Process feature identification method, apparatus, device, storage medium, and program product

Abstract

This invention relates to the field of computer technology, providing a method, apparatus, device, storage medium, and program product for process feature recognition. The method includes: inputting a three-dimensional model into a process feature recognition model to obtain recognition results output by the model; for any given process feature, determining the geometric type of each base surface based on the recognition results; determining the parameter calculation method for each base surface based on its geometric type; and calculating the geometric parameters of the process feature based on the parameter calculation method. This invention can automatically calculate geometric parameters, thereby improving the efficiency and reducing the cost of geometric parameter calculation; and by determining corresponding parameter calculation methods for different base surface geometric types, it can more accurately calculate the geometric parameters of the process feature, thus improving the accuracy of geometric parameter calculation.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, storage medium, and program product for identifying process features. Background Technology

[0002] Computer-Aided Process Planning (CAPP) refers to the use of computer hardware and software technology and supporting environments to formulate machining processes for parts through numerical calculations, logical judgments, and reasoning. It is a key component of modern intelligent manufacturing systems. CAPP connects with Computer-Aided Design (CAD) at the top and Computer-Aided Manufacturing (CAM) at the bottom, serving as a bridge between design and manufacturing. It transforms the geometric information in the design model into an executable machining process plan. Process feature identification is a crucial step in CAPP, aiming to automatically identify manufacturing semantics-based process features from the design model, such as holes, slots, bends, and louvers, providing a basis for subsequent process decisions. Therefore, process feature identification of the design model is necessary.

[0003] Traditional process feature recognition schemes are mainly based on predefined rules and geometric pattern matching. However, this traditional approach has significant limitations when dealing with complex 3D models; for example, it is very difficult and ineffective when handling cases with blurred boundaries and complex and variable topologies caused by feature intersections.

[0004] Currently, deep learning methods can learn feature representations from a large number of samples, overcoming the insufficient generalization ability of rule-based methods, especially performing well when dealing with complex features with highly intersecting geometric shapes. However, these methods only identify process features and do not address the extraction of the geometric parameters of these features. These geometric parameters are crucial for supporting intelligent process planning and cost analysis, but currently, they still rely on manual measurement, resulting in high extraction costs and low efficiency. Summary of the Invention

[0005] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a process feature identification method that can automatically calculate geometric parameters, thereby improving the calculation efficiency and reducing the calculation cost of geometric parameters; and determines the corresponding parameter calculation method for different bottom surface geometric types to more accurately calculate the geometric parameters of process features, thereby improving the calculation accuracy of geometric parameters.

[0006] The present invention also provides a process feature identification device, an electronic device, a storage medium, and a computer program product.

[0007] The process feature identification method according to a first aspect embodiment of the present invention includes:

[0008] The three-dimensional model is input into the process feature recognition model to obtain the recognition result output by the process feature recognition model. The recognition result includes the feature type of each process feature in the three-dimensional model, the constituent surfaces of each process feature, and the detection result of each constituent surface in the three-dimensional model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the three-dimensional model.

[0009] For any of the aforementioned process features, based on the identification results, the geometric type of each bottom surface included in the process feature is determined;

[0010] Based on the geometric type of each of the bottom surfaces, the parameter calculation method for each of the bottom surfaces is determined respectively;

[0011] Based on the parameter calculation method of each of the bottom surfaces, the geometric parameters of the process features are calculated;

[0012] The process feature recognition model is trained based on the sample 3D model and the recognition result labels corresponding to the sample 3D model.

