Device, system and method for sampling the airfoil profile at the notch location

Abstract

This invention discloses a device, system, and method for sampling the airfoil profile at the notch location, including: aligning the measured point cloud with the theoretical model through progressive spatial registration for initial coarse registration, and dynamically adjusting the root mean square error threshold and the rate of change of curvature threshold based on curvature sensitivity analysis to achieve adaptive segmented interruption control; constructing a three-dimensional hash mapping engine to store the original point cloud data, combined with an incremental data access mode and iterative data structure, including a historical transformation matrix stack, B-spline control points, and an 8-bit state register, to ensure data processing integrity and efficiency; using a genetic algorithm to optimize the segmented B-spline surface fitting, simultaneously optimizing the intake and exhaust edge profiles through lateral and longitudinal segmentation; and finally verifying through composite convergence. This method significantly improves the detection accuracy and computational efficiency of complex surfaces through dynamic threshold control and a hybrid optimization strategy.

Description

Technical Field

[0001] This invention relates to the field of precision manufacturing technology for aero-engines, specifically to a device, system, and method for sampling the profile of the blade at the notch position. Background Technology

[0002] In equipment such as aero-engines and gas turbines, precision-cast blades are core components, and their blade profile directly affects the performance and efficiency of the equipment. Especially at the exhaust edge notch of the blade, due to its complex structure and large curvature variations, traditional methods for sampling blade profiles are insufficient to meet the high-precision measurement requirements. Existing technologies for this location suffer from insufficient fitting accuracy, low data processing efficiency, and an inability to effectively handle curvature-sensitive areas, resulting in significant measurement errors that affect the subsequent machining and assembly quality of the blade. Traditional measurement methods face three major technical bottlenecks at the blade exhaust edge notch: First, the measurement point distribution strategy relies on manual experience, easily creating blind spots in areas of abrupt curvature changes, leading to the loss of key feature points during profile fitting; second, when using cubic spline interpolation or polynomial fitting, the response to changes in the second derivative of curvature in curvature-sensitive areas lags, causing a systematic deviation between the fitted curve and the actual surface; third, the data processing flow lacks a curvature-adaptive sampling density mechanism, often resulting in segmented fitting distortion in transition areas with curvature radii less than 2 mm. These defects not only cause the airfoil profile error to exceed the ±0.03mm manufacturing standard, but also lead to a decrease in aerodynamic performance, directly affecting the combustion chamber thermal efficiency and turbine mechanical power output. Therefore, a new method for determining the airfoil profile at the notch location has become a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] In view of this, the present invention provides a device, system and method for sampling the airfoil profile at the notch position, in order to solve the problems raised in the background art.

[0004] The technical solution adopted is as follows:

[0005] The notch location blade profile sampling device has the following features:

[0006] The processor and memory, wherein the memory has a contour-based point-taking program that causes the processor to perform the following operations:

[0007] Registration Department: Installs a progressive spatial registration model to align the measured point cloud with the theoretical model. Specifically, it includes: using an improved iterative nearest neighbor algorithm to perform coarse registration on the measured point cloud, and constructing a KD-Tree to accelerate nearest neighbor search through privileged weight allocation.

[0008] The curvature field of the point cloud is calculated in real time by the curvature-sensitive region identification unit, and the registration parameters are dynamically adjusted.

[0009] Segmentation Processing Unit: Based on the alignment data output by the registration unit, it implements segmentation and interruption control using dynamically adjusted root mean square error thresholds and curvature change rate thresholds, generating classified segmented contours. Specific triggering conditions are as follows:

[0010] When the root mean square error exceeds 0.5% of the theoretical radius of curvature, lateral segmentation is triggered, and the segmentation boundary is dynamically adjusted by the interruption coupling controller;

[0011] The rate of change of curvature of adjacent point clouds exceeds 0.01 mm. -1 When the longitudinal segmentation is triggered, B-spline surface segmentation reconstruction is performed on the curvature abrupt change interval through progressive segmented fitting unit;

[0012] The optimization fitting section employs a genetic algorithm to optimize the piecewise B-spline surface fitting. Based on the piecewise contours, the contour data of the intake and exhaust sides are simultaneously optimized, configured as follows:

[0013] Encode the segment boundary points into binary gene sequences;

[0014] The fitness function is used to comprehensively evaluate the fitting accuracy and curvature continuity.

[0015] Verification Output Unit: Using the data obtained from the optimization fitting unit, the geometric accuracy and surface smoothness are verified by the spatial residual monitoring unit and the curvature continuity analysis unit, and the final contour point set is output.

[0016] Non-destructive data management module: used to build a 3D hash index structure to store raw point cloud data;

[0017] Iterative state memory: used to store the stack of historical transformation matrices, B-spline control points, and convergence flags.

[0018] Furthermore, the coarse registration operation performed by the registration unit includes:

[0019] During the iteration process, a dynamic interruption strategy is adopted, which adaptively adjusts the segment length according to the curvature and error distribution. When the curvature is large or the error is large, the segment length is shortened, and vice versa.

[0020] Incremental fitting retains the condition number of the Jacobian matrix from the previous fitting, and the LM algorithm is used to update the parameters. The iteration terminates when the error is ≤0.001mm.

