Method and system for automated machining path planning based on machine vision

Abstract

The application provides a kind of automatic machining path planning method and system based on machine vision, it is related to machine vision technical field.First, the multi-angle vision data set of processing object is acquired, then the three-dimensional structure feature extraction is carried out to multi-angle vision data set, obtains contour boundary, surface curvature and regional depression and other surface structure features, based on above feature identification processing constraint region feature, including non-machinable region boundary, precision machining area and machining path turning limit feature, then processing constraint region feature is associated with preset machining process parameter fusion, generates the path planning constraint feature set containing regional machining priority, path continuity constraint and machining precision matching feature, finally, the path generation processing of path planning constraint feature set is carried out by calling pre-training model, generates the automatic machining path instruction set containing machining path coordinate sequence, path execution order and machining parameter configuration, realizes efficient, accurate automatic machining path planning.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and more specifically, to an automated processing path planning method and system based on machine vision. Background Technology

[0002] In the field of automated machining, machining path planning is a crucial step in ensuring machining quality and efficiency. Traditional machining path planning methods often rely on human experience or simple two-dimensional drawing information. Human experience-based path planning suffers from high subjectivity, low efficiency, and is prone to errors, especially for complex-shaped objects, where it's difficult to comprehensively consider various machining constraints, leading to unstable machining quality. Methods based on two-dimensional drawing information lack accurate understanding of the three-dimensional structure of the object, failing to effectively identify complex surface features such as surface curvature variations and regional depressions. This makes it difficult to accurately determine key constraints such as unmachinable areas, precision machining areas, and machining path turning restrictions. Furthermore, existing methods fail to deeply integrate the characteristics of machining constraint areas with preset machining process parameters during path planning, resulting in deficiencies in areas such as regional machining priority, path continuity, and machining accuracy matching, failing to meet the high precision, high efficiency, and high stability requirements of modern automated machining. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an automated processing path planning method based on machine vision, the method comprising: A multi-angle visual data set of the processing object is obtained, wherein the multi-angle visual data set includes surface images of the processing object acquired from different viewpoints and corresponding viewpoint parameters; The multi-angle visual data set is subjected to three-dimensional structural feature extraction processing to obtain the surface structural features of the processed object. The surface structural features include the contour boundary features, surface curvature features and regional concavity features of the processed object. Based on the surface structure features, the machining constraint region features of the workpiece are identified. The machining constraint region features include the boundary features of the unmachinable region, the features of the precision machining region, and the turning restriction features of the machining path. The processing constraint region features are associated and fused with preset processing process parameters to generate a path planning constraint feature set, which includes regional processing priority features, path continuity constraint features, and processing accuracy matching features. The pre-trained processing path planning model is invoked to perform path generation processing on the path planning constraint feature set, generating an automated processing path instruction set containing processing path coordinate sequences, path execution order, and processing parameter configurations.

[0004] In another aspect, embodiments of the present invention also provide an automated processing path planning system based on machine vision, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0005] Based on the above, this embodiment of the invention, by acquiring a multi-angle visual data set of the processing object, can comprehensively and accurately capture the three-dimensional structural information of the processing object. By performing three-dimensional structural feature extraction processing on the multi-angle visual data set, it can accurately obtain surface structural features such as contour boundary features, surface curvature features, and regional concavity features of the processing object. Based on the surface structural features, it can identify processing constraint region features, accurately determine key information such as the boundaries of unprocessable areas, precision processing areas, and processing path turning restrictions, effectively avoiding collisions, overcutting, and other problems during processing, ensuring processing safety and quality. The processing constraint region features are correlated and fused with preset processing process parameters to generate a path planning constraint feature set containing regional processing priority features, path continuity constraint features, and processing accuracy matching features. This allows path planning to fully consider processing process requirements and achieve reasonable optimization of the processing path. Finally, a pre-trained processing path planning model is called to generate a path from the path planning constraint feature set. The generated automated processing path instruction set includes a processing path coordinate sequence, path execution order, and processing parameter configuration, which can guide automated processing equipment to complete processing tasks efficiently and accurately, significantly improving processing efficiency and quality, and reducing processing costs. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the automated processing path planning method based on machine vision provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the automated processing path planning system based on machine vision provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an automated processing path planning method based on machine vision provided in one embodiment of the present invention. The automated processing path planning method based on machine vision will be described in detail below.

[0009] Step S110: Obtain a multi-angle visual data set of the processing object, wherein the multi-angle visual data set includes surface images of the processing object collected from different viewpoints and corresponding viewpoint parameters.

[0010] In this embodiment, a mechanical part is selected as the processing object. This mechanical part typically has a complex curved surface structure and multiple holes. To obtain its multi-angle visual data set, an image acquisition system consisting of multiple industrial cameras needs to be deployed. These industrial cameras are fixed at different positions and angles, distributed around the processing object to ensure that the processing object can be photographed from multiple directions.

[0011] During the acquisition process, each industrial camera can simultaneously capture images of the surface of the workpiece at preset time intervals. Each time an image is captured, the corresponding viewing angle parameters are automatically recorded. These parameters include the camera's position coordinates, shooting angle, focal length, and intrinsic parameter matrix. For example, one camera might be positioned directly in front of the workpiece, its position coordinates determined by a 3D coordinate system, its shooting angle perpendicular to the front of the workpiece, its focal length set to a value that clearly captures the entire front area, and its intrinsic parameter matrix being the inherent parameters calibrated at the factory.

[0012] By taking multiple photos using the above method, a large number of images of the surface of the object being processed from different perspectives can be obtained, which, together with the corresponding perspective parameters, constitute a multi-angle visual data set of the object being processed.

[0013] Step S120: Perform three-dimensional structural feature extraction processing on the multi-angle visual data set to obtain the surface structural features of the processed object. The surface structural features include the contour boundary features, surface curvature features, and regional concavity features of the processed object.

[0014] After obtaining a multi-angle visual data set, three-dimensional structural features can be extracted from it. This process is to restore the three-dimensional structural information of the object being processed from the two-dimensional image data, and then extract the required surface structural features.

[0015] Step S121: Input the multi-angle visual data set into the multi-view feature encoding layer of the three-dimensional structural feature extraction network, perform depth feature extraction processing on the surface images of the processed object from different perspectives, and generate a multi-view feature map containing perspective difference information.

[0016] In this embodiment, the multi-view feature encoding layer of the 3D structural feature extraction network consists of multiple convolutional layers and pooling layers. First, each surface image of the processed object in the multi-angle visual dataset is input into the multi-view feature encoding layer.

[0017] For each image, the first convolutional layer uses a predetermined number of convolutional kernels. Each kernel performs a sliding convolution operation on the image, extracting basic features such as edges and textures. During the convolution operation, the kernel moves across the image with a preset stride. At each position, it is multiplied element-wise with the corresponding image region and summed to obtain a feature value. Through these operations, a corresponding feature map is generated.

[0018] Next, the pooling layer downsamples the feature map output by the convolutional layer. For example, it uses max pooling to divide the feature map into multiple non-overlapping regions. The maximum value of each region is taken as the feature value of that region, thereby reducing the dimensionality of the feature data while retaining important feature information.

[0019] After alternating processing by multiple convolutional and pooling layers, the multi-view feature encoding layer further fuses features extracted from images from different viewpoints to reflect the differences between viewpoints. For example, by comparing feature values ​​at corresponding locations under different viewpoints, the degree of difference between them is calculated, and this difference information is encoded into the final feature map, generating a multi-view feature map containing viewpoint difference information. This multi-view feature map is a multi-dimensional feature set, with each dimension corresponding to different feature information, which can comprehensively reflect the surface features and differences of the processed object under different viewpoints.

[0020] Step S122: The feature matching module of the three-dimensional structural feature extraction network performs cross-view feature point association processing on the multi-view feature map to generate a set of feature point matching pairs with spatial correspondence.

[0021] The feature matching module first extracts feature points from the multi-view feature map from each viewpoint. For each feature point, feature information within a defined area around it can be extracted as a descriptor for that feature point. These descriptors contain information such as the grayscale changes and texture structure of the feature point, which are used to characterize the uniqueness of that feature point.

[0022] Then, the feature matching module uses a feature point descriptor similarity comparison method to associate feature points across different viewpoints. For feature points from different viewpoints, a similarity metric is calculated between their descriptors, for example, by calculating the Euclidean distance or cosine similarity between the two descriptors. When the similarity metric meets a preset threshold, the two feature points are considered to be matched, belonging to the projections of the same spatial point on the surface of the processed object from different viewpoints.

[0023] During the matching process, some constraints are used to filter features to improve accuracy. For example, based on the epipolar constraint principle, for a feature point in one view, the matching feature point in another view must lie on the epipolar line corresponding to that feature point. This constraint can eliminate some obviously mismatched feature point pairs.

[0024] Through the above processing, a set of feature point matching pairs with spatial correspondence is finally generated. Each element in this set consists of two feature points from different viewpoints, and there is a clear spatial correspondence between them, which can reflect the features of the same point on the surface of the processed object under different viewpoints.

[0025] Step S123: Based on the feature point matching pair set and the corresponding view parameters, calculate the three-dimensional coordinate information of the feature points on the surface of the processing object through triangulation processing, and construct the point cloud data set of the processing object.

[0026] Triangulation is performed using the pixel coordinates of the two feature points in the corresponding image for each matching pair in the feature point matching pair set, as well as the camera intrinsic and extrinsic parameters (position and pose) in the corresponding viewpoint parameters.

[0027] For each feature point matching pair, the projection matrices of the two cameras (calculated from the camera intrinsic and extrinsic parameters) and the pixel coordinates of the feature point on the image planes of the two cameras are known. According to the triangulation principle, the projections of a point in space onto the image planes of the two cameras satisfy a set geometric relationship. By solving the system of equations corresponding to this geometric relationship, the three-dimensional coordinate information of the point in space can be obtained.

