Metal pipe inner wall antirust agent AI spraying track planning method and system based on visual guidance

Abstract

The application provides a metal pipe inner wall rust inhibitor AI spraying track planning method and system based on visual guidance, relates to the field of industrial automatic spraying technology, and the method comprises the following steps: acquiring original sequence images of a metal pipe inner wall collected by an endoscopic visual probe, performing optical distortion correction and multi-view point cloud space registration on the original sequence images to obtain an inner wall three-dimensional topological grid; extracting node coordinates and surface normal vectors of the inner wall three-dimensional topological grid, and calculating local curvature gradient distribution of the node coordinates and the surface normal vectors; performing threshold segmentation and feature boundary extraction on the local curvature gradient distribution to obtain a geometric feature partition graph of the spatial range of the thread groove bottom, the windward surface of the reinforcing rib and the weld reinforcement area. The application improves the uniformity of the coating and the protection reliability.

Description

Technical Field

[0001] This invention relates to the field of industrial automated spraying technology, and in particular to a vision-guided AI spraying trajectory planning method and system for rust inhibitors on the inner wall of metal pipes. Background Technology

[0002] As a core component for industrial fluid transportation, metal pipelines have their inner walls exposed to corrosive media for extended periods. Rust inhibitor spraying is a crucial process for ensuring the service life of the pipe walls. Existing methods for planning the spraying trajectory of rust inhibitors on the inner walls of metal pipes generally employ equal-pitch spiral feeding or uniform scanning with fixed step lengths. The spraying trajectory is uniformly distributed on the inner surface of the pipe wall, without considering the differences in the geometric characteristics of the inner wall.

[0003] The inner walls of actual metal pipes often possess complex geometric features such as threaded grooves, reinforcing ribs, and weld reinforcement. These regions exhibit significant curvature abrupt changes and varying surface angles, leading to fundamental differences in the deposition behavior and adhesion requirements of rust inhibitor droplets in different areas: the bottom space of threaded grooves is narrow and obstructed by the groove sidewalls, making it difficult for the rust inhibitor to effectively penetrate to the bottom; the windward side of the reinforcing ribs is significantly disturbed by airflow within the pipe, reducing droplet deposition efficiency; and the weld reinforcement area suffers from abrupt changes in surface roughness and microstructure, making it difficult to guarantee coating thickness uniformity. Although the failure modes of these various regions differ, they all stem from the same technical contradiction: the mismatch between the assumption of uniform distribution of the spray trajectory and the reality of non-uniform distribution of the inner wall's geometric features. Traditional uniform trajectory planning methods are insufficient for differentiated trajectory density adjustments in these geometric regions, resulting in insufficient rust inhibitor coverage or localized over-accumulation in critical protection areas, thus restricting the overall rust prevention quality and service life of the metal pipe's inner wall. Summary of the Invention

[0004] This invention provides a vision-guided AI spraying trajectory planning method and system for rust inhibitors on the inner wall of metal pipes, which improves coating uniformity and protective reliability.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a visually guided AI-based trajectory planning method for spraying rust inhibitors onto the inner wall of metal pipes, the method comprising: The original sequence images of the inner wall of the metal tube were acquired by the endoscopic vision probe. Optical distortion correction and multi-view point cloud spatial registration were performed on the original sequence images to obtain the three-dimensional topological mesh of the inner wall. Extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and calculate the local curvature gradient distribution using the node coordinates and surface normal vectors; perform threshold segmentation and feature boundary extraction on the local curvature gradient distribution to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. Based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map, a constrained model of the aerodynamic settling path of the rust inhibitor droplets is constructed. The rheological dynamic parameters are inverted and the gravity deflection compensation is calculated through the constrained model to obtain the target flow distribution matrix. The flow density gradient data inside the analytical matrix is ​​used to map the distribution rules of trajectory points along the pipe axis advancement direction based on the flow density gradient data, and a preliminary path polyline is obtained. The preliminary path polyline is then projected onto the polygon boundary plane corresponding to the geometric feature partition map, and the intersection coordinates between the polygon boundary and each segment of the polyline are solved by iterating through the polygon boundary. Perform out-of-bounds segment clipping and topological continuity reconstruction of internal valid segments on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; Based on the spatial pose coordinates of the initial trajectory points, combined with the tube shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, radial cumulative deviation is calculated, and an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal tube is output.

[0006] Secondly, a vision-guided AI-based trajectory planning system for spraying rust inhibitors onto the inner walls of metal pipes includes: The acquisition module is used to acquire the original sequence images of the inner wall of the metal tube collected by the endoscopic vision probe, perform optical distortion correction and multi-view point cloud spatial registration on the original sequence images, and obtain the three-dimensional topological mesh of the inner wall. The extraction module is used to extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and calculate the local curvature gradient distribution using the node coordinates and surface normal vectors; the local curvature gradient distribution is segmented by threshold and the feature boundary is extracted to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. The module is used to construct a constrained model of the aerodynamic settling path of rust inhibitor droplets based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map. The rheological dynamic parameters are inverted and the gravity deflection compensation is calculated through the constrained model to obtain the target flow distribution matrix. The mapping module is used to parse the flow density gradient data inside the matrix, and based on the flow density gradient data, map the density distribution rules of the trajectory points along the pipe axis advancement direction to obtain the preliminary path polyline; project the preliminary path polyline onto the polygon boundary plane corresponding to the geometric feature partition map, and traverse and solve the intersection coordinates between the polygon boundary and each segment of the polyline; The reconstruction module is used to perform outbound line segment clipping and internal valid line segment topological continuity reconstruction on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; The calculation module is used to calculate the radial cumulative deviation based on the spatial pose coordinates of the initial trajectory point series, combined with the tube shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, and output an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal tube.

[0007] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0008] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0009] The above-described solution of the present invention has at least the following beneficial effects: The topological mesh of the pipe inner wall was obtained by endoscopic vision 3D reconstruction. Based on the segmentation and extraction of local curvature gradient distribution, a geometric feature partition map was obtained, marking the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess area. Then, using the spatial tilt angle and curvature radius parameters of each partition as constraints, a rust inhibitor droplet aerodynamic sedimentation path constraint model was constructed. Through rheological dynamic parameter inversion and gravity deflection compensation calculation, a target flow distribution matrix and partition contour constraint trajectory coupled with the geometric feature partition of the pipe inner wall were generated. This achieved adaptive differential matching of the spray trajectory density and flow parameters with the geometric features of the pipe inner wall, ensuring that the bottom of the threaded groove is fully penetrated by spraying, the windward side of the reinforcing rib overcomes airflow disturbance to achieve effective deposition, and the weld excess area is uniformly covered. This solved the technical problem of uneven rust inhibitor adhesion caused by complex inner wall structure, improved coating uniformity and protective reliability, and effectively extended the service life of metal pipelines. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the AI ​​spraying trajectory planning method for rust inhibitors on the inner wall of metal pipes based on visual guidance, provided in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of a vision-guided AI spraying trajectory planning system for rust inhibitors on the inner wall of metal pipes, provided in an embodiment of the present invention. Detailed Implementation

[0012] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0013] like Figure 1As shown, embodiments of the present invention propose an AI-based method for planning the spraying trajectory of rust inhibitors on the inner wall of metal pipes based on visual guidance. The method includes the following steps: Step 1: Obtain the original sequence images of the inner wall of the metal tube captured by the endoscopic vision probe, perform optical distortion correction and multi-view point cloud spatial registration on the original sequence images to obtain the three-dimensional topological mesh of the inner wall. Step 2: Extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and calculate the local curvature gradient distribution using the node coordinates and surface normal vectors; perform threshold segmentation and feature boundary extraction on the local curvature gradient distribution to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. Step 3: Construct a constrained model for the aerodynamic settling path of the rust inhibitor droplets based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map. Use the constrained model to perform rheological parameter inversion and gravity deflection compensation calculation to obtain the target flow distribution matrix. Step 4: Analyze the flow density gradient data inside the matrix, and based on the flow density gradient data, map the distribution rules of trajectory points along the pipe axis advancement direction to obtain the preliminary path polyline; project the preliminary path polyline onto the polygon boundary plane corresponding to the geometric feature partition map, and iterate through and solve the intersection coordinates between the polygon boundary and each segment of the polyline. Step 5: Perform boundary segment clipping and topological continuity reconstruction of internal valid line segments on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; Step 6: Based on the spatial pose coordinates of the initial trajectory points, combined with the pipe shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, perform radial cumulative deviation calculation and output an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal pipe.

[0014] In this embodiment of the invention, by employing a full-chain technical approach including visual 3D reconstruction, curvature gradient partitioning, rheological dynamics inversion, flow density mapping, partition boundary constraints, and radial deviation compensation, the technical problem of insufficient rust inhibitor coverage or local over-accumulation in key protective areas caused by neglecting the differences in the geometric features of the inner wall of the pipe, as in traditional uniform trajectory planning methods, is overcome. This achieves automated, precise, and differentiated adaptive matching of spray trajectory density and flow parameters with complex geometric features such as the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area, thereby improving the technical effect of coating uniformity and protective reliability.

[0015] In a preferred embodiment of the present invention, step 1 above may include: Step 11: Acquire the original sequence of image frames continuously captured by the endoscopic vision probe along the tube axis; based on the original sequence of image frames, extract the camera intrinsic parameter calibration data and lens radial distortion coefficients and perform pixel-level mapping correction to obtain a standard two-dimensional image sequence with barrel distortion eliminated. Specifically, this includes: acquiring the original sequence of image frames continuously captured by the endoscopic vision probe along the tube axis at a constant advancing speed on the inner wall of the metal tube; based on the original sequence of image frames, extracting the camera intrinsic parameter matrix and lens radial distortion coefficient vector calibrated by the endoscopic vision probe at the factory, and obtaining the camera imaging model parameter set required for distortion correction. The complete expression of the camera intrinsic parameter matrix is: ; In the formula, The equivalent focal length along the horizontal axis of the image plane, in pixels; The equivalent focal length along the vertical axis of the image plane, in pixels; The horizontal axis pixel coordinates of the principal point in the image plane; Let be the vertical axis pixel coordinates of the principal point in the image plane. Based on the radial distortion coefficient vector in the camera imaging model parameter set, a radial distortion polynomial mapping model is constructed from the distorted pixel coordinates to the corrected pixel coordinates, yielding the distortion correction mapping function. The complete expression of the radial distortion polynomial mapping model is: ; ; ; In the formula, The horizontal pixel coordinates of the distorted pixels in the original sequence image frame; These are the vertical pixel coordinates of the distorted pixels in the original sequence image frames; The horizontal pixel coordinates of the corresponding pixel in the corrected standard image; The vertical pixel coordinates of the corresponding pixel in the corrected standard image; This represents the radial distance of the distorted pixel relative to the principal point. The first-order radial distortion coefficient; It is the second-order radial distortion coefficient; It is the third-order radial distortion coefficient; The horizontal axis pixel coordinates of the principal point in the image plane; Given the vertical axis pixel coordinates of the principal point in the image plane, the original sequence image frames are subjected to inverse mapping and bilinear grayscale interpolation operations pixel by pixel according to the distortion correction mapping function to obtain a standard two-dimensional image sequence with barrel distortion eliminated.