[0021] This invention provides a device for sampling the airfoil profile at the notch location, comprising:

[0022] The contour sampling module is used to acquire the measured point cloud data of the precision-cast blades;

[0023] The processing feedback module is used to dynamically adjust the blade processing parameters based on the contour sampling result set to form a detection-feedback-processing closed loop;

[0024] The evaluation module interacts in a closed loop with the processing feedback module.

[0025] The dynamic fitting execution module interacts with the non-destructive data management module via a high-speed data bus, including:

[0026] The curvature-sensitive region identification unit is used to calculate the curvature field of the point cloud in real time, mark the curvature change interval, and send segmented signals to the progressive segmented fitting unit.

[0027] The progressive piecewise fitting unit is used to perform piecewise reconstruction of B-spline surfaces on the marked intervals, and calls the genetic algorithm to optimize the control point weights to ensure the continuity of the second derivative of the surface.

[0028] The interrupted coupling controller is used for bidirectional communication with the progressive piecewise fitting unit. It dynamically adjusts the segment boundaries based on the set error threshold and curvature threshold, and triggers the incremental data access interface to load the corresponding hash unit point cloud.

[0029] The non-destructive data management module includes:

[0030] A 3D hash mapping engine is used to divide the point cloud according to spatial hash values ​​and transmit curvature field data to the curvature-sensitive region identification unit in real time.

[0031] The incremental data access interface is used to dynamically push local point cloud data streams based on segmented requests, and supports fast retrieval based on hash key values;

[0032] The iterative state memory is used to store the IterationData structure, which provides transformation data and fitting parameters for the dual convergence criterion evaluation module;

[0033] The dual convergence criterion evaluation module includes:

[0034] The spatial residual monitoring unit is used to read the transformation matrix from the iterative state memory and calculate the distance between the actual point and the fitted surface.

[0035] The curvature continuity analysis unit is used to receive curvature field data and verify the continuity of the rate of change of curvature and the second derivative of adjacent segments;

[0036] The composite conditional decision maker is used to trigger threshold adjustment through an interrupt mechanism. When both criteria are met, it outputs a convergence signal; otherwise, it generates a new transformation matrix and writes it into the iterative state memory.

[0037] Furthermore, the collaborative mechanism between the dynamic fitting execution module and the non-destructive data management module is as follows:

[0038] The 3D hash mapping engine divides the original point cloud into hash units, and each unit stores the curvature value of the corresponding point cloud.

[0039] The curvature-sensitive region identification unit quickly locates curvature abrupt change units using hash key values, triggering the progressive piecewise fitting unit to call the incremental data access interface to load the point cloud of that unit, with a data loading delay of ≤5ms.

[0040] Furthermore, the closed-loop control path of the dual convergence criterion evaluation module is as follows:

[0041] The composite conditional decision-maker outputs a convergence signal → the coupled controller is interrupted → the segment boundary and fitting parameters are adjusted.

[0042] If a single condition is not met, the historical transformation matrix stack is updated through the iterative state memory, and a new round of genetic algorithm optimization is started.

[0043] Furthermore, the interaction between the processing feedback module and the contour measurement device includes:

[0044] Receive the contour error report from the verification output unit and automatically generate machining compensation instructions;

[0045] The processed measured point cloud data is fed back to the evaluation module, forming a detection-processing iterative optimization link.

[0046] This invention also provides a method for taking points for the airfoil profile at the notch location, including the following steps:

[0047] Step S1: Dynamic Fitting - Coupling Breaking Stage: Based on the progressive spatial registration of the theoretical model and the measured point cloud, a curvature-driven adaptive segmentation strategy is adopted. Segmentation breaking control is implemented through dynamically adjusted root mean square error threshold and curvature change rate threshold, including:

[0048] The coarse registration step is used to project the measured point cloud onto the theoretical model coordinate system, and an improved iterative nearest point algorithm is used for initial registration.

[0049] The curvature sensitivity analysis step is used to calculate the curvature value of each point cloud using parametric surface differential geometry to dynamically adjust the registration parameters until the root mean square error of each sub-segment meets the set conditions.

[0050] The segmented trigger step is used to trigger lateral segmentation when the root mean square error of the current segment fitting exceeds 0.5% of the theoretical radius of curvature, and when the rate of change of curvature of adjacent point clouds exceeds 0.01 mm. -1 The longitudinal segmentation is triggered at any time, and the segmentation boundary is dynamically adjusted by the interruption coupling controller;

[0051] Step S2: Non-destructive data architecture phase: Create a 3D hash mapping engine to store the raw point cloud data, adopt an incremental data access mode to preserve data integrity, and synchronously establish an iterative data structure, including:

[0052] The original point cloud is divided into spatial hash units using a 3D hash mapping function, and only the local point cloud data associated with the current segment is loaded;

[0053] Establish an iterative data structure that includes a pointer to the original data, a stack of historical transformation matrices, a set of B-spline control points, and an 8-bit convergence flag register;

[0054] Step S3: Dual-objective optimization fitting stage: A genetic algorithm is used to optimize the piecewise B-spline surface fitting, simultaneously controlling the lateral and longitudinal segmentation of the intake and exhaust sides, including:

[0055] The chromosome encoding step is used to encode the segment boundary point index number into a binary gene sequence, and an elite preservation strategy is used to prevent the loss of the optimal solution.

