupling interference of multiple notches. When inspecting the profile of the double-notch region on the intake / exhaust edge, existing methods (such as discrete sampling on a three-dimensional coordinate machine) 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 results may deviate from the true value by 0.1mm, which is unacceptable for components such as precision-forged aircraft engine blades, which require extremely high precision. Second, there is a lack of data correlation. Current commercial software (such as PolyWorks 2023) lacks topological correlation analysis when processing data in the double-notch region. Topological correlation analysis is crucial for understanding the relationship between the intake / exhaust edge data. This lack of analysis results in high error rates when matching the intake / exhaust edge data, exceeding 8%. This means that the intake / exhaust edge data cannot be accurately aligned and integrated during data processing, affecting the accuracy of the overall profile inspection. Third, the algorithm lacks adaptability. The existing RANSAC algorithm suffers from excessively large fitting residuals when processing regions with sudden changes in double curvature (when the curvature difference is greater than 0.1 mm⁻¹). The fitting residual, a key metric for evaluating the algorithm's performance, exceeded 0.1 mm in this case, failing to meet the requirements of the ASME B89.3.4-2020 standard. This indicates that the existing algorithm cannot effectively fit the data for the unique curvature variations found in the double-notched areas of the inlet and exhaust edges of precision-forged blades, thus affecting the accuracy of profile detection. Summary of the Invention

[0004] In order to solve the problems of the prior art, the present invention provides a method for detecting the profile of a double-notched blade profile on the intake and exhaust edges of a precision forged blade, comprising the following steps:

[0005] a) Based on the dual-domain decoupling strategy, the intake and exhaust edge regions are divided into independent coordinate systems to suppress multi-notch coupling interference;

[0006] b) Implement multi-level noise suppression processing on noisy point cloud data, including dynamic clustering analysis and deformation compensation mechanism;

[0007] c) A bidirectional collaborative iterative optimization algorithm is used to simultaneously perform forward fitting on the intake side and reverse fitting on the exhaust side, and convergence control is achieved through cross-domain residual verification;

[0008] d) Adaptive segmentation strategy based on feature-sensitive areas to dynamically adjust the data fitting interval;

[0009] e) Build an incremental data management module to achieve versioned storage and compression optimization of the iterative process.

[0010] Furthermore, the dual-domain decoupling strategy includes: establishing an independent coordinate system based on the geometric characteristics and functional weights of the intake / exhaust edges; dynamically allocating the intake / exhaust edge area weights through a curvature weighted model, where the intake edge weight range is 0.5-0.6 and the exhaust edge weight range is 0.4-0.5.

[0011] Furthermore, the multi-level noise suppression process specifically includes:

[0012] First layer: A clustering algorithm that dynamically adjusts density based on material properties to suppress hardened layer noise;

[0013] The second layer: Combining the compensation function of temperature and deformation parameters to correct the plastic deformation error;

[0014] The neighborhood radius parameter of the clustering algorithm is dynamically calculated based on the hardness and surface roughness of the forging, and the expression is: Eps=k1(1+k2), where k1 is the basic neighborhood radius and k2 is the material property correction coefficient;

[0015] The noise suppression process further includes: a noise classification module that identifies the noise type based on surface roughness and material hardness; uses Gaussian filtering for high-frequency noise and median filtering for low-frequency noise;

[0016] The input parameters of the noise classification module include: forging surface hardness value range 30-60HRC; roughness value range Ra=0.8-2.5μm.

[0017] Furthermore, the bidirectional collaborative iterative optimization algorithm meets the following convergence conditions: spatial residual threshold ≤ 0.03 mm; curvature gradient continuity meets the C² continuity standard;

[0018] The triggering conditions of the cross-domain residual check include: performing a residual comparison every N iterations, where N is a preset positive integer; and the check threshold is dynamically adjusted based on the intake / exhaust edge weight, with a range of ≤0.01 mm.