[0028] Specifically, let the three-dimensional coordinates of a spatial point be (X, Y, Z), its pixel coordinates in the first camera image be (u1, v1), and its pixel coordinates in the second camera image be (u2, v2). The projection matrices of the two cameras are P1 and P2, respectively. According to the projection relationship, there exist normalized forms of the equations u1 = P1 * [X, Y, Z, 1]^T and u2 = P2 * [X, Y, Z, 1]^T. By solving the system of equations consisting of these two equations, the values ​​of X, Y, and Z can be obtained, which are the three-dimensional coordinates of the feature point.

[0029] The above triangulation process is performed on all matching pairs in the feature point matching pair set to obtain the three-dimensional coordinate information corresponding to each feature point. Combining these three-dimensional coordinate information pieces constructs a point cloud dataset of the processed object. This point cloud dataset consists of a large number of three-dimensional coordinate points, each representing a feature point on the surface of the processed object, and can initially reflect the three-dimensional shape of the processed object.

[0030] Step S124: Perform surface reconstruction processing on the point cloud data set to generate a three-dimensional mesh model of the object to be processed. The three-dimensional mesh model includes vertex coordinate information, facet connection relationship and normal vector parameters.

[0031] First, the point cloud dataset is preprocessed, including the removal of noise and outliers. A statistical filtering method is used to calculate the average distance from each point to other points in its neighborhood. Based on a preset standard deviation threshold, points whose average distance exceeds the threshold are identified as noise or outliers and removed to improve the quality of the point cloud data.

[0032] Next, point cloud registration is performed to unify the point cloud data acquired from different perspectives into the same coordinate system. By using an iterative nearest-point algorithm, corresponding point pairs between two point clouds are continuously found, and the transformation matrix is ​​calculated to make the two point clouds overlap as much as possible in space. This process is repeated multiple times until the convergence condition is met.

[0033] Then, surface reconstruction is performed. The Poisson surface reconstruction algorithm is used. First, the normal vector of the point cloud data is calculated. Then, the covariance matrix of the neighborhood points of each point is calculated by principal component analysis to obtain eigenvalues ​​and eigenvectors. The eigenvector corresponding to the smallest eigenvalue is the normal vector direction of that point.

[0034] After obtaining the normal vector, an implicit function is constructed to represent the surface of the object being processed. This implicit function is positive inside the point cloud, negative outside, and zero at the surface. The implicit function is obtained by solving the Poisson equation, and then the moving cube algorithm is used to extract isosurfaces from the implicit function. These isosurfaces constitute the three-dimensional mesh model of the object being processed.

[0035] The generated 3D mesh model contains vertex coordinate information, with each vertex having corresponding 3D coordinates; patch connectivity, i.e., which vertices form a triangular patch; and normal vector parameters, with each patch having a normal vector to represent the orientation of the patch.

[0036] Step S125: Extract contour boundary features from the three-dimensional mesh model. The contour boundary features are described by the edge vertex sequence of the mesh model and the angle parameter between the normal vectors of adjacent facets.

[0037] Traverse all edges in the 3D mesh model and determine whether each edge is a boundary edge. If an edge is shared by only one triangular facet, then that edge is a boundary edge and forms part of the contour boundary.

[0038] Collect all the vertices of the boundary edges and arrange them according to their spatial connection order to form an edge vertex sequence. This edge vertex sequence can reflect the shape and orientation of the contour boundary.

[0039] Simultaneously, for each vertex on a boundary edge, calculate the angle between the normal vectors of its two adjacent faces. Specifically, take the normal vectors of the two adjacent faces of the vertex (for boundary vertices, one face may not exist; in this case, take the existing face and its adjacent non-boundary face), and then calculate the angle between these two normal vectors.

[0040] The edge vertex sequence and the corresponding angle parameters of the normal vectors of adjacent faces are combined to form the contour boundary features of the processed object. For example, the edge vertex sequence can be a series of vertex coordinates arranged in order, such as (V1x, V1y, V1z), (V2x, V2y, V2z), ..., (Vnx, Vny, Vnz), while the angle parameters of the normal vectors of adjacent faces are the sequence of angle values ​​corresponding to these vertices, such as θ1, θ2, ..., θn.

[0041] Step S126: Calculate the curvature value of each vertex in the three-dimensional mesh model and generate a surface curvature feature distribution map containing principal curvature values, curvature directions and curvature change rates.

[0042] For each vertex in the 3D mesh model, select a set number of neighboring vertices around it, for example, find the k nearest vertices to that vertex using the k nearest neighbor algorithm.

[0043] Calculate the covariance matrix of the vertex, and use the coordinate offsets of the neighboring vertices relative to this vertex as data points. Calculate the covariance matrix of these data points to reflect the distribution of the neighboring points.

[0044] The covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues ​​λ1, λ2, and λ3, where λ1 ≥ λ2 ≥ λ3. The principal curvature values ​​are then calculated based on these eigenvalues, where the maximum principal curvature k1 = (λ2 + λ3) / (λ1 + λ2 + λ3) and the minimum principal curvature k2 = (λ1 + λ3) / (λ1 + λ2 + λ3) (this is a simplified description of the calculation logic; actual calculations must follow strict curvature calculation standards).

[0045] The principal curvature directions are determined by the corresponding eigenvectors. The direction of the maximum principal curvature is the direction of the eigenvector corresponding to the eigenvalue λ1, and the direction of the minimum principal curvature is the direction of the eigenvector corresponding to the eigenvalue λ2.

[0046] The rate of curvature change is obtained by calculating the average difference between the principal curvature values ​​of the vertex and its neighboring vertices, reflecting how the curvature changes around the vertex.

[0047] The principal curvature values, curvature directions, and rates of curvature change of each vertex are arranged according to their positions in the 3D mesh model to generate a surface curvature feature distribution map. This surface curvature feature distribution map can display the curvature characteristics of different locations on the surface of the machined object.

[0048] Step S127: Identify regions in the three-dimensional mesh model that meet the preset depression depth threshold, and extract the boundary contour coordinates, depression depth parameters, and region area information of the depression regions as region depression features.

[0049] First, determine the preset indentation depth threshold, which is set according to the actual situation of the workpiece and the processing requirements.

[0050] For each vertex in the 3D mesh model, calculate its indentation depth. By comparing the positional relationship between the vertex and the mean plane of the face it belongs to, if the vertex is below the mean plane (determined by the direction of the normal vector), then calculate the distance from the vertex to the mean plane as a preliminary value of the indentation depth.

[0051] Then, the depth of the concavity of each vertex is determined. If it is greater than the preset concavity depth threshold, the vertex is marked as a concavity point.

[0052] Connecting adjacent depressions forms continuous regions, which are the depression regions that meet the conditions. For each depression region, extract its boundary contour coordinates, which constitute the vertex coordinate sequence of the depression region boundary.

[0053] Calculate the depression depth parameter of the depression region, and take the average value of the depression depth of all vertices in the region as the depression depth parameter of the region.

[0054] At the same time, the area of ​​the concave region is calculated, and the total area of ​​the region is obtained by adding the areas of the patches containing the concave region.

[0055] The boundary contour coordinates of the depression region, the depression depth parameter, and the region area information are combined to form the regional depression feature.

[0056] Step S128: Perform feature fusion processing on the contour boundary features, surface curvature feature distribution map and regional concavity features to generate surface structure features containing spatial geometric attributes.

[0057] The contour boundary features, surface curvature feature distribution maps, and regional concavity features are standardized to fuse them at the same scale. For example, the values ​​in each feature are mapped to the [0, 1] interval according to a set ratio.

[0058] Then, feature splicing is used for fusion. The edge vertex sequence and the angle parameter between the normal vectors of adjacent facets in the contour boundary features, the principal curvature value, curvature direction, and rate of change of curvature in the surface curvature feature distribution map, and the boundary contour coordinates, depression depth parameters, and area information in the region depression features are spliced ​​together in a set order to form a comprehensive feature vector. This vector contains spatial geometric attribute information of the surface of the processed object, such as contour, curvature, and depressions, which is the generated surface structure feature.

[0059] Step S130: Identify the processing constraint region features of the processing object based on the surface structure features. The processing constraint region features include the boundary features of the non-processable region, the features of the precision processing region, and the turning restriction features of the processing path.

[0060] By utilizing the various spatial geometric properties contained in the surface structure features, machining constraint regions on the workpiece can be identified. These constraint regions have a significant impact on subsequent machining path planning and need to be accurately identified.

[0061] Step S131: Call the region segmentation module of the processing constraint region recognition model to perform semantic segmentation processing on the surface structure features, and divide the three-dimensional surface of the processing object into multiple semantic region units, each semantic region unit containing a corresponding region type label.

[0062] The processing constraint region identification model is a pre-trained deep learning model whose region segmentation module is specifically designed for semantic segmentation of the surface structural features of the processing object.

[0063] Step S1311: Input the surface structure features into the feature enhancement layer of the region segmentation module, and perform feature enhancement processing on the contour boundary features, surface curvature features and region concavity features through the residual connection structure to generate enhanced surface structure features.

[0064] The feature enhancement layer consists of multiple residual blocks, each containing two convolutional layers and a skip connection. Surface structure features are input into the first convolutional layer for feature extraction, and then fed into the second convolutional layer for further processing.

[0065] Skip connections concatenate the original features input to the residual block with the features output from the second convolutional layer. This avoids the vanishing gradient problem that occurs in deep networks while preserving important information from the original features.

[0066] By processing multiple residual blocks, key information in contour boundary features, surface curvature features, and regional concavity features is enhanced, making these features more prominent in subsequent processing and generating enhanced surface structure features.

[0067] Step S1312: Perform three-dimensional convolution processing on the enhanced surface structure features to extract multi-scale spatial features under different receptive fields. The multi-scale spatial features include local detail features and global context features.

