An aircraft skin surface waviness measurement method based on adaptive band separation

Abstract

The application discloses an aircraft skin surface waviness measurement method based on adaptive frequency band separation, comprising: acquiring complete three-dimensional point cloud data of the aircraft skin surface and performing pretreatment; fitting a local reference surface to the pretreated point cloud, calculating the deviation value of each point relative to the local reference surface, and obtaining a topography signal; constructing a graph Laplacian matrix based on the pretreated point cloud, the point cloud neighborhood relationship, the spatial distance between points and the normal similarity; performing spectral domain decomposition on the topography signal by using the graph Laplacian matrix to obtain surface topography components under different frequency bands; reconstructing the surface topography components in the target frequency band range by using an adaptive frequency band separation network to obtain a surface waviness component; and calculating the aircraft skin surface waviness parameter according to the surface waviness component to obtain a waviness measurement result. The application can effectively improve the accuracy, stability and adaptability of the aircraft skin surface waviness measurement.

Description

Technical Field

[0001] This invention belongs to the field of surface morphology detection and quality assessment technology, specifically relating to a method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation. Background Technology

[0002] As a crucial component of an aircraft's aerodynamic shape, the surface quality of aircraft skin directly impacts its aerodynamic performance, structural service performance, and aesthetic appearance. During aircraft manufacturing, assembly, and maintenance, factors such as material forming, riveting, stress release, machining errors, and the service environment can easily cause varying degrees of waviness and undulation on the skin surface. Surface waviness not only disrupts the continuity and smoothness of the skin surface but can also lead to increased local aerodynamic drag, changes in airflow separation characteristics, and localized stress concentration, thereby affecting the overall performance and operational safety of the aircraft. Therefore, accurate measurement and evaluation of aircraft skin surface waviness are of paramount importance.

[0003] Currently, the detection of waviness on aircraft skin surfaces mainly employs methods such as rulers, templates, manual visual inspection, contact contour measurement, and 3D optical scanning. While manual visual inspection and template comparison methods are simple to implement, the results rely heavily on operator experience, are highly subjective, and are difficult to quantify. Contact measurement methods, although possessing a certain level of accuracy, have low detection efficiency and are easily limited by the number of measurement points, measurement path, and contact conditions, making them unsuitable for rapid detection of large-area, complex curved skin surfaces. 3D optical scanning can efficiently acquire point cloud data of the skin surface, but due to the overall curvature, local assembly deformation, and random noise inherent in aircraft skin, accurately separating the components truly representing waviness characteristics from the complex surface morphology remains a challenge in current technology.

[0004] Existing waviness measurement methods typically extract waviness information using surface fitting, filtering decomposition, or fixed-band analysis. However, the morphological signals of aircraft skin surfaces often simultaneously contain overall trend terms, waviness terms, and noise terms. The spatial scale and frequency distribution of waviness vary in different regions and structural locations. Using fixed parameters or manually set frequency bands makes it difficult to balance the accuracy and adaptability of waviness identification under different operating conditions, easily leading to problems such as inaccurate waviness component extraction, unclear abnormal area location, and insufficient stability of measurement results. Therefore, there is an urgent need to provide a surface waviness measurement method that can automatically identify waviness-related frequency bands, effectively separate waviness components, and achieve quantitative measurement for the complex surface morphological characteristics of aircraft skin, in order to improve the accuracy, stability, and intelligence level of aircraft skin surface waviness detection. Summary of the Invention

[0005] To address the shortcomings of the prior art, this application provides a method for measuring the waviness of aircraft skin surfaces based on adaptive frequency band separation. The method aims to accurately extract and quantitatively characterize the waviness on aircraft skin surfaces, automatically identify the target frequency band corresponding to the waviness in the surface morphology signal, and effectively locate abnormal waviness regions. This solves the problems of existing technologies, such as reliance on manual experience for waviness component separation, fixed frequency band selection, insufficient measurement accuracy, and difficulty in accurately identifying abnormal regions. It improves the accuracy, adaptability, and detection efficiency of aircraft skin surface waviness measurement.

[0006] To achieve the above-mentioned technical objectives, the present invention provides the following technical solution: A method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation, which specifically includes the following steps: S1. Obtain complete 3D point cloud data of the aircraft skin surface; S2. Preprocess the complete 3D point cloud data, including denoising, outlier removal, normal estimation, and point cloud smoothing. S3. Fit the preprocessed point cloud to a local reference surface, calculate the deviation of each point relative to the local reference surface, and obtain the morphology signal characterizing the surface undulation features of the skin. S4. Based on the preprocessed point cloud, point cloud neighborhood relations, spatial distance between points, and normal similarity, construct a weighted undirected graph of the point cloud on the aircraft skin surface; then construct a graph Laplacian matrix based on the weighted undirected graph. S5. Use the Graph Laplacian matrix to perform spectral domain decomposition on the topography signal to obtain the surface topography components in different frequency bands; S6. Extract local geometric features and spectral component features of the point cloud and input them into an adaptive frequency band separation network to adaptively determine the target frequency band range corresponding to the surface ripples of the aircraft skin; reconstruct the surface topography components within the target frequency band range to obtain the surface ripple components; S7. Calculate the surface waviness parameters of the aircraft skin based on the surface waviness components to obtain the waviness measurement results.