[0019] Furthermore, the adaptive segmentation strategy includes: a dynamic interruption mechanism based on curvature mutation, with the curvature change threshold set to 0.05-0.15mm -1 ; The cumulative error threshold triggers segmented reconstruction, and the error range is ≤0.03mm;

[0020] The dynamic interruption mechanism includes: using morphological filtering to eliminate isolated noise points, the filter kernel size is dynamically selected according to the feature sensitive area; and retaining high curvature feature points through bidirectional geometric constraints.

[0021] Furthermore, the incremental data management module includes: based on the differential coding model, only recording the data differences between iterations; compression rate evaluation model, real-time calculation of storage optimization efficiency;

[0022] The differential coding model is implemented by: performing lossless compression storage on the initial data set; in subsequent iterations, only the difference matrix between the current data set and the previous version is stored;

[0023] The calculation formula of the compression rate evaluation model is: compression rate = 1-(incremental data size / original data size), and the compression rate threshold is set to ≥50%.

[0024] Furthermore, the method further includes: achieving initial alignment of the measured point cloud with the theoretical model based on an improved ICP algorithm, and allocating feature weights in the order of leading edge > exhaust edge > blade back / basin;

[0025] The improved ICP algorithm is optimized by using KD-Tree to accelerate the nearest neighbor search with a node size of 0.05-0.15 mm.

[0026] Furthermore, a non-rigid deformation field is constructed based on thin plate spline function to compensate for the nonlinear displacement during forging processing;

[0027] The construction of the non-rigid deformation field includes: defining a four-dimensional deformation tensor based on the material constitutive relationship; using the rigid translation and the nonlinear displacement components as initial constraints of the thin plate spline function;

[0028] The deformation tensor model includes:

[0029] Linear component: strain tensor (ε x ,ε y ,γ xy ) and the rigid translation (δ x ,δ y ); Nonlinear component: displacement field interpolation based on thin plate spline function;

[0030] The tensile or compressive strain of the material in the x-axis direction is recorded as ε x , the tensile or compressive strain in the y-axis direction is recorded as ε yThe shear slip deformation of the material in the xy plane is recorded as γxy, the rigid translation of the material in the x-axis direction is recorded as δx, and the rigid translation of the material in the y-axis direction is recorded as δy.

[0031] Furthermore, the curvature calculation in the method adopts: local curvature is mapped to the feature protection module, and for high curvature areas (κ>0.1mm -1 ) Disable compensation; medium and high curvature areas (0.05mm -1 <κ<0.1mm -1 ) enables curvature adaptive filtering.

[0032] Furthermore, based on the adaptive update mechanism of the damping factor of the LM algorithm, the update formula is:

[0033] ;

[0034] in, is the regulating factor, For the Iteration residual;

[0035] The LM algorithm is optimized as follows: parameter update under the constraint of Jacobian matrix condition number; historical Jacobian matrix reuse mechanism to reduce repeated calculations.

[0036] Furthermore, the weight distribution model satisfies the following conditions: the intake edge weight is positively correlated with the local curvature; the exhaust edge weight is negatively correlated with the theoretical model deviation.

[0037] Furthermore, the minimum entropy compensation scheme is searched based on the simulated annealing algorithm to optimize the residual distribution of the deformation field;

[0038] The search process of the minimum entropy compensation scheme includes: taking the entropy value of the residual field as the objective function; and controlling the convergence speed of the simulated annealing by the temperature attenuation coefficient.

[0039] Furthermore, data preprocessing includes: implementing radius compensation on the laser scanning point cloud to eliminate the inherent measurement error of the equipment; converting the theoretical CAD model into a NURBS subdivision surface with a subdivision level ≥ Lv4.

[0040] Furthermore, the detection performance indicators include: the time consumption of dual-notch simultaneous detection is ≤1.5 times that of single-side detection; the intake / exhaust edge matching error rate is ≤1.2%; the residual RMS value is ≤0.02mm.

[0041] Furthermore, a real-time data quality monitoring module detects point cloud density anomalies and triggers data re-collection;

[0042] The triggering conditions of the data quality monitoring module include: the local point cloud density is lower than the preset threshold (≤150 points / mm²); the coverage rate of the curvature mutation area is <90%.

