Composite material skin grinding amount calculation method based on clustering analysis and symbolic distance field

Abstract

The invention relates to the technical field of composite material digital detection, in particular to a composite material skin grinding amount calculation method based on clustering analysis and a symbolic distance field, which specifically comprises the following steps: acquiring three-dimensional measurement data of the surface of a composite material skin, and designing a three-dimensional measurement point feature descriptor; based on the feature descriptors, a clustering analysis algorithm is designed, and a grinding area three-dimensional measuring point class and a non-grinding area three-dimensional measuring point class are divided; aiming at the three-dimensional measuring points of the non-grinding area, carrying out implicit curved surface reconstruction based on a radial basis function; aiming at the three-dimensional measuring point of the grinding area, a grinding area symbol distance field is constructed, and the symbol distance is the grinding amount of the measuring point; according to the method, the problem that in the prior art, profile tolerance estimation cannot be obtained during aircraft CAD mathematical model is solved, the grinding allowance of the aircraft skin repairing area can be accurately analyzed, and aircraft maintenance personnel are guided to conduct grinding. Compared with a traditional method for judging the profile tolerance based on manual touch, the method has the advantages that interference of human subjective factors is eliminated, the grinding allowance is quantified, and the precision is higher.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of digital detection of composite materials, in particular to a composite skin grinding amount calculation method based on cluster analysis and a signed distance field. BACKGROUND

[0002] With the continuous development of composite forming technology, advanced composite materials have been widely used in high-end equipment such as aerospace and rail transportation, which can effectively reduce the overall weight of the aircraft and improve the performance of the aircraft. Although the composite material has good mechanical properties, its impact resistance is poor, and the composite structure may have more or less structural damage during manufacturing, installation and use.

[0003] With the development of composite material repair technology, the results have shown that the composite component can be effectively repaired and even restored to the original strength by using appropriate repair technology. Among them, the most popular for strength recovery is the oblique joint patching, and the patching repair has the advantages of high strength after repair, maintaining the original structural aerodynamic shape, etc., and has broad development prospects and popularization value in the repair application of aerospace composite structures.

[0004] However, the anisotropic shrinkage effect of the resin-based composite curing process causes a 0.1-1.5mm layered step difference protrusion in the repair area, which causes the following problems: (1) aerodynamic performance degradation: local protrusions destroy the laminar flow state and increase the turbulent friction resistance; (2) attenuation of stealth characteristics: surface discontinuity leads to enhanced electromagnetic wave scattering; (3) structural stress concentration: step difference edges form fatigue crack initiation points, reducing service life.

[0005] The traditional grinding process relies on manual tactile experience to judge the grinding depth, and has the problems of low efficiency, uncontrollable precision, and difficult quality quantification and traceability. Therefore, it is urgent to design a composite skin grinding amount automatic calculation method based on high-precision sensing measurement to realize accurate quantification of the geometric deviation of the repair area and assist in generating a gradient grinding path to suppress subsurface fiber damage. SUMMARY

[0006] To solve the above technical problems, the application provides a composite skin grinding amount calculation method based on cluster analysis and a signed distance field. The purpose is to accurately analyze the grinding allowance of the repair area of the aircraft skin and guide the aircraft maintenance personnel to grind.

[0007] The technical problems to be solved by the application are solved by the following technical solutions:

[0008] The composite skin grinding amount calculation method based on cluster analysis and a signed distance field comprises the following steps:

[0009] S1, obtain three-dimensional measurement data of the composite material skin surface, and define a feature descriptor of each three-dimensional measurement point according to the obtained measurement data;

[0010] S2, based on the feature descriptor of each three-dimensional measurement point, design a clustering analysis algorithm to distinguish three-dimensional measurement point classes of the grinding area and three-dimensional measurement point classes of the non-grinding area in a multidimensional space;

[0011] S3, for the three-dimensional measurement points of the non-grinding area, an implicit surface is constructed based on a radial basis function;

[0012] S4, for the three-dimensional measurement points of the grinding area, a grinding area symbolic distance field is constructed, and the directed distance of each grinding area three-dimensional measurement point to the implicit surface is calculated, i.e. the grinding amount at the measurement point is obtained.