[0016] Step 12: Based on the standard two-dimensional image sequence, extract the feature key points within the overlapping field of view of adjacent image frames and calculate the relative pose transformation matrix; transform the pixel depth information in the standard two-dimensional image sequence to a unified world coordinate system using the relative pose transformation matrix to obtain the initial discrete point cloud dataset of the inner wall of the tube. Specifically, this includes: extracting the maximum pixel gradient magnitude points in each overlapping region as scale-invariant feature key points based on the overlapping field of view of two adjacent frames in the standard two-dimensional image sequence, and generating corresponding multi-dimensional feature descriptor vectors to obtain a set of candidate feature point pairs for adjacent frames; performing coarse matching and filtering on the candidate feature point pairs using the Euclidean distance ratio criterion, removing mismatched point pairs with a distance ratio greater than a preset threshold to obtain a set of finely matched feature point pairs for adjacent frames; converting the pixel coordinates of each matched feature point into normalized image coordinates using the camera intrinsic parameter matrix, substituting them into the epipolar geometric constraint equation to calculate the essential matrix, and obtaining the epipolar geometric constraint relationship between adjacent frames. The complete expression of the epipolar geometric constraint equation is: ; In the formula, For the first The first frame Normalized homogeneous coordinate vectors of matching feature points; For the first The normalized homogeneous coordinate vector of the corresponding matching feature point in the frame; For the first Frame and the The essential matrix between frames has a dimension of three rows and three columns; superscript This represents the matrix transpose operation.

[0017] Based on the essential matrix, singular value decomposition is performed. The relative rotation matrix and relative translation vector are extracted from the decomposition results and assembled to obtain the relative pose transformation matrix between adjacent frames. The complete expression for singular value decomposition is: ; ; In the formula, This is a three-row, three-column left singular vector matrix obtained from the singular value decomposition of the essential matrix; It is a 3x3 right singular vector matrix; It is a singular value diagonal matrix; It is the first singular value; It is the second singular value; The formulas for extracting relative rotation matrices and relative translation vectors, representing the diagonal matrix construction operation, are as follows: ; ; ; In the formula, For the first Frame to the The relative rotation matrix of the frame has dimensions of three rows and three columns; It is a relative translation vector with dimensions of three rows and one column; This is an auxiliary matrix for orthogonalizing the rotation matrix; For matrix The third column vector; the positive and negative signs are determined by the positive depth constraint of the triangulated spatial point. Based on the relative rotation matrix and the relative translation vector, the relative pose transformation matrix of adjacent frames is assembled, and the expression is: ; In the formula, For the first Frame to the The four-row, four-column homogeneous pose transformation matrix of the frame; Given a 3D zero row vector, based on the relative pose transformation matrix of adjacent frames, and using the camera coordinate system of the first frame as the world reference coordinate system, the pixel depth estimates of each frame image are transformed to a unified world coordinate system through chain-accumulation operation of the pose transformation matrix, resulting in the initial discrete point cloud dataset of the inner wall of the pipe. The complete expression for chain-accumulation pose transformation is: ; In the formula, For the first The three-dimensional spatial coordinates of a point in a unified world coordinate system; For the first The image frame index number from which each spatial point originates; The pixel coordinates of the point in the image frame; The depth estimate for this pixel is obtained through multi-view stereo matching or structured light measurement. This is the inverse of the camera intrinsic parameter matrix; From frame 1 to frame 2 The cumulative pose transformation matrix product of frames.

[0018] Step 13: Based on the initial discrete point cloud dataset of the pipe inner wall, calculate the consistency of the neighborhood normal vectors of spatial points and remove outlier noise points to obtain clean point cloud data after removing outlier noise points; perform local surface fitting and triangular meshing on the clean point cloud data to obtain the three-dimensional topological mesh of the inner wall representing the continuous morphology of the pipe inner wall. Specifically, this includes: based on a specified number of nearest neighbor point sets for each spatial point in the initial discrete point cloud dataset of the pipe inner wall, calculating the covariance matrix of the neighborhood point sets and performing eigenvalue decomposition, extracting the eigenvector corresponding to the smallest eigenvalue as the normal vector estimate of that point, and obtaining the neighborhood normal vector set of each spatial point. The complete expression of the neighborhood covariance matrix is: ; ; In the formula, The number of points to search for nearest neighbors, ranging from fifteen to thirty; For the first The first spatial point Three-dimensional coordinate vectors of the nearest neighbor spatial points; Let be the centroid coordinate vector of the nearest neighbor set; Given a 3x3 neighborhood covariance matrix, the eigenvalue decomposition expression for the normal vector extraction is: ; ; In the formula, The first covariance matrix is ​​the first... There are eigenvalues ​​that satisfy... ; Eigenvalues The corresponding three-row, one-column feature vector; The eigenvector corresponding to the minimum eigenvalue represents the optimal estimate of the local neighborhood surface normal at that point; This operation represents the index corresponding to the minimum value. Based on the neighborhood normal vector set of each spatial point, the cosine value of the angle between the current point's normal vector and the normal vectors of its nearest neighbors is calculated point by point. The proportion of neighborhood points whose cosine value is lower than the consistency threshold is counted. Spatial points whose proportion exceeds the preset outlier criterion are marked as outlier noise points and removed, resulting in clean point cloud data after removing outlier noise points. The complete expression of the normal consistency criterion is: ; In the formula, The normal vector of the current point and the first The inner product of the normal vectors of the neighboring points, i.e., the cosine value of the included angle; This is the normal consistency threshold; This is an indicator function that takes the value of one when the condition is true, and zero otherwise. Given the normal uniformity ratio within the neighborhood of the current point, based on the nearest neighbor set of each spatial point in the clean point cloud data, a quadratic surface polynomial is fitted on the local tangent plane using the moving least squares method. This yields the local surface fitting function and its parameter set at each point. The complete expression for the moving least squares local surface fitting is: ; ; ; In the formula, The horizontal coordinate components in the local tangent plane coordinate system with the current point as the origin and the normal vector as the vertical axis; These are the perpendicular coordinate components in the local tangent plane coordinate system; For locally fitted surfaces at points The offset along the normal direction; to These are the components of the six-dimensional fitting coefficient vector; This is a vector of fitting coefficients; Let be a vector of basis functions of a quadratic polynomial, whose elements are... ; This indicates taking parameter values ​​that minimize the objective function; This is a Gaussian-weighted kernel function; The Euclidean distance from the neighboring points to the current point; The bandwidth parameter of the weighted kernel function is one-third of the maximum radius of the neighborhood point set; For the first The normal offset of each neighborhood point relative to the local tangent plane is calculated based on the local surface fitting parameter set at each point. Regular sampling grid points are then generated on the local tangent plane, and their corresponding normal offsets are calculated. These local regular grid points are then mapped back to the world coordinate system through an inverse transformation, resulting in a denser set of local surface sampling points. The complete expression for the world coordinate mapping of the local regular grid points is: ; In the formula, , which is the unit orthogonal basis vector along the horizontal axis of the local tangent plane coordinate system; , which are unit orthogonal basis vectors in the direction perpendicular to the axis of the local tangent plane coordinate system; The normal vector of the current point; To map the sampled grid points back to the world coordinate system in 3D, a global dense point set is constructed based on the union of the sampled point sets of each locally dense surface. A 3D spatial triangulation algorithm is then used to perform triangulation on the global dense point set. Furthermore, the grid edges in the transition regions between adjacent local surfaces are smoothed to eliminate seams, resulting in a 3D topological mesh representing the continuous morphology of the pipe's inner wall. The iterative formula for the smoothing process is as follows: ; In the formula, For the first During the nth iteration The three-dimensional coordinates of each grid vertex; For the first The set of vertices in the ring neighborhood of each vertex; The number of vertices in the neighborhood of a ring; The Laplacian smoothing relaxation factor for the mesh vertices is used to move each mesh vertex towards the arithmetic mean of its neighboring vertices. Through iteration, high-frequency geometric noise and local sharp undulations on the mesh surface are eliminated, making the reconstructed inner wall surface of the pipe smoother. Controlling the vertex displacement step size is a mesh smoothing operation belonging to the geometry modeling layer; For the first Vertex coordinates after smoothing in the next iteration.

[0019] In this embodiment of the invention, a progressive three-dimensional topography reconstruction technique is adopted, which proceeds from camera imaging model distortion correction to multi-view point cloud spatial registration, and then to point cloud purification and mesh reconstruction. Specifically, pixel-level inverse mapping based on the radial distortion coefficients of the camera's internal components is used to eliminate barrel distortion of wide-angle lenses; feature matching of adjacent frames and epipolar geometric constraints and pose transformation chain accumulation are used to achieve multi-frame depth information world coordinate system fusion; and neighborhood covariance feature decomposition and normal consistency criteria and moving least squares local surface fitting are used to denoise the point cloud and perform triangular mesh partitioning. Therefore, this technique overcomes a series of problems such as geometric distortion of edge regions caused by imaging distortion of endoscopic vision probes, lack of three-dimensional information in single-frame images and inconsistency of coordinate systems in multiple frames making it difficult to construct a complete topography, and low reconstruction quality caused by the triple interference of sensor noise, mismatched outliers and surface discontinuities in the original point cloud. Thus, it achieves the technical effect of reconstructing a three-dimensional topological mesh of the inner wall of the tube with accurate geometric relationships, thorough noise filtering, smooth and continuous local surfaces, complete triangular mesh topology, and a true representation of the continuous three-dimensional topology of the inner wall of the tube from the original sequence images.

