A Propeller Blade Quality Assessment Method Based on Regional Differences

Abstract

This invention discloses a method for quality assessment of propeller blades based on regional difference partitioning. The steps are as follows: triangulate the initial surface of the propeller CAD model and extract the vertex set and its neighborhood feature set; analyze the geometric topology information of the vertex set's neighborhood to calculate vertex curvature features and extract seed point sets; calculate the surface vertex feature voting attributes based on the vertex set; perform regional energy clustering segmentation based on feature voting based on the seed point set; plan local measurement points based on the surface measurement area; perform chord tolerance sampling on each measurement point set to extract the locally optimal registration point set; perform nonlinear region accuracy hierarchical registration based on the locally optimal measurement point set and the vertex set; perform quality assessment based on propeller measurement standards according to the registration results; and output processing guidance data in a tabular format based on the quality assessment results. This invention solves the problems of insufficient measurement point description, low registration accuracy, and poor adaptability in the current propeller blade surface quality assessment stage caused by regional surface shape differences.

Description

Technical Field

[0001] This invention relates to the quality assessment of propeller blades, and in particular to a method for assessing the quality of propeller blades based on regional differences. Background Technology

[0002] With the rapid development of the shipbuilding industry, the manufacturing of high-quality marine equipment components has received increasing attention. Propellers, as representatives of complex and multifaceted parts, are widely used in various engineering industries, and a fast, stable, and accurate quality assessment method is crucial to determining the manufacturing quality of their blades. Quality assessment consists of two key technologies: measurement point planning and registration assessment. Currently, the quality assessment of propeller blade data in the industry still requires manual intervention, resulting in accuracy fluctuations and insufficient measurement point descriptions. Some institutions have conducted technical research on integrated measurement point planning and data registration to address this issue, but the problem remains unresolved.

[0003] The prior art that is similar to this invention includes: (1) Patent application number CN202011370807.6 proposes "A method for evaluating the quality of propeller blades based on surface registration", which takes the intersection of the stacking radius and the radial line as the result of the measurement point layout and combines data registration to achieve quality evaluation, thus solving the problem of evaluation accuracy caused by human factors. However, the method is based on the overall parameters of the propeller blade surface and does not consider the differences in the surface shape of the propeller. The description of the measurement points in the early planning is low and the subsequent registration accuracy is difficult to guarantee. (2) Patent application number CN201410821784.4 proposes "A method for measuring propeller blades", which collects measurement data by installing a steering gear on the pitch gauge, converts the data into a coordinate point format, and then backfits the surface mesh and compares it with the theoretical model to output the quality evaluation result, which improves the measurement accuracy and the device is also flexible. However, the measurement point planning still relies on the traditional method and there is still a problem with the measurement point description, which affects the subsequent quality evaluation. (3) The patent application number CN201910618649.2 proposed a method for evaluating the surface profile of propeller blades based on matrix rank operation. It calculates whether the measuring points are located in the contour error envelope plane after the measurement points are transformed and aligned with the theoretical model to determine the error information. It has high accuracy and strong observability, but it has low computational efficiency and the measurement point planning still does not consider the multi-faceted problem. It is very important to propose a high-precision quality evaluation method suitable for multi-faceted surfaces of propeller blades by combining measurement point planning and data registration. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a propeller blade quality assessment method based on regional difference division, thereby solving the problems of insufficient measurement point description, low registration accuracy and poor adaptability in the current propeller blade surface quality assessment stage caused by regional surface shape differences.

[0005] Technical solution: The present invention provides a method for evaluating the quality of propeller blades based on regional differences, comprising the following steps:

[0006] (1) Triangulate the surface of the initial propeller CAD model and extract the vertex set and its neighborhood feature set;

[0007] (2) Analyze the geometric topology information of the vertex set neighborhood, calculate the vertex curvature features, and extract the seed point set;

[0008] (3) Calculate the voting attributes of surface vertex features based on the vertex set;

[0009] (4) Perform region energy clustering segmentation based on feature voting according to the seed point set;

[0010] (5) Plan local measurement points based on the surface measurement area;

[0011] (6) Extract the local optimal registration point set by chord tolerance sampling for each measurement point set;

[0012] (7) Perform nonlinear region accuracy level registration based on the local optimal measurement point set and vertex set;

[0013] (8) Based on the registration results, perform a quality assessment using propeller measurement standards;

[0014] (9) Output the processing guidance data in tabular form based on the quality assessment results.

[0015] The specific steps (1) are as follows:

[0016] (1.1) Input the propeller CAD model, analyze the topological relationship between the model's geometric features and core parameters, generate a high-precision NURBS mesh for the surface, and then triangulate the mesh to generate a high-precision triangular mesh for the surface.

[0017] (1.2) Based on the high-precision triangular mesh of the propeller surface obtained, extract the mesh nodes and the set of neighboring triangular facets as the vertex set P and feature set E.

[0018] Step (2) specifically involves:

[0019] (2.1) Analyze the geometric topological relationships between each vertex in the vertex set P and its neighboring points, and calculate the target vertex p. i Covariance matrix C:

[0020]

[0021] In the formula, the target point p i ∈P, p G The centroids of the neighborhood points in the target point m are obtained by decomposing the covariance matrix C, and the equation is:

[0022] Cα=ξα (2)

[0023] The three eigenvalues ​​ξ1, ξ2, ξ3 and the corresponding eigenvectors α1, α2, α3 can be obtained by using the above equation (2). By comparing the three eigenvalues, the unit vector in the direction of the eigenvector corresponding to the smallest eigenvalue is taken as the normal vector N.