[0068] Three-dimensional convolution processing of enhanced surface structure features is performed using convolution kernels of different sizes. For example, using a smaller convolution kernel (such as 3x3x3) can extract local detail features of the processed object's surface, such as small-scale curvature changes and subtle depressions.

[0069] Using a larger convolution kernel (such as 7x7x7) for convolution can extract global contextual features over a wider range, such as the direction of the entire contour and the relationship between large concave regions and surrounding structures.

[0070] Through the above processing, feature maps under different receptive fields are obtained. These feature maps are combined together to form a multi-scale spatial feature that includes local detail features and global context features.

[0071] Step S1313: The multi-scale spatial features are processed by channel weight allocation through the channel attention mechanism to strengthen the feature channels that play an important role in the identification of processing constraint regions, suppress redundant feature channels, and obtain the weighted multi-scale spatial features.

[0072] The channel attention mechanism first performs global average pooling on the multi-scale spatial features, converting the feature map of each channel into a numerical value that represents the global information of the channel features.

[0073] These values ​​are then fed into a network consisting of fully connected layers, which learns the weight values ​​for each channel. The magnitude of the weight values ​​reflects the importance of the channel's features for identifying processing constraint regions.

[0074] The obtained weight values ​​are multiplied with the corresponding channel features to enhance important feature channels (with larger weight values) and suppress redundant feature channels (with smaller weight values), thus obtaining the weighted multi-scale spatial features.

[0075] Step S1314: Input the weighted multi-scale spatial features into the graph convolutional network layer of the region segmentation module to construct graph structure data with vertices of the 3D mesh model as nodes and adjacency relationships as edges, and perform spatial relationship modeling processing.

[0076] The graph convolutional network layer first constructs graph structure data based on the connection relationships between vertices in the 3D mesh model. Each vertex is treated as a node in the graph, and if an edge connects two vertices, an edge is created in the graph to represent their adjacency relationship.

[0077] The feature vector corresponding to each vertex in the weighted multi-scale spatial features is used as the feature attribute of the corresponding node in the graph.

[0078] Through graph convolution operations, each node aggregates the feature information of its neighboring nodes, thereby modeling the spatial relationships between vertices. For example, the new feature of each node is obtained by concatenating its own features with the features of its neighboring nodes according to a set weight, which can reflect the mutual influence and spatial association between vertices.

[0079] Step S1315: Perform node classification processing on the graph structure data through the graph convolutional network layer, and assign a corresponding region type label to each vertex. The region type label includes unprocessable region label, precision processing region label and ordinary processing region label.

[0080] After modeling spatial relationships, the graph convolutional network layers input the processed node features into a classifier, which consists of multiple fully connected layers and activation functions. First, the node features are input into the first fully connected layer, where a linear transformation is performed through the calculation of the weight matrix. Then, non-linear processing is applied through an activation function (such as the ReLU function) to increase the model's expressive power.

[0081] After processing by multiple fully connected layers, the output dimension of the last fully connected layer corresponds to the number of region type labels. The output is converted into a probability distribution by the Softmax function, with each probability value corresponding to the probability of a region type label.

[0082] The region type label with the highest probability value is selected as the classification result for that node, and a corresponding region type label is assigned to each vertex. For example, if a vertex is assigned the label "unprocessable region," it means that the region where the vertex is located is not suitable for processing operations; if a vertex is assigned the label "precision processing region," the region where the vertex is located needs to be processed with high precision; and if a vertex is assigned the label "normal processing region," the region where the vertex is located can be processed according to the standard processing requirements.

[0083] Step S1316: Based on the region type labels of vertices, perform region partitioning on the 3D mesh model, dividing a set of vertices with the same region type label and spatial continuity into a semantic region unit, and generating a set of semantic region units containing region identifiers, vertex index lists and region type labels.

[0084] Traverse all vertices in the 3D mesh model and group them according to the region type label of each vertex. For vertices with the same region type label, determine whether they are spatially continuous, i.e., by checking whether there are edges connecting the vertices.

[0085] A set of vertices that are spatially contiguous and have the same region type label is divided into a semantic region unit, and each semantic region unit is assigned a unique region identifier to distinguish different region units.

[0086] Simultaneously, a list of vertex indices contained in each semantic region unit is recorded. The indices in this list correspond to the vertex numbers in the 3D mesh model, allowing for the rapid retrieval of all vertices within the region unit using the indices.

[0087] The region identifier, vertex index list, and corresponding region type label are combined to form a set of semantic region units. For example, a semantic region unit may have a region identifier of R1, a vertex index list of [1, 2, 3, ..., 100], and a region type label of "precision processing region," indicating that the region is a continuous region that requires precision processing.

[0088] Step S132: Select semantic region units marked as unprocessable regions according to the region type labels, extract the boundary contour coordinates and boundary smoothness parameters of the unprocessable regions, and generate the boundary features of the unprocessable regions.

[0089] From the set of semantic region units, select semantic region units with the region type label "unprocessable region". For each of the above semantic region units, obtain its vertex index list, and extract the boundary contour coordinates of the unprocessable region based on the coordinate information of these vertices in the 3D mesh model.

[0090] The process of extracting boundary contour coordinates is similar to the method of extracting contour boundary features in step S125, that is, finding the boundary edges within the region unit, collecting the vertices of the boundary edges and arranging them in order to form a sequence of boundary contour coordinates.

[0091] Calculate the boundary smoothness parameters for unchemifiable regions. The smoothness of the boundary is measured by calculating the rate of change of distance and the rate of change of angle between adjacent vertices in the boundary contour coordinate sequence. For example, the distance change rate is obtained by calculating the ratio of the distance between two adjacent vertices to the distance between the previous set of adjacent vertices; the angle change rate is obtained by calculating the angle change of the angle formed by three adjacent vertices. These rates of change are then statistically analyzed, such as calculating the mean and variance, and used as boundary smoothness parameters.

[0092] The boundary contour coordinates and boundary smoothness parameters of the unelectable region are combined to generate the unelectable region boundary feature. This unelectable region boundary feature can accurately describe the location, shape, and smoothness of the unelectable region's boundary.

[0093] Step S133: Identify the target region marked as a precision machining region from the semantic region unit, and extract the surface roughness parameters, flatness error values ​​and dimensional tolerance requirements of the target region as features of the precision machining region.

[0094] In the set of semantic region units, identify the semantic region units whose region type label is "precision processing region" and use them as the target region.

[0095] For each target region, its surface roughness parameter is calculated. By obtaining the normal vector information of all vertices within the region, the angle change between the normal vectors of adjacent vertices is calculated. Combined with the distance between vertices, the surface roughness is estimated. For example, the larger the change in the angle between the normal vectors and the more uneven the distance between vertices, the larger the surface roughness parameter value, indicating a rougher surface.

[0096] Calculate the flatness error value of the target area. If the target area is approximately planar, select multiple vertices within the area, fit an ideal plane, and then calculate the distance from each vertex to the ideal plane. Take the maximum value among these distances as the flatness error value, which reflects the degree of deviation between the actual surface and the ideal plane.

[0097] Dimensional tolerance requirements are predetermined based on the design specifications of the workpiece. For precision-machined areas, the allowable deviation range in dimensions such as length, width, and height needs to be clearly defined. For example, the length dimensional tolerance requirement for a certain precision-machined area is that a certain range of deviation is allowed based on the design length.

[0098] The surface roughness parameters, flatness error values, and dimensional tolerance requirements of the target area are extracted and combined to form the features of the precision machining area.

[0099] Step S134: Analyze the relative positional relationship between the precision machining area and the non-machinable area, calculate the minimum distance parameter and relative azimuth information, and generate positional constraint features between areas.

[0100] For each precision-machined area, determine its relative positional relationship with all unmachined areas. Calculate the center coordinates of the precision-machined area and the center coordinates of the unmachined areas, and then calculate the straight-line distance between them using these two center coordinates. Select the minimum value among all straight-line distances as the minimum distance parameter, which represents the closest distance between the precision-machined area and the unmachined areas.

[0101] To calculate the relative azimuth information, a local coordinate system is established with the center of the precision-machined area as the origin. The azimuth of the unmachined area's center is then calculated based on its position within this local coordinate system. The azimuth can be obtained through trigonometric functions; for example, the angle between the center's x and y coordinates (in the local coordinate system) and the positive x-axis is calculated, which is the relative azimuth.

[0102] By combining the minimum distance parameter and relative azimuth information, an inter-regional positional constraint feature is generated. This feature describes the spatial relationship between the precision machining area and the non-machinable area. This relationship needs to be considered during path planning to ensure that the machining path can meet the precision machining requirements while avoiding the non-machinable area.

[0103] Step S135: Based on the surface curvature feature distribution map of the three-dimensional mesh model, identify areas where the curvature change rate exceeds a preset threshold as path turning restriction areas, and extract the path direction change range parameters and minimum turning radius requirements of the turning restriction areas as processing path turning restriction features.

[0104] Referring to the surface curvature feature distribution map of the 3D mesh model, we focus on the distribution of the rate of change of curvature. A preset threshold for the rate of change of curvature is established, determined based on the performance of the machining tool and the requirements of the machining process.

[0105] The process iterates through all regions of the 3D mesh model, identifying areas where the rate of curvature change exceeds a preset threshold as path turning restriction regions. Due to the significant changes in surface curvature in these regions, the machining tool will face certain limitations when performing turning operations.

[0106] For each path turning restriction region, the path direction change range parameter is extracted. By analyzing the normal vector direction and principal curvature direction of the vertices within this region, the allowed direction change range of the machining path within this region is determined. For example, based on the distribution of the principal curvature direction, it is determined that the path direction can only change within a set angle range.

[0107] The minimum turning radius requirement is determined based on the structure and performance of the machining tool; that is, it is the minimum turning radius that the machining tool must meet when turning in that area. For example, if a certain machining tool has a fixed minimum turning radius, the turning radius of the machining path cannot be less than this value within the path turning restriction area.