[0020] In a preferred embodiment of the present invention, step 2 above may include: Step 21: Extract the vertex coordinates of the triangular facets and the normal vectors of adjacent facets from the three-dimensional topological mesh of the inner wall. Based on these coordinates, construct a local differential geometric neighborhood. Perform principal curvature calculation and gradient field fitting on the local differential geometric neighborhood to obtain the local curvature gradient distribution tensor characterizing the surface undulation trend. Specifically, this includes: extracting the spatial coordinates of the three vertices of each triangular facet and the unit normal vectors of its adjacent triangular facets from the three-dimensional topological mesh of the inner wall; constructing a local differential geometric neighborhood by extending a specified number of rings outward along the topological adjacent edges, centered on each mesh vertex, thus obtaining a set of triangular facets in the local surface neighborhood centered on the vertex; fitting a local quadratic parametric surface using the weighted least squares method within the local tangent plane coordinate system of the vertex, calculating the first and second fundamental form coefficients of the local parametric surface, and obtaining the metric and curvature characteristics of the local differential geometry. The complete expression for the first fundamental form coefficient is: , , ; In the formula, For local parametric surfaces along the parametric direction The first-order partial derivative vector; For local parametric surfaces along the parametric direction The first-order partial derivative vector; For parameter direction The metric coefficient; For parameter direction and The coupling metric coefficient; For parameter direction The metric coefficient; symbol The complete expression for the coefficients of the second fundamental form of the vector dot product operation is: , , ; In the formula, This is the unit normal vector of the local surface at the current vertex; For local parametric surfaces along the parametric direction The second-order partial derivative vector; For local parametric surfaces along the parametric direction and The mixed second-order partial derivative vector; For local parametric surfaces along the parametric direction The second-order partial derivative vector; For parameter direction The second-order bending coefficient; For parameter direction and The coupling bending coefficient; For parameter direction The second-order curvature coefficients are obtained by assembling the Weingarten mapping matrix, i.e., the shape operator, based on the first and second fundamental form coefficients of the local parametric surface. Eigenvalue decomposition is then performed on the shape operator to extract two eigenvalues ​​as the first and second principal curvatures of the current vertex, yielding the local principal curvature scalar field for each mesh vertex. The complete expression for the shape operator matrix is: ; In the formula, It is a 2x2 shape operator matrix; , , The coefficients are the first fundamental form coefficients; , , The coefficient of the second fundamental form; superscript The complete expression for the principal curvature eigendecomposition, representing the matrix inversion operation, is: , ; In the formula, This is the first principal curvature, which is also the maximum principal curvature; This is the second principal curvature, which is also the minimum principal curvature. This is the 2-row, 1-column principal direction feature vector corresponding to the first principal curvature; Let be the principal direction eigenvector corresponding to the second principal curvature. Based on the local principal curvature scalar field of each grid vertex, calculate the average curvature and Gaussian curvature vertex by vertex. Perform a second-order Taylor expansion fitting on the average curvature in the local tangent plane to extract the first-order gradient vector and the second-order Hessian matrix. Assemble these to obtain the local curvature gradient distribution tensor characterizing the trend of surface undulation. The complete expressions for the average curvature and Gaussian curvature are: , ; In the formula, The average curvature describes the average degree of curvature of a local surface. The Gaussian curvature describes the intrinsic bending properties of a local surface. The first principal curvature; Given the second principal curvature, the complete expression for the local curvature gradient distribution tensor is: ; ; In the formula, For the first The two-row, two-column local curvature gradient distribution tensor at each grid vertex is the mean curvature Hessian matrix; For the first The average curvature gradient vector at each grid vertex (two rows and one column); The mean curvature along the local tangent plane parameter direction The first-order partial derivative; The mean curvature along the local tangent plane parameter direction The first-order partial derivative; The mean curvature along the parameter direction The second-order partial derivative; The mean curvature along the parameter direction and The mixed second-order partial derivatives; The mean curvature along the parameter direction The second-order partial derivative; Indicates that the partial derivatives are in the th case. The value is taken at each vertex.

[0021] Step 22: Based on the local curvature gradient distribution tensor, extract the curvature extrema points of each grid cell and set a high / low curvature segmentation threshold; use the high / low curvature segmentation threshold to perform interval mapping and connected component labeling to obtain a curvature classification mask that initially distinguishes between convex ridges and concave grooves. Specifically, this includes: calculating the curvature gradient magnitude vertex by vertex based on the local curvature gradient distribution tensor and its corresponding average curvature gradient vector, detecting local maxima of the gradient magnitude, marking the local maxima points as curvature extrema feature points, and obtaining a set of curvature extrema feature points. The complete expression for the curvature gradient magnitude is: ; In the formula, For the first The average curvature gradient magnitude at each grid vertex; The mean curvature along the parameter direction The first-order partial derivative; The mean curvature along the parameter direction Based on the first-order partial derivatives and the statistical distribution characteristics of the set of curvature extremum feature points, the Otsu thresholding method or the adaptive percentile method are used to calculate the high curvature segmentation threshold and the low curvature segmentation threshold, thus obtaining the high-low curvature segmentation threshold pair that distinguishes between convex features, concave features, and flat transition regions. The complete expression for the high-low curvature segmentation threshold is: , ; In the formula, A high curvature segmentation threshold is used to identify convex ridge regions; A low curvature segmentation threshold is used to identify recessed trench areas; This represents the mean of the average curvature of all vertices in the grid. The standard deviation of the average curvature of all grid vertices; This is the high threshold adjustment coefficient, with a value ranging from 0.8 to 1.5. The low threshold adjustment coefficient, with a value ranging from 0.8 to 1.5, is used to perform a three-interval threshold mapping on all grid vertices based on the high and low curvature segmentation threshold pairs. Vertices with an average curvature higher than the high threshold are marked as convex candidate regions, vertices with an average curvature lower than the low threshold are marked as concave candidate regions, and vertices in between are marked as flat transition regions. This yields a ternary initial curvature classification label field. The complete expression for the three-interval threshold mapping is: ; In the formula, For the first Curvature classification label value of each grid vertex; For the first Average curvature value of each grid vertex; Indicates the raised candidate region marker; Marking a flat transition region; The candidate region of the depression is marked. Based on the ternary initial curvature classification label field, the connected component labeling algorithm is performed on the spatially adjacent vertices with the same label value. Vertices that are spatially continuous and have the same label value are aggregated into independent connected regions. Fragmented connected regions with an area smaller than the preset minimum area threshold are removed to obtain a curvature classification mask that initially distinguishes between convex ridges and depressions.

[0022] Step 23: Extract the curvature abrupt transition zone at the boundary of the curvature classification mask and perform boundary smoothing filtering to obtain the smoothed boundary contour. Based on the smoothed boundary contour and the projection relationship of the pipe axis centerline, perform spatial region mapping and feature label assignment to obtain a geometric feature partition map with the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area marked. Specifically, this includes: extracting the spatial interface between the raised and recessed areas as the curvature abrupt transition zone according to the adjacency relationship between the raised and recessed areas in the curvature classification mask, obtaining the boundary contour line of the curvature classification mask; performing Laplacian smoothing filtering on the coordinates of each vertex on the boundary contour along the pipe wall surface according to the boundary contour line; iteratively correcting the boundary vertex position using the geometric centroid of the one-ring neighborhood vertices of the boundary vertex to eliminate the sawtooth fluctuations and local irregular undulations of the boundary contour, obtaining the smoothed boundary contour. The complete expression of the boundary Laplacian smoothing filter is: ; In the formula, For the first The three-dimensional spatial coordinates of the boundary vertices at the next iteration; For the set of vertices in a ring neighborhood of the boundary vertex on the smoothed boundary contour; The number of vertices on the boundary of a ring neighborhood; The trajectory curvature smoothing relaxation factor controls the intensity of the boundary nodes pulling back towards the contour. It is usually set to 0.5 to 0.9 to ensure that the boundary trajectory does not go out of bounds. For the first After smoothing the boundary vertex coordinates in the next iteration, the spatial linear equation parameters of the tube axis centerline are extracted based on the smoothed boundary contour. Each boundary vertex and mesh vertex is then projected into a cylindrical coordinate system with the tube axis centerline as the vertical axis, yielding the axial position parameters and circumferential angle parameters of each spatial point. The complete expression for the cylindrical coordinate projection is: ; ; ; In the formula, For the first The axial position coordinates of a spatial point, that is, the directional projection distance along the centerline of the tube axis; For the first The circumferential angular coordinates of a spatial point range from zero to twice the value of pi. For the first A three-dimensional Cartesian coordinate vector of a spatial point; The coordinate vector of the reference origin of the tube axis centerline; is the unit direction vector of the tube axis centerline; For the first The coordinates of the orthogonal projection of a spatial point onto the center line of the tube axis; It is the first orthogonal basis vector perpendicular to the centerline of the tube axis; It is the second orthogonal basis vector perpendicular to the centerline of the tube axis. and An orthogonal coordinate system constituting the plane of the pipe cross-section; Using the four-quadrant arctangent function, based on the axial position parameters and circumferential angle parameters of each spatial point, and combined with the curvature classification mask boundary contour after smoothing filtering, the raised ridge region and the recessed groove region are mapped to the corresponding spatial range of the cylindrical development surface of the pipe wall. The bottom region of the threaded groove is assigned the first feature label, the windward side region of the reinforcing rib is assigned the second feature label, the weld excess height region is assigned the third feature label, and the transition region of the flat pipe wall is assigned the fourth feature label, resulting in a geometric feature partition map labeled with the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height region.

[0023] In this embodiment of the invention, a progressive geometric feature partitioning and recognition technology is adopted, which proceeds from the construction of local differential geometric neighborhoods to the calculation of curvature gradient distribution tensors, and then to the generation of curvature classification masks and boundary smoothing mapping. Specifically, the technology uses local quadratic surface fitting of the vertex-ring neighborhood of the topological mesh to calculate the Weingarten shape operator and principal curvature feature decomposition to obtain the curvature gradient distribution tensor; Otsu adaptive threshold segmentation and connected domain labeling to distinguish between convex ridges and concave grooves; and Laplacian boundary smoothing and cylindrical coordinate projection of the tube axis to realize spatial region mapping and multi-category feature label assignment. This overcomes the three major technical problems of the blurred and difficult-to-distinguish convex and concave geometric feature boundaries in the three-dimensional topological mesh of the inner wall, the sawtooth fluctuations of the curvature abrupt transition zone causing the partition contour to be unsmooth, and the lack of a unified spatial reference system causing the spatial position relationship of feature regions to be unable to be accurately defined. Thus, the technology achieves the technical effect of automatically identifying and accurately marking the geometric feature partitioning map of the bottom of the thread groove, the windward side of the reinforcing rib, and the spatial range of the weld excess height area from the three-dimensional topological mesh of the inner wall.