[0056] The fitness function optimization step is used to comprehensively evaluate the fitting accuracy and curvature continuity through the fitness function, and performs selection, crossover, and mutation operations, with a maximum number of iterations of 1000.

[0057] The piecewise reconstruction step is used to perform piecewise reconstruction of B-spline surfaces in curvature abrupt change intervals to ensure that the fitted surface has the continuity of the second derivative.

[0058] Step S4: Composite Convergence Verification Stage: When both geometric accuracy and surface smoothness conditions are satisfied, output the final contour point set, including:

[0059] The geometric accuracy verification procedure is used to verify accuracy with a spatial residual threshold of ≤0.03 mm.

[0060] The surface smoothness verification procedure requires that the rate of change of curvature between adjacent segments be ≤0.01 mm. -1 And the second derivative is continuous;

[0061] If a single condition is not met, the segment boundary redefinition and fitting parameter iterative optimization will be automatically triggered until both criteria are met.

[0062] Furthermore, in the iterative process of the progressive spatial registration, a dynamic interruption strategy and incremental fitting are adopted:

[0063] Dynamically adjust the segment length to shorten the segment length when the curvature is large or the error is large, and extend it when the curvature is small or the error is large.

[0064] Incremental fitting retains the condition number of the previous Jacobian matrix, and the LM algorithm is used to update the parameters. The iteration terminates when the error is ≤0.001 mm.

[0065] Furthermore, curvature continuity analysis and piecewise reconstruction are achieved through the following collaborative mechanism:

[0066] The curvature-sensitive region identification unit marks curvature abrupt change intervals in real time;

[0067] The progressive piecewise fitting unit calls the point cloud data of the corresponding hash unit according to the marked interval, performs B-spline surface piecewise reconstruction, and optimizes the control point weights through a genetic algorithm.

[0068] Furthermore, the process of the evaluation module and the processing feedback module forming a closed loop interaction is as follows:

[0069] Generate blade machining parameter adjustment instructions based on the final contour point set;

[0070] After receiving the processed measured point cloud data, progressive spatial registration and segmented fitting are re-executed to form an iterative optimization link of detection-processing-re-detection.

[0071] The beneficial effects of this application are as follows: The method and system for obtaining the profile of blades at the notch location described in this invention significantly improve the accuracy and efficiency of profile detection for blades with complex structures through multi-stage collaborative control and intelligent algorithm optimization. Specific technical advantages are as follows: In the dynamic fitting-decoupling stage, based on a curvature-driven adaptive segmentation strategy and progressive spatial registration technology, a dual mechanism of dynamically adjusting the root mean square error threshold and the rate of change of curvature threshold is used to achieve accurate identification and segmented control of curvature-sensitive areas such as the blade exhaust edge notch. Compared to traditional fixed-threshold segmentation methods, this scheme dynamically optimizes the segment boundary division based on the actual geometric characteristics of the point cloud data, effectively avoiding fitting distortion caused by curvature abrupt changes. It is particularly suitable for the local geometric distortion area at the exhaust edge notch of precision-cast blades caused by the casting process, ensuring that the segmented fitting process can accurately characterize the microscopic features of the blade profile.

[0072] The non-destructive data architecture phase integrates a 3D hash index structure with an incremental data access mechanism to build an efficient data retrieval and iterative update system while maintaining the topological integrity of the original point cloud data. By designing a 3D hash mapping function, rapid spatial partitioning of the point cloud data is achieved. Combined with the incremental access mode, only the local point cloud data associated with the current segment needs to be loaded during dynamic fitting, reducing data processing volume by 60%-75% and significantly improving system response efficiency. The historical transformation matrix stack and B-spline control point set configured in the iterative data structure provide a parameter traceability benchmark for multiple rounds of fitting iterations, ensuring verifiable stability of the fitting process.

[0073] In the dual-objective optimization fitting stage, a genetic algorithm is used to achieve global optimization of the piecewise B-spline surface fitting. By encoding the segment boundary points with binary gene sequences and constructing a comprehensive fitness function that integrates fitting accuracy and curvature continuity indices, collaborative optimization of lateral and longitudinal segments is achieved. Compared with fitting methods guided by single geometric error, this dual-objective optimization mechanism simultaneously satisfies the geometric tolerance requirements of the leaf profile and the surface smoothness index. When segmented control is executed simultaneously on the inlet / outlet sides, it can effectively eliminate the abrupt changes in surface continuity caused by single-dimensional threshold-triggered segmentation, ensuring that the fitted surface in the notch region has second-order derivative continuity and significantly enhancing the engineering adaptability of the profile point selection results.

[0074] The composite convergence verification stage sets a joint verification standard of spatial residual threshold and curvature continuity criterion. The final result is output only when both indicators simultaneously meet the accuracy specifications, completely eliminating the risk of false convergence due to single-condition judgment. In a typical scenario, when the spatial residual meets the standard but the curvature change rate exceeds the limit, the system will automatically trigger a dynamic adjustment mechanism for the fitting parameters. Iterative optimization is implemented through redefining the segmented boundaries and redistributing the fitting weights until both criteria meet the engineering standards, thus mathematically ensuring the geometric accuracy and surface quality of the contour point acquisition results.