[0043] Furthermore, the dynamic parameter adjustment includes: the minimum number of neighborhood points (MinPts) of the clustering algorithm is dynamically set based on data density; the number of fitting iterations is adaptively allocated according to curvature complexity;

[0044] The dynamic setting formula of the minimum number of neighborhood points is: .

[0045] Furthermore, the multi-sensor data fusion module integrates laser scanning and three-dimensional coordinate measurement data; it eliminates the systematic errors between sensors through weighted averaging;

[0046] The weight distribution of the sensor data fusion is based on: the confidence of the laser scanning data (positively correlated with the surface reflectivity); the spatial resolution of the three-dimensional coordinate measurement data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of the profile comparison provided by the present invention;

[0048] Figure 2 A method flow chart is provided for the present invention. DETAILED DESCRIPTION

[0049] 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.

[0050] The present invention provides a method for detecting the profile of a double-notched blade profile on the intake and exhaust edges of a precision forged blade, comprising the following steps:

[0051] a) Based on the dual-domain decoupling strategy, the intake and exhaust edge regions are divided into independent coordinate systems to suppress multi-notch coupling interference;

[0052] b) Implement multi-level noise suppression processing on noisy point cloud data, including dynamic clustering analysis and deformation compensation mechanism;

[0053] c) A bidirectional collaborative iterative optimization algorithm is used to simultaneously perform forward fitting on the intake side and reverse fitting on the exhaust side, and convergence control is achieved through cross-domain residual verification;

[0054] d) Adaptive segmentation strategy based on feature-sensitive areas to dynamically adjust the data fitting interval;

[0055] e) Build an incremental data management module to achieve versioned storage and compression optimization of the iterative process.

[0056] Furthermore, the dual-domain decoupling strategy includes: establishing an independent coordinate system based on the geometric characteristics and functional weights of the intake / exhaust edges; dynamically allocating the intake / exhaust edge area weights through a curvature weighted model, where the intake edge weight range is 0.5-0.6 and the exhaust edge weight range is 0.4-0.5.

[0057] Furthermore, the multi-level noise suppression process specifically includes:

[0058] First layer: A clustering algorithm that dynamically adjusts density based on material properties to suppress hardened layer noise;

[0059] The second layer: Combining the compensation function of temperature and deformation parameters to correct the plastic deformation error;

[0060] The neighborhood radius parameter of the clustering algorithm is dynamically calculated based on the hardness and surface roughness of the forging, and the expression is: Eps=k1(1+k2), where k1 is the basic neighborhood radius and k2 is the material property correction coefficient;

[0061] The noise suppression process further includes: a noise classification module that identifies the noise type based on surface roughness and material hardness; uses Gaussian filtering for high-frequency noise and median filtering for low-frequency noise;

[0062] The input parameters of the noise classification module include: forging surface hardness value range 30-60HRC; roughness value range Ra=0.8-2.5μm.

[0063] Furthermore, the bidirectional collaborative iterative optimization algorithm meets the following convergence conditions: spatial residual threshold ≤ 0.03 mm; curvature gradient continuity meets the C² continuity standard;

[0064] The triggering conditions of the cross-domain residual check include: performing a residual comparison every N iterations, where N is a preset positive integer; and the check threshold is dynamically adjusted based on the intake / exhaust edge weight, with a range of ≤0.01 mm.

[0065] Furthermore, the adaptive segmentation strategy includes: a dynamic interruption mechanism based on curvature mutation, with the curvature change threshold set to 0.05-0.15mm -1 ; The cumulative error threshold triggers segmented reconstruction, and the error range is ≤0.03mm;

[0066] The dynamic interruption mechanism includes: using morphological filtering to eliminate isolated noise points, the filter kernel size is dynamically selected according to the feature sensitive area; and retaining high curvature feature points through bidirectional geometric constraints.