[0007] Furthermore, step S1 specifically includes: The original point cloud data of the aircraft skin surface is acquired using a structured light scanner. When the coverage is incomplete in a single scan, the scanning position of the structured light scanner or the attitude of the aircraft skin is adjusted to acquire original point cloud data from multiple perspectives. The original point cloud data from multiple perspectives is registered and the coordinates are unified to obtain complete three-dimensional point cloud data of the aircraft skin surface.

[0008] Furthermore, step S2 specifically includes: For complete 3D point cloud data any point in Find its neighborhood point set Based on neighboring points and The spatial distance between them forms a weighting coefficient. The coordinates are weighted and smoothed to obtain a smoothed point cloud after denoising. ; For smooth point clouds For each point in the dataset, the spatial geometric distribution characteristics of its local neighborhood point cloud are statistically analyzed. These characteristics include the distance distribution of neighboring points, the point density distribution, and the continuity characteristics of the local surface. Based on these spatial geometric characteristics, outliers that deviate from the local surface distribution are removed to obtain the effective measurement point cloud. ; For effective measurement point cloud any point in ,exist Construct the covariance matrix in the local neighborhood and take The eigenvector corresponding to the smallest eigenvalue is used as the normal vector of that point. ; Combined with normal vector Then measure the effective point cloud Perform bilateral filtering smoothing based on normal constraints: determine the smoothing weights of neighboring points to the target point by using the spatial distance between points and the normal similarity; Based on smoothing weights, the effective measured point cloud The coordinates of each point are updated, and the normal vectors of each point are re-estimated after the coordinates are updated, finally obtaining the preprocessed point cloud of the aircraft skin surface. ;in, Indicates the points after preprocessing. express The normal vector, Represents a point index. This indicates the number of points after preprocessing.

[0009] Furthermore, step S3 specifically includes: Point clouds on the pre-processed aircraft skin surface any point in Find its neighborhood point set , with point Establish a local coordinate system with the origin; the normal direction of the local coordinate system originates from the point... normal vector It is determined that the directions of the two coordinate axes of the tangential plane originate from point [point missing]. neighborhood point set The two principal directions obtained after principal component analysis are determined. Will Each neighborhood point in the graph is transformed from global coordinates to a local coordinate system. The transformed neighborhood point set is then used to fit a local reference surface to obtain the points. The corresponding local reference surface ; Calculation points Along the normal vector Direction relative to local reference surface normal deviation value Point clouds on the pre-processed aircraft skin surface Repeat the above process for all points to obtain the normal deviation value of each point from its local reference surface, and statistically obtain the morphology signal of the aircraft skin surface. , Represents a point index. This indicates the number of points after preprocessing.

[0010] Furthermore, step S4 specifically includes: Point clouds on the pre-treated aircraft skin surface Each point in the graph is used as a node in the graph model, and each point is represented by a neighborhood relationship. Establish a set of adjacent points This allows for the construction of a weighted undirected graph of the aircraft skin surface. ,in, Represents a set of nodes. Denotes the set of edges. Represent the set of edge weights; For any adjacent nodes and Calculate the spatial distance between the two points as follows: and the dot product of normal vectors The edge weights are constructed by comprehensively considering the spatial distance between points and the inner product of normal vectors. The formula is expressed as: ; in, This represents a spatial distance scale parameter. Indicates the normal constraint scale parameter; Construct a weighted adjacency matrix based on the edge weights between all adjacent points. When node and When they are not adjacent, ; By weighted adjacency matrix Construct the degree matrix degree matrix Let be a diagonal matrix, and its first... The diagonal elements are represented as , , Represents a point index. This indicates the number of points after preprocessing. According to the degree matrix and weighted adjacency matrix Construct the Graph Laplace matrix .

[0011] Furthermore, step S5 specifically includes: Eigenvalue decomposition of the graph Laplacian matrix yields... ,in, The eigenvector matrix, For the eigenvalue matrix, Describe the first of the graph Laplace matrix L The eigenvector, i.e. the th eigenvector... Frequency basis functions of each frequency component in the spectral domain Indicates the first Eigenvalues ​​of frequency components in the spectral domain; The total number of frequency components in the spectral domain; morphological signals Projected onto the eigenvector matrix In Zhang Cheng's spectral domain space, the spectral domain response coefficient matrix of the morphology signal is obtained. , in , indicating the first The response coefficients of the frequency components in the spectral domain; Each spectral frequency component is represented as... ; Based on eigenvalues The magnitude of the frequency components in the spectral domain is used to adaptively decouple and layer them, grouping them into different frequency bands. Frequency components within the same group have eigenvalues ​​in the same numerical range. Inverse transforms are then performed on each group of frequency components to obtain surface topography components in different frequency bands. The formula for these surface topography components is as follows: ; in, Indicates the first Surface topography components in each frequency band Indicates the first The set of indices for the frequency components in the group spectral domain. G represents the total number of frequency bands.