[0013] According to the process feature recognition method of the present invention, a three-dimensional model is input into a process feature recognition model to obtain the recognition result output by the process feature recognition model. The process feature recognition model is trained based on a sample three-dimensional model and the corresponding recognition result labels, thereby overcoming the problem of insufficient generalization ability of rule-based methods through deep learning, and thus improving the accuracy of process feature recognition. The recognition result includes the feature type of each process feature in the three-dimensional model, the constituent surfaces included in each process feature, and the detection results of each constituent surface in the three-dimensional model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature. The bottom surface is the geometric surface used to define the size of the three-dimensional model, thereby enabling accurate identification and positioning of process features and their related constituent surfaces, and thus improving the accuracy of process feature recognition. The accuracy of process feature recognition is improved, and the recognition of process features and the detection of key geometric attributes (bottom surfaces) are completed in a unified model, directly outputting recognition results that can be used for geometric parameter calculation. This breaks through the process bottleneck of separating process feature recognition and geometric parameter calculation in traditional solutions. It can automatically calculate geometric parameters, thereby improving the calculation efficiency of geometric parameters and reducing the calculation cost. For any process feature, based on the recognition results, the geometric type of each bottom surface included in the process feature is determined. Based on the geometric type of each bottom surface, the parameter calculation method of each bottom surface is accurately determined. Based on the parameter calculation method of each bottom surface, the geometric parameters of the process feature are accurately calculated. Thus, the corresponding parameter calculation method is determined for different bottom surface geometric types, so as to calculate the geometric parameters of the process feature more accurately and improve the calculation accuracy of geometric parameters.

[0014] According to one embodiment of the present invention, the process feature recognition model is trained based on a training dataset, the training dataset including multiple sample 3D models and recognition result labels corresponding to each sample 3D model;

[0015] The three-dimensional model of any of the samples is constructed based on the following method:

[0016] Based on the process scenario to be constructed, a preset set of process features is determined;

[0017] Several preset process features are determined from the preset process feature set, and the geometric parameters of the several preset process features are randomly determined based on the preset parameter ranges corresponding to the several preset process features respectively.

[0018] Based on the aforementioned preset process features and their geometric parameters, a three-dimensional model of the sample is constructed.

[0019] According to one embodiment of the present invention, after constructing the three-dimensional model of the sample based on the plurality of preset process features and the geometric parameters of the plurality of preset process features, the method further includes:

[0020] Display the feature names of each sample process feature in the sample 3D model, and display the sample 3D model;

[0021] The name of the first target feature selected by the user is determined based on the first feature selection instruction;

[0022] Based on the feature type-color mapping relationship, the label color corresponding to the first target feature name is determined; the feature type-color mapping relationship is used to indicate the one-to-one correspondence between the feature type corresponding to each feature name and each color.

[0023] The annotation color is rendered on each sample constituent surface of the sample process feature corresponding to the first target feature name in the sample 3D model, and a color different from the annotation color is rendered on the sample bottom surface of the sample process feature corresponding to the first target feature name in the sample 3D model.

[0024] After receiving the color update instruction, the color of the sample constituent surface indicated by the color update instruction is updated to obtain the color of each sample constituent surface included in the sample process feature corresponding to the first target feature name.

[0025] Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, the recognition result label corresponding to the three-dimensional model of the sample is determined.

[0026] According to one embodiment of the present invention, after determining the name of the first target feature selected by the user based on the first feature selection instruction, the method further includes:

[0027] After receiving the face selection instruction, the sample process features corresponding to the first target feature name are determined based on the face selection instruction, including each sample constituent face.

[0028] Accordingly, determining the recognition result label corresponding to the three-dimensional model of the sample based on the color of each sample constituent face in the sample three-dimensional model includes:

[0029] Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, and the constituent surfaces of each sample included in the process features of each sample in the three-dimensional model of the sample, the identification result label corresponding to the three-dimensional model of the sample is determined.

[0030] According to an embodiment of the present invention, after obtaining the face selection instruction and determining the sample constituent faces included in the sample process feature corresponding to the first target feature name based on the face selection instruction, the method further includes:

[0031] Upon receiving the annotation application instruction, construct the unique identifier information of the sample process feature corresponding to the first target feature name;

[0032] The unique identifier information of the sample process feature corresponding to the first target feature name is bound to each sample constituent surface included in the sample process feature corresponding to the first target feature name.

[0033] After determining the recognition result label corresponding to the three-dimensional model of the sample based on the color of each sample constituent surface in the sample three-dimensional model and the sample constituent surfaces included in each sample process feature in the sample three-dimensional model, the method further includes:

[0034] Based on the preset parameter ranges corresponding to the process features of each sample in the three-dimensional model of the sample, the geometric parameters of the process features of each sample in the three-dimensional model of the sample are updated to obtain a new three-dimensional model of the sample.