[0075] The supporting point acquisition system constructs a fully automated system for curvature-sensitive region identification, segmented fitting, and convergence verification through closed-loop collaboration of a dynamic fitting execution module, a non-destructive data management module, and a dual convergence criterion evaluation module. A high-speed data bus architecture and priority interrupt mechanism ensure data transmission latency between modules is ≤3ms. The collaborative design of the 3D hash mapping engine and incremental data access interface enables the system to handle point cloud datasets of ≥10^7. Attached Figure Description

[0076] Figure 1 This is a schematic diagram of contour comparison provided by the present invention;

[0077] Figure 2 A schematic diagram of the notch location leaf profile sampling system provided by the present invention;

[0078] Figure 3 This is a schematic diagram of the notch location leaf profile sampling device module provided by the present invention. Detailed Implementation

[0079] 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, and 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.

[0080] A notch-position leaf profile sampling device includes a processor and a memory. The memory contains a profile sampling program that causes the processor to perform the following operations: A registration unit: This unit installs a progressive spatial registration model to align the measured point cloud with the theoretical model. Specifically, it includes: performing coarse registration of the measured point cloud using an improved iterative nearest-neighbor algorithm; constructing a KD-Tree to accelerate nearest-neighbor search through privileged weight allocation; and dynamically adjusting registration parameters by calculating the point cloud curvature field in real time using a curvature-sensitive region identification unit. A segmentation processing unit: Based on the alignment data output by the registration unit, it implements segmentation interruption control using dynamically adjusted root mean square error thresholds and curvature change rate thresholds to generate classified segmented profiles. The specific triggering conditions are: lateral segmentation is triggered when the root mean square error exceeds 0.5% of the theoretical curvature radius, and the segmentation boundary is dynamically adjusted by the interruption coupling controller; and the curvature change rate of adjacent point clouds exceeds 0.01 mm. -1 When the longitudinal segmentation is triggered, B-spline surface segmentation reconstruction is performed on the curvature abrupt change interval through progressive segmented fitting unit;

[0081] Optimization Fitting Unit: A genetic algorithm is used to optimize the fitting of the piecewise B-spline surface. The contour data of the intake and exhaust sides are optimized simultaneously based on the piecewise contours. The configuration is as follows: the segment boundary points are encoded as binary gene sequences; the fitting accuracy and curvature continuity are comprehensively evaluated through a fitness function. Validation Output Unit: Using the data obtained from the optimization fitting unit, the geometric accuracy and surface smoothness are verified through the spatial residual monitoring unit and the curvature continuity analysis unit, and the final contour point set is output.

[0082] This application analyzes the curvature distribution characteristics of the exhaust edge notch region of the blade, recognizing the necessity of dynamically adjusting registration parameters and adopting an adaptive segmentation strategy. Based on the iterative nearest-point algorithm to optimize registration efficiency, and combined with real-time curvature field calculation to establish a segmentation triggering mechanism, a genetic algorithm is proposed to be introduced into the surface reconstruction process, forming a closed-loop quality control system.

[0083] The progressive spatial registration model refers to a coordinate transformation model that implements point cloud alignment in stages. It can be implemented using an improved iterative nearest-point algorithm combined with a privileged weight allocation mechanism to improve initial registration efficiency. The curvature-sensitive region identification unit is a processing module that calculates point cloud curvature based on parametric surface differential geometry. For example, it identifies abrupt change regions by calculating principal curvature values, used to dynamically adjust registration weights. The dynamically adjusted root mean square error threshold is an error trigger value dynamically set based on the theoretical radius of curvature. Binary gene sequence encoding converts the spatial location information of segmented boundary points into binary strings that can be processed by genetic algorithms, for example, using Gray code encoding, to optimize surface reconstruction parameters.

[0084] Specifically, the registration unit accelerates the matching of point cloud correspondences using a KD-Tree, while the curvature-sensitive region identification unit updates registration weight coefficients in real time, prioritizing registration accuracy in high-curvature regions. The segmentation processing unit dynamically adjusts the horizontal and vertical segment boundaries based on real-time calculated error distribution and curvature gradients, interrupting the coupling controller to coordinate segment reconstruction and data loading. The optimization fitting unit iteratively optimizes control point positions using a genetic algorithm, while the fitness function simultaneously evaluates geometric deviation and curvature continuity indices, ensuring the reconstructed surface meets both measurement accuracy and smoothness requirements. The verification output unit employs a dual verification mechanism: a spatial residual monitoring unit detects the distance deviation between the point cloud and the surface, and a curvature continuity analysis unit verifies the curvature transition characteristics of adjacent segments.

[0085] The notch location leaf profile sampling device also includes: a non-destructive data management module for building a 3D hash mapping engine to store the original point cloud data; and an iterative state memory for storing the historical transformation matrix stack, B-spline control points, and convergence flags.

[0086] The 3D hash mapping engine divides the 3D space into equally sized cubic units, each mapped to a unique key value via a hash function. This can be achieved using a spatial grid partitioning algorithm, generating hash keys by calculating the quotient of point cloud coordinates and grid resolution. This structure optimizes data access paths through the principle of spatial locality, addressing the efficiency bottleneck of traditional linear storage structures when retrieving complex surface data. The historical transformation matrix stack stores the sequence of spatial transformation parameters generated during each iteration of the registration process. This can be implemented using a dynamic array structure, saving the historical state of the transformation matrix through push operations, providing a backtracking basis for parameter optimization. B-spline control points store the coordinate data of control points used in piecewise surface reconstruction. This can be implemented using a sparse matrix storage format, ensuring geometric continuity during piecewise reconstruction by preserving control point parameters. The convergence flag records whether the iteration process has reached the termination condition, implemented using an iteration state register, achieving adaptive iteration termination control through state bit judgment.