[0012] Furthermore, step S6 specifically includes: For any point in the preprocessed point cloud Extract the local geometric feature vectors of its point cloud. ,in They are respectively Three-axis spatial coordinates, Divided into The three-axis normal vector components, for The local surface curvature, for The significance of anisotropy; Point exist The amplitudes of surface topography components from different frequency bands are pieced together to form a point. Surface topography component vector , Point In the Surface topography components at each scale; Local geometric feature vectors and surface topography component vector Perform point-by-point concatenation to construct a joint feature matrix; An adaptive frequency band separation network is constructed, which includes a spatial feature extraction subnetwork, a spectral domain feature extraction subnetwork, a feature fusion and discrimination subnetwork, and a multilayer perceptron (MLP). The joint feature matrix is ​​input into the spatial feature extraction subnetwork, and the surface topography component matrix is ​​input into the spectral domain feature extraction subnetwork to extract the global spatial geometric feature vector and the spectral domain frequency response feature matrix, respectively. The global spatial geometric feature vector and the spectral domain frequency response feature matrix are input into the feature fusion discriminant subnetwork. Feature fusion is performed through a cross-attention mechanism. The fused features are then passed through a multilayer perceptron (MLP) and the output is... Each frequency band corresponds to a frequency band weight vector. ; Mapping the band weights back to the frequency components in each spectral domain yields the final band selection coefficient vector. ; The spectral domain response coefficients are corrected using frequency band selection coefficients as weighting coefficients, and the surface ripple components are reconstructed by inverse graphical Fourier transform. ,in, Indicates the first Band selection coefficients for each frequency component in the spectral domain Indicates the first Spectral domain response coefficients of each spectral domain frequency component Indicates the first Frequency basis functions of frequency components in the spectral domain.

[0013] Furthermore, the step of inputting the joint feature matrix into the spatial feature extraction subnetwork and the surface topography component matrix into the spectral domain feature extraction subnetwork to extract the global spatial geometric feature vector and the spectral domain frequency response feature matrix, respectively, is as follows: The spatial feature extraction subnetwork employs a dynamic graph convolutional network to first perform local processing on the joint feature matrix. The neighborhood graph topology is constructed and the center-neighborhood difference edge feature is mapped, and then global max pooling is performed to obtain the global spatial geometric feature vector; The spectral domain feature extraction subnetwork adopts a one-dimensional convolutional network to perform sliding convolution on the surface topography components in the surface topography component matrix along the frequency domain axis, capture the energy coupling between adjacent frequency bands, and output the spectral domain frequency response feature matrix.

[0014] Furthermore, the adaptive frequency band separation network is trained using a composite loss function consisting of mean square error loss and smoothing L1 loss, the formula of which is expressed as: ; ; in, This represents the surface ripple component at the i-th point output by the adaptive frequency band separation network. Let be the true measured value of the surface ripple component at the i-th point. To select coefficient vectors for frequency band The concave function has a bipolar sparsity penalty term; For balance coefficient, This indicates the number of points after preprocessing. This is the index of the preprocessed point.

[0015] Furthermore, step S7 specifically includes: Based on surface ripple components Calculate the root mean square height parameter of the corrugation on the aircraft skin surface. , ,in Let i be the surface ripple component at the i-th point. This indicates the number of points after preprocessing. The index of the preprocessed point; For surface ripple components Perform spatial spectrum analysis to extract the main peak frequency. Calculate the dominant wavelength parameters ; For surface ripple components Energy statistics are performed in different directions to determine the principal direction parameters. , Indicates direction The energy distribution of the ripples on the surface; ripple height parameter With preset allowable threshold Compare and count the number of points exceeding the allowable threshold within the measurement area. And calculate the regional deviation rate.

[0016] By employing the above technical solution, the present invention has at least the following beneficial effects:

[0017] 1. This invention constructs a topographic signal from a three-dimensional point cloud of an aircraft skin surface and combines it with map decomposition to achieve multi-scale surface composition expression, which can effectively characterize the overall trend, local ripples and random noise features of the skin surface.

[0018] 2. This invention introduces an adaptive frequency band separation mechanism, which automatically identifies the target frequency band corresponding to the ripple by combining the local geometric features and spectral domain features of the point cloud. This avoids the problem of poor adaptability of traditional fixed frequency band or manual experience setting methods, and improves the accuracy and robustness of ripple component extraction.