[0067] Furthermore, the incremental data management module includes: based on the differential coding model, only recording the data differences between iterations; compression rate evaluation model, real-time calculation of storage optimization efficiency;

[0068] The differential coding model is implemented by: performing lossless compression storage on the initial data set; in subsequent iterations, only the difference matrix between the current data set and the previous version is stored;

[0069] The calculation formula of the compression rate evaluation model is: compression rate = 1-(incremental data size / original data size), and the compression rate threshold is set to ≥50%.

[0070] Furthermore, the method further includes: achieving initial alignment of the measured point cloud with the theoretical model based on an improved ICP algorithm, and allocating feature weights in the order of leading edge > exhaust edge > blade back / basin;

[0071] The improved ICP algorithm is optimized by using KD-Tree to accelerate the nearest neighbor search with a node size of 0.05-0.15 mm.

[0072] Furthermore, a non-rigid deformation field is constructed based on thin plate spline function to compensate for the nonlinear displacement during forging processing;

[0073] The construction of the non-rigid deformation field includes: defining a four-dimensional deformation tensor based on the material constitutive relationship; using the rigid translation and the nonlinear displacement components as initial constraints of the thin plate spline function;

[0074] The deformation tensor model includes: linear components: strain tensors (εx, εy, γxy) and rigid translations (δx, δy); nonlinear components: displacement field interpolation based on thin plate spline functions;

[0075] The tensile or compressive strain of the material in the x-axis direction is recorded as ε x , the tensile or compressive strain in the y-axis direction is recorded as ε y The shear slip deformation of the material in the xy plane is recorded as γxy, the rigid translation of the material in the x-axis direction is recorded as δx, and the rigid translation of the material in the y-axis direction is recorded as δy.

[0076] Furthermore, the curvature calculation in the method adopts: local curvature is mapped to the feature protection module, and for high curvature areas (κ>0.1mm -1 ) Disable compensation; medium and high curvature areas (0.05mm -1 <κ<0.1mm -1 ) enables curvature adaptive filtering.

[0077] Furthermore, based on the adaptive update mechanism of the damping factor of the LM algorithm, the update formula is:

[0078] ;

[0079] in, is the regulating factor, For the Iteration residual;

[0080] The LM algorithm is optimized as follows: parameter update under the constraint of Jacobian matrix condition number; historical Jacobian matrix reuse mechanism to reduce repeated calculations.

[0081] Furthermore, the weight distribution model satisfies the following conditions: the intake edge weight is positively correlated with the local curvature; the exhaust edge weight is negatively correlated with the theoretical model deviation.

[0082] Furthermore, the minimum entropy compensation scheme is searched based on the simulated annealing algorithm to optimize the residual distribution of the deformation field;

[0083] The search process of the minimum entropy compensation scheme includes: taking the entropy value of the residual field as the objective function; and controlling the convergence speed of the simulated annealing by the temperature attenuation coefficient.

[0084] Furthermore, data preprocessing includes: implementing radius compensation on the laser scanning point cloud to eliminate the inherent measurement error of the equipment; converting the theoretical CAD model into a NURBS subdivision surface with a subdivision level ≥ Lv4.

[0085] Furthermore, the detection performance indicators include: the time consumption of dual-notch simultaneous detection is ≤1.5 times that of single-side detection; the intake / exhaust edge matching error rate is ≤1.2%; the residual RMS value is ≤0.02mm.

[0086] Furthermore, a real-time data quality monitoring module detects point cloud density anomalies and triggers data re-collection;

[0087] The triggering conditions of the data quality monitoring module include: the local point cloud density is lower than the preset threshold (≤150 points / mm²); the coverage rate of the curvature mutation area is <90%.

[0088] Furthermore, the dynamic parameter adjustment includes: the minimum number of neighborhood points (MinPts) of the clustering algorithm is dynamically set based on data density; the number of fitting iterations is adaptively allocated according to curvature complexity;

[0089] The dynamic setting formula of the minimum number of neighborhood points is: .