[0013] As a further improvement of the application, the feature descriptor of each three-dimensional measurement point in step S1 is specifically defined as:

[0014] A six-dimensional feature vector including curvature, normal direction angle, local height difference, point cloud color intensity, density weight and distance weight, denoted as F;

[0015] Wherein, the curvature represents the bending degree of the local surface, the normal direction angle represents the angle between the normal of the current measurement point and the average normal of the neighborhood, the local height difference represents the height difference between the current measurement point and its neighborhood points, the point cloud color intensity represents the point cloud RGB component of the current measurement point, the density weight represents the local point cloud density around the current measurement point, and the distance weight represents the distance of the current measurement point to its neighborhood fitting plane.

[0016] As a further improvement of the application, the curvature represents the bending degree of the local surface, i.e. the minimum eigenvalue of the neighborhood point covariance matrix of the current measurement point based on principal component analysis (PCA) is calculated, and then the value after normalization.

[0017] As a further improvement of the application, step S2 specifically includes:

[0018] S21, normalization and weighting of the feature descriptor:

[0019] The six-dimensional feature vector in the feature descriptor F is respectively subjected to Min-Max normalization to eliminate the dimensional difference, and the feature descriptor F after Min-Max normalization is obtained, and a weight coefficient is dynamically assigned to each dimensional feature according to the importance, and the weight coefficient of the kth dimensional feature is denoted as w k ,k∈{1,2,...,6};

[0020] S22, define a multidimensional distance metric:

[0021] The formula is:

[0022]

[0023] wherein r i , r j represent any i-th and j-th three-dimensional measuring point respectively, f k (.) is the k-th dimension feature after Min-Max normalization, D(r i , r j ) represents the multi-dimensional distance metric between r i , r j ;

[0024] S23, setting clustering parameters:

[0025] including setting clustering threshold ∈ as N times of three-dimensional laser sensor resolution and setting neighborhood expected point number threshold minPts;

[0026] S24, clustering execution and category determination:

[0027] calculating multi-dimensional distance values between all three-dimensional measuring points, if the distance value is within the threshold range, then clustering into a category, if the point number of three-dimensional measuring points in the category ≥ minPts, then calculating the mean of Min-Max normalized feature descriptors of all measuring points in the category, denoted as F', setting the virtual measuring point with F' as the feature descriptor as the new clustering center, clustering the points in the category with point number less than minPts again, repeatedly iterating this process until the last clustering result is divided into grinding area three-dimensional measuring point category and non-grinding area three-dimensional measuring point category, and adding category labels as category 1 and category 0 respectively.

[0028] As a further improvement of the present application, in step S24, if there are still three-dimensional measuring points not classified as category 1 or category 0 after multiple iterations, they are marked as noise and discarded.

[0029] As a further improvement of the present application, step S3 specifically includes:

[0030] S31, for point cloud data and normal vector information M is the point number in the non-grinding area, randomly sampling L points as the center points of the radial basis function, obtaining the center point set of the radial basis function

[0031] S32, based on the implicit surface modeling principle of the radial basis function, for any point p i in P, it needs to satisfy g(p i ) = 0, wherein:

[0032]

[0033] wherein g(pi ) is an implicit curve equation at p i , γ is a hyperparameter set to a value, is a coefficient to be solved;

[0034] S33, construct a linear equation group:

[0035] S331,

[0036] S332, ensure the uniqueness of the solution, and the orthogonality condition needs to be met:

[0037]

[0038] S34, convert the linear equation group in step S33 into a pathological matrix form, and obtain:

[0039]

[0040] Ψi, j, i∈{1, 2,..., M}, j∈{1, 2,..., L}, ij represents the element in the i-th row and the j-th column of the Ψ matrix;

[0041] S35, add a regularization term to the obtained pathological matrix, and modify the equation to:

[0042] (Ψ+λI)α+Pβ=0,

[0043] wherein λ is a hyperparameter designed value, and I is a unit matrix;

[0044] S36, based on singular value decomposition, the coefficients α and β in the obtained equation are solved, and the implicit surface expression of the non-grinding area is determined.