[0108] By combining the parameters of the path direction change range and the minimum turning radius requirement, a processing path turning restriction feature is formed, thus providing clear constraints for the planning of the processing path within the turning restriction area.

[0109] Step S136: The non-processable area boundary features, precision processing area features, and processing path turning restriction features are structurally integrated to generate a set of processing constraint area features containing spatial constraint relationships.

[0110] Standardize the data format for non-processable region boundary features, precision processing region features, and processing path turning constraint features to ensure they can be effectively integrated. For example, convert the parameters in each feature to the same data type and dimension.

[0111] A structured approach is used for integration, organizing these features according to a predetermined logical order to form a set of processing constraint region features. For example, taking each semantic region unit as a unit, the boundary features of the non-processable region (if it is a non-processable region), the features of the precision processing region (if it is a precision processing region), and the processing path turning restriction features related to its location are associated with that region unit. The integrated set of processing constraint region features completely contains the constraint information of each region on the processing object and the spatial constraint relationships between them.

[0112] Step S140: The processing constraint region features are associated and fused with the preset processing process parameters to generate a path planning constraint feature set, which includes regional processing priority features, path continuity constraint features, and processing accuracy matching features.

[0113] By associating and integrating the characteristics of the processing constraint region with the preset processing parameters, the processing constraints and process requirements can be combined to form a comprehensive set of constraint features for path planning.

[0114] Step S141: Analyze the preset machining process parameters, extract the machining tool parameters, machining sequence rules and accuracy requirement indicators, and generate a structured set of process parameters.

[0115] Preset processing parameters are a series of parameters set in advance based on the material, shape, and processing requirements of the object being processed. These parameters are then analyzed to separate different types of parameters.

[0116] Machining tool parameters include the type, size, and cutting performance parameters of the machining tool. For example, the diameter, length, and hardness of the tool material.

[0117] The processing sequence rule refers to the requirement for the order of processing of each area during the processing process, such as processing planes before processing curved surfaces, or roughing before finishing.

[0118] The accuracy requirements include the overall machining accuracy level, the maximum allowable surface roughness, and the dimensional tolerance range, thus providing standards for accuracy control during the machining process.

[0119] The extracted machining tool parameters, machining sequence rules, and precision requirement indicators are organized according to a set structure to generate a structured set of process parameters. For example, different types of parameters can be categorized and stored in the form of a dictionary or list to facilitate subsequent correlation and fusion processing.

[0120] Step S142: Perform compatibility analysis on the boundary features of the unmachinable region in the processing constraint region features and the processing tool parameters in the process parameter set, and calculate the minimum safe distance parameter between the movement trajectory of the processing tool and the boundary of the unmachinable region.

[0121] By analyzing the boundary contour coordinates and boundary smoothness parameters in the boundary features of the unmachinable area, as well as the tool dimensions (such as diameter and length) in the machining tool parameters, it can be determined whether the machining tool will collide with the unmachinable area during its movement.

[0122] Calculate the minimum safe distance parameters based on the size of the machining tool and the shape of the unmachinable area boundary. For example, considering the maximum radius of the machining tool and the smoothness of the unmachinable area boundary, determine the minimum distance that the machining tool's trajectory must maintain between the unmachinable area boundary to ensure that the machining tool does not touch the unmachinable area.

[0123] During the calculation process, it is necessary to comprehensively consider the movement of the machining tool in different directions to ensure that the minimum safe distance requirement is met in all possible movement directions. The generated minimum safe distance parameter will serve as an important constraint in path planning.

[0124] Step S143: Match the features of the precision machining area with the accuracy requirement indicators in the process parameter set to generate accuracy matching features that include the area accuracy level, allowable error range and surface quality requirements.

[0125] Compare the surface roughness parameters, flatness error values, and dimensional tolerance requirements in the precision machining area with the accuracy requirements in the process parameter set.

[0126] Based on the comparison results, a corresponding regional accuracy level is determined for each precision-machined area. For example, if the surface roughness parameter of a certain precision-machined area is lower than the maximum allowable surface roughness value in the process parameters, and the flatness error value and dimensional tolerance requirements are within the process requirements range, then a higher accuracy level is assigned to it.

[0127] The permissible error range is determined based on the accuracy requirements of the process parameter set, combined with the actual characteristics of the precision machining area. For example, for dimensional tolerance requirements, the permissible error range for that area is determined by further subdividing it within the total permissible tolerance range of the process, according to the size of the area and the machining difficulty.

[0128] Surface quality requirements are determined by the surface roughness requirements in the comprehensive process parameters and the surface characteristics of the precision machining area, specifying the quality standards such as the surface smoothness that the area should achieve after machining.

[0129] By combining the regional accuracy level, allowable error range, and surface quality requirements, a precision matching feature is generated, which achieves an effective match between the features of the precision machining area and the process accuracy requirements.

[0130] Step S144: Based on the turning constraint features of the processing path and the processing sequence rules in the process parameter set, construct a direction constraint graph model of the processing path. The nodes in the direction constraint graph model represent processing regions, and the directed edges represent the processing sequence and turning constraint relationships between regions.

[0131] Step S1441: Parse the processing sequence rules and extract the regional processing sequence relationship, parallel processing restrictions, and path backtracking prohibition rules.

[0132] A detailed analysis of the processing sequence rules clarifies which areas need to be processed first and which need to be processed later, establishing a processing priority relationship between the areas. For example, if the rule stipulates that area A must be processed before area B can be processed, then there is a priority relationship between area A and area B.

[0133] Parallel processing constraints refer to which areas can be processed simultaneously and which cannot. For example, some areas cannot be processed in parallel due to proximity or limitations of processing tools.

[0134] The path backtracking prohibition rule means that once a processing path has passed through a certain area and completed processing, it is prohibited to return to that area for further processing, in order to avoid duplicate processing and interference.

[0135] Step S1442: Use each semantic region unit of the processing object as a node of the orientation constraint graph model, and initialize the node attribute parameters, which include region area, accuracy level and processing time estimation parameters.

[0136] Each semantic region unit is treated as a node in the orientation constraint graph model, and each node is assigned initial attribute parameters. The region area can be calculated based on the sum of the areas of the faces formed by the vertices contained in that region unit.

[0137] The accuracy level is determined based on the type of the area; for example, the accuracy level of a precision machining area is higher than that of a regular machining area.

[0138] The processing time estimation parameters are estimated based on factors such as the area, accuracy level, and processing efficiency of the processing tools, and are used to roughly predict the time required to process the area.

[0139] Step S1443: Calculate the path turning cost parameters between adjacent semantic region units based on the path direction change range parameter and minimum turning radius requirement in the processing path turning restriction feature. The path turning cost parameters are positively correlated with the direction change range and turning radius.

[0140] For two adjacent semantic region units, obtain the processing path turning constraint features of the connecting region between them, namely the path direction change range parameter and the minimum turning radius requirement.

[0141] The greater the range of path direction changes, the greater the angle the machining tool needs to adjust when turning, the higher the turning difficulty, and the greater the path turning cost parameter. Conversely, the larger the minimum turning radius, the greater the space required for the machining tool to turn, and the higher the turning cost. Therefore, the path turning cost parameter is positively correlated with these two parameters.

[0142] The path turning cost parameters between adjacent semantic region units are calculated by performing a function operation (such as weighted summation) that sets the path direction change range parameter and the minimum turning radius requirement.

[0143] Step S1444: Based on the processing sequence relationship of the region, add directed edges to the direction constraint graph model. The weight value of the directed edge is calculated by multiplying the path turning cost parameter and the processing sequence priority coefficient.

[0144] Based on the processing order of regions, for two region units with a processing order (such as region A being processed before region B), a directed edge is added to the directional constraint graph model from node A in region B to node B in region A.

[0145] The processing order priority coefficient is set based on the importance and urgency of processing in different regions; the higher the priority, the larger the coefficient. Multiplying the path turning cost parameter by the processing order priority coefficient yields the weight value of the directed edge, which reflects the cost and priority of processing from one region to another.

[0146] Step S1445: According to the parallel processing constraints, mark the set of nodes that can be processed in parallel in the directional constraint graph model, add undirected connecting edges between the nodes that can be processed in parallel, and the weight value of the edge represents the synergy coefficient of parallel processing.

[0147] For a set of nodes that meet the parallel processing constraints (i.e., region units that can be processed simultaneously), they are marked in the directional constraint graph model.

[0148] Undirected edges are added between these nodes that can be processed in parallel, with the weight of each edge being a coordination coefficient for parallel processing. The coordination coefficient is determined based on factors such as the positional relationship between regions and the degree of sharing of processing tools. A higher coordination coefficient indicates higher efficiency and better coordination when these regions process in parallel.

[0149] Step S1446: Convert the path backtracking prohibition rule into the edge constraint condition of the direction constraint graph model, prohibiting the addition of reverse directed edges between the processed region nodes and the subsequent processed region nodes.

[0150] According to the path backtracking prohibition rule, once a directed edge from node A to node B exists (indicating that B is processed after A has finished processing), it is prohibited to add a reverse directed edge from node B to node A to ensure that the processing path does not backtrack to the already processed area.

[0151] Step S1447: Perform topological sorting verification on the constructed directional constraint graph model to determine whether there is a circular dependency. If there is a circular dependency, adjust the weight values ​​of the directed edges of the relevant nodes until the topological sorting requirements are met.

[0152] The topological sorting algorithm is used to process the directional constraint graph model to check whether there are cyclic dependencies in the graph (i.e., starting from a certain node, passing through a series of directed edges and returning to the same node).

[0153] If a circular dependency exists, it indicates a contradiction in the processing order rules, requiring adjustment of the directed edge weights of the relevant nodes. For example, increasing the weight of a directed edge in a circular path reduces its likelihood of being selected, thus breaking the circular dependency until the directional constraint graph model satisfies the topological sorting requirements, resulting in a reasonable processing order sequence.