[0024] (2.2) Based on the normal vector N obtained in step (2.1), the target point p is... i The average curvature is obtained by representing the surface near the target point as s = s(u,v), and the average curvature is expressed as:

[0025]

[0026] In the formula, the parameters are passed through the first matrix. With the second matrix This indicates that N is the normal vector at that point, and the characteristic curvature H at that point can be further obtained through the average curvature. s :

[0027]

[0028] Traverse all vertices in the vertex set P and calculate the characteristic curvature of the vertices using equation (4);

[0029] (2.3) Based on the characteristic curvature H of each vertex obtained in step (2.2) s Traverse the vertex set P, sort the vertex set by gradient according to the curvature, and then extract the first n vertices with larger feature curvature values ​​as the seed set to provide an initial domain reference for subsequent surface clustering and segmentation.

[0030] Step (3) specifically involves:

[0031] (3.1) Analyze the geometric and topological relationship between the vertex set P and the neighborhood feature set E. Integrate the vertex and neighborhood feature sets, calculate the vertex feature voting matrix as a reference for subsequent surface measurement region clustering and segmentation. Define the feature voting attribute of the surface vertices as:

[0032]

[0033] Where T(i) is the triangle facet where the current vertex is located, and n T(i) It is the unit normal vector corresponding to the triangle, and the matrix. And coefficient Area A area(t) It is the sum of the areas of all triangular faces within the neighborhood of the current vertex, A. area(max) Let z represent the triangle with the maximum area in the neighborhood, σ represent the minimum bounding box side length of the target point's neighborhood, and z represent the triangle with the maximum area in the neighborhood. * T(i)It is the centroid of the triangular face T(i);

[0034] (3.2) According to the feature voting attribute calculation method described in formula (5) in step (3.1), traverse the vertex set and assign feature voting attribute values ​​to each vertex to provide feature clustering reference for subsequent surface measurement region segmentation.

[0035] Step (4) specifically involves:

[0036] (4.1) Based on the point set P and neighborhood feature set E generated in step (2), the vertices and neighborhood features contained in the seed point set are used as the initial growth domain M for surface segmentation. i (i = 1...n), in the initial state, other vertex sets P and feature sets E have no home region;

[0037] (4.2) Based on the initial growth domain partitioning state generated in step (4.1), perform the clustering segmentation of vertex set P and feature set E. Traverse vertex set P, skipping if the current vertex is a seed point; if the currently traversed vertex has a neighboring point belonging to the initial growth domain M. i If a vertex and its feature set are assigned to a given domain, then that vertex and its feature set are assigned to that domain. If multiple neighboring vertices of the currently traversed vertex have a domain or none have a domain, the domain assignment is determined based on the energy clustering model under feature voting. By calculating the energy function model result, the vertex is assigned to the domain where the energy value is minimum. After traversing all vertices and determining their domains, the initial surface measurement region clustering and segmentation are completed. The energy clustering model function is defined as follows:

[0038]

[0039] In the formula, E i It is an energy model function, T j It is the triangular facet where the current calculation point is located, s j The area of ​​the triangular facet is obtained through Calculate ρ(x), where ρ(x) is a constant function. Decompose the characteristic voting matrix (5) to obtain the eigenvalues ​​ξ. K1 ,ξ K2 ,ξ K3 V i =(1,ξ K2 / ξ K1 ,ξ K3 / ξ K1 () is a vector obtained by solving for normalized eigenvalues;

[0040] (4.3) Based on the initial surface measurement region clustering and segmentation results in step (4.2), traverse all vertices and adjust the neighboring domain M according to equation (6). i and M jThe vertex and feature assignment of {i,j=1,...n and i≠j} are updated by minimizing the region energy function value;

[0041] (4.4) If the attribution judgment in (4.3) does not change, the function converges and outputs the surface measurement region after clustering and segmentation; if the vertex and feature are updated after the energy value is minimized according to equation (6), repeat step (4.2) to update the vertex and feature attribution according to the case of minimum energy value, until the function converges and the model does not produce any attribution changes, and outputs the segmentation result to generate each local measurement region.

[0042] Step (5) specifically involves:

[0043] (5.1) Generate local spatial grid stacking lines based on the divided measurement area:

[0044]

[0045] In the formula, N i,p (u) and N j,p (v) are the spline basis functions that make up the surface mesh, p ij These are spline control nodes, formed by staggered stacking of surface spline basis functions at equal intervals from both u and v directions to create a spatial mesh.

[0046]

[0047] In the formula, n is the number of grid stack lines, and R i,n (t) is the comprehensive expression of spline basis functions, and finally the spatial approximation grid of the local measurement area is formed by equation (8);

[0048] (5.2) Map the spatial mesh onto the surface of the propeller theoretical model. The intersection of the mesh nodes and the surface projection is the local measurement point. Piece together the local measurement points to form the initial measurement point set P. o .

[0049] Step (6) specifically involves:

[0050] (6.1) Calculate the chord tolerance based on the local measurement points generated in step (5). Referring to the local surface measurement areas generated in step (4), calculate the chord tolerance on each measurement point generated by the spatial spline mesh projection in the local area. Then calculate the mean chord tolerance of the local measurement point set. Arrange the local measurement areas in ascending order according to the mean chord tolerance result. The chord tolerance calculation is as follows:

[0051]

[0052] In the formula, the node sequence U = {μ} i}, B(μ) is the spline basis function in equation (7), and ε is the chord tolerance value.