[0035] Based on the unique identifier information of each sample process feature in the new sample 3D model, and the color of each sample constituent surface in the new sample 3D model, the identification result label of the new sample 3D model is determined.

[0036] According to one embodiment of the present invention, after constructing the three-dimensional model of the sample based on the plurality of preset process features and the geometric parameters of the plurality of preset process features, the method further includes:

[0037] If the topological rationality verification of the sample 3D model is passed, the sample 3D model is added to the training dataset;

[0038] The training dataset is optimized based on at least one of the following methods:

[0039] If the number of sample 3D models in the training dataset is different from the number of identification result labels in the training dataset, delete the first abnormal sample 3D model and / or the first abnormal identification result label in the training dataset; the first abnormal sample 3D model is a sample 3D model without a corresponding identification result label, and the first abnormal identification result label is an identification result label without a corresponding sample 3D model.

[0040] If a second abnormal sample 3D model and a second abnormal identification result label exist in the training dataset, delete the second abnormal sample 3D model and the second abnormal identification result label from the training dataset; the exported file name of the second abnormal sample 3D model is different from the exported file name of the second abnormal identification result label;

[0041] If a 3D model of a third anomalous sample and a label of a third anomalous identification result exist in the training dataset, delete the 3D model of the third anomalous sample and the label of the third anomalous identification result from the training dataset; the number of constituent faces of the 3D model of the third anomalous sample is different from the number of constituent faces indicated by the label of the third anomalous identification result.

[0042] According to one embodiment of the present invention, the step of inputting the three-dimensional model into the process feature recognition model and obtaining the recognition result output by the process feature recognition model includes:

[0043] The 3D model is converted into a face adjacency graph; the nodes of the face adjacency graph represent the constituent faces, and the edges of the face adjacency graph represent the connecting edges between two constituent faces.

[0044] Based on the geometric information of the 3D model, the attribute information of the 3D model, and the face adjacency graph, an attribute adjacency graph is constructed; the geometric information includes face geometric information and edge geometric information, the attribute information includes face attribute information and edge attribute information, the face geometric information is used to be attached to the nodes of the face adjacency graph, the edge geometric information is used to be attached to the edges of the face adjacency graph, the face attribute information is used to be attached to the nodes of the face adjacency graph, and the edge attribute information is used to be attached to the edges of the face adjacency graph;

[0045] The attribute adjacency graph is input into the process feature recognition model to obtain the recognition result output by the process feature recognition model.

[0046] According to one embodiment of the present invention, when at least one bottom surface of the process feature is located or spans a bending region of the three-dimensional model, the geometric parameters of the process feature are calculated based on the following:

[0047] The process features are projected to obtain a two-dimensional contour graphic;

[0048] Based on the two-dimensional contour graphic, the geometric parameters of the process feature are determined.

[0049] According to one embodiment of the present invention, after inputting the three-dimensional model into the process feature recognition model and obtaining the recognition result output by the process feature recognition model, the method further includes:

[0050] Based on the feature type of each process feature, the parameter calculation tool corresponding to each process feature is scheduled respectively;

[0051] The parameter calculation tool corresponding to any of the process features is used to calculate the geometric parameters of the process feature.

[0052] According to one embodiment of the present invention, the recognition result includes semantic segmentation result, instance segmentation result, and bottom surface detection result;

[0053] The semantic segmentation result is used to indicate the feature type of each constituent surface in the three-dimensional model, the instance segmentation result is used to indicate the constituent surfaces included by each process feature in the three-dimensional model, and the bottom surface detection result includes the detection results of each constituent surface in the three-dimensional model;

[0054] The feature types of each process feature in the three-dimensional model are determined based on the following method:

[0055] Based on the semantic segmentation results and the instance segmentation results, the feature types of each process feature in the three-dimensional model are determined.

[0056] According to one embodiment of the present invention, after obtaining the geometric parameters of each process feature in the three-dimensional model, the method further includes:

[0057] Display the feature names of each process feature in the 3D model, and display the 3D model;

[0058] The name of the second target feature selected by the user is determined based on the second feature selection instruction;

[0059] Display the geometric parameters of the process feature corresponding to the second target feature name, and render the process feature corresponding to the second target feature name on the three-dimensional model using a preset rendering method; the preset rendering method is used to distinguish the process feature corresponding to the second target feature name from other process features in the three-dimensional model besides the process feature corresponding to the second target feature name.