[0019] 3. This invention can quantitatively calculate key parameters such as corrugation height, dominant wavelength, and dominant direction of aircraft skin surface, and locate and proportionally evaluate out-of-tolerance areas, thereby meeting the needs of aircraft skin surface quality inspection and manufacturing evaluation. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of an aircraft skin surface waviness measurement method based on adaptive frequency band separation proposed in this invention; Figure 2 This is a schematic diagram illustrating the acquisition of raw point cloud data in an embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] Although the steps in this invention are arranged by reference numerals, this is not intended to limit the order of the steps. Unless the order of the steps is explicitly stated or the execution of a step requires other steps as a basis, the relative order of the steps can be adjusted. It is understood that the term "and / or" as used herein refers to and covers any and all possible combinations of one or more of the associated listed items.

[0023] like Figure 1 As shown, this invention proposes a method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation, which specifically includes the following steps: S1. Obtain complete 3D point cloud data of the aircraft skin surface; In a preferred embodiment, step S1 specifically comprises: like Figure 2 As shown, a structured light scanner is used to acquire raw point cloud data of the aircraft skin surface. When a single scan does not cover the entire surface, the scanning position of the structured light scanner or the attitude of the aircraft skin is adjusted to acquire raw point cloud data from multiple perspectives. The raw point cloud data from multiple perspectives are then registered and their coordinates are unified to obtain complete 3D point cloud data of the aircraft skin surface. .

[0024] S2. Preprocess the complete 3D point cloud data, including denoising, outlier removal, normal estimation, and point cloud smoothing. In a preferred embodiment, step S2 specifically includes: For complete 3D point cloud data any point in Find its neighborhood point set Based on neighboring points and The spatial distance between them forms a weighting coefficient. The coordinates are weighted and smoothed to obtain a smoothed point cloud after denoising. ; For smooth point clouds For each point in the dataset, the spatial geometric distribution characteristics of its local neighborhood point cloud are statistically analyzed. These characteristics include the distance distribution of neighboring points, the point density distribution, and the continuity characteristics of the local surface. Based on these spatial geometric characteristics, outliers that deviate from the local surface distribution are removed to obtain the effective measurement point cloud. ; For effective measurement point cloud any point in ,exist Construct the covariance matrix in the local neighborhood and take The eigenvector corresponding to the smallest eigenvalue is used as the normal vector of that point. ; Combined with normal vector Then measure the effective point cloud A bilateral filtering smoothing process based on normal constraints is performed: the smoothing weights of neighboring points to the target point are determined by the spatial distance between points and the normal similarity. These smoothing weights are used to further suppress random noise while preserving the true morphological features of the aircraft skin surface. Specifically, for the effective measurement point cloud... any point in the middle its neighboring points Smoothing weights for that point Represented as: ; in, Indicates spatial distance control parameters. Indicates the normal constraint control parameters; Based on smoothing weights, the effective measured point cloud The coordinates of each point in the middle are updated, and the formula is expressed as: ; After updating the coordinates, the normal vectors of each point are re-estimated, and finally the pre-processed point cloud of the aircraft skin surface is obtained. ;in, Indicates the points after preprocessing. express The normal vector, Represents a point index. This indicates the number of points after preprocessing.

[0025] In the preprocessing of the method proposed in this application, two smoothing operations are designed: the first smoothing focuses on reducing random noise during scanning and improving the stability of subsequent outlier detection; the second smoothing focuses on using normal information to perform feature-preserving smoothing, thereby improving the accuracy of subsequent frequency band separation and waviness measurement while retaining the real waviness structure.

[0026] S3. Fit the preprocessed point cloud to a local reference surface, calculate the deviation of each point relative to the local reference surface, and obtain the morphology signal characterizing the undulation features of the skin surface. In a preferred embodiment, step S3 specifically includes: Point clouds on the pre-processed aircraft skin surface any point in Find its neighborhood point set , with point Establish a local coordinate system with the origin; the normal direction of the local coordinate system originates from the point... normal vector It is determined that the directions of the two coordinate axes of the tangential plane originate from point [point missing]. neighborhood point set The two principal directions obtained after principal component analysis are determined. Will Each neighborhood point in the graph is transformed from global coordinates to a local coordinate system. The transformed neighborhood point set is then used to fit a local reference surface to obtain the points. The corresponding local reference surface ; Calculation points Along the normal vector Direction relative to local reference surface normal deviation value Point clouds on the pre-processed aircraft skin surface Repeat the above process for all points to obtain the normal deviation value of each point from its local reference surface, and statistically obtain the morphology signal of the aircraft skin surface. , Represents a point index. This indicates the number of points after preprocessing.

[0027] S4. Based on the preprocessed point cloud, point cloud neighborhood relations, spatial distance between points, and normal similarity, construct a weighted undirected graph of the point cloud on the aircraft skin surface; then construct a graph Laplacian matrix based on the weighted undirected graph. In a preferred embodiment, step S4 specifically includes: Point clouds on the pre-treated aircraft skin surface Each point in the graph is used as a node in the graph model, and each point is represented by a neighborhood relationship. Establish a set of adjacent points This allows for the construction of a weighted undirected graph of the aircraft skin surface. ,in, Represents a set of nodes. Denotes the set of edges. Represent the set of edge weights; For any adjacent nodes and Calculate the spatial distance between the two points as follows: and the dot product of normal vectors Based on the traditional heat kernel function, a normal consistency constraint is introduced, and edge weights are constructed by comprehensively considering the spatial distance between points and the inner product of normal vectors. The formula is expressed as: ; in, This represents a spatial distance scale parameter. The normal constraint scale parameter is represented; when the spatial distance between two points is close and the normal directions are consistent, the edge weight is larger; when the spatial distance increases or the difference in normal directions increases, the edge weight decreases.