[0024] In a preferred embodiment of the present invention, step 3 above may include: Step 31: Based on the geometric feature partitioning map, extract the spatial tilt angle parameters and radius of curvature parameters of each partition to obtain the partition morphology feature dataset. Specifically, this includes: based on the spatial range of each feature region marked on the geometric feature partitioning map, extracting the spatial coordinates and normal vectors of all grid vertices within the thread groove bottom region, the stiffener windward side region, the weld reinforcement area region, and the flat pipe wall transition region, respectively, to obtain the vertex coordinate set and normal vector set of each feature region. Based on the vertex coordinate set and normal vector set of each feature region, calculate the spatial tilt angle of the normal vector relative to the orthogonal section of the pipe axis centerline in a cylindrical coordinate system frame, to obtain the spatial tilt angle parameter of each grid vertex in each feature region. The complete expression of the spatial tilt angle parameter is: ; In the formula, For the first The spatial tilt angle parameter of each grid vertex, in radians, represents the degree of tilt of the pipe wall surface from the orthogonal section of the pipe axis; For the first The unit normal vector of each grid vertex; is the unit direction vector of the tube axis centerline; It is the inverse cosine function; This represents the absolute value operation; To represent pi, based on the vertex coordinate set and normal vector set of each feature region, and using the pipe axis centerline as a reference, the radius of curvature of the pipe wall surface along the normal section is calculated vertex by vertex. This yields the radius of curvature parameter for each grid vertex within each feature region. The complete expression for the radius of curvature parameter is: ; ; In the formula, For the first The radius of curvature parameter at each grid vertex, in millimeters, represents the average radius of curvature of the local bending of the pipe wall; For the first The first principal curvature at each grid vertex; For the first Second principal curvature at each grid vertex; constraint conditions This indicates that the Gaussian curvature at the vertex is positive, meaning the local surface is an ellipse and an effective radius of curvature can be defined. Based on the spatial tilt angle and radius of curvature parameters of each grid vertex within each feature region, the mean spatial tilt angle, standard deviation of spatial tilt angle, mean radius of curvature, and gradient of radius of curvature variation are statistically analyzed according to the feature label category. These are then assembled into a structured dataset using the feature labels as index keys, resulting in a partitioned topographic feature dataset containing the spatial tilt angle and radius of curvature distributions of each feature region. The statistical expressions for the partitioned topographic feature dataset are as follows: , ; , ; In the formula, For the first Mean spatial tilt angle of the class feature region; For the first Standard deviation of spatial tilt angle of the class feature region; For the first Mean radius of curvature of the feature region; For the first The gradient of the radius of curvature of a feature region is the difference between the maximum and minimum radius of curvature within the region. For the first The total number of grid vertices contained in the class feature region; For the first The set of vertex indices for the class feature region; The values ​​range from one to four, corresponding to the bottom area of ​​the threaded groove, the windward side area of ​​the reinforcing rib, the weld reinforcement area, and the transition area of ​​the flat pipe wall, respectively. and These represent operations to retrieve the maximum and minimum values, respectively.

[0025] Step 32: Based on the partitioned topography feature dataset and the physical properties of the rust inhibitor droplets, construct an aerodynamic settling path constraint model to obtain a fluid dynamic solution domain containing the differential equation of droplet trajectory and pipe wall boundary conditions. Specifically, this includes: based on the spatial tilt angle and radius of curvature parameters in the partitioned topography feature dataset, combined with the physical property parameters of the rust inhibitor liquid atomized droplets, including the rust inhibitor droplet density, droplet dynamic viscosity, droplet surface tension coefficient, and initial median droplet size, establish an aerodynamic settling path constraint model describing the droplet trajectory in the near-wall region of the pipe wall. The complete differential equation expression for the droplet trajectory in the aerodynamic settling path constraint model is: ; ; In the formula, For droplets in The three-dimensional spatial position vector at time; For droplets in The three-dimensional velocity vector at time t; The mass of a single droplet is determined by the droplet density and volume. The vector of the aerodynamic drag force acting on the droplet; Let be the vector of gravity acting on the droplet; Let be the vector of the aerodynamic lift force acting on the droplet; The complete expression for the aerodynamic drag vector, where is the thermophoretic force vector acting on the droplet, is: ; ; In the formula, The diameter of the droplet is in millimeters. The density of the air inside the tube; This is the aerodynamic drag coefficient; The Reynolds number represents the relative motion of the droplets. is the local airflow velocity vector at the spatial location of the droplet; To determine the aerodynamic viscosity inside the pipe, based on the partitioned topographic feature dataset, the spatial tilt angle parameter and the radius of curvature parameter are introduced into the pipe wall geometric boundary conditions to construct the hydrodynamic solution domain from the injection port to the near-wall settlement zone of each feature region. The pipe wall boundary of the solution domain is described by a surface implicit equation. The non-flat topographic features in the geometric feature partition map are embedded into the boundary constraints. The complete expression of the pipe wall surface boundary constraint is as follows: ; , , ; In the formula, For the first Implicit function description of the pipe wall surface in the class feature region; To find the three-dimensional coordinate vector of any point in space within the domain; For the curved surface of the pipe wall at point The local normal vector at a given location is obtained by local interpolation of the normal vectors of the mesh vertices in that region from the partition topography feature dataset; For the curved surface of the pipe wall at point The local radius of curvature at a given location is obtained by local interpolation of the radius of curvature parameter of that region in the partitioned topographic feature dataset. The complete expression for the inlet boundary condition of the fluid dynamics solution domain is: ; ; In the formula, To find the boundary surface at the entrance of the solution domain; Let be the airflow velocity scalar at the inlet section; Given the static pressure of the airflow at the inlet section, the complete expression for the no-slip boundary condition of the pipe wall in the fluid dynamics solution domain is: , ; In the formula, the airflow velocity vector at the curved surface of the pipe wall is always zero, which represents the no-slip condition of the viscous fluid at the solid wall.

[0026] Step 33: Based on the fluid dynamics solution domain, substitute the spatial tilt angle parameter and the radius of curvature parameter to perform inverse parameter identification and viscous rheological inversion, obtaining the rheological compensation coefficients corresponding to each feature region. Specifically, this includes: based on the differential equation of droplet trajectory and pipe wall boundary conditions in the fluid dynamics solution domain, and under the premise of known physical properties of the rust inhibitor droplets, using the spatial tilt angle parameter and radius of curvature parameter in the partitioned morphology feature dataset as input conditions, constructing an inverse inversion model with the rheological compensation coefficients as the parameters to be identified. The construction process of the inverse inversion model is as follows: integrate the differential equation of droplet trajectory along the settling path to obtain the theoretical settling time and theoretical wall-attaching position of the droplet from the injection port to the near wall surface of the pipe; use the deviation between the theoretical wall-attaching position and the actual expected wall-attaching position as the objective function, and use a sequential quadratic programming optimization algorithm to adjust the rheological compensation coefficients to minimize the objective function. The complete expression of the objective function is: ; In the formula, For the first Inverse identification target function for class feature regions; For the first The rheological dynamic compensation coefficient vector of the class feature region; In the compensation coefficient Next The theoretical contact position vector of the droplet at each vertex; For the first The droplet at each vertex expects to hit the wall position vector, which is the target landing point determined based on the geometric feature partition map and actual spraying requirements. In the compensation coefficient The droplet reaches the first Theoretical settling time for each vertex; Let be the desired settlement time constant; As the time deviation weighting coefficient, based on the inverse inversion model and the objective function, the Levenberg-Marquardt algorithm is used to iteratively optimize the objective function. In each iteration, the Jacobian matrix and the approximate Hessian matrix of the objective function with respect to the rheological compensation coefficient vector are calculated. The iterative increment of the compensation coefficient vector is updated along the negative gradient direction to obtain the optimal rheological compensation coefficient that makes the objective function converge to a minimum. The complete expression for the Levenberg-Marquardt iterative update is: ; ; In the formula, For the first During the nth iteration The rheological dynamic compensation coefficient vector of the class feature region; This is the updated compensation coefficient vector; The residual vector of the objective function For the compensation coefficient vector The Jacobian matrix; For the first The damping factor in each iteration gradually decreases as the iteration converges. Local quadratic convergence is fast but unstable and related to the gradient descent direction. It is globally stable but converges slowly; As the identity matrix, the converged optimal rheological compensation coefficient vector is stored according to the feature label category, resulting in the rheological compensation coefficient dataset corresponding to each feature region. The complete expression of the rheological compensation coefficient vector is: ; In the formula, For the first for the second The optimal rheological dynamic compensation coefficient vector for the class-feature region; The viscosity compensation factor characterizes the amount of correction to the trajectory of rust inhibitor droplets by viscous dissipation during settling, and its value is usually in the range of 0.8 to 1.5. The elastic compensation factor characterizes the amount by which the elastic recovery of the droplet corrects the stress distribution after impacting the pipe wall, and its value ranges from 0.2 to 0.8. The plastic thixotropic compensation factor characterizes the effect of thixotropic thinning on wetting behavior during the spreading of rust inhibitors on the wall surface, with a value ranging from 0.5 to 15.0.

[0027] Step 34: Based on the rheological compensation coefficient, calculate the superposition effect of gravity settling deflection and slit air resistance to obtain the gravity deflection compensation matrix for each spraying node. Specifically, this includes: based on the rheological compensation coefficient corresponding to each characteristic region, and combined with the component of the gravity vector along the tangential plane of the pipe wall in the differential equation of the droplet trajectory, calculating the cumulative offset effect of gravity settling deflection on the droplet's wall contact position. The complete expression for the offset of gravity settling deflection is: ; ; ; In the formula, For the effect of gravity settling and deflection on the first The three-dimensional offset vector of each spraying node; This is the estimated settling time for the droplets from the nozzle to the pipe wall; For the injection nozzle to the first Spatial distance between target points on the pipe wall; This represents the axial velocity component of the droplet along the tube axis. This is the scalar value of gravitational acceleration; For the first Spatial tilt angle parameters at each spraying node; This is the unit projection direction vector of gravity along the tangent plane of the pipe wall; It is the gravity vector; For the first Unit normal vector of the pipe wall at each spraying node; The viscosity compensation factor characterizes the correction of the rust inhibitor droplet's trajectory by viscous dissipation during settling. Its value typically ranges from 0.8 to 1.5. Based on the rheological compensation coefficients of each characteristic region and the geometric constraints of the pipe wall boundary, the superimposed air resistance effect experienced by the rust inhibitor droplet when passing through the air resistance layer near the pipe wall slit is calculated. The complete expression for the superimposed air resistance effect of the slit is: ; ; In the formula, For the first The additional drag vector of droplets due to the air resistance of the slits at each spraying node; The dynamic viscosity of the air inside the pipe; The diameter of the droplet; For the first The equivalent slit gap thickness for droplet near-wall settling at each spraying node is determined by the radius of curvature parameter and the spray cone angle. Here is the local airflow velocity vector near the wall; Here is the near-wall droplet velocity vector; For the first Curvature radius parameter at each spraying node; The spray cone angle; The elastic compensation factor characterizes the correction of stress distribution by elastic recovery after droplet impact on the pipe wall, with a value ranging from 0.2 to 0.8. Based on the superposition of the gravity settling deflection effect and the slit air resistance, the two are combined into a comprehensive gravity deflection compensation amount, which is then assembled into a matrix according to the array order of the spraying nodes on the cylindrical surface of the pipe wall. This yields the gravity deflection compensation amount matrix for each spraying node. The complete expression of the gravity deflection compensation amount matrix is ​​as follows: ; ; In the formula, for OK The column's gravity deflection compensation matrix; For the first The first axial segment, the first The scalar value of gravity deflection compensation for each circumferential segmented spraying node; This represents the total number of axial segments of the spraying nodes along the pipe axis. This represents the total number of circumferential segments of the spraying nodes along the pipe wall; Index With matrix row and column coordinates correspond; It is the equivalent spring stiffness coefficient when rust inhibitor droplets impact the wall surface, used to convert the dimension of resistance into the dimension of displacement compensation.