[0060] A process feature identification device according to a second aspect of the present invention includes:

[0061] The feature recognition module is used to input a 3D model into a process feature recognition model and obtain the recognition result output by the process feature recognition model. The recognition result includes the feature type of each process feature in the 3D model, the constituent surfaces included in each process feature, and the detection result of each constituent surface in the 3D model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the 3D model.

[0062] A type determination module is used to determine the geometric type of each bottom surface included in any of the process features based on the identification result;

[0063] The method determination module is used to determine the parameter calculation method for each of the bottom surfaces based on the geometric type of each bottom surface;

[0064] The parameter calculation module is used to calculate the geometric parameters of the process features based on the parameter calculation method of each of the bottom surfaces;

[0065] The process feature recognition model is trained based on the sample 3D model and the recognition result labels corresponding to the sample 3D model.

[0066] An electronic device according to a third aspect of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the process feature identification method as described above.

[0067] According to a fourth aspect of the present invention, a non-transitory computer-readable storage medium is provided thereon storing a computer program that, when executed by a processor, implements the process feature identification method as described above.

[0068] A computer program product according to a fifth aspect of the present invention includes a computer program that, when executed by a processor, implements the process feature identification method as described above.

[0069] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0070] A 3D model is input into a process feature recognition model to obtain the recognition results output by the model. This model is trained based on sample 3D models and their corresponding recognition result labels, thus overcoming the insufficient generalization ability of rule-based methods through deep learning, thereby improving the accuracy of process feature recognition. The recognition results include the feature type of each process feature in the 3D model, the constituent surfaces of each feature, and the detection results of each constituent surface. The detection result of any constituent surface indicates whether it is the base surface used to calculate the geometric parameters of the process feature. The base surface is the geometric surface used to define the dimensions of the 3D model, thereby enabling precise identification and positioning of process features and their related constituent surfaces, further improving the accuracy of process feature recognition. Furthermore, it integrates process feature identification and key geometric attribute (bottom surface) detection into a unified model, directly outputting identification results for geometric parameter calculation. This overcomes the bottleneck of separating process feature identification and geometric parameter calculation in traditional solutions, enabling automatic calculation of geometric parameters, thereby improving calculation efficiency and reducing calculation costs. For any process feature, based on the identification results, it determines the geometric type of each bottom surface included in the process feature. Based on the geometric type of each bottom surface, it accurately determines the parameter calculation method for each bottom surface, and accurately calculates the geometric parameters of the process feature. Thus, it determines the corresponding parameter calculation method for different bottom surface geometric types to more accurately calculate the geometric parameters of the process feature, thereby improving the accuracy of geometric parameter calculation.

[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 This is one of the flowcharts of the process feature recognition method provided in the embodiments of the present invention.

[0074] Figure 2 This is a schematic diagram of the geometric parameter calculation method provided in the embodiment of the present invention.

[0075] Figure 3 This is the second flowchart of the process feature identification method provided in the embodiments of the present invention.

[0076] Figure 4 This is a schematic diagram of the labeling method provided in an embodiment of the present invention.

[0077] Figure 5 This is one of the schematic diagrams of the three-dimensional model of the sample provided in the embodiments of the present invention.

[0078] Figure 6 This is the second schematic diagram of the three-dimensional model of the sample provided in the embodiment of the present invention.

[0079] Figure 7 This is a schematic diagram of the three-dimensional model provided in the embodiment of the present invention.

[0080] Figure 8 This is one of the schematic diagrams showing the geometric parameters provided in the embodiments of the present invention.

[0081] Figure 9 This is the second schematic diagram showing the geometric parameters provided in the embodiment of the present invention.

[0082] Figure 10 This is a schematic diagram of the process feature recognition device provided in an embodiment of the present invention.

[0083] Figure 11 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0085] The present invention proposes the following embodiments. The process feature recognition method provided by the embodiments of the present invention is described below with reference to the accompanying drawings. The executing entity of this process feature recognition method can be a process feature recognition system, a server, a desktop computer, a laptop computer, or a user terminal, including but not limited to mobile phones, tablet computers, vehicle terminals, and smart home appliances.