[0090] Furthermore, the multi-sensor data fusion module integrates laser scanning and three-dimensional coordinate measurement data; it eliminates the systematic errors between sensors through weighted averaging;

[0091] The weight distribution of the sensor data fusion is based on: the confidence of the laser scanning data (positively correlated with the surface reflectivity); the spatial resolution of the three-dimensional coordinate measurement data.

[0092] In some embodiments, a dual-domain decoupling strategy is adopted to overcome multi-notch coupling interference, including: regional decoupling is achieved by establishing independent coordinate systems on the intake side and the exhaust side, the weight of the intake side is set to 0.55, and the weight of the exhaust side is set to 0.45, and 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. In order to overcome the problem of multi-notch coupling interference, the present invention adopts a dual-domain decoupling strategy. Specifically, regional decoupling is achieved by establishing independent coordinate systems for the intake / exhaust sides. When determining the weight distribution of the independent coordinate system, the weight of the intake side is set to 0.55, and the weight of the exhaust side is set to 0.45 according to the characteristics of the intake / exhaust side. This weight distribution is based on a comprehensive consideration of multiple factors such as the importance of the intake / exhaust side in the entire blade structure and its own geometric characteristics.

[0093] The weights are obtained based on a mathematical model, which is expressed as:

[0094] ;

[0095] ;

[0096] 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.

[0097] 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.

[0098] The initialization of dual-domain decoupling includes: First, it is necessary to load the forging point cloud data. In this embodiment, the data in .asc format is used, and its data density is 200 points / mm². This data format and density are determined according to the actual detection needs and equipment characteristics, and can meet the subsequent processing requirements. Secondly, when performing regional segmentation, it is necessary to set the recognition threshold of the intake side and the exhaust side. For the intake side, the recognition threshold is set to curvature ≥ 0.08mm -1 ; For the exhaust edge, the recognition threshold is set to curvature ≥ 0.07mm -1 These thresholds were determined by analyzing a large number of precision-forged blade samples and considering the geometric characteristics of the intake and exhaust edges. They can accurately distinguish the intake and exhaust edges from the overall data, laying the foundation for the subsequent dual-domain decoupling strategy.

[0099] The improved clustering algorithm used to process noisy cloud data includes: Anti-noise point cloud preprocessing using an improved DBSCAN clustering algorithm, with parameters set to Eps = 0.05 mm and MinPts = 6. To address the noise issue caused by forging surface characteristics, the present invention utilizes the improved DBSCAN clustering algorithm. This algorithm, with specific parameter settings (Eps = 0.05 mm, MinPts = 6), effectively processes noisy point cloud data. This algorithm achieves a noise removal rate of ≥95%, significantly improving the quality of the point cloud data.

[0100] Anti-noise point cloud processing involves the following steps: The input data is raw data containing 15% noise. This high-noise data is common in actual inspection processes and reflects the impact of forging surface characteristics on scanned data. The parameters for the improved DBSCAN clustering algorithm are set to Eps=0.05mm and MinPts=6. These parameters were determined through extensive experimentation and optimization, enabling the algorithm to achieve optimal results when processing this noisy data. After the above processing, the output is valid point cloud data with a noise content of ≤5%. This indicates that after anti-noise point cloud preprocessing, the quality of the point cloud data has been significantly improved, meeting the data quality requirements of subsequent contour detection.

[0101] A bidirectional progressive fitting method is used to process the data of the intake and exhaust sides. The forward iteration of the intake side and the reverse iteration of the exhaust side are performed simultaneously. A cross-domain residual check is performed every five iterations. The convergence conditions of the bidirectional progressive fitting are that the spatial residual satisfies ≤ 0.02 mm and the curvature gradient is continuous.

[0102] The present invention utilizes a bidirectional progressive fitting method to process the intake and exhaust data. Its core principle is to perform iterative fitting on both the intake and exhaust sides simultaneously, in different directions. The intake side uses a forward iteration approach, performing fitting operations sequentially according to the natural order of the data; the exhaust side uses a reverse iteration approach, starting from the end of the data and working backwards. This bidirectional approach fully accounts for the unique characteristics of both sides, such as differences in airflow effects and geometric shapes. Bidirectional fitting can more accurately capture these differences, improving fitting accuracy.