[0028] Traditional heat kernel functions rely solely on spatial distance to construct edge weights. When micro-ripples, bends, or machining marks appear on the boundaries of aerospace skin, the spatial distance between points on either side of the abrupt change is extremely close. Traditional methods assign excessively large edge weights, leading to the forced smoothing of feature boundaries and morphological adhesion during subsequent graph domain decomposition. This application introduces a normal consistency constraint term on top of the traditional heat kernel. Its core advantage lies in the fact that even if two points are extremely close in spatial distance, as long as there are abrupt changes in their micro-morphology causing normal inconsistency, this composite heat kernel function can sensitively detect and rapidly decay the edge weights. This transforms the graph Laplacian matrix into an adaptive anisotropic topological operator, focusing on noise suppression in flat regions and dynamically protecting and enhancing the morphological contour boundaries in ripple regions. This fundamentally eliminates geometric boundary adhesion and significantly improves the decoupling signal-to-noise ratio of the target ripple component during graph domain decomposition.

[0029] Construct a weighted adjacency matrix based on the edge weights between all adjacent points. When node and When they are not adjacent, ; By weighted adjacency matrix Construct the degree matrix degree matrix Let be a diagonal matrix, and its first... The diagonal elements are represented as , , Represents a point index. This indicates the number of points after preprocessing. According to the degree matrix and weighted adjacency matrix Construct the Graph Laplace matrix .

[0030] S5. Use the Graph Laplacian matrix to perform spectral domain decomposition on the topography signal to obtain the surface topography components in different frequency bands; In a preferred embodiment, step S5 specifically includes: Eigenvalue decomposition of the graph Laplacian matrix yields... ,in, The eigenvector matrix, For the eigenvalue matrix, Describe the first of the graph Laplace matrix L The eigenvector, i.e. the th eigenvector... Frequency basis functions of each frequency component in the spectral domain Indicates the first Eigenvalues ​​of frequency components in the spectral domain; The total number of frequency components in the spectral domain; morphological signals Projected onto the eigenvector matrix In Zhang Cheng's spectral domain space, the spectral domain response coefficient matrix of the morphology signal is obtained. , in , indicating the first The response coefficients of the frequency components in the spectral domain; Each spectral frequency component is represented as... ; Based on eigenvalues The magnitude of the value adaptively decouples and layers the frequency components in the spectral domain, grouping them into different frequency bands. The frequency components in the same group have characteristic values ​​in the same numerical range.

[0031] In this embodiment, G=3 is selected, meaning the frequency components are divided into three different groups in the spectral domain, specifically: Based on the eigenvalue matrix numerical distribution range , The low-frequency cutoff threshold is defined as the maximum eigenvalue. The high-frequency cutoff threshold is , The quantization range of the value is , The quantization range of the value is The index boundaries of the sets of spectral frequency components obtained by dividing the spectral domain into different frequency bands are: Low-frequency component index set : Satisfies eigenvalues The corresponding spectral frequency components characterize the overall macroscopic shape trend of the aircraft skin surface; Intermediate Frequency Component Index Set : Satisfies eigenvalues The corresponding spectral frequency components characterize the target frequency components containing the corrugated structure of the aircraft skin surface. High-frequency component index set : Satisfies eigenvalues The corresponding frequency components in the spectral domain characterize the local microscopic noise and detail variations on the aircraft skin surface.

[0032] Inverse transforms are performed on each group of spectral frequency components to obtain surface topography components in different frequency bands. The formula for surface topography components is as follows: ; in, Indicates the first Surface topography components in each frequency band Indicates the first The set of indices for the frequency components in the group spectral domain. G represents the total number of frequency bands.