[0086] Figure 1 This is one of the flowcharts illustrating the process feature recognition method provided in this embodiment of the invention, such as... Figure 1 As shown, the process feature identification method includes steps 110, 120, 130 and 140.

[0087] Step 110: Input the 3D model into the process feature recognition model to obtain the recognition result output by the process feature recognition model.

[0088] Here, a 3D model refers to a digital model created using computer software to represent the three-dimensional geometry and structure of an object; it is a model that requires process feature identification. In one embodiment, the 3D model can be a 3D CAD (Computer-Aided Design) model. For example, the 3D model could be a model of a piping or sheet metal part.

[0089] Here, the process feature recognition model is a machine learning model, and more specifically, a deep learning model, which is used to recognize process features on a 3D model.

[0090] In one embodiment, the process feature recognition model is a multi-task model, meaning it can include a semantic segmentation layer, an instance segmentation layer, and a bottom surface detection layer. The semantic segmentation layer performs semantic segmentation on the 3D model to obtain semantic segmentation results, which indicate the feature types of each constituent surface in the 3D model. The instance segmentation layer performs instance segmentation on the 3D model to obtain instance segmentation results, which indicate the constituent surfaces included in each process feature of the 3D model. The bottom surface detection layer performs bottom surface detection on the 3D model to obtain bottom surface detection results, which include the detection results of each constituent surface in the 3D model. Based on this, through the above three tasks, process features and their related constituent surfaces can be accurately identified and located, providing crucial support for downstream operations and thus improving the accuracy of process feature recognition. Furthermore, this invention proposes an end-to-end solution that integrates process feature recognition (semantic segmentation, instance segmentation) and key geometric attribute (bottom surface) detection into a unified deep learning model, directly outputting structured feature instances (recognition results) that can be used for geometric parameter calculation. This breaks through the process bottleneck of separating process feature recognition and geometric parameter calculation in traditional solutions, thereby improving the calculation efficiency of geometric parameters and reducing the calculation cost of geometric parameters.

[0091] In one embodiment, the process feature recognition model can extract local and global features of a three-dimensional model, and can fuse local and global features to obtain recognition results based on the fused features.

[0092] The process feature recognition model is trained based on the sample 3D model and the corresponding recognition result label. Further, the initial network model is trained and optimized, and finally, the training effect is evaluated using metrics such as accuracy, F1 score, and Mean Intersection over Union (MIoU), with the optimal weights used as model parameters.

[0093] The identification results include the feature type of each process feature in the three-dimensional model, the constituent surfaces of each process feature, and the detection results of each constituent surface in the three-dimensional model; the detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the three-dimensional model.

[0094] Here, process features are also called machining features, such as crimping, upsetting, flaring, square holes, straight pipes, bends, and perforations. Flaring is the feature name corresponding to this process feature. The feature type of a process feature is also called a feature category. In one embodiment, the feature type of a process feature can be identified and distinguished by a numerical value, for example, a feature category ID.

[0095] It should be understood that a process feature includes multiple feature faces (constituent faces or surfaces). The bottom face is a specific geometric face used to define the critical dimensions of the 3D model; more specifically, the bottom face is a specific geometric face used to define the critical dimensions of the process feature. The definition of the bottom face is related to the feature type and machining datum of the process feature. By labeling the bottom face detection results and using deep learning methods, the detection of a specific bottom face can be achieved. A process feature typically includes at least one bottom face, but it can also include multiple bottom faces.

[0096] Step 120: For any of the process features, based on the identification results, determine the geometric type of each bottom surface included in the process feature.

[0097] It should be noted that similar processing is required for each process feature in the 3D model. For ease of explanation, we will focus on one of the process features here.

[0098] Since the identification results include the constituent surfaces of each process feature and the detection results of each constituent surface in the 3D model, the bottom surfaces included in the process feature can be determined; and since the identification results include the feature types of each process feature in the 3D model, the geometric types of the bottom surfaces included in the process feature can be determined.