[0103] During the iteration process, a cross-domain residual check is performed every 5 iterations. The residual refers to the difference between the fitting result and the actual data. The cross-domain residual check is to check the difference between the fitting results of the intake side and the exhaust side. By performing cross-domain residual checks regularly, deviations that may occur in the fitting process can be discovered in a timely manner. For example, if after several iterations, there is a large inconsistency between the fitting results of the intake side and the exhaust side, it means that there may be a fitting deviation. At this time, the fitting process can be adjusted and corrected to ensure that the final fitting result is more in line with the actual situation.

[0104] The convergence conditions include two aspects: first, the spatial residual needs to be ≤0.02mm. This standard ensures the accuracy of the fitting results in terms of spatial position; second, the curvature gradient must be continuous (C² continuity). This condition mathematically guarantees the smoothness of the fitting curve and avoids situations such as sudden curvature changes that do not conform to the actual blade shape.

[0105] Convergence conditions include: Spatial residual condition: The spatial residual must be ≤0.02mm. Spatial residual reflects the degree of deviation between the spatial position of the fitting result and the actual data. This criterion is set to ensure high spatial accuracy of the fitting result. For example, in blade design and manufacturing, the spatial position accuracy of the inlet / exhaust edge directly affects the aerodynamic performance of the blade. Excessive spatial residual can cause problems such as airflow turbulence during operation. Curvature gradient continuity condition: Curvature gradient continuity (C² continuity) must be ensured. Curvature describes the curvature of a curve. Curvature gradient continuity means that the curve changes smoothly, without sudden changes in curvature. In actual blade shapes, the curves of the inlet / exhaust edge typically have smooth transitions, and sudden changes in curvature are not realistic. From a mathematical perspective, C² continuity ensures the continuity of the curve's second-order derivative, making the fitting curve mathematically smoother and more consistent with the actual blade shape requirements.

[0106] Bidirectional progressive fitting involves using the LM algorithm for fitting the inlet edge using a forward iteration approach, with μ = 0.005. This parameter value is determined based on the specific data characteristics and fitting requirements of the inlet edge, ensuring that the true contour curve is effectively approximated during the forward iteration process.

[0107] For the exhaust edge, a reverse iteration approach was used, also using the LM algorithm, but with μ = 0.008. This parameter was set to account for differences in geometry and data distribution between the exhaust and intake edges. By adjusting μ to accommodate the characteristics of the exhaust edge, the accuracy of the reverse iteration was ensured.

[0108] During the bidirectional progressive fitting process, a cross-domain residual check is performed every five iterations. The check threshold is set to ΔRMS ≤ 0.005mm. This regular check mechanism promptly detects and corrects any deviations that may occur during the fitting process on the intake and exhaust sides, ensuring that the entire fitting process proceeds accurately.

[0109] The process of processing the data of the intake and exhaust edges by the bidirectional progressive fitting method includes: 1. Data preprocessing: loading the measured point cloud data (.las1.4 format) that has been radius compensated. The point cloud data is a collection of a large number of discrete points obtained by a three-dimensional measuring device (such as a laser scanner). These points record the three-dimensional coordinate information of the object surface. During the measurement process, radius errors may exist due to the characteristics of the measuring equipment and other reasons. Radius compensation is to correct these errors so that the point cloud data more accurately reflects the actual shape of the intake / exhaust edge. .las1.4 is a common point cloud data storage format. It is efficient and flexible and can store a large amount of point cloud data and related attribute information, such as the color and intensity of the points. Loading data in this format is the basis for subsequent processing.

[0110] Import the theoretical CAD model and convert it into a subdivision NURBS surface (subdivision level: Lv5). The theoretical CAD model is a 3D model of the blade created based on design requirements, representing the ideal shape of the blade. This model is then imported into the processing system as a reference for comparison and fitting with the measured point cloud data.