[0053] (6.2) Based on the results of sorting the measurement points in ascending order of the mean chord tolerance in step (6.1), select the set of measurement points in the local measurement area with the smallest mean chord tolerance as the optimal set of measurement points P. m .

[0054] The specific steps (7) are as follows:

[0055] (7.1) Based on the optimal measurement point set P in step (6) m As the point cloud to be registered, the vertex set P extracted in step (1) is used as the target point cloud, and a registration model E(R,t) is established, where R and t are the rotation matrix and the translation matrix, respectively.

[0056] (7.2) First, the registration model is transformed into a problem of solving the registration transformation parameters between the optimal measurement point set and the vertex set:

[0057]

[0058] Where n is the number of point clouds to be registered, p i Let R be the transformation point, and let R = (α, β, γ) be the three rotational degrees of freedom in space, and t = (t x ,t y ,t z ) represents the three degrees of freedom of spatial displacement, a = (α, β, γ, t) x ,t y ,t z Equation (10) is obtained by substituting... It can be transformed into:

[0059] E(a+u)=e T e+u T J T e+u T J T Ju (11)

[0060] Where J = ▽e, step size u = -(J T J) -1 J T After repeatedly optimizing the step size and iterating to solve for R and t in objective a, and updating u, E(a+u) can be further transformed into:

[0061]

[0062] The coefficient λ is iteratively calculated in real time, involving the gradient descent weight λ / (λ+1) and the Newton's method weight 1 / (λ+1). When λ = 1, the function performs both calculations simultaneously. When this coefficient is small, the descent weight dominates, with gradient descent taking precedence, and coefficient α approaches the optimal solution. When this coefficient is large, the Newton's method weight dominates, and coefficient α moves away from the optimal solution, with Newton's method taking precedence. This approach helps the function converge quickly while avoiding local optima during iteration. An error threshold ω is set. If the current calculation result is better than the previous one, an error threshold ω is used to determine whether to output the result. If it is not better than the previous result, the step size is updated, and the nonlinear calculation steps are repeated iteratively until the result is better than the previous one. When the final calculation result meets the error threshold ω, the iteration is complete, and the registration pose transformation result E is obtained. T (R,t);

[0063] (7.3) Obtain the registration pose transformation result E according to step (7.2). T (R,t), for the initial set of measurement points P o The registration result P is obtained based on this matrix transformation. t The registration transformation can be expressed as:

[0064] P t =P o ·E T (R,t)=R·P o +t (13)

[0065] In the formula, R and t are the rotation matrix and the translation matrix, respectively, and E T (R,t) is the set of key matrices for pose transformation, P o It is the initial set of measurement points, P t The point set after pose transformation. Step (8) specifically includes:

[0066] (8.1) Based on the point set P after pose transformation t The root mean square error (RMS) between the vertex set P and the model error distance (DRMS) of a single measurement point are used as quality assessment criteria.

[0067]

[0068] Where, P(x,y,z) t Let P(x,y,z) be the point set P after pose transformation. t The three-dimensional coordinates of a point in the vertex set P are given by dist(P(x,y,z)). t -P(x,y,z)) represents the distance between points.

[0069] (8.2) Referring to the surface accuracy requirements of propeller blades in GBT 12916-2010 Technical Conditions for Marine Metal Propellers, the overall measurement point accuracy error is required to be less than 0.5mm. Compare with the results in (8.1) and output the comparison judgment results. First, evaluate whether the overall blade error meets the standard, and then output the position coordinate information of the error position that is lower than the standard accuracy to complete the quality assessment.

[0070] The specific steps (9) are as follows: based on the quality assessment results, data is output for secondary processing guidance; for positions with lower-than-standard precision, error judgment is made, and the positive and negative values ​​and magnitudes of the processing error are tabulated for subsequent processing guidance.

[0071] Beneficial effects: Compared with existing technologies, this paper has the following advantages:

[0072] 1. A region energy clustering segmentation method based on feature voting is proposed to segment the measurement region and to plan measurement points based on the local measurement region. This solves the problem that the existing measurement point planning methods have insufficient descriptiveness due to the concentration of multifaceted features on the surface of propeller blades. At the same time, it provides a technical reference for the segmentation of complex surfaces.

[0073] 2. The planned measurement points are not only highly descriptive, but also have few redundant measurement points. This improves the ability to describe measurement points while optimizing actual measurement efficiency, achieving a two-way balance between accuracy and efficiency.

[0074] 3. By calculating the mean chordal tolerance of the measurement area, the set of measurement points in the area with the smallest mean chordal tolerance is registered with the vertex set in a nonlinear region accuracy hierarchy. This solves the problem of error distribution effect in the application of existing registration methods for propeller blade quality assessment and achieves high-precision data registration.

[0075] 4. The registration problem is optimized by nonlinear methods, which achieves higher accuracy and faster calculation speed. Nonlinear calculation also has anti-interference properties, thus solving the problem of initial pose interference in registration.

[0076] 5. It has strong visualization capabilities for quality assessment results and provides follow-up guidance, solving the problem of poor observability of measurement results in manual quality assessment. Attached Figure Description

[0077] Figure 1 This is a flowchart illustrating the overall quality assessment process described in this invention.