[0099] The geometry type can include, but is not limited to, at least one of the following: planar, cylindrical, toroidal, and hybrid types. For example, a process feature flare, which includes five constituent surfaces (two conical surfaces, two cylindrical surfaces, and one toroidal surface), and the corresponding flare can include a bottom surface (an inner cylindrical surface).

[0100] Step 130: Based on the geometric type of each bottom surface, determine the parameter calculation method for each bottom surface.

[0101] Here, the parameter calculation method refers to the geometric parameter calculation method, which specifies the calculation method for the geometric parameters of the corresponding process features. Different geometric types correspond to different parameter calculation methods; that is, one geometric type corresponds to one parameter calculation method.

[0102] For example, if the geometry of the bottom surface is planar, its corresponding parameters are calculated as follows: by traversing the topological relationships of the 3D model (e.g., by characterizing the model using B-Rep data), parallel surfaces of the bottom surface are found and their normal distances are calculated to obtain the length and width; or adjacent surfaces are found through the shared edges of the bottom surface and the angle between their normal vectors is calculated to obtain the angle. In other words, for process features dominated by planar surfaces such as louvers, square holes, and cross holes, the parallel surface distance measurement method is used to calculate the surface spacing, and the maximum / minimum distance is taken as the length / width, respectively; for bending angle calculation, the adjacent surface angle method is introduced, and adjacent planes are located through shared edges, and the angle between their normal vectors in space is calculated.

[0103] For example, if the geometry of the bottom surface is a curved surface (such as a cylindrical surface or a torus), the corresponding parameter calculation method is as follows: the radius of the cylindrical surface and the large radius of the torus are directly read through the geometry adapter, and the diameter, length, bending angle, and spatial angle are obtained by calculating the span of the surface parameter domain or the geometric relationship of the endpoints. In other words, when processing cylindrical / torus surface features such as circular holes and bent pipes, the cylindrical surface radius method is used for circular hole features: locate the cylindrical surface → extract the radius value → calculate the diameter; the parameter domain difference method is used for tubular features: obtain the UV domain of the surface → calculate the V-direction span → output the length; the composite calculation is used for bent pipe features: the bending angle (angle between the center and the endpoint) is calculated using the endpoint geometry method, combined with the spatial direction method (comparing the acute angles of adjacent bent pipe axes).

[0104] For example, if the geometry of the bottom surface is a mixed type, the corresponding parameter calculation method is as follows: Dynamic local coordinate system construction: A temporary local coordinate system is automatically determined and constructed based on the main direction of the process feature (such as the longest side and the normal of adjacent faces); Coordinate transformation and bounding box analysis: All relevant faces of the feature are transformed to the above local coordinate system, and their axial bounding boxes are calculated to extract the length, width, and height dimensions; Multi-method fusion: The geometric relationship analysis method (for angle calculation) and the parameter direct extraction method (for radius calculation) are used in combination to calculate other geometric parameters. In other words, for complex contour features such as bridge holes, ribs, and bends, a coordinate system transformation bounding box method is developed to establish a coordinate system (X-axis = the direction of the longest straight edge, Z-axis = the normal of adjacent faces), implement local coordinate transformation → calculate the bounding box after transformation → decompose the length / width / height dimensions.

[0105] It should be understood that, based on the geometry type of each base surface, it is automatically routed to the corresponding layered calculation strategy, that is, the parameter calculation method of each base surface is determined separately, thereby improving the calculation efficiency of geometric parameters and reducing the calculation cost of geometric parameters.

[0106] Step 140: Calculate the geometric parameters of the process features based on the parameter calculation method of each bottom surface.

[0107] It should be noted that the bottom surface is used to calculate the geometric parameters of the process feature. Therefore, the geometric parameters of the process feature are calculated based on the parameter calculation method of each bottom surface. These geometric parameters may include, but are not limited to: length, width, radius, diameter, inner diameter, angle, bending angle, spatial angle, etc.