[0111] NURBS (Non-Uniform Rational B-Splines) surfaces are mathematical models that accurately represent complex geometries, such as the inlet and exhaust edges of blades. Subdivision further refines NURBS surfaces, improving their accuracy and detail. Subdivision level Lv5 represents five levels of subdivision. Higher subdivision levels increase surface accuracy and detail, enabling more accurate matching and fitting with measured point cloud data.

[0112] 2. Initial Coordinate System Alignment: Execute the improved ICP algorithm, assigning feature weights of 0.6 for the leading edge, 0.3 for the exhaust edge, and 0.1 for the blade back / basin. The ICP (Iterative Closest Point) algorithm is a commonly used point cloud registration algorithm for aligning point cloud data from different coordinate systems into a common coordinate system. The improved ICP algorithm is an optimization of the traditional ICP algorithm, achieving more efficient and accurate registration.

[0113] Feature weighting is determined based on the importance of the intake / exhaust edges, the leading edge, and the blade back / basin within the blade. The leading edge plays a key role in guiding airflow; its shape and position directly impact the blade's intake efficiency and aerodynamic performance, so it is assigned a higher weight of 0.6. The exhaust edge significantly influences airflow discharge and is weighted 0.3. The blade back / basin, while having a relatively minor direct impact on airflow, is weighted 0.1. This weighting allows for greater emphasis on the alignment of important features during registration, improving registration accuracy.

[0114] Using a KD-Tree to accelerate the closest point search (node ​​size: 0.1mm): In the ICP algorithm, it is necessary to continuously search for the closest point pairs in the point cloud. The KD-Tree (K-Dimensional Tree) is an efficient spatial index structure that can partition and organize points in high-dimensional space, allowing rapid retrieval of the closest point to a given reference point.

[0115] The node size is set to 0.1mm. This parameter affects the KD-Tree's partitioning accuracy. A smaller node size improves search accuracy but increases tree construction and search time; a larger node size has the opposite effect. Choosing a node size of 0.1mm strikes a balance between search accuracy and efficiency. While ensuring search accuracy, it speeds up the nearest point search, thereby improving the efficiency of the entire registration process.

[0116] 3. Iterative optimization process: Dynamic interruption strategy: Adaptive segmentation based on curvature and error.

[0117] Based on the curvature change and error accumulation, the intake and exhaust edge data are adaptively segmented, and the dynamic interruption strategy is used to optimize the iterative process;

[0118] The formula for curvature mutation detection is:

[0119] ;

[0120] Where, Represents the change in curvature between the i-th point and the i+1-th point; represents the curvature of the i-th point; represents a pre-set curvature change threshold;

[0121] The error accumulation judgment formula is:

[0122] ;

[0123] Indicates the coordinates of the jth actual measurement point, which represents the real physical location data in the point cloud data;

[0124] is the fitting function value about j, that is, the theoretical coordinate value at the corresponding position obtained by the curve fitting algorithm; Indicates the preset error threshold.

[0125] 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. A method for detecting the profile of a double-notch blade on the intake and exhaust edges of a precision forged blade, characterized in that: The following steps are involved: a) Based on the dual-domain decoupling strategy, the intake and exhaust edge regions are divided into independent coordinate systems to suppress multi-notch coupling interference; b) Implement multi-level noise suppression processing on noisy point cloud data, including dynamic clustering analysis and deformation compensation mechanism; c) A bidirectional collaborative iterative optimization algorithm is used to simultaneously perform forward fitting on the intake side and reverse fitting on the exhaust side, and convergence control is achieved through cross-domain residual verification; d) Adaptive segmentation strategy based on feature-sensitive areas to dynamically adjust the data fitting interval; e) Build an incremental data management module to achieve versioned storage and compression optimization of the iterative process.

2. The method according to claim 1, characterized in that The dual-domain decoupling strategy includes: establishing an independent coordinate system based on the geometric characteristics and functional weights of the intake / exhaust edges; and dynamically allocating intake / exhaust edge regional weights through a curvature weighted model, where the intake edge weight range is 0.5-0.6 and the exhaust edge weight range is 0.4-0.