[0078] Figure 2 This is a schematic diagram of the surface mesh of the model described in this invention;

[0079] Figure 3 This is a flowchart of the vertex seeding process described in this invention;

[0080] Figure 4This is a flowchart of the surface clustering and segmentation process described in this invention;

[0081] Figure 5 This is a diagram showing the attribution of surface features as described in this invention;

[0082] Figure 6 This is a schematic diagram of the local measurement point results described in this invention;

[0083] Figure 7 This is a schematic diagram of the planning and measurement point results described in this invention;

[0084] Figure 8 This is a flowchart of the nonlinear region accuracy level registration process described in this invention. Detailed Implementation

[0085] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0086] like Figure 1 As shown, this invention provides a method for evaluating the quality of propeller blades based on regional difference partitioning. First, a 3D CAD model of the propeller blade to be evaluated is input. The topological correlation and geometric features of the model's surface parameters are analyzed, a high-precision NURBS mesh is generated and triangulated, and then the vertex set and feature set of the mesh are extracted. The feature information in the vertex set and its neighboring area is analyzed, vertex curvature features are calculated, and a seed point set is extracted to provide an initial domain reference for subsequent segmentation. Vertex feature voting attributes are calculated, and values ​​are assigned sequentially to each vertex as a clustering segmentation reference. Based on the feature voting decomposition normalization results and the seed point set, a regional energy model is established and an initial growth domain is generated, realizing regional energy clustering driven by feature voting. The process involves: dividing the surface measurement area; generating a spatial grid model for each local measurement area, mapping it to form local measurement points, and stitching them together to generate an initial set of measurement points; calculating the mean chord tolerance of each local measurement point set according to the chord tolerance criterion, and selecting the measurement points in the measurement area with the smallest mean chord tolerance as the optimal set of measurement points; solving the registration problem using a nonlinear method based on the generated optimal set of measurement points to achieve hierarchical registration of nonlinear regions and obtain the registration results; obtaining error information based on the registration results and comparing it with the national standard propeller technical requirements to output the quality assessment results; finally, generating a data table based on the quality assessment results to guide secondary processing. The specific steps are as follows:

[0087] Step 1: Input the 3D CAD model of the propeller blade, solve the surface NURBS mesh, then triangulate the model mesh and analyze and extract the vertex set and neighborhood feature set;

[0088] As a multifaceted and complex surface part, the propeller model can retain its geometric features and parameter information well by triangulating the surface with a high-precision NURBS mesh. At the same time, the triangulated mesh data has the advantages of lightweight data and wide interface applicability, making it very suitable for propeller model description.

[0089] Step 1.1: Export the propeller CAD model and perform surface parameter analysis to calculate and generate a high-precision NURBS surface mesh, and then triangulate the mesh to generate a high-precision surface triangular mesh.

[0090] Step 1.2: Since the surface triangular mesh is composed of nodes and triangular faces, based on the obtained... Figure 2 A high-precision triangular mesh is obtained from the propeller surface. The geometric parameters of the mesh nodes and other triangular facets in the neighborhood, excluding the nodes, are extracted and used as the vertex set P and the neighborhood feature set E.

[0091] Step 2: Based on the vertex set P and neighborhood feature set E obtained in Step 1, calculate the vertex curvature features and obtain the seed point set under gradient sorting.

[0092] In triangular mesh representations, the vertex set and neighborhood features possess good parametric properties, while the selection of seed points directly affects the quality of subsequent surface segmentation. Areas with prominent surface geometric features often exhibit large feature curvature; vertices at these locations are highly discriminative and representative, making them suitable as seed points for surface clustering and segmentation, guiding the determination of initial domains. By analyzing the topological relationships between vertices and calculating feature curvature, the most prominent points are selected as the seed point set, guiding the production of the optimal initial domains for subsequent surface segmentation.

[0093] Specifically, it includes:

[0094] Step 2.1: Based on the vertex set P, analyze the topological properties of the vertices and use the characteristic curvature to describe the vertex features as the basis for selecting the seed point. Calculate the target point p. i The covariance matrix in the neighborhood yields the vertex feature normal N, and the covariance matrix C is defined as follows:

[0095]

[0096] Wherein, the current target point is p. i ∈P, p G The centroid of the parametric neighborhood formed by m neighboring points near the target point is called the m-neighbor centroid. The equation is obtained by decomposing the covariance matrix C:

[0097] Cα=ξα (2)

[0098] The covariance matrix is ​​a positive semi-definite matrix. By decomposing it, we can obtain three eigenvalues ​​ξ1, ξ2, ξ3 and the corresponding eigenvectors α1, α2, α3. We select the unit normal vector in the direction of the eigenvector corresponding to the smallest eigenvalue as the normal vector N of the target vertex.

[0099] Step 2.2: Propeller blades are typical complex surfaces with significant variations in surface curvature. Based on this property, the curvature of each vertex is calculated as the basis for selecting the seed point set required for subsequent surface segmentation. First, based on the normal vector N of the surface vertices obtained in Step 2.1, the target point p is solved. i The characteristic curvature is calculated after averaging the curvature. Let the target point be p. i The nearby surface is s = s(u, v), and the mean curvature can be represented by two matrix parameters:

[0100]

[0101] Wherein, the first parameter matrix is matrix The second type of parameter matrix is ​​N, which is the normal vector of the target point obtained in step 2.1. The curvature feature H in the m-neighborhood of the target point is then obtained by using the average curvature. s :

[0102]

[0103] Step 2.3: Based on Step 2.2, calculate the feature curvature of all vertices in the vertex set P, then sort them by gradient according to the curvature magnitude. Select the top n vertices with larger curvature features from the sorted set as the seed point set, providing an initial reference for the number of partitioned regions for subsequent surface segmentation. The process is as follows: Figure 3 .