[0087] The non-destructive data management module uses a 3D hash mapping engine to cluster and store the original point cloud according to its spatial location. When accessing data in a specific area is needed, the corresponding cubic unit is quickly located using the hash key, avoiding the delay caused by full data scanning. The iterative state memory continuously updates the historical transformation matrix stack during registration. New transformation matrices generated in each iteration are pushed onto the top of the stack, forming a traceable parameter evolution trajectory. B-spline control points are dynamically updated during the surface reconstruction stage. The control point parameters generated by each piecewise fitting are incrementally stored, ensuring geometric continuity between adjacent segments. The convergence flag is updated in real time by the dual convergence criterion evaluation module. When both spatial residuals and curvature continuity meet the criteria, a state flip is triggered, terminating the current iteration process.

[0088] The coarse registration operation performed by the registration unit includes: adopting a dynamic interruption strategy during the iteration process, adaptively adjusting the segment length according to the curvature and error distribution, shortening the segment length when the curvature is large or the error is large, and extending it when the curvature is small; retaining the condition number of the Jacobian matrix of the previous fitting in incremental fitting, using the LM algorithm to update the parameters, and terminating the iteration when the error is ≤0.001mm.

[0089] The dynamic segmentation strategy refers to a control mechanism that dynamically adjusts the segment length of the point cloud based on the real-time calculated curvature field and error distribution. Specifically, it can be implemented by the collaborative work of a curvature-sensitive region identification unit and an error space residual monitoring unit. When the curvature value exceeds a preset threshold or the local root mean square error increases, segment boundary adjustment is immediately triggered. This strategy achieves dynamic matching between segment length and surface complexity through a dual feedback mechanism of curvature and error.

[0090] Incremental fitting is an optimization method that retains the condition number of the Jacobian matrix from the previous iteration for parameter updates. Specifically, it can be implemented using a matrix condition number caching mechanism in conjunction with the LM algorithm. In each iteration, the previously calculated matrix condition number is reused as the initial value, and updates are only performed on the incrementally changing portion. This approach avoids resource waste caused by repeated calculations by inheriting the historical iteration state.

[0091] The present invention also provides a notch position blade profile measurement system, including the notch position blade profile measurement device of any one of the above, comprising: a profile measurement module for acquiring measured point cloud data of precision-cast blades; a machining feedback module for dynamically adjusting blade machining parameters according to the profile measurement result set to form a detection-feedback-machining closed loop; and an evaluation module for forming a closed loop interaction with the machining feedback module.

[0092] The dynamic fitting execution module interacts with the non-destructive data management module via a high-speed data bus. It includes: a curvature-sensitive region identification unit for real-time calculation of the point cloud curvature field, marking curvature abrupt change intervals, and sending segmentation signals to the progressive segmented fitting unit; a progressive segmented fitting unit for performing B-spline surface segmentation reconstruction on the marked intervals, calling a genetic algorithm to optimize control point weights, and ensuring the continuity of the surface's second derivative; and a break coupling controller for bidirectional communication with the progressive segmented fitting unit, dynamically adjusting segmentation boundaries based on set error and curvature thresholds, and triggering the incremental data access interface to load the corresponding hash unit point cloud. The non-destructive data management module includes: a 3D hash mapping engine for dividing the point cloud according to spatial hash values ​​and transmitting curvature field data to the curvature-sensitive region identification unit in real-time; an incremental data access interface for dynamically pushing local point cloud data streams based on segmentation requests, supporting fast retrieval based on hash keys; and an iterative state memory for storing IterationData. The structure provides transformation data and fitting parameters for the dual convergence criterion evaluation module. The dual convergence criterion evaluation module includes: a spatial residual monitoring unit, used to read the transformation matrix from the iterative state memory and calculate the distance between the actual point and the fitted surface; a curvature continuity analysis unit, used to receive curvature field data and verify the continuity of the curvature change rate and second derivative of adjacent segments; and a composite condition decision-maker, used to trigger threshold adjustment through an interrupt mechanism, outputting a convergence signal when both criteria are met, otherwise generating a new transformation matrix and writing it into the iterative state memory.

[0093] The contour sampling module is a device used to acquire and process three-dimensional coordinate data of the blade surface. Specifically, it can be implemented using a laser scanner combined with a point cloud preprocessing algorithm, generating a standardized measured point cloud dataset through filtering, noise reduction, and coordinate transformation. The machining feedback module is the control unit connecting the measurement system and the machining equipment. Specifically, it can be implemented using an industrial Ethernet communication protocol, converting the contour error vector into toolpath compensation instructions for the CNC machine tool. The evaluation module and the machining feedback module form a closed-loop interaction. Based on the final contour sampling result set, it generates blade machining parameter adjustment instructions; receives the measured point cloud data after machining, and re-executes progressive spatial registration and piecewise fitting, forming an iterative optimization link of detection-machining-re-detection.