[0028] Step 35: Based on the gravity flow compensation matrix and the zoning curvature radius parameters, perform flow demand mapping and numerical normalization to obtain the target flow distribution matrix covering the characteristic areas of the entire pipe inner wall. Specifically, this includes: extracting the compensation values ​​corresponding to the spraying nodes in each characteristic area based on the gravity flow compensation matrix; calculating the basic flow demand value for each spraying node according to the direct proportional mapping relationship between the compensation amount and the rust inhibitor mass flow rate demand. The complete expression for the basic flow demand value is: ; ; In the formula, For the first The basic flow rate requirement for each spraying node is expressed in grams per second. This is the reference mass flow rate under standard operating conditions in a flat pipe wall region; The gain coefficient for mapping compensation quantity to flow rate; For the first The gravity flow compensation amount corresponding to each node; The arithmetic mean of the total mesh compensation; Given the total number of spraying nodes, based on the basic flow requirement value of each spraying node, the curvature radius parameter from the partition morphology feature dataset is introduced to perform surface coverage area correction on the basic flow requirement value. The additional flow required for curved surfaces is then adjusted inversely proportionally to the curvature radius to obtain the curvature-corrected flow requirement value for each spraying node. The complete expression for the curvature-corrected flow requirement value is as follows: ; In the formula, For the first Curvature correction flow requirement for each spraying node; The curvature radius influence weighting factor controls the compensation intensity for flow rate when the local curvature radius of the pipe wall deviates from the nominal radius. The larger the value, the stronger the additional flow compensation obtained in the small curvature radius region, in order to offset the film thickness reduction effect caused by the increase in the coating area of ​​the curved surface. The value ranges from 0.2 to 1.0. For the first The radius of curvature parameter at each node; Using the radius of curvature as a reference, the value is taken as the nominal inner radius of the pipe wall. Based on the plastic thixotropic compensation factor in the rheological compensation coefficient of each characteristic region, thixotropic spreading margin compensation is applied to the curvature correction flow rate requirement value to obtain the final target mass flow rate value for each spraying node. The complete expression for the final target mass flow rate value is as follows: ; In the formula, For the first The final target mass flow rate value for each spraying node; The plastic thixotropic compensation factor characterizes the effect of thixotropic thinning on wetting behavior during the spreading of rust inhibitors on the wall surface, with a value ranging from 0.5 to 15.0. The duration of rust inhibitor droplet spread on the wall surface; The time constant of the thixotropic characteristic of the rust inhibitor; The hyperbolic tangent function is used to describe the nonlinear saturation characteristics of the thixotropic thinning effect. Based on the final target mass flow rate value of each spraying node, the flow rate value of the entire grid is numerically normalized to make the total mass flow rate equal to the rated total flow rate of the spraying system. The matrix is ​​then assembled according to the axial and circumferential segmented array format of the cylindrical surface of the pipe wall to obtain the target flow rate distribution matrix covering the characteristic area of ​​the entire inner wall of the pipe.

[0029] In this embodiment of the invention, by employing a differentiated flow control technology that involves extracting zonal morphology parameters, constructing a fluid dynamics solution domain, and then applying reverse rheological compensation and a three-layer progressive flow distribution, the problems of poor adaptability of traditional uniform spraying to non-flat feature areas such as threaded grooves, reinforcing ribs, and welds, gravity-induced flow deviation and slit air resistance coupling leading to wall offset and uneven coating, and the lack of a geometric-rheological differential compensation model resulting in parameter dependence on empirical trial and error are overcome. This achieves the technical effect of differentiated and precise flow distribution for different feature areas, providing a directly usable target flow distribution matrix for spraying robot path planning and real-time control of spray guns.

[0030] In a preferred embodiment of the present invention, step 4 above may include: Step 41: Based on the target flow distribution matrix, analyze the internal flow density gradient data to obtain the flow rate change distribution sequence along the pipe axis advancement direction. Specifically, this includes: performing a first-order forward difference operation on each circumferential column vector of the matrix along the pipe axis advancement direction based on the target flow distribution matrix, calculating the mass flow rate change between adjacent axial segments column by column, and obtaining the axial flow gradient matrix. The complete expression of the axial flow gradient matrix is: , ; ; ; In the formula, The axial flow gradient matrix is ​​the first Line number Column element, representing the first The circumferential column is in the... The rate of change of flow rate along the pipe axis for each axial segment, in grams per second per millimeter; Assign the first element in the target flow matrix Line number Column elements; The equivalent axial element spacing between adjacent axial segments, in millimeters; This is the total length of the tube shaft; This represents the total number of axial segments; Given the total number of circumferential segments, based on the axial flow gradient matrix, the root mean square value of the gradient at each axial segment position is taken in the circumferential dimension. The gradient fluctuations in each column of the circumferential direction are combined into a flow rate change index for a single axial position. At the same time, the mean and peak values ​​of the circumferential flow gradient are calculated as auxiliary reference quantities to obtain a one-dimensional flow rate change distribution sequence along the pipe axis advancement direction. The complete expression of the flow rate change distribution sequence is: , ; ; ; In the formula, For the first The root mean square value of the flow rate change rate of each axial segment, in grams per second per millimeter; The arithmetic mean of the root mean square values ​​of the total pipe axis flow rate change is used as the reference normalization benchmark for mapping. For the first The maximum absolute value of the flow gradient in each column of the circumferential direction of each axial segment; This indicates the operation of finding the maximum value. This indicates the operation of taking the absolute value.

[0031] Step 42: Based on the flow rate change distribution sequence, map the density distribution rules of the trajectory points and interpolate to generate a spatial node sequence, obtaining a preliminary path polyline extending along the pipe axis. Specifically, this includes: based on the flow rate change distribution sequence, using an inversely proportional decay mapping function to convert the flow rate change rate of each axial segment into the axial spacing of the trajectory points, making the trajectory points dense at the boundary of the characteristic area with drastic flow changes and sparse inside the flat area with gentle flow changes, thus obtaining the axial spacing sequence of trajectory points along the pipe axis. The complete expression of the inversely proportional decay mapping function is: ; In the formula, For the first The axial spacing between the trajectory points corresponding to each axial segment is in millimeters. The preset maximum trajectory point spacing corresponds to low-rate-of-change working conditions in flat areas; The minimum trajectory point spacing is preset to correspond to high rate of change of feature boundaries. This is the sensitivity adjustment coefficient for the density mapping, with a value ranging from 0.5 to 2.0; To obtain the root mean square value of the flow rate change, based on the axial spacing sequence of the trajectory points, starting from the beginning of the pipe shaft, a cumulative summation is performed on the spacing sequence to generate a sequence of axial position coordinates of the trajectory points. The number of axial position coordinates is automatically determined by the condition that the cumulative sum does not exceed the total length of the pipe shaft. Furthermore, circumferential node angles are determined at fixed equal angular intervals at each axial position, resulting in a cylindrical coordinate system. The complete expression for the cumulative axial position coordinates of the following spatial node array is: ; ; ; In the formula, For the first The axial position coordinates of each trajectory point Corresponding to the starting end of the tube shaft; The total number of trajectory points along the axial direction, summed by the spacing, does not exceed the total length of the tube axis. The maximum sequence number is determined by incrementing one; This indicates the operation that takes the largest ordinal number that makes the expression true. For the first The angular coordinates of each circumferential node are in radians. To determine the total number of segments with equal angles in the circumferential direction, based on the spatial node array in cylindrical coordinates, the coordinates of each node are... Transforming from cylindrical coordinates to a three-dimensional Cartesian coordinate system, and sequentially connecting the nodes along each axis at the same circumferential angle to form a longitudinal polyline, yields the result... The complete expression for converting the initial set of polylines along the tube axis from cylindrical coordinates to Cartesian coordinates is: ; ; In the formula, For the first The first axial position, the first A three-dimensional Cartesian coordinate vector of a trajectory node with a circumferential angle; The coordinate vector of the reference origin for the tube axis centerline; The unit direction vector of the tube axis centerline; The nominal inner radius of the pipe wall; and These are orthogonal basis vectors perpendicular to the centerline of the tube axis; For the first The ordered sequence of nodes of the longitudinal path, i.e., the first node in the initial path polyline set. A polyline.

[0032] Step 43: Based on the preliminary path polylines, extract the spatial endpoint coordinates of each line segment and perform orthogonal projection operations to obtain the two-dimensional projection trajectory projected onto the polygon boundary plane corresponding to the geometric feature partition map. Specifically, this includes: extracting spatial line segments formed by connecting adjacent nodes in each polyline from the preliminary path polyline set; recording the starting and ending three-dimensional coordinates of each spatial line segment to obtain the set of endpoint coordinates for the path spatial line segments; and, based on this set, orthogonally projecting the three-dimensional Cartesian coordinates of each line segment endpoint onto the cylindrical unfolding parameter plane where the geometric feature partition map is located, using the tube axis centerline as the projection reference axis. The projection operation preserves the axial position and maps the circumferential spatial position to radian angles, resulting in the projection trajectory of the geometric feature partition map. The complete expression for orthogonal projection operation on the set of projection trajectory line segments represented by two-dimensional coordinates is: ; ; ; ; ; In the formula, For the first The first axial position, the first The axial projection coordinates of the trajectory nodes of each circumferential angle on the cylindrical unfolded surface; The circumferential angular projection coordinates of the corresponding node on the cylindrical unfolded surface, with values ​​ranging from zero to twice the value of pi; The coordinate vector of the orthogonal projection point of the trajectory node on the center line of the tube axis; The two-dimensional projection coordinates of the trajectory node on the cylindrical unfolding parameter plane; For the first The set of two-dimensional projection trajectory lines of a longitudinal path on the unfolded surface of a cylinder; symbol This represents a directed line segment from the starting point's projected coordinates to the ending point's projected coordinates.