[0045] As a further improvement of the present application, step S4 specifically comprises:

[0046] S41, for any one measuring point q i in the grinding area k , based on the Newton iteration method, find a point p k on the implicit surface expression g of the non-grinding area i , so that p i is closest to q i in the Euclidean distance;

[0047] Specifically, first assume Then iteratively update: Until

[0048] S42, define the signed distance of the measuring point q i to the implicit surface as sdf i = sign × ||qi -p k ||,

[0049] wherein, the obtained signed distance field sdf i is the grinding amount at the measuring point q. i .

[0050] The beneficial effects of the present application are:

[0051] The present application provides a composite skin grinding amount calculation method based on cluster analysis and signed distance field. The multi-dimensional feature descriptor can quickly and accurately distinguish the three-dimensional measuring points of the grinding and non-grinding areas, overcoming the inefficiency and subjectivity of traditional manual discrimination. Based on cluster analysis, the implicit surface can accurately represent the surface features of any topological structure, effectively improving the geometric reduction degree of the target surface, providing a high-fidelity reference benchmark for grinding amount calculation, and avoiding grinding amount deviation caused by model error. In addition, combined with the signed distance field to quantify the spatial relationship between the grinding area measuring point and the implicit surface, the grinding amount is converted into a directional numerical field, so that the method proposed by the present application discards the traditional iterative matching or projection calculation process, and realizes the "one-step" direct output of the grinding amount. Compared with the traditional method of judging the profile based on manual touch, the present method eliminates the interference of human subjective factors, quantizes the grinding allowance, and has higher precision. BRIEF DESCRIPTION OF DRAWINGS

[0052] The present application will be further described below in conjunction with the drawings and examples:

[0053] Figure 1 is the flow chart of the present application;

[0054] Figure 2 is the signed distance representation map of any three-dimensional measuring point in the grinding area. DETAILED DESCRIPTION

[0055] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the present application will be further described below in conjunction with the drawings and examples.

[0056] The present application defines a feature descriptor including a plurality of dimensional measuring point features, combines cluster analysis and signed distance field, realizes accurate distinction between grinding area and non-grinding area of composite skin, can accurately analyze the grinding allowance of the aircraft skin repair area, and guides the aircraft maintenance personnel to grind. Compared with the traditional method of judging the profile based on manual touch, the present method eliminates the interference of human subjective factors, quantizes the grinding allowance, and has higher precision. As shown in Figure 1 The composite skin grinding amount calculation method based on cluster analysis and signed distance field specifically includes the following steps:

[0057] S1, obtain the three-dimensional measurement data of the composite skin surface, and define the feature descriptor of each three-dimensional measurement point according to the obtained measurement data. In the embodiment, a high-resolution laser scanner is used as a data acquisition instrument.

[0058] As a preferred embodiment, the feature descriptor of each three-dimensional measurement point in step S1 is specifically defined as:

[0059] A six-dimensional feature vector including curvature, normal direction angle, local height difference, point cloud color intensity, density weight and distance weight, denoted as F;

[0060] Wherein, the curvature represents the bending degree of the local surface, and the curvature is taken as a dimensional feature because the grinding area usually has a sudden change in curvature due to material defects or processing marks; in the embodiment, the minimum eigenvalue of the covariance matrix of the neighborhood points of the current measurement point calculated based on principal component analysis (PCA) is taken as the curvature (range [0, 1]) after normalization.

[0061] The normal direction angle represents the angle between the normal of the current measurement point and the average normal of the neighborhood, and the normal direction angle of the non-grinding area is continuous, while the angle of the grinding area increases significantly; therefore, the angle cosine value (range [-1, 1]) between the normal of the current measurement point and the average normal of the neighborhood points is taken as another dimensional feature in the embodiment.

[0062] The local height difference represents the height difference between the current measurement point and its neighborhood points, and the surface of the grinding area is uneven with large fluctuations in height difference, while the non-grinding area is relatively smooth; therefore, the standard deviation of the Z-axis coordinates of the current measurement point and its neighborhood points is selected as the local height difference (unit: mm) in the embodiment.