[0094] The curvature-sensitive region identification unit in the dynamic fitting execution module is a real-time computation unit based on differential geometry principles. Specifically, it can be implemented using a Gaussian curvature algorithm combined with a sliding window mechanism, identifying curvature abrupt change regions by calculating the curvature values ​​of local surfaces in the point cloud. The progressive piecewise fitting unit is the computational unit that performs surface reconstruction. Specifically, it can be implemented using a non-uniform rational B-spline algorithm combined with a genetic algorithm optimizer, ensuring the continuity of the second derivative of the surface by adjusting the control point weights. The 3D hash mapping engine in the non-destructive data management module is a spatial data indexing device, specifically implemented using a 3D mesh partitioning algorithm combined with a hash function, dividing the point cloud data into quickly searchable storage units based on spatial location. The composite condition decision-maker in the dual convergence criterion evaluation module is a multi-condition logic judgment device, specifically implemented using a combination of a threshold comparator and logic gate circuits, triggering iterative optimization instructions by parallel verification of geometric accuracy and surface smoothness indices.

[0095] The collaborative mechanism between the dynamic fitting execution module and the non-destructive data management module is as follows: the 3D hash mapping engine divides the original point cloud into hash units, and each unit stores the curvature value of the corresponding point cloud; the curvature sensitive region identification unit quickly locates the curvature change unit through the hash key value, triggers the progressive segmented fitting unit to call the incremental data access interface to load the point cloud of that unit, and the data loading delay is ≤5ms.

[0096] The 3D hash mapping engine is an indexing engine that generates hash keys based on spatial coordinates. Specifically, it uses a 3D mesh partitioning algorithm to divide the point cloud space into equal-volume cubic units. Each unit stores the curvature value of the point cloud within that space, and fast retrieval is achieved by establishing a mapping relationship between spatial location and hash key values. The curvature-sensitive region identification unit is a feature detection module based on curvature threshold judgment. Specifically, it uses Gaussian curvature calculation methods to analyze the curvature distribution of the point cloud in real time, and quickly locates the spatial units corresponding to regions of abrupt curvature changes through hash key value matching. The incremental data access interface is a transmission channel for loading local data on demand. Specifically, it uses memory mapping technology to establish a direct mapping between hash unit addresses and physical storage locations, loading only the point cloud data blocks associated with the currently processed region.

[0097] The closed-loop control path of the dual convergence criterion evaluation module is as follows: the composite condition decision maker outputs a convergence signal → the coupled controller is interrupted → the segment boundary and fitting parameters are adjusted; if a single condition is not met, the historical transformation matrix stack is updated through the iterative state memory, and a new round of genetic algorithm optimization is started.

[0098] A composite conditional decision unit is a logical unit that determines convergence based on two indices: geometric accuracy and surface smoothness. It can be implemented using a multi-threaded parallel computing architecture and is used for real-time comparison of spatial residual thresholds and curvature continuity indices. An iterative state memory is a data storage unit that stores a stack of historical transformation matrices. It can be implemented using non-volatile memory and is used to save the parameter trajectory of the previous optimization process. Genetic algorithm optimization is a global optimization method based on the principles of biological evolution. It can be implemented using an elite retention strategy and an adaptive crossover mechanism to overcome the limitations of local optima.

[0099] The interaction between the processing feedback module and the contour measurement device includes: receiving the contour error report from the verification output unit and automatically generating processing compensation instructions; feeding back the measured point cloud data after processing to the evaluation module, forming a detection-processing iterative optimization link.

[0100] The contour error report refers to a set of quantified error data verified by the spatial residual monitoring unit and the curvature continuity analysis unit. Specifically, it can be generated by combining a three-dimensional coordinate deviation matrix and a curvature gradient change matrix, used to accurately characterize the deviation between the actual machined surface and the theoretical model. Machining compensation commands are process parameter correction commands derived inversely from the error vector field. They can be converted into machine tool motion trajectory compensation quantities through the CNC system interface protocol, achieving online compensation of machining errors. The detection-machining iterative optimization link refers to a bidirectional feedback channel between measurement data and machining parameters. Specifically, a high-speed industrial bus is used to establish a real-time data pipeline to ensure that machining compensation commands complete a closed-loop response within a single production cycle.

[0101] This invention also provides a method for taking points for the airfoil profile at the notch location, including the following steps:

[0102] Step S1: Dynamic Fitting-Coupled Breaking Stage: Based on the progressive spatial registration of the theoretical model and the measured point cloud, a curvature-driven adaptive segmentation strategy is adopted. Segmentation breaking control is implemented through dynamically adjusted root mean square error thresholds and curvature change rate thresholds, including: a coarse registration step, used to project the measured point cloud onto the theoretical model coordinate system, using an improved iterative nearest-point algorithm for initial registration; a curvature sensitivity analysis step, used to calculate the curvature value of each point cloud using parametric surface differential geometry to dynamically adjust the registration parameters until the root mean square error of each sub-segment meets the set conditions; and a segmentation triggering step, used to trigger lateral segmentation when the root mean square error of the current segment exceeds 0.5% of the theoretical curvature radius, and when the curvature change rate of adjacent point clouds exceeds 0.01 mm. -1 The longitudinal segmentation is triggered at any time, and the segmentation boundary is dynamically adjusted by the interruption coupling controller;

[0103] Step S2: Non-destructive data architecture stage: Create a 3D hash mapping engine to store the original point cloud data. Use an incremental data access mode to preserve data integrity and simultaneously establish an iterative data structure, including: using a 3D hash mapping function to divide the original point cloud into spatial hash units, loading only the local point cloud data associated with the current segment; and establishing an iterative data structure containing original data pointers, historical transformation matrix stacks, B-spline control points, and an 8-bit convergence flag register.