[0154] Step S145: Perform feature fusion processing on the minimum safety distance parameter, accuracy matching feature and direction constraint map model to generate regional processing priority features. The regional processing priority features are calculated by comprehensively considering the regional area, accuracy requirements and positional relationship with other regions.

[0155] First, assign appropriate weighting coefficients to factors such as area, accuracy requirements, and positional relationship with other areas. The magnitude of these weighting coefficients is determined based on their influence on processing priority. For example, areas with higher accuracy requirements have larger weighting coefficients.

[0156] For each region unit, the area score is calculated based on its region area. The larger the area (within a reasonable range), the higher the score may be. The accuracy score is calculated based on the region accuracy level in the accuracy matching feature. The higher the accuracy level, the higher the score. The positional relationship score is calculated based on the positional relationship between the region and other regions in the directional constraint graph model (such as the weight of directed edges, whether parallel processing is possible, etc.). The closer the region is to important regions, the higher the score may be.

[0157] The area score, accuracy score, and positional relationship score are multiplied by their respective weighting coefficients and then concatenated (rather than added) to obtain the comprehensive score of the region unit. This comprehensive score constitutes the region processing priority feature. The higher the comprehensive score, the higher the processing priority of the region.

[0158] For example, the area score of a certain region is S1, with a corresponding weight coefficient of W1. Multiplying the two gives S1×W1; the accuracy score is S2, with a weight coefficient of W2, and multiplying them gives S2×W2; the positional relationship score is S3, with a weight coefficient of W3, and multiplying them gives S3×W3. These three results are concatenated in the form [S1×W1, S2×W2, S3×W3] to form the comprehensive score of that region, which is the region's processing priority feature. Through this method, each region has a feature that comprehensively reflects its processing priority.

[0159] Step S146: Analyze the path connection relationship between adjacent processing areas, extract path smoothness requirements, continuity threshold and transition area length parameters, and generate path continuity constraint features.

[0160] Examine each adjacent processing area unit one by one and analyze the possible path connections between them. For each pair of adjacent area units, determine the possible path connections based on their 3D mesh model shape, boundary contour features, and processing path turning constraint features.

[0161] Path smoothness requirement refers to the smoothness of the path curve when the processing path transitions from one region to another. By analyzing the curvature changes and normal vector directions at the boundaries of adjacent regions, the range of tangent direction changes at the connection points of the path is determined. The smaller the range of change, the higher the path smoothness requirement.

[0162] Continuity thresholds are metrics used to measure the continuity of a path, including positional continuity thresholds and directional continuity thresholds. Positional continuity thresholds define the maximum allowable deviation in position between adjacent path segments at their connection points; directional continuity thresholds define the maximum allowable angle between the tangent directions of adjacent path segments at their connection points.

[0163] The transition region length parameter refers to the length of the area used for path transition between the boundaries of one region and another. This region length needs to be determined based on the distance between the two regions, the size of the machining tool, and the path smoothness requirements to ensure that the machining tool can smoothly transition from one region to another.

[0164] By combining path smoothness requirements, continuity thresholds, and transition region length parameters, a path continuity constraint feature is generated. This feature ensures that the connection between different regions of the processing path remains smooth and continuous, avoiding abrupt changes or breaks, thereby guaranteeing the stability and quality of the processing.

[0165] Step S147: The regional processing priority features, path continuity constraint features, and accuracy matching features are structurally integrated to generate a path planning constraint feature set containing multi-dimensional constraint information.

[0166] Data standardization is performed on regional processing priority features, path continuity constraint features, and accuracy matching features to ensure consistency in data format and value range, facilitating integration. For example, parameter values ​​in each feature are mapped to the same interval.

[0167] Based on the division of processing area units, the processing priority features, path continuity constraints with other areas, and accuracy matching features of each area unit are associated. For example, for area unit R1, its processing priority features, path continuity constraints with adjacent areas such as R2 and R3, and its own accuracy matching features are integrated together.

[0168] These associated features are organized using a structured data format (such as a multidimensional array or linked list) to form a path planning constraint feature set. This set of path planning constraint features contains multi-dimensional constraint information related to priority, continuity, and accuracy involved in the processing, comprehensively reflecting the various conditions that path planning needs to meet.

[0169] Step S150: Call the pre-trained processing path planning model to perform path generation processing on the path planning constraint feature set, and generate an automated processing path instruction set containing processing path coordinate sequence, path execution order and processing parameter configuration.

[0170] By utilizing a pre-trained processing path planning model and combining it with a set of path planning constraint features, processing path instructions that meet various constraints can be automatically generated, thus achieving automated planning of processing paths.

[0171] Step S151: Input the set of path planning constraint features into the constraint feature encoding layer of the processing path planning model for feature vectorization processing to generate a fixed-dimensional constraint feature vector.

[0172] The constraint feature encoding layer consists of multiple embedding layers and fully connected layers. First, different types of features in the path planning constraint feature set (such as region processing priority features, path continuity constraint features, and accuracy matching features) are input into the corresponding embedding layers.

[0173] The embedding layer transforms these non-vector features or high-dimensional features into low-dimensional vector representations. For example, it converts the comprehensive score sequence in the region processing priority feature into a fixed-length vector. Then, the vectors output from each embedding layer are input into a fully connected layer, where they are fused into a fixed-dimensional constrained feature vector through linear transformations and non-linear activation functions (such as the tanh function).

[0174] The generated constraint feature vector can condense the multi-dimensional information in the path planning constraint feature set into a unified vector representation, which facilitates subsequent processing and analysis of the path planning model. For example, for a processing object containing multiple regional units, the constraint feature vector can comprehensively reflect information such as the priority, path connection requirements, and accuracy requirements of each region.

[0175] Step S152: The constraint feature vector is processed by the graph attention network module of the processing path planning model to perform regional association modeling, enhance the feature representation of important processing areas, and generate a weighted constraint feature map.

[0176] The graph attention network module first constructs a graph structure based on the connection relationships between processing region units (such as adjacency relationships and processing order relationships), where each node represents a processing region unit, and the features of the node are the constraint feature vectors generated by the constraint feature encoding layer.

[0177] Then, the graph attention network module calculates the attention weights between each node and other nodes. The attention weights are calculated based on the feature vectors of two nodes. After the node features are processed by a multilayer perceptron, they are normalized by the Softmax function, reflecting the importance of the association between the two nodes.

[0178] Based on the attention weights, the feature vector of each node is concatenated with the feature vectors of other nodes according to their weights, thereby achieving region association modeling. For important processing regions (such as regions with high priority or high accuracy requirements), their corresponding nodes will receive higher attention weights, which will strengthen their feature representation in the weighted constrained feature map.

[0179] The generated weighted constraint feature map not only retains the constraint feature information of each region, but also reflects the relationship between regions, and highlights the features of important regions.

[0180] Step S153: Input the weighted constraint feature map into the path generation network of the processing path planning model. The path generation network includes a region ranking subnetwork and a path coordinate prediction subnetwork.

[0181] The path generation network is the core component of the processing path planning model, responsible for generating specific processing path information based on the weighted constraint feature map. The region ranking subnetwork focuses on determining the processing order of each processing region, while the path coordinate prediction subnetwork is responsible for generating specific path coordinates within each region.

[0182] The weighted constraint feature map is simultaneously input into the region sorting subnetwork and the path coordinate prediction subnetwork. The two subnetworks work in parallel, handling different tasks respectively, but sharing some feature information to ensure that the generated processing order and path coordinates can match each other.

[0183] Step S154: Perform sequential planning processing on the processing regions through the region sorting sub-network, and generate a region processing sequence based on the region processing priority features.

[0184] Step S1541: Input the weighted constraint feature map into the self-attention mechanism layer of the region ranking sub-network, perform correlation modeling on the feature representations of each processing region, and generate an inter-regional attention weight matrix.

[0185] The self-attention mechanism layer captures the correlation between regions by calculating the similarity between each region node and all other region nodes in the weighted constrained feature map. Specifically, for the feature vector of each region node, a query vector, key vector, and value vector are generated through linear transformation.

[0186] The similarity between the query vector and the key vectors of all other region nodes is calculated. After normalization using the Softmax function, the attention weights of each region node to other region nodes are obtained. These attention weights are arranged according to the order of the region nodes to form an inter-region attention weight matrix. Each element in this inter-region attention weight matrix represents the association strength between two corresponding regions; the larger the value, the stronger the association between the two regions.

[0187] Step S1542: Based on the inter-regional attention weight matrix, calculate the global importance score of each processing region. The global importance score is obtained by weighted summation of attention weight values ​​and regional processing priority features.

[0188] For each processing region, the row vector corresponding to that region in the inter-region attention weight matrix (i.e., the attention weight of that region with all other regions) is weighted and summed with the region processing priority features of other regions.

[0189] For example, the global importance score of region i is equal to the sum of the attention weights of regions i and j multiplied by the region processing priority feature score of region j for all other regions j, and then all these products are added together. In this way, the global importance score of each region not only considers its own priority but also integrates the degree of correlation with other regions, thus reflecting the importance of that region in the overall processing more comprehensively.

[0190] Step S1543: Initialize the region processing sequence to an empty sequence, select the processing region with the highest global importance score as the starting processing region, and add it to the region processing sequence.

[0191] Initially, the region processing sequence does not contain any processing regions. Iterate through the global importance scores of all processing regions, find the region with the highest score, determine it as the starting processing region, and add it to the beginning of the region processing sequence.

[0192] The selection of the initial processing area is crucial to the rationality of the entire processing sequence. Usually, the area with the highest global importance score is selected because this area is often the key part of the processing.

[0193] Step S1544: From the remaining processing regions, calculate the selection probability of the candidate processing region based on the path turning cost parameter between the selected processing region and the candidate processing region and the inter-region attention weight value.