[0063] The point cloud color intensity represents the RGB component of the point cloud of the current measurement point, and since the grinding area and the non-grinding area present certain color difference, the gray value of the current measurement point is selected as the point cloud color intensity (range [0, 255]) in the embodiment.

[0064] The density weight represents the local point cloud density around the current measurement point, and the grinding area and the non-grinding area present certain point cloud density difference; in the embodiment, the density weight is specifically the normalized value of the number of points in the neighborhood of the current measurement point (range [0, 1]).

[0065] The distance weight represents the distance from the current measurement point to its neighborhood fitting plane, and is specifically the projection distance (unit: mm) from the current measurement point to the initial coarse fitting plane calculated after the initial coarse fitting plane.

[0066] In addition, in the embodiment, the neighborhood of the current measuring point is specifically a unit sphere neighborhood with the current measuring point as the center of the sphere when defining the feature descriptor. For each three-dimensional measuring point, the present application extracts features from multiple dimensions in the unit sphere neighborhood of the measuring point, and integrates the features to define the feature descriptor as the basis for subsequent clustering analysis; on the one hand, the extraction of features from multiple dimensions enriches the measuring point information and ensures the accuracy of the clustering analysis, and on the other hand, the analysis in combination with the neighborhood of the current measuring point emphasizes the difference between the current measuring point and the neighborhood points and avoids the loss of local details; therefore, the method proposed in the present application can quickly and accurately distinguish the three-dimensional measuring points of the grinding and non-grinding regions.

[0067] S2, based on the feature descriptor of each three-dimensional measuring point, designing a clustering analysis algorithm to distinguish the three-dimensional measuring point class of the grinding region and the three-dimensional measuring point class of the non-grinding region in a multidimensional space.

[0068] As a preferred embodiment, step S2 specifically includes:

[0069] S21, normalization and weighting of the feature descriptor.

[0070] The six-dimensional feature vectors in the feature descriptor F are subjected to Min-Max normalization respectively to eliminate the dimensional differences, and a feature descriptor F subjected to Min-Max normalization is obtained, and a weight coefficient is dynamically assigned to each dimensional feature according to the importance, and the weight coefficient of the kth dimensional feature is denoted as w k. k ,k∈{1,2,...,6}.

[0071] S22, defining a multidimensional distance metric. The formula is expressed as:

[0072]

[0073] wherein r i and r j respectively represent any ith and jth three-dimensional measuring point, f k (r i) and f k (r j) are the kth dimensional features subjected to Min-Max normalization, and D(r i, r j) represents the multidimensional distance metric between r i and r j. i j k i j i j

[0074] S23, setting clustering parameters.

[0075] including setting a clustering threshold ∈ to be N times of the resolution of the three-dimensional laser sensor and setting a neighborhood expected point number threshold minPts. In the present embodiment, N=2 and minPts=15.

[0076] S24, clustering execution and category determination.

[0077] ​​​​​​​Calculate the multi-dimensional distance value between all three-dimensional measurement points, if the distance value is within the threshold range, then it is clustered into a class, if the number of three-dimensional measurement points clustered into a class is greater than or equal to minPts, then the mean value of the feature descriptor of all measurement points in the class after Min-Max normalization is calculated, denoted as Set to The virtual measurement point of the feature descriptor is the new cluster center, and the points in the class whose point number is less than minPts are clustered again, and this process is iterated multiple times until the final clustering result is divided into grinding area three-dimensional measurement point class and non-grinding area three-dimensional measurement point class, and the class labels are added as class 1 and class 0 respectively.

[0078] In step S24, it should be noted that the present application finally needs to be clustered and classified into two categories, and in the clustering process, it is inevitable to be divided into multiple different clusters; therefore, the present application sets a clustering threshold and a neighborhood expected point number threshold, and iteratively clusters the classes that do not meet the point number requirement, and the cluster center in the iterative update process is also integrated with the feature descriptor of the cluster, which ensures that each iteration can be more detailed and can aggregate more points. In addition, if there are still three-dimensional measurement points that have not been classified into class 1 or class 0 after multiple iterations, they are marked as noise and discarded. This ensures the accuracy and reliability of the clustering analysis.

[0079] S3, for non-grinding area three-dimensional measurement points, an implicit surface is constructed based on a radial basis function.