[0104] Step S3: Dual-objective optimization fitting stage: A genetic algorithm is used to optimize the piecewise B-spline surface fitting, simultaneously controlling the lateral and longitudinal segmentation of the intake and exhaust sides. This includes: a chromosome encoding step, used to encode the segment boundary point index numbers into binary gene sequences, employing an elite retention strategy to prevent the loss of the optimal solution; a fitness function optimization step, used to comprehensively evaluate the fitting accuracy and curvature continuity through the fitness function, performing selection, crossover, and mutation operations, with an upper limit of 1000 iterations; and a piecewise reconstruction step, used to perform piecewise reconstruction of the B-spline surface in curvature abrupt change intervals to ensure that the fitted surface has second-order derivative continuity.

[0105] Step S4: Composite Convergence Verification Stage: When both geometric accuracy and surface smoothness conditions are met, the final contour point set is output, including: geometric accuracy verification step, which uses a spatial residual threshold of ≤0.03 mm as the accuracy verification standard;

[0106] The surface smoothness verification procedure requires that the rate of change of curvature between adjacent segments be ≤0.01 mm. -1 And the second derivative is continuous;

[0107] If a single condition is not met, the segment boundary redefinition and fitting parameter iterative optimization will be automatically triggered until both criteria are met.

[0108] During the iterative process of progressive spatial registration, a dynamic interruption strategy and incremental fitting are adopted: the segment length is dynamically adjusted to shorten the segment length when the curvature is large or the error is large, and to extend it otherwise; the incremental fitting retains the condition number of the previous Jacobian matrix, and the LM algorithm is used to update the parameters. The iteration termination condition is that the error is ≤0.001 mm.

[0109] Curvature continuity analysis and segmented reconstruction are achieved through the following collaborative mechanism: the curvature sensitive region identification unit marks the curvature change interval in real time; the progressive segmented fitting unit calls the point cloud data of the corresponding hash unit according to the marked interval, performs B-spline surface segmented reconstruction, and optimizes the control point weights through a genetic algorithm.

[0110] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A device for taking points to measure the profile of the blade at the notch location, characterized in that, It includes: a processor; and a memory, wherein the memory has a contour-based point acquisition program that causes the processor to perform the following operations: A registration unit configured to: install a progressive spatial registration model for aligning the measured point cloud with the theoretical model, specifically including: performing coarse registration of the measured point cloud using an improved iterative nearest-neighbor algorithm; constructing a KD-Tree to accelerate nearest-neighbor search through privileged weight allocation; and dynamically adjusting registration parameters by calculating the point cloud curvature field in real time through a curvature-sensitive region identification unit; and a segmentation processing unit that, based on the alignment data output by the registration unit, implements segmentation interruption control using dynamically adjusted root mean square error thresholds and curvature change rate thresholds to generate classified segmented contours. Specifically, the triggering conditions are: lateral segmentation is triggered when the root mean square error exceeds 0.5% of the theoretical curvature radius, and the segmentation boundaries are dynamically adjusted by the interruption coupling controller; and the curvature change rate of adjacent point clouds exceeds 0.

01. The system triggers longitudinal segmentation and performs B-spline surface segmentation reconstruction on the curvature abrupt change interval through a progressive segmented fitting unit. The optimization fitting unit uses a genetic algorithm to optimize the segmented B-spline surface fitting, simultaneously optimizing the contour data of the intake and exhaust sides based on the segmented contours. The configuration involves encoding the segment boundary points into binary gene sequences and comprehensively evaluating the fitting accuracy and curvature continuity through a fitness function. The verification output unit uses the data obtained from the optimization fitting unit, verifies the geometric accuracy and surface smoothness through a spatial residual monitoring unit and a curvature continuity analysis unit, and outputs the final contour point set. A non-destructive data management module is used to construct a 3D hash index structure to store the original point cloud data. An iterative state memory is used to store the historical transformation matrix stack, B-spline control points, and convergence flags.

2. A notch location airfoil profile measurement system, comprising the notch location airfoil profile measurement device as described in claim 1, characterized in that, include: The contour sampling module is used to acquire measured point cloud data of precision-cast blades; the machining feedback module is used to dynamically adjust blade machining parameters based on the contour sampling result set to form a detection-feedback-machining closed loop; the evaluation module interacts with the machining feedback module in a closed loop. The dynamic fitting execution module interacts with the non-destructive data management module via a high-speed data bus. It includes: a curvature-sensitive region identification unit, used to calculate the point cloud curvature field in real time, mark curvature abrupt change intervals, and send segmentation signals to the progressive piecewise fitting unit; a progressive piecewise fitting unit, used to perform B-spline surface segmentation reconstruction on the marked intervals, calling a genetic algorithm to optimize control point weights and ensure the continuity of the surface's second derivative; and a break coupling controller, used for bidirectional communication with the progressive piecewise fitting unit, dynamically adjusting segmentation boundaries based on set error and curvature thresholds, and triggering the incremental data access interface to load the corresponding hash unit point cloud. The non-destructive data management module includes: a 3D hash mapping engine, used to divide the point cloud according to spatial hash values ​​and send segmentation signals to the curvature-sensitive region identification unit. The system includes: a real-time transmission of curvature field data; an incremental data access interface for dynamically pushing local point cloud data streams based on segment requests, supporting fast retrieval based on hash keys; an iterative state memory for storing IterationData structures to provide transformation data and fitting parameters for the dual convergence criterion evaluation module; and a dual convergence criterion evaluation module comprising: a spatial residual monitoring unit for reading the transformation matrix from the iterative state memory and calculating the distance between the actual point and the fitted surface; a curvature continuity analysis unit for receiving curvature field data and verifying the continuity of the rate of change of curvature and the second derivative of adjacent segments; and a composite condition decision-maker for triggering threshold adjustment through an interrupt mechanism, outputting a convergence signal when both criteria are met, otherwise generating a new transformation matrix and writing it into the iterative state memory.