[0194] The remaining processing regions refer to the processing regions that have not yet been added to the region processing sequence. For each candidate processing region (i.e., one of the remaining processing regions), the path turning cost parameters (from the direction constraint graph model) and the inter-region attention weight values ​​(from the inter-region attention weight matrix) between the selected processing region (i.e., the last region in the region processing sequence) and the candidate processing region are obtained.

[0195] When calculating the selection probability, the path turning cost parameter is first normalized to ensure it falls within the same range as the inter-region attention weight value. Then, appropriate weights are assigned to these two parameters, and their weighted sum is processed by the Sigmoid function to obtain the selection probability of the candidate processing region. A higher selection probability indicates a greater likelihood that the candidate region will be selected as the next processing region in the current step.

[0196] Step S1545: Select the candidate processing region with the highest selection probability and add it to the region processing sequence. Repeat the above selection process until all processing regions have been added to the sequence.

[0197] In each selection step, the region with the highest selection probability is chosen from the remaining candidate processing regions and added to the end of the region processing order sequence. Then, the selected processing region is updated to the newly added region, and steps S1544 and S1545 are repeated until all processing regions are included in the region processing order sequence.

[0198] The generated regional processing sequence can be arranged according to a set priority order, taking into account the inter-regional relationships and turning costs, thus ensuring the efficiency and rationality of the processing process.

[0199] Step S1546: Perform local optimization on the generated regional processing sequence, use the neighborhood exchange algorithm to adjust the order of adjacent regions, calculate the total path cost before and after adjustment, and retain the regional processing sequence with the minimum total path cost.

[0200] The neighborhood exchange algorithm is a local search algorithm used to optimize the initially generated sequence of region processing orders. First, the total path cost of the initial region processing order sequence is calculated. The total path cost is calculated based on factors such as the path turning cost parameters between adjacent regions in the sequence and the estimated processing time parameters, reflecting the total cost required to process according to this order.

[0201] Then, adjacent regions in the processing sequence are swapped pairwise to generate multiple neighborhood solutions. For example, swapping regions at position i and i+1 in the sequence yields a new sequence. The total path cost is also calculated for each neighborhood solution.

[0202] Compare the total path cost of all neighborhood solutions with the initial solution, and select the sequence with the minimum total path cost as the optimized region processing sequence. If the total path cost of all neighborhood solutions is greater than that of the initial solution, then retain the initial sequence.

[0203] Local optimization can further improve the rationality of the regional processing sequence, reduce the total processing cost, and improve processing efficiency.

[0204] Step S155: Input the processing sequence of the region into the path coordinate prediction subnetwork, and combine it with the path continuity constraint features to generate the path coordinate sequence within each processing region. The path coordinate sequence is described by the ordered arrangement of three-dimensional spatial coordinate points.

[0205] Step S1551: Input the 3D mesh model of the current processing area and the path endpoint coordinates of the previous processing area in the processing sequence of the area into the path starting point determination module of the path coordinate prediction sub-network.

[0206] The path start point determination module receives two key inputs: a 3D mesh model of the processing area for which the path needs to be planned, which contains geometric information such as vertex coordinates and facet connection relationships of the area; and the path end point coordinates of the previous processing area in the processing sequence of the area, which indicate the position of the processing tool when the previous area finishes processing.

[0207] Step S1552: Based on the path smoothness requirement in the path continuity constraint feature, calculate the optimal path start coordinates of the current processing area. The distance between the optimal path start coordinates and the path end coordinates of the previous processing area should meet the smooth transition threshold requirement.

[0208] The path start point determination module first determines the possible candidate locations of the path start point on the boundary of the current processing area based on the 3D mesh model of the current area. These candidate locations are usually distributed near the boundary between the current area and the previous area.

[0209] Then, the distance and direction of the connecting line between each candidate position and the endpoint coordinates of the previous processing area path are calculated. Based on the path smoothness requirements in the path continuity constraint features, candidate positions whose distance and direction changes meet the smooth transition threshold requirements are selected. For example, the distance cannot exceed a preset maximum distance threshold, and the angle between the connecting line direction and the tangent direction of the previous path endpoint cannot exceed a preset angle threshold.

[0210] Among the qualified candidate locations, select the location that will make subsequent path planning smoothest as the optimal path starting coordinates. For example, choose a location located deep within the current area that can cover more processing areas as the starting point.

[0211] Step S1553: On the surface of the 3D mesh model of the current processing area, an initial set of path points is generated using an equidistant sampling algorithm. The distribution density of the initial path points is positively correlated with the accuracy level of the current processing area.

[0212] The equidistant sampling algorithm first determines the sampling range on the surface of the 3D mesh model of the current processing area, which typically covers the entire processing area. Then, sampling is performed within this range according to a preset sampling interval to generate a series of initial path points.

[0213] The sampling interval is determined based on the accuracy level of the current processing area. The higher the accuracy level, the smaller the sampling interval and the greater the distribution density of the initial path points, so as to ensure that the fine features of the area surface can be captured and meet the requirements of high-precision processing. When the accuracy level is low, the sampling interval can be appropriately increased to reduce the number of path points and improve the efficiency of path planning.

[0214] The generated initial path point set is a collection of three-dimensional spatial coordinate points, which are uniformly distributed on the surface of the current processing area.

[0215] Step S1554: The initial set of path points is optimized by the path optimization layer of the path coordinate prediction sub-network. The positions of the path points are adjusted based on the surface curvature feature distribution map to obtain the optimized path points.

[0216] The path optimization layer comprises multiple convolutional layers and recurrent neural network layers. First, the initial set of path points is converted into a feature matrix and input into the convolutional layers. The convolutional layers extract local spatial features of the initial path points using a sliding window operation, capturing the local relationships between the path points.

[0217] Then, a recurrent neural network layer (such as an LSTM layer) processes the features output by the convolutional layer, considering the sequential information of the path points and analyzing the overall trend of the path. Simultaneously, a surface curvature feature distribution map of the current processing area is introduced, and the positions of the path points are adjusted based on the curvature values ​​at different locations in the map.

[0218] For regions with high curvature, the number of path points can be increased and their positions adjusted to better conform the path to the surface shape. For flat regions with low curvature, the number of path points can be reduced to simplify the path. Through these adjustments, optimized path points are obtained.

[0219] Step S1555: Based on the continuity threshold in the path continuity constraint features, perform connection processing on the optimized path points to generate an initial path coordinate sequence.

[0220] Following the optimized distribution order of the path points within the processing area, these path points are connected sequentially. During the connection process, it is checked whether the changes in distance and direction between adjacent path points meet the continuity threshold requirements in the path continuity constraint feature.

[0221] If the distance between adjacent path points exceeds the positional continuity threshold, a new path point is inserted between them to keep the distance between adjacent points within the threshold range; if the direction change exceeds the direction continuity threshold, the position of the path point is adjusted or a transition point is added to make the direction change meet the requirements.

[0222] Through the above connection process, an initial path coordinate sequence is formed. This initial path coordinate sequence is a series of ordered three-dimensional spatial coordinate points, reflecting the initial movement trajectory of the machining tool in the current area.

[0223] Step S1556: The initial path coordinate sequence is smoothed using a B-spline curve fitting algorithm to generate a smooth path curve containing a control point sequence and curve parameters.

[0224] The B-spline curve fitting algorithm first determines the order and node vectors of the B-spline curve based on the initial path coordinate sequence. The order is usually determined by the complexity of the path; the more complex the path, the higher the order. The node vectors are set according to the distribution of path points.

[0225] Then, the control point sequence of the B-spline curve is solved using the least squares method, so that the curve can approximate the initial path coordinate sequence as closely as possible. During the solution process, it is necessary to ensure the smoothness of the curve and avoid sharp bends or jitters.

[0226] The generated smooth path curve contains a sequence of control points and curve parameters (such as order and node vectors). This smooth path curve is smoother than the initial path coordinate sequence, which can reduce the vibration and impact of the machining tool during the movement.

[0227] Step S1557: Sample the smooth path curve at a preset step size to obtain the final path coordinate sequence, the path coordinate sequence including three-dimensional coordinate values ​​and corresponding curve tangent direction vectors.

[0228] The sampling step size is set based on the required machining accuracy and the motion performance of the machining tool. A smaller sampling step size results in a denser path coordinate sequence and higher machining accuracy, but also a larger data volume; a larger sampling step size results in a sparser path coordinate sequence and a smaller data volume, but may affect machining accuracy.

[0229] Following a preset sampling step size, uniform sampling is performed on the smooth path curve to obtain the three-dimensional coordinate values ​​of a series of sampling points. Simultaneously, the tangent direction vector of the curve at each sampling point is calculated, indicating the direction of movement of the machining tool at that point.

[0230] The final generated path coordinate sequence is an ordered sequence composed of the three-dimensional coordinate values ​​of these sampling points and the corresponding tangent direction vectors, which accurately describes the movement trajectory of the machining tool within the current machining area.

[0231] Step S156: Based on the accuracy matching features, configure corresponding machining parameters for each path coordinate point. The machining parameters include feed rate, depth of cut, and spindle speed.

[0232] For each point in the path coordinate sequence, the machining parameters are configured based on the accuracy matching characteristics of the machining area where that point is located. The area accuracy level, allowable error range, and surface quality requirements in the accuracy matching characteristics jointly determine the machining parameter configuration for that point.

[0233] If the path coordinates are located in an area with a high precision level, a small allowable error range, and strict surface quality requirements, a lower feed rate is necessary. A lower feed rate allows the machining tool to maintain contact with the workpiece for a longer period, resulting in smoother cutting, thus reducing machining errors and ensuring surface quality. For example, at path coordinates in precision machining areas, the feed rate would be set to a relatively low value to meet the demands of high-precision machining.