[0033] Step 44: Based on the two-dimensional projection trajectory, calculate the intersection coordinates of the projection line segments and the boundaries of each polygon to obtain the set of cross-boundary clipping intersection points. Specifically, this includes: extracting the parameter coordinates of the two endpoints of each projection line segment from the set of two-dimensional projection trajectory line segments, representing the projection line segments as parameterized vectors; simultaneously extracting all boundary line segments of the polygon boundaries of each feature region from the geometric feature partition map to obtain the set of boundary line segments; performing parameterized intersection detection operation on each pair of projection line segments and boundary line segments based on the parameterized vectors of the projection line segments and the set of boundary line segments; solving the line segment parameters of the intersection point by simultaneously solving the parametric equations of the two line segments on the parameter plane to obtain the set of cross-boundary clipping intersection points. The complete expression for parameterized intersection detection is: ; ; ; In the formula, and The parameters for the projected line segment and the boundary line segment are respectively, if and only if and When two line segments intersect effectively; and for axial coordinates and circumferential angular coordinates; and for The coordinates; and The starting point of the boundary line segment The coordinates; and End point of the boundary line segment The coordinates, according to the parametric intersection detection, satisfy... and For each pair of projected line segments and boundary line segments under the condition, calculate the intersection point. The precise coordinates on the parameter plane are recorded, along with the boundary labels of the feature regions to which the intersection points belong and the original 3D path segment indices corresponding to the intersection points. This yields the set of out-of-bounds clipping intersection points. The complete expression for calculating the intersection point coordinates is: ; ; In the formula, For the first The two-dimensional coordinates of the intersection of the out-of-bounds cuts on the cylindrical unfolded surface; For the first The parameter values ​​of the projection line segments corresponding to each intersection point; For the first The boundary labels of the feature regions to which each intersection point belongs include the bottom boundary of the thread groove, the windward boundary of the reinforcing rib, the boundary of the weld reinforcement area, and the boundary of the flat area. For the first The index number of the original 3D path segment corresponding to each intersection point; This represents the total number of intersections where the boundary was crossed.

[0034] In this embodiment of the invention, a progressive adaptive path planning technique is adopted, which involves flow gradient analysis, density mapping, orthogonal projection dimensionality reduction, and parameterized intersection clipping. The flow rate change rate is extracted by axial difference and circumferential root mean square to drive inverse attenuation mapping to achieve adaptive distribution of trajectory point density. Orthogonal projection of cylindrical expansion and parametric intersection detection of line segments are used to achieve accurate comparison and clipping of the path and feature partition boundaries. Therefore, it overcomes the problems of traditional equidistant paths being unable to adapt to drastic changes in flow demand leading to uneven coating thickness, lack of correlation between three-dimensional path and two-dimensional partition map leading to out-of-bounds mis-spraying and missed spraying, and lack of a calculable mapping method from flow gradient to trajectory density leading to reliance on manual experience. Thus, it achieves the technical effect of adaptive polyline path where the density of the spraying path is adaptively adjusted according to flow demand and the trajectory is aligned with the boundary of the feature area without going out of bounds.

[0035] In a preferred embodiment of the present invention, step 5 above may include: Step 51: Based on the set of cross-boundary clipping intersections, locate the redundant line segment intervals outside the polygon boundary and perform endpoint truncation processing to obtain the set of valid line segments retained within the partition contour. Specifically, this includes: based on the set of cross-boundary clipping intersections, traversing each original path line segment intercepted by the partition polygon boundary, and denoting the set of original path line segments as: ; In the formula, The original path segment set, containing Line segment; For the first The original path segment, from its starting coordinate vector and endpoint coordinate vector Defined as follows: For each original line segment, determine the positional affiliation of its two endpoints relative to the boundary of the partitioned polygon. The function for determining the affiliation of a point as inside or outside the polygon is: ; In the formula: For point Compared to the interior determination function of a partitioned polygon, if a point lies within an open region of the polygon... The value is 1 if the endpoints are true and 0 otherwise. Based on the combination of internal and external attribution of the two endpoints, the following four truncation processing logics are executed: ; In the formula: For the first The valid sub-segments retained after the original line segment is truncated may be an empty set; For the first The coordinate vectors of the cross-boundary clipping intersection points are located on the boundary of the partitioned polygon. superior; For line segments The total number of intersections with the boundary of the partitioned polygon, after performing the above judgment and truncation processing on all original line segments, the union of all non-empty retained sub-line segments is collected to form the initial set of valid line segments: Perform endpoint overlap detection on the initial set of valid line segments, and merge line segments with common endpoints whose endpoint spacing is less than the preset tolerance: Among them, This is the initial set of valid line segments; This is the final set of valid line segments after endpoint merging and simplification. For endpoint merging distance tolerance; This is an endpoint merging operator with a value range of 0.1 to 0.5 mm, which merges nodes with spacing smaller than the tolerance. The line segments with common endpoints are merged into a shared endpoint.

[0036] Step 52: Based on the set of valid line segments, identify the topological gaps between the endpoints of adjacent line segments and insert transition connection nodes. Perform path continuity stitching operation to obtain the topologically connected contour path within the partition. Specifically, this includes: constructing the undirected graph adjacency relationship of the endpoints of valid line segments based on the set of valid line segments, treating the start and end points of each line segment as graph nodes, and constructing the endpoint set. ; In the formula, For the set of endpoints, containing Each valid line segment contributes two endpoints: a starting point and an ending point. For the first If the distance between the two endpoints of a given set of endpoint coordinate vectors is less than the preset adjacency tolerance, then the vectors are considered topologically adjacent and a graph edge connection is established. In the formula, It is the set of adjacent edges; Using the adjacency tolerance threshold, an endpoint adjacency graph is constructed. Perform connected component analysis on the adjacency graph: In the formula, Let be the set of connected components of an adjacency graph. A connected component extraction operator for topology analysis of undirected graphs takes a constructed endpoint adjacency graph as input. Each connected component corresponds to a set of topologically continuous but disconnected path segments. The operator traverses endpoint pairs between different connected components and identifies break pairs where the distance along the boundary of the partitioned polygon is less than the preset maximum allowable gap. ; In the formula: It is a set of break pairs; and These are the two endpoints of the break belonging to different connected components; This indicates that there is no connected path between the two endpoints in the adjacency graph; For the boundary of the partitioned polygon Geodesic distance from one endpoint to the other; To determine the maximum allowable gap threshold, for each fracture pair, curvature adaptive interpolation is performed along the boundary of the partitioned polygon to generate transition connection nodes. The number of transition nodes is determined by the geodesic distance along the boundary and the curvature of the boundary of that section. ; In the formula: This represents the total number of transition nodes that need to be inserted between the break pairs; Use the reference insertion spacing; This is the average curvature weighting factor for the boundary segment. The larger the curvature, the larger the value. The value ranges from 0.1 to 0.9, controlling the conformal-smoothing weight allocation for trajectory curvature smoothing. To use the rounding notation, each transition connection node is generated along the boundary according to an equal parameter ratio: ; In the formula, The first interpolation generated along the boundary The coordinate vectors of each transition connection node; For the interpolation operator along the boundary of the partitioned polygon, intermediate points are generated on the boundary according to the parameter ratio, and the sequence of transition connection nodes is inserted between the corresponding breaks to perform path stitching: ; In the formula, The contour path within the topologically connected partition obtained after stitching; The path stitching operator integrates the set of valid line segments with the entire sequence of transition nodes into a globally continuous path.

[0037] Step 53: Based on the topologically connected contour path within the partition, extract discrete spatial coordinate nodes along the path extension direction at preset process intervals to obtain an initial trajectory point sequence constrained within the partition contour. Specifically, this includes: calculating the cumulative arc length parameterized expression along the path extension direction based on the topologically connected contour path within the partition. Using the path start point as zero, sequentially accumulate the Euclidean distance between adjacent nodes along the path. ; In the formula, For the first on the path The cumulative arc length of each discrete node; For the first topologically connected path The spatial coordinate vector of each node; Given the total number of nodes in the topologically connected path, perform monotonically cubic spline interpolation on the cumulative arc length sequence to construct a continuous mapping between the arc length parameter and the path space coordinates: ; In the formula, Arc length parameter The path space coordinate vector at that location, This is a monotonic cubic spline interpolation operator, with inputs consisting of a discrete arc length-coordinate node pair set and the arc length parameter. The output is the path space coordinate vector corresponding to the arc length position; the mapping constructed by this operator satisfies the following conditions: it is continuous everywhere, the first derivative is continuous, the second derivative is continuous and monotonically increasing. The sampling step size is determined according to the preset spraying process spacing parameters. The process spacing is jointly determined by the effective spraying width of the spray gun and the coating overlap rate between adjacent path strips. ; In the formula, The preset process spacing is the arc length interval between adjacent trajectory points; This defines the effective spray width of the spray gun. This represents the coating overlap rate between adjacent path strips, and its value is a decimal between zero and one. The average angle between the path tangent and the partition boundary direction is used to determine the arc length position of each sampling point at equal intervals along the arc length parameter direction, starting from the path start point: ; ; In the formula: For the first The arc length parameter value corresponding to each sampling point; This represents the total number of initial trajectory points; To use the floor notation, for each sampling arc length position, the corresponding spatial coordinates and path tangent vector are extracted using a spline mapping function: ; ; In the formula, For the first The three-dimensional spatial coordinate vector of an initial trajectory point; For the first The path unit tangent vector at each initial trajectory point is obtained by taking the first derivative of the spline mapping function with respect to the arc length parameter and normalizing it. All extracted nodes are arranged in ascending order of arc length to form a sequence of initial trajectory points constrained within the partition contour. In the formula, An initial trajectory point list constrained within the partition contour, each element containing trajectory point coordinates and path tangent vector.

[0038] In this embodiment of the invention, by employing a three-step progressive processing approach—including cross-boundary intersection truncation and endpoint merging to extract effective line segments, adjacency graph connectivity component analysis to identify breaks and adaptive interpolation along boundary curvature to stitch the path, and arc length parameterization—the invention overcomes three technical problems: uncut redundant line segments across boundaries leading to mis-spraying, unstitched topological breaks leading to empty travel and film breakage, and uneven spacing leading to deviations from the overlap rate specifications. This achieves the technical effect of strictly constraining the trajectory within the partition, ensuring full connectivity of the path without breaks, and uniformly controlling the spacing between trajectory points to meet the requirements of the overlap process.