[0108] For example, such as Figure 2 As shown, the 3D model is a piping model, which includes three types of process features: flared pipes, bends, and straight pipes. For flared pipes, the length and inner diameter need to be calculated; for bends, the bend radius, bending angle, and spatial angle need to be calculated; and for straight pipes, the length needs to be calculated. The bottom surfaces of all three process features are curved surfaces. The inner diameter specification of the flared pipe is calculated using the cylindrical surface radius method: the cylindrical surface is positioned as the inner wall of the flared pipe → the radius value is extracted → the calculated diameter is 12.7 mm. The flared pipe length and straight pipe length are calculated using the parameter domain difference method: the UV domain of the curved surface is obtained → the V-direction span is calculated → the flared pipe length is output as 10 mm, and the straight pipe length is 66 mm. For bends, a composite calculation is used: the bending angle is calculated as 90° by the angle between the center and the two endpoints; the spatial angle is calculated by comparing the acute angles of adjacent bend axes using the spatial direction method.

[0109] Here, geometric parameters can include attribute names, values, and units. Furthermore, while calculating geometric parameters, the coordinates of process features can also be obtained, thus acquiring geometric information. Based on this, ready-to-use and accurate geometric information is provided for downstream applications such as CAPP (Computer-Aided Process Planning) intelligent process planning and cost analysis.

[0110] It should be understood that in existing technologies, process design still primarily uses two-dimensional engineering drawings, while product design often employs three-dimensional models. Therefore, manual extraction and conversion of the three-dimensional model (design data) are required, leading to low efficiency in process design and a disconnect between product design and process design. Based on this, the embodiments of this invention can accurately identify and locate process features and their related constituent surfaces, and complete process feature identification and key geometric attribute (bottom surface) detection within a unified model, directly outputting identification results suitable for geometric parameter calculation. This overcomes the bottleneck of separating process feature identification and geometric parameter calculation in traditional solutions, automatically calculating geometric parameters without requiring manual extraction and conversion of design data, thus improving efficiency and reducing costs.

[0111] It should be understood that since the automatic recognition of 3D process features directly expresses the 3D model, the CAPP system has realized the 3D process design capability. The development can express the design intent with a visual and interactive 3D process model, output 3D process guidance documents to the downstream manufacturing end, and improve the intuitiveness and accuracy of process design expression.

[0112] It should be understood that in existing technologies, process design relies heavily on engineers' experience. This involves visually identifying process features in a 3D model, manually searching and comparing them with existing process design experience (process templates, rules, etc.), and compiling process plans through copying and template calling. This process involves significant human intervention and reliance on individual engineers' abilities, making it difficult to guarantee the accuracy and efficiency of process design. Therefore, the embodiments of this invention can automatically identify process features and automatically calculate their geometric parameters, thereby reducing reliance on engineers and improving the accuracy and efficiency of process design.

[0113] It should be understood that in existing technologies, when process engineers estimate product processing costs, they need to manually extract relevant attribute data from the model and find relevant cost calculation formulas. This requires detailed breakdown and precise calculation, resulting in high complexity and a large workload for cost analysis. Therefore, the embodiments of this invention can automatically identify process features and automatically calculate their geometric parameters, thereby reducing reliance on process engineers and improving the accuracy and efficiency of cost calculation.

[0114] The process feature recognition method provided in this invention inputs a 3D model into a process feature recognition model to obtain the recognition result output by the model. The process feature recognition model is trained based on sample 3D models and the corresponding recognition result labels, thereby overcoming the insufficient generalization ability of rule-based methods through deep learning, thus improving the accuracy of process feature recognition. The recognition result includes the feature type of each process feature in the 3D model, the constituent surfaces of each process feature, and the detection results of each constituent surface in the 3D model. The detection result of any constituent surface indicates whether it is the bottom surface used to calculate the geometric parameters of the process feature. The bottom surface is the geometric surface used to define the dimensions of the 3D model, thereby enabling accurate identification and positioning of process features and their related constituent surfaces, thus improving the accuracy of process feature recognition. The accuracy of process feature recognition is improved, and the recognition of process features and the detection of key geometric attributes (bottom surfaces) are completed in a unified model, directly outputting recognition results that can be used for geometric parameter calculation. This breaks through the process bottleneck of separating process feature recognition and geometric parameter calculation in traditional solutions. It can automatically calculate geometric parameters, thereby improving the calculation efficiency of geometric parameters and reducing the calculation cost. For any process feature, based on the recognition results, the geometric type of each bottom surface included in the process feature is determined. Based on the geometric type of each bottom surface, the parameter calculation method of each bottom surface is accurately determined. Based on the parameter calculation method of each bottom surface, the geometric parameters of the process feature are accurately calculated. Thus, the corresponding parameter calculation method is determined for different bottom surface geometric types, so as to calculate the geometric parameters of the process feature more accurately and improve the calculation accuracy of geometric parameters.