[0080] S31, for point cloud data in the non-grinding area and normal vector information M is the number of points in the non-grinding area, L points are randomly sampled as the center points of the radial basis function, and the center point set of the radial basis function is obtained

[0081] S32, based on the implicit surface modeling principle of the radial basis function, for any point p in P i , it needs to satisfy g(p i ) = 0, where:

[0082]

[0083] where g(p i ) is the implicit curve equation at p i , γ is a hyperparameter set to a value of is the coefficient to be solved.

[0084] S33, construct a linear equation set:

[0085] S331,

[0086] S332, to ensure the uniqueness of the solution, the orthogonal condition needs to be met:

[0087]

[0088] S34, the linear equation set in step S33 is converted into a form of a pathological matrix, and the following is obtained:

[0089]

[0090] Ψ = {Ψ ij}, i ∈ {1, 2,..., M}, j ∈ {1, 2,..., L}, represents the element in the i-th row and the j-th column of the Ψ matrix.

[0091] S35, a regularization term is added to the obtained pathological matrix, and the equation is modified as:

[0092] (Ψ+λI)α+Pβ=0,

[0093] where λ is a value designed for the hyperparameter, and I is a unit matrix.

[0094] S36, based on singular value decomposition, the coefficients α and β in the obtained equation are solved, and the implicit surface expression of the non-grinding area is determined.

[0095] S4, for the three-dimensional measurement points of the grinding area, a symbolic distance field of the grinding area is constructed, and the directed distance from each three-dimensional measurement point of the grinding area to the implicit surface is calculated, that is, the grinding amount at the measurement point is obtained.

[0096] As a preferred embodiment, step S4 specifically comprises:

[0097] S41, for any measurement point q i in the grinding area , based on the Newton iteration method, a point p k is found on the implicit surface expression g of the non-grinding area k , so that p i is closest to q i in the Euclidean distance.

[0098] Specifically, first assume Then update iteratively: Until

[0099] S42, as shown in Figure 2 , the signed distance of the measurement point q i to the implicit surface is defined as sdf i = sign × ||q i -p k ||,

[0100] wherein The obtained signed distance field sdf i is the grinding amount at the measuring point q. i is the grinding amount at the measuring point q.

[0101] In summary, the composite skin grinding amount calculation method based on cluster analysis and signed distance field provided by the present application discards the traditional iterative matching or projection calculation process, realizes the "one-step" direct output of the grinding amount, can accurately analyze the grinding allowance of the aircraft skin repair area, and guides the aircraft maintenance personnel to perform grinding. Compared with the traditional method based on manual touch to judge the profile, the present method eliminates the interference of human subjective factors, quantifies the grinding allowance, and has higher precision.

[0102] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for calculating composite skin grinding loss based on cluster analysis and signed distance field, characterized by: The following steps are involved: S1. Obtain three-dimensional measurement data of the composite skin surface, and define a feature descriptor for each three-dimensional measurement point based on the obtained measurement data; S2. Based on the feature descriptor of each 3D measuring point, a cluster analysis algorithm is designed to distinguish the 3D measuring points in the grinding area and the 3D measuring points in the non-grinding area in the multi-dimensional space; S3, for the three-dimensional measurement points in the non-grinding area, implicit surfaces are constructed based on radial basis functions; S4. For the three-dimensional measuring points in the grinding area, a signed distance field of the grinding area is constructed, and the signed distance from each three-dimensional measuring point in the grinding area to the implicit surface is calculated, that is, the grinding amount at the measuring point is obtained.

2. The method for calculating composite skin grinding amount based on cluster analysis and signed distance field according to claim 1, characterized in that: The feature descriptor of each 3D measurement point in step S1 is specifically defined as: A six-dimensional feature vector, denoted as F, including curvature, normal direction angle, local height difference, point cloud color intensity, density weight and distance weight; Among them, the curvature represents the degree of curvature of the local surface, the normal direction angle represents the angle between the normal of the current measuring point and the average normal of the neighborhood, the local height difference represents the height difference between the current measuring point and its neighborhood points, the point cloud color intensity represents the RGB component of the point cloud of the current measuring point, the density weight represents the local point cloud density around the current measuring point, and the distance weight represents the distance from the current measuring point to the fitting plane of its neighborhood.