[0104] Step 3: Perform neighborhood triangular facet parameter analysis based on the vertex set described in Step 1, calculate the surface feature voting attribute based on the vertex and neighborhood parameters, and assign values ​​to each vertex.

[0105] Feature voting attributes have good feature description ability and feature preservation stability in the model's geometric space. This method can effectively distinguish the vertex differences between different face shapes, providing a good reference for similar attribute clustering for subsequent surface clustering and segmentation.

[0106] Step 3.1: Analyze the vertex set P and neighborhood feature set E obtained in Step 1, and combine this with the target vertex p. i The target vertex feature is calculated by voting on the properties of the neighboring triangles, and the feature voting matrix is ​​defined as follows:

[0107]

[0108] Where T(i) represents the target vertex p i Triangular facets in the neighborhood, n T(i) It is the normal vector on the triangular patch of the neighborhood of the target point, and the matrix. And coefficient Area parameter A area(t) The parameter A represents the total area of ​​all triangular faces within the neighborhood of the target point.area(max) Let z represent the triangle with the maximum area among all triangles in the neighborhood, σ represent the minimum bounding box side length formed by the neighborhood of the target point, and z represent the triangle with the maximum area among all triangles in the neighborhood. * T(i) It is the centroid of the triangular face T(i) currently being calculated.

[0109] Step 3.2: Based on the feature voting matrix defined in Step 3.1, calculate the feature voting description value for each vertex and assign the retained result as a clustering reference for subsequent region energy segmentation.

[0110] Step 4: Use the feature votes obtained in Step 3 as the basis for clustering, and perform region energy clustering segmentation under feature voting to divide the measurement area;

[0111] Propeller blades have complex surface shapes with concentrated polyhedral features and significant differences in characteristics between different facets. Segmenting the measurement area before subsequent processing not only improves the accuracy and descriptive power of subsequent surface measurement points but also provides an optimized reference for subsequent registration processing, effectively improving the quality assessment results. The segmentation steps are as follows: Figure 4 .

[0112] Step 4.1: Based on the n-point set obtained in Step 2, use the vertices and neighborhood features contained in it as the initial number of growing regions M. i (i = 1...n), at this time, the vertex set P of the surface excluding the seed point and the neighborhood feature set E do not belong to any initial domain.

[0113] Step 4.2: Based on the initial domain set in Step 4.1, divide the feature set E (excluding the seed point) and the vertex set P into their respective domains. Traverse all vertices; if a point is a seed point, skip the domain determination; if the currently traversed point has a single neighboring point, assign it to M. i If the traversed point has no assigned abode, then the point is assigned to M. i If a traversed point has multiple neighboring points with assigned affiliations or none with assigned affiliations, the region affiliation of that point is determined using the region energy clustering model based on feature voting. The affiliation of a point depends on the calculated value of the energy function; if a point is within a region, the point with the lowest energy function value is assigned to that region. After traversing all vertices and determining their affiliations, the initial surface clustering segmentation is complete. The energy function is defined as:

[0114]

[0115] Among them, E i It is the solution value of the energy model function, T j It is the triangle face where the current traversal point is located, s j The area of ​​the triangular facet is obtained through It is found that ρ(x) is usually a constant function and can be regarded as a constant. The characteristic voting matrix obtained in step 3.1 is a positive semi-definite matrix, and its eigenvalues ​​ξ can be obtained through matrix decomposition. K1 ,ξ K2 ,ξ K3 , and V i =(1,ξ K2 / ξ K1 ,ξ K3 / ξ K1 ) is a vector obtained by normalizing the eigenvalues ​​obtained through decomposition.

[0116] Step 4.3: Determine the energy function value based on the preliminary division results in Step 4.2, such as... Figure 5 The energy function is solved for vertex P and feature set E, and the case with the minimum energy value is selected for the neighboring region M. i and M j Determine the attribution of vertices and their neighborhood features in {i,j=1,...n and i≠j} to achieve the update of the attribution of surface vertices and feature sets.

[0117] Step 4.4: Analyze the judgment results in Step 4.3. If the surface assignment in Step 4.3 no longer changes, the assignment judgment is completed, and the segmented surface measurement region is output. If the surface features continue to generate assignment updates at this time, repeat Step 4.2 until the function value converges and no longer changes, output the surface measurement region segmentation result, and generate each local measurement region.

[0118] Step 5: Plan local measurement points based on the measurement area divided in Step 4;

[0119] Measurement point layout planning is an important technique for quality assessment. Its results directly affect subsequent registration processing and influence the overall quality assessment results. As a typical component with a concentrated multifaceted surface, the propeller can have better surface description properties for the measurement points through measurement area division and measurement point layout planning, making the actual measurement results more reliable.

[0120] Step 5.1. Based on the surface measurement area results divided in Step 4.4, spatial mesh stacking is performed on the local measurement area by generating spatial mesh stacking lines. The spatial mesh stacking lines are defined as follows:

[0121]

[0122] Where, N i,p (u) and N j,p (v) are the spline basis functions that make up the surface space mesh, p ij These are the control nodes for the surface mesh spline stacking lines. The desired mesh spline stacking lines are arranged equidistantly and staggered in both the u and v directions to form a spatial mesh for the local measurement area.

[0123]

[0124] Where n is the number of stacked lines, and this coefficient determines the mesh density, R i,n (t) is the comprehensive expression of spline functions. It forms an approximate grid for the local measurement area by stacking n numbers of splines in space according to an equidistant distribution.