[0039] In a preferred embodiment of the present invention, step 6 above may include: Step 61: Based on the spatial pose coordinates of the initial trajectory point sequence, the tube shaft assembly tolerance parameters and the crawling mechanism micro-vibration parameters are superimposed to simulate pose disturbance. The radial deviation of each trajectory node relative to the theoretical tube shaft center is calculated to obtain the radial cumulative deviation distribution sequence. Specifically, this includes: based on the spatial pose coordinates of each trajectory point in the initial trajectory point sequence, the tube shaft assembly tolerance parameters and the crawling mechanism micro-vibration parameters are introduced to construct a joint disturbance model. The tube shaft assembly tolerance parameters are modeled as radial offset envelope functions distributed along the tube shaft axis. ; In the formula: Axial position Assembly tolerance radial offset at the location; The reference assembly offset constant; The linear rate of change of assembly offset along the axial direction; This represents the periodic fluctuation amplitude of the assembly deviation. Wavelength is a characteristic wavelength. As the initial phase angle, the micro-vibration parameters of the crawling mechanism are modeled as a high-frequency perturbation function coupled in the time and space domains: ; In the formula, For a moment In axial position The radial offset of the micro-vibration at the location; Used as a reference for vibration amplitude; This is the damping coefficient of the vibration along the axial direction; This is the dominant frequency of vibration; As a spatial phase function related to the axial position, the radial offsets of the two perturbation models are spatiotemporally superimposed synchronously. For the first... A trajectory point, and its corresponding time. and axial position Extract the radial deviation of the trajectory point relative to the theoretical tube axis center from the coordinates of the trajectory point: ; In the formula, For the first Radial cumulative deviation of each trajectory point; This is the random disturbance intensity coefficient; For a standard normally distributed random variable, the radial deviation is calculated for each trajectory point to construct a radial cumulative deviation distribution sequence: ; In the formula, The radial cumulative deviation distribution sequence contains There are 10 deviation values, and each deviation value corresponds to a trajectory point. Given the sequence length, the radial cumulative deviation distribution sequence composed of the radial deviations of each trajectory node relative to the theoretical tube axis center is obtained.

[0040] Step 62: Based on the radial cumulative deviation distribution sequence, extract the deviation fluctuation trend features within the sequence, and perform frequency domain filtering and high-frequency noise stripping on the deviation fluctuation trend features to obtain a smooth radial pose compensation vector sequence corresponding to each trajectory node. Specifically, this includes: extracting the deviation fluctuation trend features within the sequence based on the radial cumulative deviation distribution sequence, treating the discrete deviation sequence as a time-domain sampled signal, and performing a discrete Fourier transform on it. ; In the formula, For the deviation sequence in the th Fourier coefficients at each frequency domain index; The imaginary unit; Given the sequence length, a low-pass filter transfer function is constructed in the frequency domain to filter out high-frequency noise components and retain the low-frequency deviation trend that reflects the actual pipe wall morphology fluctuations: ; ; In the formula, This is the transfer function for the low-pass filter; This is the frequency domain index corresponding to the cutoff frequency; The cutoff frequency is taken as the dominant vibration frequency of the crawling mechanism. 0.3 to 0.5 times; Using the spatial sampling frequency of the trajectory points, perform an inverse Fourier transform on the filtered frequency domain signal to reconstruct the smoothed deviation sequence: ; In the formula: For the first The radial deviation value of each trajectory point after filtering and smoothing is used to determine the spatial orientation of the compensation vector based on its radial deviation direction. The compensation direction is taken as the projection direction of the outward normal direction of the inner wall surface of the pipe at that point onto the cross-section. Let the first... The unit normal vector of the inner wall surface at each trajectory point is: Then the radial compensation unit direction vector is: ; In the formula: For the first Radial compensation unit direction vector of each trajectory point; This is the unit normal vector of the inner wall surface at that point; The path tangent vector at this point is combined with the smoothed radial deviation magnitude and the compensation direction to form the smoothed radial pose compensation vector for each trajectory point: ; ; In the formula: For the first Smooth radial pose compensation vector for each trajectory point; This is a smooth radial pose compensation vector sequence.

[0041] Step 63: Based on the smooth radial pose compensation vector sequence, extract the compensation vector components of each node in the sequence, superimpose the compensation vector components along the normal of the inner wall surface onto the corresponding spatial coordinates of the initial trajectory point set, and perform adaptive interpolation correction calculation to obtain the corrected trajectory point set that conforms to the actual three-dimensional shape of the inner wall of the pipe. Specifically, this includes: based on the smooth radial pose compensation vector sequence, extract the compensation vector components of each node in the sequence, and superimpose each compensation vector along the normal of the inner wall surface onto the corresponding spatial coordinates of the initial trajectory point set. ; In the formula, For the first The spatial coordinate vectors of the trajectory points after preliminary compensation and correction; Given the original initial trajectory point coordinate vector, a uniformity check of the spacing between adjacent points is performed on the pre-compensated trajectory point sequence. If the spacing between adjacent corrected points exceeds the preset tolerance range, adaptive interpolation is performed between the two points. The criterion for determining the spacing between adjacent points is as follows: ;like Then insert between the two points. Intermediate nodes: ; ; In the formula, The distance between adjacent correction points; This is the maximum allowable spacing threshold; The target is uniform spacing; The number of nodes to be inserted; In the first and the Inserting the first point between points Given the coordinates of several intermediate nodes, perform linear interpolation on the compensation vector at the inserted node: ; The initial compensation points and interpolation points are merged sequentially to form a set of corrected trajectory points that conform to the actual three-dimensional morphology of the inner wall of the pipe: ; In the formula, To correct the trajectory point set, including A trajectory point; and These are the first interpolation points after merging. Spatial coordinates and tangent vectors of each corrected trajectory point.

[0042] Step 64: Based on the corrected trajectory point set, and combined with the kinematic boundary conditions of the end effector of the spraying actuator, perform trajectory curvature smoothing and continuity verification to obtain an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal pipe. Specifically, this includes: performing trajectory curvature smoothing based on the corrected trajectory point set and the kinematic boundary conditions of the end effector of the spraying actuator, and calculating the discrete curvature point by point of the corrected trajectory point set. ; In the formula: For the first Discrete curvature values ​​at each trajectory point; For vector cross product operation, the curvature values ​​at each point are compared with the maximum permissible curvature at the end of the spraying actuator. Comparisons are made, and local iterative smoothing is performed on points where curvature exceeds limits: ; ; In the formula, For the first The coordinates of the trajectory points after one smoothing iteration; As a boundary smoothing relaxation factor, boundary trajectory nodes that may cross the boundary due to the superposition of compensation vectors are pulled back into the partition contour, ensuring that the spraying trajectory is strictly constrained within the effective working area. Controlling the pullback strength is a boundary compliance correction operation of the trajectory constraint layer; The maximum permissible curvature of the actuator; To maximize the number of iterations, a continuity check is performed on the smoothed trajectory point list, examining the path tangential change rate between adjacent trajectory points segment by segment: ; like Then, a tangent vector smoothing process based on spherical linear interpolation is performed between these segments, and the spatial positions of the trajectory points are adjusted accordingly. After all curvature smoothing and continuity checks pass, the final AI spraying trajectory is output: ; In the formula: The final output is an AI spraying trajectory that conforms to the three-dimensional contours of the inner wall of the metal pipe, including One trajectory node; For the first The spatial coordinates of the trajectory points; For the first The path tangent vector of each trajectory point; For the first The end-effector velocity vector of the spray gun actuator at each trajectory point; This represents the maximum permissible rate of tangential change.

[0043] In this embodiment of the invention, a four-step progressive processing method is adopted, which involves extracting radial cumulative deviation by jointly modeling and simulating disturbances based on assembly tolerance and crawling micro-vibration, removing high-frequency noise by frequency domain low-pass filtering to generate a smooth compensation vector, correcting the fit of the real wall morphology by normal superposition and adaptive interpolation, and adapting the kinematic boundary of the actuator by curvature constraint smoothing and continuity verification. This overcomes four technical problems: deviation of trajectory points from the real wall due to coupling between pipe shaft assembly deviation and mechanism vibration, inaccurate oscillation of compensation vector due to high-frequency noise interference, loss of wall undulation features due to insufficient interpolation, and singular motion of actuator due to excessive curvature. As a result, the technical effects of the trajectory point radial pose closely fitting the three-dimensional undulation contour of the inner wall of the pipe after compensation, the compensation vector being smooth, stable and oscillating without oscillation, the complete preservation of the curved surface morphology features, and the continuous and executable trajectory curvature satisfying the end kinematic constraints of the spraying mechanism are achieved.

[0044] like Figure 2 As shown, embodiments of the present invention also provide a vision-guided AI spraying trajectory planning system for rust inhibitors on the inner wall of metal pipes, including: The acquisition module is used to acquire the original sequence images of the inner wall of the metal tube collected by the endoscopic vision probe, perform optical distortion correction and multi-view point cloud spatial registration on the original sequence images, and obtain the three-dimensional topological mesh of the inner wall. The extraction module is used to extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and calculate the local curvature gradient distribution using the node coordinates and surface normal vectors; the local curvature gradient distribution is segmented by threshold and the feature boundary is extracted to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. The module is used to construct a constrained model of the aerodynamic settling path of rust inhibitor droplets based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map. The rheological dynamic parameters are inverted and the gravity deflection compensation is calculated through the constrained model to obtain the target flow distribution matrix. The mapping module is used to parse the flow density gradient data inside the matrix, and based on the flow density gradient data, map the density distribution rules of the trajectory points along the pipe axis advancement direction to obtain the preliminary path polyline; project the preliminary path polyline onto the polygon boundary plane corresponding to the geometric feature partition map, and traverse and solve the intersection coordinates between the polygon boundary and each segment of the polyline; The reconstruction module is used to perform outbound line segment clipping and internal valid line segment topological continuity reconstruction on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; The calculation module is used to calculate the radial cumulative deviation based on the spatial pose coordinates of the initial trajectory point series, combined with the tube shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, and output an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal tube.