3. The method for calculating composite skin grinding loss based on cluster analysis and signed distance field according to claim 2, characterized in that: The curvature characterizes the degree of curvature of the local surface, that is, the minimum eigenvalue of the covariance matrix of the neighborhood points of the current measuring point calculated based on principal component analysis (PCA), and then the normalized value.

4. The method for calculating composite skin grinding loss based on cluster analysis and signed distance field according to claim 1, characterized in that: Step S2 specifically includes: S21. Normalization and weighting of feature descriptors: The six-dimensional feature vectors in the feature descriptor F are Min-Max normalized to eliminate the dimension difference and obtain the Min-Max normalized feature descriptor F. The weight coefficient is dynamically assigned to the features of each dimension according to the importance. The weight coefficient of the feature of the kth dimension is w k ,k∈{1,2,...,6}; S22. Define multi-dimensional distance metrics: The formula is: Among them, r i 、r j represent any i-th and j-th three-dimensional measurement points, respectively, f k (.) is the feature of the k-th dimension after Min-Max normalization, D(r i , r j ) represents r i 、r j Multi-dimensional distance measurement between; S23. Set clustering parameters: This includes setting the clustering threshold ∈ to N times the resolution of the three-dimensional laser sensor and setting the neighborhood expected point number threshold minPts; S24. Clustering execution and category determination: Calculate the multi-dimensional distance value between all three-dimensional measurement points. If the distance value is within the threshold range, they are clustered into one category. If the number of three-dimensional measurement points clustered into one category is ≥ minPts, calculate the mean of the Min-Max normalized feature descriptors of all measurement points in the category, which is recorded as Set to The virtual measurement points of the feature descriptor are used as new cluster centers, and the points in the remaining categories whose number of points is less than minPts are clustered again. This process is iterated multiple times until the clustering results are finally divided into the grinding area 3D measurement point class and the non-grinding area 3D measurement point class, and the category labels are added as category 1 and category 0 respectively.

5. The method for calculating composite skin grinding loss based on cluster analysis and signed distance field according to claim 4, characterized in that: In step S24 , if there are still three-dimensional measurement points that are not classified into category 1 or category 0 after multiple iterations, they are marked as noise and discarded.

6. The method for calculating composite skin grinding loss based on cluster analysis and signed distance field according to claim 1, characterized in that: Step S3 specifically includes: S31, point cloud data in non-grinding area and normal vector information M is the number of points in the non-grinding area, and L points are randomly sampled as the center points of the radial basis function to obtain the center point set of the radial basis function. S32, Based on the radial basis function implicit surface modeling principle, for any point p in P i , must satisfy g(p i )=0, where: Among them, g(p i ) is p i The implicit curve equation at , γ is the value set by the hyperparameter, is the coefficient to be solved; S33. Construct a system of linear equations: S331、 S332. To ensure the uniqueness of the solution, the orthogonality condition must be met: S34, converting the linear equations in step S33 into an ill-conditioned matrix form to obtain: Where Ψ={Ψ ij }, i∈{1, 2, ..., M}, j∈{1, 2, ..., L}, represents the element in row i and column j of the Ψ matrix; S35. Add a regularization term to the obtained ill-conditioned matrix and modify the equation to: (Ψ+λI)α+Pβ=0, Where λ is the designed value of the hyperparameter and I is the identity matrix; S36. Based on the singular value decomposition, the coefficients α and β in the equation are obtained to determine the implicit surface expression of the non-grinding area.

7. The method for calculating composite skin grinding loss based on cluster analysis and signed distance field according to claim 1, characterized in that: Step S4 specifically includes: S41, for grinding area Any measuring point q i , based on the Newton iteration method, find a point p on the implicit surface expression g in the non-grinding area k , so that p k With q i The European style is the closest; Specifically, assume Then iterate and update: until S42, define measuring point q i The signed distance to the implicit surface is sdf i =sign×‖q i -p k ‖, in, The resulting signed distance sdf i That is the measuring point q i The amount of grinding.

Images