[0125] Step 5.2: Obtain surface mesh B according to 5.1 w (u w ,v w Measurement point generation is performed by mapping the spatial grid of the local measurement area onto the surface of the blade to be measured, such as... Figure 6 The intersection of the mesh nodes and the surface projections forms local measurement points, generating a set of local measurement points. These local measurement point sets are then stitched together to form the final measurement point set. Figure 7 The initial measurement point set P on the propeller surface o .

[0126] Step 6: Calculate the chord tolerance of the local measurement point set mentioned in Step 5, and extract the measurement point set of the local measurement area with the minimum mean chord tolerance as the local optimal registration point set;

[0127] Data registration is a core technology in quality assessment, and its results directly determine the quality of the assessment. Due to the concentrated polyhedral shape of the propeller surface, there are differences in the accuracy of measuring points between measurement areas. The chord tolerance criterion can effectively reflect the accuracy of each measuring point. Extracting the measuring points in the measurement area with the smallest mean chord tolerance as the locally optimal registration point set effectively avoids the error distribution phenomenon in overall registration caused by different blade shape parameters in different measurement areas of the surface, thus helping to improve the accuracy of measurement quality assessment.

[0128] Step 6.1: Based on the local measurement point set obtained in Step 5, traverse all measurement points and calculate the chord tolerance for all measurement points in the point set according to the chord tolerance criterion.

[0129]

[0130] Among them, U=[0,0,0,0,μ4,...,μ n [,1,1,1,1] is the node sequence, B(μ i ) is the spline basis function obtained in step 5.1, and ε is the chord tolerance value. The chord tolerance values ​​of each measuring point are calculated and assigned and saved in sequence.

[0131] Step 6.2: Based on the chord tolerance values ​​of each measuring point obtained in Step 6.1, calculate the mean chord tolerance value of each measurement area containing the measuring points, and select the set of measuring points in the local measurement area with the smallest local average chord tolerance value as the local optimal registration point set P. m .

[0132] Step 7: Based on the locally optimal registration point set P obtained in Step 6 m Using the vertex set P obtained in step 1, the registration problem is calculated using a nonlinear method. The pose transformation matrix is ​​solved to align the two point clouds, achieving hierarchical registration with nonlinear region accuracy.

[0133] The essence of quality assessment is to evaluate the registration results between the measurement point set and the vertex set and output error information. The computational efficiency and accuracy of registration are crucial. Using nonlinear solution methods to solve the registration problem can reduce registration calculation errors, improve registration solution efficiency, and optimize the quality assessment results.

[0134] Step 7.1: Based on the results in Step 6, set the optimal measurement point set P. m Using the vertex set P as the point cloud to be registered, establish an initial registration model E(R,t), where R and t are the rotation matrix and the translation matrix, respectively.

[0135] Step 7.2: Based on the initial registration model described in Step 7.1, transform the model into the optimal measurement point set P. m Solving the registration transformation parameters between vertex set P and vertex set P:

[0136]

[0137] Where n is the number of point clouds to be registered, generally referring to the optimal measurement point set, and p i For points in the measurement point set, the rotation matrix R = (α, β, γ) represents the rotational degrees of freedom of the three coordinate axes in three-dimensional space, and t = (t x ,t y ,t z ) represents the displacement degrees of freedom on the three coordinate axes in three-dimensional space, and the coefficient a = (α, β, γ, t) x ,t y ,t z This integrates the six degrees of freedom into a single matrix representation by merging all point problems in the point set into a single matrix. Substituting the values ​​and performing a Taylor expansion, we get:

[0138] E(a+u)=e T e+u T J T e+u T J T Ju (11)

[0139] Where J = ▽e, and the forward step length u = -(J T J) -1 J TLet R and t in objective a be solved by step-size optimization through nonlinear optimization iteration. By further optimizing and updating u, E(a+u) can be transformed into:

[0140]

[0141] The coefficient λ is used to correct the function iteration process in real time, preventing it from getting trapped in local optima. The gradient descent weight λ / (λ+1) and the Newton's method weight 1 / (λ+1) are calculated. When λ = 1, the function performs both calculations simultaneously. When the coefficient value is small, the function calculation mainly uses gradient descent, and the coefficient 'a' is close to the optimal solution, allowing for further calculation of the final result. Conversely, when the target 'a' is far from the optimal solution, the function mainly uses Newton's method, preventing local optima from appearing in the iterative calculation and enabling the function to converge quickly. Let the error threshold be ω. During function calculation, if the current result is better than the previous one, an error threshold check is performed to determine whether to output the current result; otherwise, the step size is updated, and the above steps are repeated for step size update iterative calculation. When the function iteration calculation is completed and the error threshold ω is satisfied, the registration pose transformation model E is obtained. T (R,t).

[0142] Step 7.3: Based on the registration pose transformation model E obtained in Step 7.2 T (R,t), for the initial set of measurement points P o Perform pose alignment transformation to obtain the final registration result P. t For subsequent quality assessment, the registration transformation can be expressed as:

[0143] P t =P o ·E T (R,t)=R·P o +t (13)

[0144] In the formula, R and t are the rotation matrix and the translation matrix, respectively, and E T (R,t) is the set of key matrices for pose transformation, P o It is the initial set of measurement points, P t Point set after pose transformation.

[0145] Step 8: Based on the registration results described in Step 7, perform a quality assessment using propeller measurement standards, and output the error analysis results. The registration process is as follows: Figure 8 ;

[0146] Step 8.1: Based on the final registration result P obtained in Step 7 t The root mean square error (RMS) and the single measurement point error (DRMS) between the vertex set P and the vertex set P are calculated and used as the quality assessment result.