[0045] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0046] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0047] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0048] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A vision-guided AI-based trajectory planning method for spraying rust inhibitors onto the inner wall of metal pipes, characterized in that, The method includes: The original sequence images of the inner wall of the metal tube were acquired by the endoscopic vision probe. Optical distortion correction and multi-view point cloud spatial registration were performed on the original sequence images to obtain the three-dimensional topological mesh of the inner wall. Extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and calculate the local curvature gradient distribution using the node coordinates and surface normal vectors; perform threshold segmentation and feature boundary extraction on the local curvature gradient distribution to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. Based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map, a constrained model of the aerodynamic settling path of the rust inhibitor droplets is constructed. The rheological dynamic parameters are inverted and the gravity deflection compensation is calculated through the constrained model to obtain the target flow distribution matrix. The flow density gradient data inside the analytical matrix is ​​used to map the distribution rules of trajectory points along the pipe axis advancement direction based on the flow density gradient data, and a preliminary path polyline is obtained. The preliminary path polyline is then projected onto the polygon boundary plane corresponding to the geometric feature partition map, and the intersection coordinates between the polygon boundary and each segment of the polyline are solved by iterating through the polygon boundary. Perform out-of-bounds segment clipping and topological continuity reconstruction of internal valid segments on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; Based on the spatial pose coordinates of the initial trajectory points, combined with the tube shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, radial cumulative deviation is calculated, and an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal tube is output.

2. The method for planning the trajectory of rust inhibitor spraying on the inner wall of metal pipes based on vision guidance according to claim 1, characterized in that, The original sequence of images of the inner wall of the metal tube acquired by the endoscopic vision probe is obtained. Optical distortion correction and multi-view point cloud spatial registration are performed on the original sequence of images to obtain a three-dimensional topological mesh of the inner wall, including: Acquire raw sequence image frames continuously captured by the endoscopic vision probe as it advances along the tube axis; based on the raw sequence image frames, extract the camera intrinsic parameter calibration data and lens radial distortion coefficient and perform pixel-level mapping correction to obtain a standard two-dimensional image sequence that eliminates barrel distortion; Based on the standard two-dimensional image sequence, feature key points within the overlapping field of view of adjacent image frames are extracted and the relative pose transformation matrix is ​​calculated. The pixel depth information in the standard two-dimensional image sequence is transformed to the unified world coordinate system through the relative pose transformation matrix to obtain the initial discrete point cloud dataset of the inner wall of the tube. Based on the initial discrete point cloud dataset of the inner wall of the pipe, the consistency of the neighborhood normal vector of the spatial points is calculated and outlier noise points are removed to obtain the clean point cloud data after removing outlier noise points; local surface fitting and triangular meshing are performed on the clean point cloud data to obtain the three-dimensional topological mesh of the inner wall that represents the continuous morphology of the inner wall of the pipe.

3. The visual guidance-based AI spraying trajectory planning method for rust inhibitors on the inner wall of metal pipes according to claim 2, characterized in that, The node coordinates and surface normal vectors of the 3D topological mesh of the inner wall are extracted, and the local curvature gradient distribution is calculated using the node coordinates and surface normal vectors. Threshold segmentation and feature boundary extraction are performed on the local curvature gradient distribution to obtain a geometric feature partition map of the spatial range including the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld reinforcement area, including: Extract the vertex coordinates of the triangular facets and the normal vectors of the adjacent facets from the three-dimensional topological mesh of the inner wall. Based on the vertex coordinates of the triangular facets and the normal vectors of the adjacent facets, construct a local differential geometric neighborhood. Perform principal curvature calculation and gradient field fitting on the local differential geometric neighborhood to obtain the local curvature gradient distribution tensor that characterizes the trend of surface undulation. Based on the local curvature gradient distribution tensor, the curvature extreme points of each grid unit are extracted and a high-low curvature segmentation threshold is set; interval mapping and connected component labeling are performed using the high-low curvature segmentation threshold to obtain a curvature classification mask that initially distinguishes between convex ridges and concave grooves. The curvature abrupt transition zone at the boundary of the curvature classification mask is extracted and boundary smoothing filtering is performed to obtain the smoothed boundary contour. Based on the smoothed boundary contour and the projection relationship of the pipe axis centerline, spatial region mapping and feature label assignment are performed to obtain a geometric feature partition map with the spatial range of the bottom of the thread groove, the windward side of the reinforcing rib, and the weld excess height area marked.

4. The visual guidance-based AI spraying trajectory planning method for rust inhibitors on the inner wall of metal pipes according to claim 3, characterized in that, Based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partitioning map, a constrained model for the aerodynamic settling path of the rust inhibitor droplets is constructed. Through the constrained model, rheological dynamic parameters are inverted and gravity flow compensation is calculated to obtain the target flow distribution matrix, including: Based on the geometric feature partitioning map, the spatial tilt angle parameter and radius of curvature parameter of each partition are extracted to obtain the partition morphology feature dataset; Based on the partition morphology feature dataset, an aerodynamic settling path constraint model is constructed by combining the physical properties of rust inhibitor droplets, and a fluid dynamics solution domain including the differential equation of droplet motion trajectory and pipe wall boundary conditions is obtained. Based on the fluid dynamics solution domain, the spatial tilt angle parameter and the radius of curvature parameter are substituted to perform inverse parameter identification and viscous rheological inversion, and the rheological compensation coefficients corresponding to each characteristic region are obtained. Based on the rheological compensation coefficient, the superposition effect of gravity settlement deflection effect and slit air resistance is calculated to obtain the gravity deflection compensation matrix for each spraying node. Based on the gravity deflection compensation matrix and the zonal curvature radius parameter, flow demand mapping and numerical normalization are performed to obtain the target flow distribution matrix covering the characteristic area of ​​the entire inner wall of the pipe.

5. The visual guidance-based AI spraying trajectory planning method for rust inhibitors on the inner wall of metal pipes according to claim 4, characterized in that, The flow density gradient data inside the analytical matrix is ​​used to map the distribution rules of trajectory points along the pipe axis advancement direction based on the flow density gradient data, and a preliminary path polyline is obtained. Project the initial path polyline onto the polygon boundary plane corresponding to the geometric feature partition map, and iterate through and solve for the intersection coordinates of the polygon boundary and each segment of the polyline, including: Based on the target flow distribution matrix, the internal flow density gradient data is analyzed to obtain the flow rate distribution sequence along the pipe axis advance direction; Based on the flow rate change distribution sequence, the density distribution rules of the trajectory points are mapped and spatial node sequences are generated by interpolation to obtain a preliminary path polyline extending along the pipe axis. Based on the preliminary path polyline, the spatial endpoint coordinates of each line segment are extracted and orthogonal projection operation is performed to obtain the two-dimensional projection trajectory projected onto the polygon boundary plane corresponding to the geometric feature partition map; Based on the two-dimensional projection trajectory, the intersection coordinates of the projection line segments and the boundaries of each polygon are calculated to obtain the set of cross-boundary clipping intersection points.

6. The visual guidance-based AI spraying trajectory planning method for rust inhibitors on the inner wall of metal pipes according to claim 5, characterized in that, By performing out-of-bounds segment clipping and topological continuity reconstruction of internal valid segments on intersecting coordinates, an initial trajectory point list constrained within the partition contour is obtained, including: Based on the set of cross-boundary clipping intersections, locate the redundant line segment interval outside the polygon boundary and perform endpoint truncation to obtain the set of valid line segments retained inside the partition contour. Based on the set of valid line segments, identify the topological gaps between the endpoints of adjacent line segments and insert transition connection nodes, perform path continuity stitching operation, and obtain the topologically connected intra-partition contour path. Based on the topologically connected contour path within the partition, discrete spatial coordinate nodes are extracted along the path extension direction at a preset process spacing to obtain an initial trajectory point sequence constrained within the partition contour.

7. The method for planning the trajectory of visually guided rust inhibitor spraying on the inner wall of metal pipes according to claim 6, characterized in that, Based on the spatial pose coordinates of the initial trajectory points, combined with the pipe shaft assembly tolerances and the micro-vibration parameters of the crawling mechanism, radial cumulative deviation is calculated, outputting an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal pipe, including: Based on the spatial pose coordinates of the initial trajectory point sequence, pose disturbance simulation is performed by superimposing the tube shaft assembly tolerance parameters and the crawling mechanism micro-vibration parameters, calculating the radial deviation of each trajectory node relative to the theoretical tube shaft center, and obtaining the radial cumulative deviation distribution sequence. Based on the radial cumulative deviation distribution sequence, the deviation fluctuation trend features within the sequence are extracted. Frequency domain filtering and high-frequency noise stripping are performed on the deviation fluctuation trend features to obtain a smooth radial pose compensation vector sequence corresponding to each trajectory node. Based on the smooth radial pose compensation vector sequence, the compensation vector components of each node in the sequence are extracted. The compensation vector components are superimposed along the inner wall surface normal to the corresponding spatial coordinates of the initial trajectory point sequence and adaptive interpolation correction operation is performed to obtain the corrected trajectory point set that conforms to the actual three-dimensional shape of the inner wall of the pipe. Based on the corrected trajectory point set, and combined with the kinematic boundary of the end effector of the spraying actuator, the trajectory curvature is smoothed and the continuity is verified to obtain an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal pipe.

8. A vision-guided AI-based trajectory planning system for spraying rust inhibitors onto the inner wall of metal pipes, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire the original sequence images of the inner wall of the metal tube collected by the endoscopic vision probe, perform optical distortion correction and multi-view point cloud spatial registration on the original sequence images, and obtain the three-dimensional topological mesh of the inner wall. The extraction module is used to extract the node coordinates and surface normal vectors of the three-dimensional topological mesh of the inner wall, and to calculate the local curvature gradient distribution using the node coordinates and surface normal vectors. Threshold segmentation and feature boundary extraction are performed on the local curvature gradient distribution to obtain a geometric feature partition map of the spatial range of the bottom of the threaded groove, the windward side of the reinforcing rib, and the weld excess height area. The module is used to construct a constrained model of the aerodynamic settling path of rust inhibitor droplets based on the spatial tilt angle and radius of curvature parameters corresponding to the geometric feature partition map. The rheological dynamic parameters are inverted and the gravity deflection compensation is calculated through the constrained model to obtain the target flow distribution matrix. The mapping module is used to parse the flow density gradient data inside the matrix, and based on the flow density gradient data, map the density distribution rules of the trajectory points along the pipe axis advancement direction to obtain the preliminary path polyline; project the preliminary path polyline onto the polygon boundary plane corresponding to the geometric feature partition map, and traverse and solve the intersection coordinates between the polygon boundary and each segment of the polyline; The reconstruction module is used to perform outbound line segment clipping and internal valid line segment topological continuity reconstruction on intersecting coordinates to obtain an initial trajectory point list constrained within the partition contour; The calculation module is used to calculate the radial cumulative deviation based on the spatial pose coordinates of the initial trajectory point series, combined with the tube shaft assembly tolerance and the micro-vibration parameters of the crawling mechanism, and output an AI spraying trajectory that conforms to the three-dimensional undulating contour of the inner wall of the metal tube.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images