5.

3. The method according to claim 1, characterized in that The multi-level noise suppression process specifically includes: First layer: A clustering algorithm that dynamically adjusts density based on material properties to suppress hardened layer noise; The second layer: Combining the compensation function of temperature and deformation parameters to correct the plastic deformation error; The neighborhood radius parameter of the clustering algorithm is dynamically calculated based on the hardness and surface roughness of the forging, and the expression is: Eps=k1(1+k2), where k1 is the basic neighborhood radius and k2 is the material property correction coefficient; The noise suppression process further includes: a noise classification module that identifies the noise type based on surface roughness and material hardness; uses Gaussian filtering for high-frequency noise and median filtering for low-frequency noise; The input parameters of the noise classification module include: forging surface hardness value range 30-60HRC; roughness value range Ra=0.8-2.5μm.

4. The method according to claim 1, wherein The bidirectional collaborative iterative optimization algorithm meets the following convergence conditions: the spatial residual threshold is ≤ 0.03 mm; the curvature gradient continuity meets the C² continuity standard; The triggering conditions of the cross-domain residual check include: performing a residual comparison every N iterations, where N is a preset positive integer; and the check threshold is dynamically adjusted based on the intake / exhaust edge weight, with a range of ≤0.01 mm.

5. The method according to claim 1, wherein The adaptive segmentation strategy includes: a dynamic interruption mechanism based on curvature mutation, with the curvature change threshold set to 0.05-0.15mm -1 ; The cumulative error threshold triggers segmented reconstruction, and the error range is ≤0.03mm; The dynamic interruption mechanism includes: using morphological filtering to eliminate isolated noise points, the filter kernel size is dynamically selected according to the feature sensitive area; and retaining high curvature feature points through bidirectional geometric constraints.

6. The method according to claim 1, wherein The incremental data management module includes: based on the differential coding model, only recording the data differences between iterations; compression rate evaluation model, real-time calculation of storage optimization efficiency; The differential coding model is implemented by: performing lossless compression storage on the initial data set; in subsequent iterations, only the difference matrix between the current data set and the previous version is stored; The calculation formula of the compression rate evaluation model is: compression rate = 1-(incremental data size / original data size), and the compression rate threshold is set to ≥50%.

7. The method according to claim 1, characterized in that The method further includes: achieving initial alignment of the measured point cloud with the theoretical model based on an improved ICP algorithm, and allocating feature weights in the order of leading edge > exhaust edge > blade back / basin; The improved ICP algorithm is optimized by using KD-Tree to accelerate the nearest neighbor search with a node size of 0.05-0.15 mm.

8. The method according to claim 1, characterized in that Constructing a non-rigid deformation field based on thin plate spline function to compensate for nonlinear displacement during forging processing; The construction of the non-rigid deformation field includes: defining a four-dimensional deformation tensor based on the material constitutive relationship; using the rigid translation and the nonlinear displacement components as initial constraints of the thin plate spline function; The deformation tensor model includes: Linear component: strain tensor (ε x ,ε y ,γ xy ) and the rigid translation (δ x ,δ y ); Nonlinear component: displacement field interpolation based on thin plate spline function; 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 shear slip deformation of the material in the xy plane is recorded as γ xy , 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。 9. The method according to claim 1, characterized in that The curvature calculation method uses: local curvature mapping to the feature protection module, for high curvature areas κ>0.1mm -1 Disable compensation; 0.05mm in medium and high curvature areas -1 <κ<0.1mm -1 Enables curvature adaptive filtering.

10. The method according to claim 1, characterized in that The method further includes: a damping factor adaptive update mechanism based on the LM algorithm, the update formula is: ； in, is the regulating factor, For the Iteration residual; The LM algorithm is optimized as follows: parameter update under the constraint of Jacobian matrix condition number; historical Jacobian matrix reuse mechanism to reduce repeated calculations.

Images