[0234] The depth of cut is configured based on the surface curvature characteristics and machining allowance at the location of the coordinate point along the path. Combining the curvature value at that point with the surface curvature characteristic distribution map, a larger curvature indicates a higher degree of surface bending. To avoid over-cutting or under-cutting of the surface by the machining tool, a smaller depth of cut will be set. Conversely, a smaller curvature indicates a relatively smooth surface, allowing for a more appropriate increase in the depth of cut. Simultaneously, the machining allowance for that area will also be considered to ensure the depth of cut is within a reasonable range, removing sufficient material without damaging the structure of the workpiece.

[0235] The spindle speed setting needs to comprehensively consider the type of machining tool, the hardness of the material being machined, and the required machining accuracy. For materials with high hardness, a higher spindle speed is set to ensure cutting efficiency and tool life. However, for areas with extremely high precision requirements, an excessively high spindle speed may cause vibration, affecting machining accuracy, so the spindle speed is appropriately reduced. Furthermore, different types of machining tools have their recommended spindle speed ranges, which are adjusted during configuration.

[0236] By using the above method, the corresponding feed rate, depth of cut, and spindle speed are configured for each point in the path coordinate sequence, so that the machining parameters of each path coordinate point can match the accuracy requirements of the area and ensure machining quality.

[0237] Step S157: Integrate the processing sequence of the region, the path coordinate sequence and the corresponding processing parameters to generate an automated processing path instruction set containing timestamps, wherein the timestamps are used to indicate the start time of execution of each path segment.

[0238] First, the processing sequence of regions is used as the overall framework. The path coordinate sequence and corresponding processing parameters of each region are processed sequentially according to the region order in the sequence. For each region, its path coordinate sequence is arranged in chronological order and associated with the corresponding processing parameters to form the processing sub-instructions for that region.

[0239] Then, the execution time for each path segment is calculated. Based on parameters such as the length of the path segment and the corresponding feed rate, the time required for the machining tool to traverse that path segment is calculated. For example, the execution time of the path segment is obtained by dividing the length of the path segment by the feed rate.

[0240] Based on the start time of the regional processing sequence, a timestamp is added to each path segment. The timestamp of the first path segment in the first region is the start time, and the timestamps of subsequent path segments are the timestamp of the previous path segment plus its execution time. This method ensures that the timestamp of each path segment accurately indicates its execution start time.

[0241] The processing sub-instructions for all regions are integrated in chronological order according to their timestamps to form a complete set of automated processing path instructions. This set of automated processing path instructions includes the processing sequence of regions, the path coordinate sequence within each region, the corresponding processing parameters, and the execution start time of each path segment, which can directly guide the processing equipment to perform automated processing operations.

[0242] Furthermore, the method may also include: step S158: performing path feasibility verification processing on the generated set of automated processing path instructions, simulating the process of the processing tool moving along the coordinate sequence of the processing path, and detecting whether there is a risk of collision with the boundary of the unprocessable area.

[0243] Using computer simulation technology, virtual models of the machining tools and the machining objects are constructed based on the path coordinate sequence in the set of automated machining path instructions and the parameters of the machining tools.

[0244] The simulated machining tool moves in a virtual environment according to a path coordinate sequence and timestamp markers, and the distance between the virtual model of the machining tool and the virtual model of the boundary of the unmachinable area is monitored in real time. During the simulation, the position and attitude of the machining tool at each path coordinate point, as well as the change in distance from the boundary of the unmachinable area, are recorded.

[0245] By setting a collision detection algorithm, when the distance between the virtual model of the machining tool and the virtual model of the boundary of the unmachinable area is less than the minimum safe distance parameter, a collision risk is determined, and the corresponding path coordinates and path segments are recorded.

[0246] Step S159: If there is a collision risk, adjust the path coordinate sequence of the corresponding area and recalculate the three-dimensional coordinate values ​​of the path points so that the distance between the processing tool and the boundary of the unprocessable area is always greater than the minimum safe distance parameter.

[0247] For path segments where collision risk is detected, analyze the boundary features of the area where the path segment is located and the corresponding non-processable area. Based on the boundary contour coordinates of the non-processable area and the minimum safe distance parameter, determine the range of paths that need to be adjusted.

[0248] A path offset method is used to shift the coordinate sequence of paths with collision risks away from the boundary of unprocessable areas by a certain distance, which is greater than the minimum safe distance parameter. During the offset process, it is necessary to ensure that the adjusted path coordinate sequence still meets processing requirements, such as path continuity and connection with other path segments.

[0249] Recalculate the 3D coordinates of the adjusted path points, ensuring that the distance between each new path coordinate point and the boundary of the unprocessable area is greater than the minimum safe distance parameter. After adjustment, verify the feasibility of the path coordinate sequence in this area again until there is no risk of collision.

[0250] Step S160: Calculate the total length of the adjusted path coordinate sequence and the rate of curvature change of each path segment to verify whether the continuity threshold requirement in the path continuity constraint feature is met.

[0251] The total length of the adjusted path coordinate sequence is calculated by summing the lengths of all path segments. Simultaneously, the rate of curvature change of each path segment is calculated by analyzing the changes in the tangent direction between adjacent path points. For example, the ratio of the angle between the tangent directions of two adjacent path points to the path segment length is the rate of curvature change of that segment.

[0252] The calculated total length and the rate of curvature change of each path segment are compared with the continuity threshold in the path continuity constraint feature. If the total length is within a reasonable range and the rate of curvature change of each path segment is less than the continuity threshold, it indicates that the adjusted path coordinate sequence meets the continuity requirement; if not, the path coordinate sequence needs to be further adjusted, such as optimizing the path orientation or adding transition path segments, until the continuity threshold requirement is met.

[0253] Step S161: Perform time parameter allocation processing on the path execution order, calculate the processing time of each region based on the processing area and processing parameters of each region, and generate a time scheduling table that includes the start and end times of regional processing.

[0254] For each region, the processing time is calculated based on its processing area and corresponding processing parameters. The larger the processing area, the longer the processing time; the lower the feed rate in the processing parameters, the longer the processing time. For example, the processing time is equal to the processing area divided by (the product of the feed rate and the cutting width). The processing time for each region is estimated using the above method.

[0255] Based on the start time of the automated processing path instruction set, and according to the processing sequence of regions, a processing start time and an end time are assigned to each region. The processing start time of the first region is the start time, and its end time is the start time plus the processing time of that region; the processing start time of the second region is the end time of the first region, and its end time is its own start time plus its processing time, and so on.

[0256] The start and end times of processing in each region are compiled into a time schedule table, which shows the processing time arrangement for each region.

[0257] Step S162: Integrate the time scheduling table with the path coordinate sequence and processing parameter configuration to generate the final set of automated processing path instructions.

[0258] The start and end times of regional processing in the time schedule table are associated with the corresponding regional path coordinate sequence and processing parameter configuration to ensure that the processing operations in each region are carried out according to the corresponding path and parameters within the specified time.

[0259] A final check is performed on the integrated content to ensure there are no conflicts between the time schedule, path coordinate sequence, and processing parameter configuration, and that all information is complete and accurate. After verification, the final set of automated processing path instructions is generated. This set of instructions can be directly read and executed by the processing equipment to achieve automated processing of the object.

[0260] Figure 2 The illustration shows exemplary hardware and software components of a machine vision-based automated processing path planning system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the machine vision-based automated processing path planning system 100 and to perform the functions in this application.

[0261] For example, the machine vision-based automated machining path planning system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the machine vision-based automated machining path planning system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The machine vision-based automated machining path planning system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0262] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned automated processing path planning method based on machine vision is implemented.

[0263] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A machine vision-based automated processing path planning method, characterized in that, The method includes: A multi-angle visual data set of the processing object is obtained, wherein the multi-angle visual data set includes surface images of the processing object acquired from different viewpoints and corresponding viewpoint parameters; The multi-angle visual data set is subjected to three-dimensional structural feature extraction processing to obtain the surface structural features of the processed object. The surface structural features include the contour boundary features, surface curvature features and regional concavity features of the processed object. Based on the surface structure features, the machining constraint region features of the workpiece are identified. The machining constraint region features include the boundary features of the unmachinable region, the features of the precision machining region, and the turning restriction features of the machining path. The processing constraint region features are associated and fused with preset processing process parameters to generate a path planning constraint feature set, which includes regional processing priority features, path continuity constraint features, and processing accuracy matching features. The pre-trained processing path planning model is invoked to perform path generation processing on the path planning constraint feature set, generating an automated processing path instruction set containing processing path coordinate sequences, path execution order, and processing parameter configurations.

2. The automated processing path planning method based on machine vision according to claim 1, characterized in that, The step of performing three-dimensional structural feature extraction processing on the multi-angle visual data set to obtain the surface structural features of the processed object includes: The multi-angle visual data set is input into the multi-view feature encoding layer of the three-dimensional structural feature extraction network to perform depth feature extraction processing on the surface images of the processed object from different perspectives, generating a multi-view feature map containing perspective difference information. The feature matching module of the three-dimensional structural feature extraction network performs cross-view feature point association processing on the multi-view feature map to generate a set of feature point matching pairs with spatial correspondence. Based on the set of feature point matching pairs and the corresponding view parameters, the three-dimensional coordinate information of the feature points on the surface of the object being processed is calculated by triangulation, and a point cloud data set of the object being processed is constructed. The point cloud dataset is subjected to surface reconstruction processing to generate a three-dimensional mesh model of the object being processed. The three-dimensional mesh model includes vertex coordinate information, facet connection relationships, and normal vector parameters. Contour boundary features are extracted from the three-dimensional mesh model. These contour boundary features are described by the edge vertex sequence of the mesh model and the angle parameter between the normal vectors of adjacent facets. Calculate the curvature value of each vertex in the three-dimensional mesh model to generate a surface curvature feature distribution map containing principal curvature values, curvature directions, and curvature change rates; Identify regions in the three-dimensional mesh model that meet a preset depression depth threshold, and extract the boundary contour coordinates, depression depth parameters, and region area information of the depression regions as region depression features; The contour boundary features, surface curvature feature distribution map, and regional depression features are fused to generate surface structure features containing spatial geometric attributes.