[0147]

[0148] In the formula, P(x,y,z) t Let P(x,y,z) be the point set P after pose transformation. t The three-dimensional coordinates of a point in the vertex set P are given by dist(P(x,y,z)). t -P(x,y,z)) represents the distance between points.

[0149] Step 8.2: Based on the results obtained in Step 8.1, and referring to the surface accuracy requirements for propeller blades in GB / T 12916-2010 Technical Conditions for Marine Metal Propellers, the overall measurement point accuracy error should be less than 0.5mm. A comparative judgment is then made. First, determine whether the overall error meets the standard requirements. Then, output the coordinate information of the error positions below the standard accuracy to complete the quality assessment.

[0150] Step 9: Output the processing guidance data in tabular form based on the quality assessment results described in Step 8;

[0151] Step 9.1: Based on the quality assessment results generated in Step 8.2, output the data for secondary processing guidance. For positions with lower-than-standard precision, make an error judgment and output the positive and negative values ​​and magnitudes of the processing allowance error in a tabular format for subsequent processing guidance.

[0152] In addition to proposing the above-mentioned quality assessment methods, it is also possible to analyze the surface morphology of the blades of different propeller models, set reasonable relevant measurement point planning and registration parameters, and realize the quality assessment of propeller blades of different models based on these parameters.

Claims

1. A method for evaluating the quality of propeller blades based on regional differences, characterized in that, Includes the following steps: (1) Triangulate the surface of the initial propeller CAD model and extract the vertex set and its neighborhood feature set; (2) Analyze the geometric topology information of the vertex set neighborhood to calculate the vertex curvature features and extract the seed point set; (3) Calculate the voting attributes of surface vertex features based on the vertex set; (4) Perform region energy clustering segmentation based on feature voting according to the seed point set; (5) Plan local measurement points based on the surface measurement area; (6) Extract the local optimal registration point set by chord tolerance sampling for each measurement point set; (7) Perform nonlinear region accuracy level registration based on the local optimal measurement point set and vertex set; (8) Based on the registration results, a quality assessment is performed using propeller measurement standards; (9) Output processing guidance data in tabular form based on the quality assessment results; Step (4) specifically involves: (4.1) Based on the point sets generated in step (2) With neighborhood feature set The vertices and neighborhood features contained in the seed point set are used as the initial growing domain for surface segmentation. , Other vertex sets in the initial state With feature set No domain; (4.2) Based on the initial growth domain partitioning state generated in step (4.1), perform vertex set... With feature set Clustering and segmentation based on ownership, traversing the vertex set. If the current vertex is a seed point, skip it; if the current vertex has a neighboring vertex belonging to the initial growth domain, skip it. If a vertex and its feature set are found to belong to a given domain, then the vertex and its feature set are assigned to that domain. If multiple neighboring vertices of the currently traversed vertex have a domain or none have a domain, the domain is determined based on the energy clustering model under feature voting. By calculating the energy function model result, the vertex is assigned to the domain where the energy value is minimized. After traversing all vertices and determining their domains, the initial surface measurement region clustering and segmentation are completed. The energy clustering model function is defined as follows: （6） In the formula, It is an energy model function. It is the triangular facet where the current calculation point is located. The area of ​​the triangular facet is obtained through calculate, It is a constant function; The eigenvalues ​​are obtained by decomposing the characteristic voting matrix (5). , It is a vector obtained by normalizing eigenvalues; (4.3) Based on the initial surface measurement region clustering and segmentation results in step (4.2), traverse all vertices and adjust the neighboring domains according to equation (6). and The vertex and feature assignments are updated by minimizing the region energy function value. (4.4) If the attribution judgment in (4.3) does not change, the function converges and outputs the surface measurement region after clustering and segmentation; if the vertex and feature are updated after the energy value is minimized according to equation (6), repeat step (4.2) to update the vertex and feature attribution according to the case of minimum energy value, until the function converges and the model does not produce any attribution changes, and outputs the segmentation result to generate each local measurement region.

2. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, The specific steps (1) are as follows: (1.1) Input the propeller CAD model, analyze the topological relationship between the model's geometric features and core parameters, generate a high-precision NURBS mesh for the surface, and then triangulate the mesh to generate a high-precision triangular mesh for the surface. (1.2) Based on the obtained high-precision triangular mesh of the propeller surface, extract the mesh nodes and the set of neighboring triangular faces as the vertex set. With feature set .

3. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, Step (2) specifically involves: (2.1) Analyze the vertex set Geometric topological relationships between each vertex and its neighboring vertices are used to calculate the target vertex. covariance matrix : （1） In the formula, the target point , The target point The centroid of a neighboring point in the neighborhood is determined by decomposing the covariance matrix. Equation obtained: （2） The three eigenvalues ​​can be obtained through the above equation (2). and corresponding feature vectors By comparing the three eigenvalues, the unit vector in the direction of the eigenvector corresponding to the smallest eigenvalue is taken as the normal vector. ; (2.2) Obtain the normal vector according to step (2.1). , target point The average curvature is obtained, and the surface near the target point is represented as... The mean curvature is expressed as: （3） In the formula, the parameters are passed through the first matrix. With the second matrix express, The normal vector at that point can be used to further determine the characteristic curvature of that point using the average curvature. : （4） Traversing the vertex set For all vertices in the equation, the characteristic curvature of each vertex is calculated using equation (4); (2.3) Based on the characteristic curvature of each vertex obtained in step (2.2) traversing the vertex set Based on the curvature magnitude, sort the vertex set by gradient, and then extract the previous vertex sets sequentially. For vertices with larger feature curvature values, a seed point set is generated to provide an initial domain reference for subsequent surface clustering and segmentation.

4. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, Step (3) specifically involves: (3.1) Analyze the vertex set With neighborhood feature set Based on the geometric topological relationships between vertices and their neighborhoods, and combining the feature sets of vertices and their neighbors, a vertex feature voting matrix is ​​calculated as a reference for subsequent surface measurement region clustering and segmentation. The feature voting attribute of surface vertices is defined as follows: in, It is the triangle face where the current vertex is located. It is the unit normal vector corresponding to the triangle, and the matrix. And coefficient ,area It is the sum of the areas of all triangular faces within the neighborhood of the current vertex, A. area(max) This represents the triangle with the maximum area among the triangular faces in its neighborhood. This represents the minimum bounding box side length of the target point's neighborhood. It is a triangular facet The center of gravity; (3.2) According to the feature voting attribute calculation method described in formula (5) in step (3.1), traverse the vertex set and assign feature voting attribute values ​​to each vertex to provide feature clustering reference for subsequent surface measurement region segmentation.

5. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, Step (5) specifically involves: (5.1) Generate local spatial grid stacking lines based on the divided measurement area: （7） In the formula, and p are the spline basis functions that make up the surface mesh. ij These are spline control nodes, equidistant from the surface spline basis functions. The two directions are stacked alternately to form a spatial grid: （8） In the formula, The number of grid stack lines, The spline basis function synthesis expression is used to finally form the local measurement area spatial approximation grid through equation (8); (5.2) Map the spatial mesh onto the surface of the propeller theoretical model. The intersection of the mesh nodes and the surface projection is the local measurement point. Piece together the local measurement points to form the initial measurement point set. .

6. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, The specific steps (6) are as follows: (6.1) Calculate the chord tolerance based on the local measurement points generated in step (5). Referring to the local surface measurement areas generated in step (4), calculate the chord tolerance on each measurement point generated by the spatial spline mesh projection in the local area. Then calculate the mean chord tolerance of the local measurement point set. Arrange the local measurement areas in ascending order according to the mean chord tolerance result. The chord tolerance calculation is as follows: In the formula, the node sequence , Let be the spline basis functions in equation (7). It is the chord tolerance value; (6.2) Based on the results of sorting the measurement points in ascending order according to the mean chord tolerance in step (6.1), select the set of measurement points in the local measurement area with the smallest mean chord tolerance as the optimal set of measurement points. .

7. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, The specific steps (7) are as follows: (7.1) Based on the optimal measurement point set in step (6) As the point cloud to be registered, the vertex set extracted in step (1) As the target point cloud, a registration model is established. , and These are the rotation matrix and the translation matrix, respectively. (7.2) First, the registration model is transformed into a problem of solving the registration transformation parameters between the optimal measurement point set and the vertex set: （10） in, The number of point clouds to be registered. For the transformation point, the rotation matrix For the three rotational degrees of freedom in space, Representing three degrees of freedom in space, Equation (10) is obtained by substituting... It can be transformed into: （11） in, Step length The objective is solved through multiple iterations of step size optimization. in and , After the update, It can be further transformed into: （12） coefficient Real-time correction function iterative calculation, calculating gradient descent weights. Weights of Newton's method ,when The time function performs two calculations simultaneously; when the coefficient is small, the descent weights dominate, and gradient descent is the primary method. In this case, the coefficient... Approaching the optimal solution; when this coefficient is large, Newton's method weights dominate, and the coefficient... Far from the optimal solution, Newton's method is mainly used, and the coefficients are optimized to ensure rapid convergence of the function while avoiding local optima during iteration; let the error threshold be... If the current calculation result is better than the previous one, then an error threshold is applied. Determine whether to output the result. If the result is not better than the previous result, update the step size and repeat the above nonlinear calculation steps, performing step size update and iterative calculations until the result is better than the previous result. The final calculation result must meet the error threshold. When required, the registration pose transformation result is obtained by iterative completion. ; (7.3) Obtain the registration pose transformation result according to step (7.2). The registration result is obtained by transforming the initial measurement point set using this matrix. The registration transformation can be expressed as: (13) In the formula, and These are the rotation matrix and the translation matrix, respectively. It is a set of key matrices for pose transformation. It is the initial set of measurement points. Point set after pose transformation.

8. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, The specific steps (8) are as follows: (8.1) Based on the point set and vertex set after pose transformation Root mean square error between Distance between a single measuring point and the model error As a quality assessment standard: （14） in, and These are the point sets after pose transformation. With vertex set The three-dimensional coordinates of the midpoint Indicates the distance between points; (8.2) Referring to the surface accuracy requirements of propeller blades in GBT 12916-2010 Technical Conditions for Marine Metal Propellers, the overall measurement point accuracy error is required to be less than 0.5mm. Compare with the results in step (8.1) and output the comparison judgment results. First, evaluate whether the overall blade error meets the standard, and then output the position coordinate information of the error that is lower than the standard accuracy to complete the quality assessment.

9. The propeller blade quality assessment method based on regional difference division according to claim 1, characterized in that, The specific steps (9) are as follows: based on the quality assessment results, data is output for secondary processing guidance. For positions with lower-than-standard precision, error judgment is made, and the positive and negative values ​​and magnitude of the processing error are tabulated for subsequent processing guidance.

Images