[0033] S6. Extract local geometric features and spectral component features of the point cloud and input them into an adaptive frequency band separation network to adaptively determine the target frequency band range corresponding to the surface ripples of the aircraft skin; reconstruct the surface topography components within the target frequency band range to obtain the surface ripple components; In a preferred embodiment, step S6 specifically includes: For any point in the preprocessed point cloud Extract the local geometric feature vectors of its point cloud. ,in They are respectively Three-axis spatial coordinates, Divided into The three-axis normal vector components, for The local surface curvature, for The significance of anisotropy; Point exist The amplitudes of surface topography components from different frequency bands are pieced together to form a point. Surface topography component vector , Point In the Surface topography components at each scale; Local geometric feature vectors and surface topography component vector Perform point-by-point concatenation to construct a joint feature matrix. , Indicates a cascade operator; An adaptive frequency band separation network is constructed, which includes a spatial feature extraction subnetwork, a spectral domain feature extraction subnetwork, a feature fusion and discrimination subnetwork, and a multilayer perceptron (MLP). The joint feature matrix is ​​input into the spatial feature extraction subnetwork, and the surface topography component matrix is ​​input into the spectral domain feature extraction subnetwork to extract the global spatial geometric feature vector and the spectral domain frequency response feature matrix, respectively. The spatial feature extraction subnetwork employs a dynamic graph convolutional network to first perform local processing on the joint feature matrix. Neighborhood graph topology construction and center-neighborhood difference edge feature mapping, followed by global max pooling, yields the global spatial geometric feature vector. , ,in This indicates max pooling. Represents a dynamic graph convolutional network; The spectral domain feature extraction subnetwork employs a one-dimensional convolutional network to perform sliding convolution on the surface topography components in the surface topography component matrix along the frequency domain axis, capturing the energy coupling between adjacent frequency bands and outputting the spectral domain frequency response feature matrix. , , Surface topography component matrix Transpose of; This represents a one-dimensional convolutional network.

[0034] The global spatial geometric feature vector and the spectral domain frequency response feature matrix are input into the feature fusion discriminant subnetwork, and feature fusion is performed through a cross-attention mechanism. In this embodiment, the query Q, key K, and value V of the cross-attention mechanism are specifically as follows: ; in, , , It is a learnable linear transformation weight matrix; Attention weight distribution matrix , For channel dimension; fused features are... .

[0035] The fused features are then processed by a multilayer perceptron (MLP) (i.e., mapped and combined through two fully connected layers). Activation function constraints), output and Each frequency band corresponds to a frequency band weight vector. The formula is expressed as: ; in, This is the mapping function for the multilayer perceptron; spectral frequency components in the same group have the same frequency band weights, avoiding the network from directly predicting too many independent weights of spectral frequency components, which could lead to the curse of dimensionality, causing the network to fail to converge and lose the feasibility of real-time measurement.

[0036] Mapping the band weights back to the frequency components in each spectral domain yields the final band selection coefficient vector. .

[0037] The spectral domain response coefficients are corrected using frequency band selection coefficients as weighting coefficients, and the surface ripple components are reconstructed by inverse graphical Fourier transform. ,in, Indicates the first Band selection coefficients for each frequency component in the spectral domain Indicates the first Spectral domain response coefficients of each spectral domain frequency component Indicates the first The frequency basis functions of each frequency component in the spectral domain. The frequency band selection in this application is not a "hard selection" that selects only a fixed frequency band and rejects other frequency bands, but a "soft selection" that uses frequency band coefficients to weight and correct each frequency band component. By using a larger selection coefficient to reflect the target frequency band, compared with the selection methods in the prior art, this application can better ensure accurate extraction of ripple components, clear location of abnormal areas, and stability of measurement results.

[0038] In a preferred embodiment, the adaptive frequency band separation network is trained using a composite loss function consisting of mean square error loss and smoothing L1 loss. The formula for the composite loss function is as follows: ; ; in, This represents the surface ripple component at the i-th point output by the adaptive frequency band separation network. Let be the true measured value of the surface ripple component at the i-th point. To select coefficient vectors for frequency band The concave function polarization sparsity penalty term is used to constrain the selection coefficients of each frequency component. Approaching or To eliminate the frequency band boundary ambiguity during adaptive reconstruction of ripple components; The balance coefficient is set to 0.1 in this embodiment.

[0039] S7. Calculate the surface waviness parameters of the aircraft skin based on the surface waviness components, and obtain the waviness measurement results. In a preferred embodiment, step S7 specifically includes: Based on surface ripple components Calculate the root mean square height parameter of the corrugation on the aircraft skin surface. , ,in For the surface ripple component at the i-th point, a square is introduced to eliminate the influence of positive and negative signs and amplify the influence of large distortions on the final result. This is suitable for scenarios such as aircraft skin where the requirements for local protrusions or depressions are extremely strict. For surface ripple components Perform spatial spectrum analysis to extract the main peak frequency. Calculate the dominant wavelength parameters The main spatial scale used to characterize ripples; For surface ripple components Energy statistics are performed in different directions to determine the principal direction parameters. , Indicates direction The energy distribution of the ripples on the surface; Used to characterize the main extension direction of the corrugations along the skin surface; ripple height parameter With preset allowable threshold Compare and count the number of points exceeding the allowable threshold within the measurement area. And calculate the regional deviation rate. This completes the measurement of the surface waviness of the aircraft skin.

[0040] In summary, this invention proposes an adaptive frequency band separation-based method for measuring the waviness of aircraft skin surfaces. This method solves the problem that fixed filtering parameters in traditional waviness measurement methods are difficult to adapt to complex curved surfaces, non-uniform sampling point clouds, and local feature variations in aircraft skin. It can effectively improve the accuracy, stability, and adaptability of aircraft skin surface waviness measurement.