3. The automated processing path planning method based on machine vision according to claim 2, characterized in that, The process of identifying the processing constraint region features of the processing object based on the surface structure features includes: The region segmentation module of the processing constraint region recognition model is invoked to perform semantic segmentation processing on the surface structure features, dividing the three-dimensional surface of the processing object into multiple semantic region units, each of which contains a corresponding region type label. Based on the region type label, semantic region units marked as unprocessable regions are selected, and the boundary vertex coordinate sequence and boundary smoothness parameters of the unprocessable regions are extracted to generate unprocessable region boundary features. Identify target regions marked as precision machining regions from the semantic region units, and extract the surface roughness parameters, flatness error values, and dimensional tolerance requirements of the target regions as features of the precision machining regions. Analyze the relative positional relationship between the precision-machined area and the non-machined area, calculate the minimum distance parameter and relative azimuth information, and generate positional constraint features between areas; Based on the surface curvature feature distribution map of the three-dimensional mesh model, regions where the curvature change rate exceeds a preset threshold are identified as path turning restriction regions. The path direction change range parameters and minimum turning radius requirements of the turning restriction regions are extracted as processing path turning restriction features. The non-processable region boundary features, precision processing region features, and processing path turning restriction features are structurally integrated to generate a set of processing constraint region features containing spatial constraint relationships.

4. The automated processing path planning method based on machine vision according to claim 3, characterized in that, The region segmentation module, which invokes the processing constraint region identification model, performs semantic segmentation processing on the surface structure features, dividing the three-dimensional surface of the processing object into multiple semantic region units, including: The surface structure features are input into the feature enhancement layer of the region segmentation module. The contour boundary features, surface curvature features and region concavity features are enhanced through the residual connection structure to generate enhanced surface structure features. The enhanced surface structure features are subjected to three-dimensional convolution processing to extract multi-scale spatial features under different receptive fields. The multi-scale spatial features include local detail features and global context features. The multi-scale spatial features are processed by channel attention mechanism to assign channel weights, which strengthens the feature channels that play an important role in the identification of processing constraint regions and suppresses redundant feature channels, resulting in weighted multi-scale spatial features. The weighted multi-scale spatial features are input into the graph convolutional network layer of the region segmentation module to construct graph structure data with vertices of the 3D mesh model as nodes and adjacency relationships as edges, and to perform spatial relationship modeling. The graph structure data is classified by the graph convolutional network layer, and a corresponding region type label is assigned to each vertex. The region type label includes unprocessable region label, precision processing region label and ordinary processing region label. Based on vertex region type labels, the 3D mesh model is divided into regions. A set of vertices with the same region type label and spatial contiguousness is divided into a semantic region unit, generating a set of semantic region units containing region identifiers, vertex index lists and region type labels.

5. The automated processing path planning method based on machine vision according to claim 1, characterized in that, The step of associating and fusing the processing constraint region features with preset processing parameters to generate a path planning constraint feature set includes: The preset machining process parameters are analyzed, and the machining tool parameters, machining sequence rules and accuracy requirements are extracted to generate a structured set of process parameters. A compatibility analysis is performed on the boundary features of the unmachinable region in the processing constraint region features and the processing tool parameters in the process parameter set to calculate the minimum safe distance parameter between the movement trajectory of the processing tool and the boundary of the unmachinable region. The precision machining area features are matched with the precision requirement indicators in the process parameter set to generate precision matching features that include the area precision level, allowable error range and surface quality requirements. Based on the machining path turning restriction features and the machining sequence rules in the process parameter set, a direction constraint graph model of the machining path is constructed. In the direction constraint graph model, the nodes represent machining regions, and the directed edges represent the machining sequence and turning restriction relationships between regions. The minimum safety distance parameter, accuracy matching feature and orientation constraint map model are subjected to feature fusion processing to generate regional processing priority features. The regional processing priority features are calculated by comprehensively considering the regional area, accuracy requirements and positional relationship with other regions. Analyze the path connectivity between adjacent processing areas, extract path smoothness requirements, continuity thresholds, and transition region length parameters, and generate path continuity constraint features; The regional processing priority features, path continuity constraint features, and accuracy matching features are structurally integrated to generate a path planning constraint feature set containing multi-dimensional constraint information.

6. The automated processing path planning method based on machine vision according to claim 5, characterized in that, The process of constructing a directional constraint graph model for the machining path based on the machining path turning constraint features and the machining sequence rules in the set of process parameters includes: The processing sequence rules are analyzed to extract the regional processing order relationship, parallel processing constraints, and path backtracking prohibition rules. Each semantic region unit of the processing object is used as a node in the orientation constraint graph model. The node attribute parameters are initialized, including the region area, accuracy level, and processing time estimation parameters. Based on the path direction change range parameter and minimum turning radius requirement in the processing path turning restriction feature, the path turning cost parameter between adjacent semantic region units is calculated. The path turning cost parameter is positively correlated with the direction change range and the turning radius. Based on the processing sequence relationship of the region, directed edges are added to the directional constraint graph model. The weight value of the directed edge is calculated by multiplying the path turning cost parameter and the processing sequence priority coefficient. Based on the parallel processing constraints, mark the set of nodes that can be processed in parallel in the directional constraint graph model, add undirected connecting edges between the nodes that can be processed in parallel, and the weight value of the edge represents the synergy coefficient of parallel processing. The path backtracking prohibition rule is transformed into an edge constraint condition of the direction constraint graph model, prohibiting the addition of reverse directed edges between nodes in the processed region and nodes in the subsequent processed region. The constructed directional constraint graph model is subjected to topological sorting verification to determine whether there are circular dependencies. If circular dependencies exist, the weight values ​​of the directed edges of the relevant nodes are adjusted until the topological sorting requirements are met.

7. The automated processing path planning method based on machine vision according to claim 1, characterized in that, The process involves calling a pre-trained processing path planning model to perform path generation processing on the path planning constraint feature set, generating an automated processing path instruction set containing a processing path coordinate sequence, path execution order, and processing parameter configuration, including: The set of path planning constraint features is input into the constraint feature encoding layer of the processing path planning model for feature vectorization processing to generate a fixed-dimensional constraint feature vector. The graph attention network module of the processing path planning model performs regional association modeling on the constraint feature vector, strengthens the feature representation of important processing areas, and generates a weighted constraint feature map. The weighted constraint feature map is input into the path generation network of the processing path planning model. The path generation network includes a region ranking subnetwork and a path coordinate prediction subnetwork. The processing regions are sequentially planned using the regional sorting sub-network, and a regional processing sequence is generated based on the regional processing priority features. The processing sequence of the region is input into the path coordinate prediction subnetwork. Combined with the path continuity constraint features, a path coordinate sequence within each processing region is generated. The path coordinate sequence is described by the ordered arrangement of three-dimensional spatial coordinate points. Based on the accuracy matching features, corresponding machining parameters are configured for each path coordinate point, including feed rate, depth of cut and spindle speed. The processing sequence of the region, the path coordinate sequence, and the corresponding processing parameters are integrated and processed to generate a set of automated processing path instructions containing timestamps. The timestamps are used to indicate the start time of execution for each path segment.

8. The automated processing path planning method based on machine vision according to claim 7, characterized in that, The step of performing sequential planning processing on the processing regions through the region sorting sub-network, and generating a region processing order sequence based on the region processing priority features, includes: The weighted constrained feature map is input into the self-attention mechanism layer of the region ranking subnetwork to perform correlation modeling on the feature representations of each processing region and generate an inter-regional attention weight matrix. Based on the inter-regional attention weight matrix, the global importance score of each processing region is calculated. The global importance score is obtained by weighted summation of the attention weight value and the region processing priority feature. The initialization process sequence is an empty sequence. The process region with the highest global importance score is selected as the starting process region and added to the process sequence. From the remaining processing regions, the selection probability of the candidate processing region is calculated based on the path turning cost parameter between the selected processing region and the candidate processing region and the attention weight value between regions; Select the candidate processing region with the highest selection probability and add it to the region processing order sequence. Repeat the above selection process until all processing regions have been added to the sequence. The generated regional processing sequence is locally optimized by using a neighborhood exchange algorithm to adjust the order of adjacent regions, calculating the total path cost before and after adjustment, and retaining the regional processing sequence with the minimum total path cost.

9. The automated processing path planning method based on machine vision according to claim 7, characterized in that, The step of inputting the processing sequence of the region into the path coordinate prediction subnetwork, and combining the path continuity constraint features to generate the path coordinate sequence within each processing region, includes: Input the 3D mesh model of the current processing area and the path endpoint coordinates of the previous processing area in the processing sequence of the area into the path starting point determination module of the path coordinate prediction sub-network; Based on the path smoothness requirement in the path continuity constraint feature, the optimal path start coordinates of the current processing area are calculated. The distance between the optimal path start coordinates and the path end coordinates of the previous processing area should meet the smooth transition threshold requirement. On the surface of the 3D mesh model of the current processing area, an initial set of path points is generated using an equidistant sampling algorithm. The distribution density of the initial path points is positively correlated with the accuracy level of the current processing area. The initial set of path points is optimized by the path optimization layer of the path coordinate prediction subnetwork, and the position of the path points is adjusted based on the surface curvature feature distribution map to obtain the optimized path points. Based on the continuity threshold in the path continuity constraint features, the optimized path points are connected to generate an initial path coordinate sequence. The initial path coordinate sequence is smoothed using a B-spline curve fitting algorithm to generate a smooth path curve containing a control point sequence and curve parameters. The final path coordinate sequence is obtained by sampling from the smooth path curve at a preset step size. The path coordinate sequence includes three-dimensional coordinate values ​​and corresponding curve tangent direction vectors.

10. An automated processing path planning system based on machine vision, characterized in that, The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the machine vision-based automated processing path planning method according to any one of claims 1-9.

Images