3. The notch location airfoil profile sampling system according to claim 2, characterized in that, The collaborative mechanism of the dynamic fitting execution module and the non-destructive data management module is as follows: the three-dimensional hash mapping engine divides the original point cloud into hash units, and each unit stores the curvature value of the corresponding point cloud. The curvature-sensitive region identification unit quickly locates curvature abrupt change units using hash key values, triggering the progressive piecewise fitting unit to call the incremental data access interface to load the point cloud of that unit, with a data loading delay of ≤5ms.

4. The notch location airfoil profile sampling system according to claim 2, characterized in that, The closed-loop control path of the dual convergence criterion evaluation module is as follows: the composite condition decision-maker outputs a convergence signal → the coupling controller is interrupted → the segment boundary and fitting parameters are adjusted; if a single condition is not met, the historical transformation matrix stack is updated through the iterative state memory, and a new round of genetic algorithm optimization is started.

5. The notch location airfoil profile sampling system according to claim 2, characterized in that, The interaction between the processing feedback module and the contour point acquisition device includes: receiving the contour error report from the verification output unit and automatically generating processing compensation instructions; feeding back the measured point cloud data after processing to the evaluation module to form a detection-processing iterative optimization link.

6. The notch location airfoil profile sampling system according to claim 5, characterized in that, The closed-loop interaction between the evaluation module and the processing feedback module is as follows: Based on the final contour point sampling results, the blade processing parameter adjustment instructions are generated, the measured point cloud data after processing is received, and the progressive spatial registration and segmented fitting are re-executed to form an iterative optimization link of detection-processing-re-detection.

7. A method for determining the profile of the blade at the notch location, characterized in that, Includes the following steps: Step S1: Dynamic Fitting-Coupled Breaking Stage: Based on the progressive spatial registration of the theoretical model and the measured point cloud, a curvature-driven adaptive segmentation strategy is adopted. Segmentation breaking control is implemented through dynamically adjusted root mean square error thresholds and curvature change rate thresholds, including: a coarse registration step, used to project the measured point cloud onto the theoretical model coordinate system, using an improved iterative nearest-point algorithm for initial registration; a curvature sensitivity analysis step, used to calculate the curvature value of each point cloud using parametric surface differential geometry to dynamically adjust the registration parameters until the root mean square error of each sub-segment meets the set conditions; and a segmentation triggering step, used to trigger lateral segmentation when the root mean square error of the current segment exceeds 0.5% of the theoretical curvature radius, and when the curvature change rate of adjacent point clouds exceeds 0.

01. The longitudinal segmentation is triggered at any time, and the segmentation boundary is dynamically adjusted by the interruption coupling controller; Step S2: Non-destructive data architecture stage: Create a 3D hash mapping engine to store the original point cloud data, adopt an incremental data access mode to preserve data integrity, and simultaneously establish an iterative data structure, including: using a 3D hash mapping function to divide the original point cloud into spatial hash units, loading only the local point cloud data associated with the current segment; establishing an iterative data structure containing original data pointers, historical transformation matrix stacks, B-spline control points, and an 8-bit convergence flag register; Step S3: Dual-objective optimization fitting stage: A genetic algorithm is used to optimize the piecewise B-spline surface fitting, simultaneously controlling the lateral and longitudinal segmentation of the intake and exhaust sides. This includes: a chromosome encoding step, used to encode the segment boundary point index numbers into binary gene sequences, employing an elite retention strategy to prevent the loss of the optimal solution; a fitness function optimization step, used to comprehensively evaluate the fitting accuracy and curvature continuity through the fitness function, performing selection, crossover, and mutation operations, with an upper limit of 1000 iterations; and a piecewise reconstruction step, used to perform piecewise reconstruction of the B-spline surface in curvature abrupt change intervals to ensure that the fitted surface has second-order derivative continuity. Step S4: Composite Convergence Verification Stage: When both geometric accuracy and surface smoothness conditions are met, the final contour point set is output, including: a geometric accuracy verification step, using a spatial residual threshold ≤ 0.03mm as the accuracy verification standard; and a surface smoothness verification step, where the rate of curvature change between adjacent segments is ≤ 0.

01. Furthermore, the second derivative is continuous; if a single condition is not met, the redefinition of the segment boundary and iterative optimization of the fitting parameters are automatically triggered until both criteria are met.

Images