[0041] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0042] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0043] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation, characterized in that, Specifically, the following steps are included: S1. Obtain complete 3D point cloud data of the aircraft skin surface; S2. Preprocess the complete 3D point cloud data, including denoising, outlier removal, normal estimation, and point cloud smoothing. S3. Fit the preprocessed point cloud to a local reference surface, calculate the deviation of each point relative to the local reference surface, and obtain the morphology signal characterizing the surface undulation features of the skin. S4. Based on the preprocessed point cloud, point cloud neighborhood relations, spatial distance between points, and normal similarity, construct a weighted undirected graph of the point cloud on the aircraft skin surface; then construct a graph Laplacian matrix based on the weighted undirected graph. S5. Use the Graph Laplacian matrix to perform spectral domain decomposition on the topography signal to obtain the surface topography components in different frequency bands; S6. Extract local geometric features and spectral component features of the point cloud and input them into an adaptive frequency band separation network to adaptively determine the target frequency band range corresponding to the surface ripples of the aircraft skin; reconstruct the surface topography components within the target frequency band range to obtain the surface ripple components; S7. Calculate the surface waviness parameters of the aircraft skin based on the surface waviness components to obtain the waviness measurement results.

2. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S1 is as follows: The original point cloud data of the aircraft skin surface is acquired using a structured light scanner. When the coverage is incomplete in a single scan, the scanning position of the structured light scanner or the attitude of the aircraft skin is adjusted to acquire original point cloud data from multiple perspectives. The original point cloud data from multiple perspectives is registered and the coordinates are unified to obtain complete three-dimensional point cloud data of the aircraft skin surface.

3. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S2 specifically includes: For complete 3D point cloud data any point in Find its neighborhood point set Based on neighboring points and The spatial distance between them forms a weighting coefficient. The coordinates are weighted and smoothed to obtain a smoothed point cloud after denoising. ; For smooth point clouds For each point in the dataset, the spatial geometric distribution characteristics of its local neighborhood point cloud are statistically analyzed. These characteristics include the distance distribution of neighboring points, the point density distribution, and the continuity characteristics of the local surface. Based on these spatial geometric characteristics, outliers that deviate from the local surface distribution are removed to obtain the effective measurement point cloud. ; For effective measurement point cloud any point in ,exist Construct the covariance matrix in the local neighborhood and take The eigenvector corresponding to the smallest eigenvalue is used as the normal vector of that point. ; Combined with normal vector Then measure the effective point cloud Perform bilateral filtering smoothing based on normal constraints: determine the smoothing weights of neighboring points to the target point by using the spatial distance between points and the normal similarity; Based on smoothing weights, the effective measured point cloud The coordinates of each point are updated, and the normal vectors of each point are re-estimated after the coordinates are updated, finally obtaining the preprocessed point cloud of the aircraft skin surface. ;in, Indicates the points after preprocessing. express The normal vector, Represents a point index. This indicates the number of points after preprocessing.

4. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S3 specifically includes: Point clouds on the pre-processed aircraft skin surface any point in Find its neighborhood point set , with point Establish a local coordinate system with the origin; the normal direction of the local coordinate system originates from the point... normal vector It is determined that the directions of the two coordinate axes of the tangential plane originate from point [point missing]. neighborhood point set The two principal directions obtained after principal component analysis are determined. Will Each neighborhood point in the graph is transformed from global coordinates to a local coordinate system. The transformed neighborhood point set is then used to fit a local reference surface to obtain the points. The corresponding local reference surface ; Calculation points Along the normal vector Direction relative to local reference surface normal deviation value Point clouds on the pre-processed aircraft skin surface Repeat the above process for all points to obtain the normal deviation value of each point from its local reference surface, and statistically obtain the morphology signal of the aircraft skin surface. , Represents a point index. This indicates the number of points after preprocessing.

5. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S4 specifically includes: Point clouds on the pre-treated aircraft skin surface Each point in the graph is used as a node in the graph model, and each point is represented by a neighborhood relationship. Establish a set of adjacent points This allows for the construction of a weighted undirected graph of the aircraft skin surface. ,in, Represents a set of nodes. Denotes the set of edges. Represent the set of edge weights; For any adjacent nodes and Calculate the spatial distance between the two points as follows: and the dot product of normal vectors The edge weights are constructed by comprehensively considering the spatial distance between points and the inner product of normal vectors. The formula is expressed as: ； in, This represents a spatial distance scale parameter. Indicates the normal constraint scale parameter; Construct a weighted adjacency matrix based on the edge weights between all adjacent points. When node and When they are not adjacent, ; By weighted adjacency matrix Construct the degree matrix degree matrix Let be a diagonal matrix, and its first... The diagonal elements are represented as , , Represents a point index. This indicates the number of points after preprocessing. According to the degree matrix and weighted adjacency matrix Construct the Graph Laplace matrix .

6. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S5 specifically includes: Eigenvalue decomposition of the graph Laplacian matrix yields... ,in, The eigenvector matrix, For the eigenvalue matrix, Describe the first of the graph Laplace matrix L The eigenvector, i.e. the th eigenvector... Frequency basis functions of each frequency component in the spectral domain Indicates the first Eigenvalues ​​of frequency components in the spectral domain; The total number of frequency components in the spectral domain; morphological signals Projected onto the eigenvector matrix In Zhang Cheng's spectral domain space, the spectral domain response coefficient matrix of the morphology signal is obtained. , in , indicating the first The response coefficients of the frequency components in the spectral domain; Each spectral frequency component is represented as... ; Based on eigenvalues The magnitude of the frequency components in the spectral domain is used to adaptively decouple and layer them, grouping them into different frequency bands. Frequency components within the same group have eigenvalues ​​in the same numerical range. Inverse transforms are then performed on each group of frequency components to obtain surface topography components in different frequency bands. The formula for these surface topography components is as follows: ； in, Indicates the first Surface topography components in each frequency band Indicates the first The set of indices for the frequency components in the group spectral domain. G represents the total number of frequency bands.

7. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S6 specifically includes: For any point in the preprocessed point cloud Extract the local geometric feature vectors of its point cloud. ,in They are respectively Three-axis spatial coordinates, Divided into The three-axis normal vector components, for The local surface curvature, for The significance of anisotropy; Point exist The amplitudes of surface topography components from different frequency bands are pieced together to form a point. Surface topography component vector , Point In the Surface topography components at each scale; Local geometric feature vectors and surface topography component vector Perform point-by-point concatenation to construct a joint feature matrix; An adaptive frequency band separation network is constructed, which includes a spatial feature extraction subnetwork, a spectral domain feature extraction subnetwork, a feature fusion and discrimination subnetwork, and a multilayer perceptron (MLP). The joint feature matrix is ​​input into the spatial feature extraction subnetwork, and the surface topography component matrix is ​​input into the spectral domain feature extraction subnetwork to extract the global spatial geometric feature vector and the spectral domain frequency response feature matrix, respectively. The global spatial geometric feature vector and the spectral domain frequency response feature matrix are input into the feature fusion discriminant subnetwork. Feature fusion is performed through a cross-attention mechanism. The fused features are then passed through a multilayer perceptron (MLP) and the output is... Each frequency band corresponds to a frequency band weight vector. ; Mapping the band weights back to the frequency components in each spectral domain yields the final band selection coefficient vector. ; The spectral domain response coefficients are corrected using frequency band selection coefficients as weighting coefficients, and the surface ripple components are reconstructed by inverse graphical Fourier transform. ,in, Indicates the first Band selection coefficients for each frequency component in the spectral domain Indicates the first Spectral domain response coefficients of each spectral domain frequency component Indicates the first Frequency basis functions of frequency components in the spectral domain.

8. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 7, characterized in that, The process of inputting the joint feature matrix into the spatial feature extraction subnetwork and the surface topography component matrix into the spectral domain feature extraction subnetwork to extract the global spatial geometric feature vector and the spectral domain frequency response feature matrix are as follows: The spatial feature extraction subnetwork employs a dynamic graph convolutional network to first perform local processing on the joint feature matrix. The neighborhood graph topology is constructed and the center-neighborhood difference edge feature is mapped, and then global max pooling is performed to obtain the global spatial geometric feature vector; The spectral domain feature extraction subnetwork adopts a one-dimensional convolutional network to perform sliding convolution on the surface topography components in the surface topography component matrix along the frequency domain axis, capture the energy coupling between adjacent frequency bands, and output the spectral domain frequency response feature matrix.

9. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 7, characterized in that, The adaptive frequency band separation network is trained using a composite loss function consisting of mean square error loss and smoothing L1 loss. The formula for the composite loss function is as follows: ； ； in, This represents the surface ripple component at the i-th point output by the adaptive frequency band separation network. Let be the true measured value of the surface ripple component at the i-th point. To select coefficient vectors for frequency band The concave function has a bipolar sparsity penalty term; For balance coefficient, This indicates the number of points after preprocessing. This is the index of the preprocessed point.

10. The method for measuring the surface waviness of aircraft skin based on adaptive frequency band separation according to claim 1, characterized in that, Step S7 specifically includes: Based on surface ripple components Calculate the root mean square height parameter of the corrugation on the aircraft skin surface. , ,in Let i be the surface ripple component at the i-th point. This indicates the number of points after preprocessing. The index of the preprocessed point; For surface ripple components Perform spatial spectrum analysis to extract the main peak frequency. Calculate the dominant wavelength parameters ; For surface ripple components Energy statistics are performed in different directions to determine the principal direction parameters. , Indicates direction The energy distribution of the ripples on the surface; ripple height parameter With preset allowable threshold Compare and count the number of points exceeding the allowable threshold within the measurement area. And calculate the regional deviation rate.

Images