[0115] Based on any of the above embodiments, in this method, the process feature recognition model is trained based on a training dataset, which includes multiple sample 3D models and recognition result labels corresponding to each sample 3D model.

[0116] Here, the training dataset includes multiple training data sets, each of which includes a sample 3D model and the corresponding recognition result label for that sample 3D model.

[0117] Figure 3 This is a second schematic flowchart of the process feature recognition method provided in this embodiment of the invention, as shown below. Figure 3 As shown, the three-dimensional model of any of the samples is constructed in the following manner.

[0118] Step 310: Based on the process scenario to be constructed, determine the preset process feature set.

[0119] Here, the process scenario can include, but is not limited to, at least one of the following: sheet metal, piping, heat exchangers, machining, etc. It should be understood that each process scenario has corresponding standard process features, that is, a corresponding set of preset process features. The preset process feature set includes multiple preset process features, thus the sample 3D model is constructed based on these standard preset process features. Based on this, it covers relevant manufacturing products, ensuring the accuracy and consistency of data from R&D to manufacturing. For example, the preset process feature set corresponding to the piping process scenario includes 12 standard preset process features, and the preset process feature set corresponding to the sheet metal process scenario includes 16 standard preset process features.

[0120] In one embodiment, the preset process feature sets corresponding to each process scenario are highly scalable, meaning that preset process features for each process scenario can be added, or modified or deleted. Based on this, the flexibility and scalability of constructing the sample 3D model are improved, thereby enhancing the training effect of the process feature recognition model and ultimately improving the accuracy of process feature recognition.

[0121] Step 320: Determine several preset process features from the preset process feature set, and randomly determine the geometric parameters of the several preset process features based on the preset parameter ranges corresponding to the several preset process features.

[0122] Here, the number of preset process features can be one or more, which can be determined randomly or selected manually. Each preset process feature has a corresponding preset parameter range, which is used to limit the geometric parameters of the preset process feature, i.e., geometric parameter constraints. Therefore, randomly determined geometric parameters are within the preset parameter range.

[0123] For example, the process scenario to be constructed is piping, and the predetermined process features include five standard features: flared end, narrow end, flanged hole, straight pipe, and bent pipe.

[0124] In one embodiment, a CSV (Comma-Separated Values) configuration file is created in advance to define the preset parameter ranges (parameter constraints) for each preset process feature.

[0125] Step 330: Based on the several preset process features and the geometric parameters of the several preset process features, construct the three-dimensional model of the sample.

[0126] In one specific embodiment, multiple expressions are constructed based on several preset process features and their geometric parameters, and a sample three-dimensional model is constructed by driving each expression.

[0127] For example, in CAD software, parametric modeling is randomly combined to create sample 3D models by driving the expressions included in the model.

[0128] In one embodiment, the exported file of the sample 3D model is a STEP (Standard for the Exchange of Product Model Data) geometry file that retains the accurate B-rep (Boundary representation) representation.

[0129] It should be understood that labeling real 3D models is very costly, and training is usually performed on computer-generated datasets. Therefore, performance degrades significantly when applied to real 3D models. Based on this, embodiments of the present invention can construct their own 3D model datasets, and can operate independently of 2D drawings, developing 3D models based on actual conditions. For example, it can automatically generate over 30,000 models each for piping and sheet metal datasets, along with corresponding identification result labels, thereby improving the robustness of the process feature recognition model and ultimately increasing the accuracy of process feature recognition.

[0130] The process feature recognition method provided in this invention determines a preset process feature set based on the process scenario to be constructed, identifies several preset process features from the preset process feature set, and randomly determines the geometric parameters of several preset process features based on the preset parameter ranges corresponding to the preset process features. Based on the several preset process features and their geometric parameters, a sample 3D model is constructed. This allows for the combination of parametric modeling of sample 3D models, automatically generating large-scale and high-quality sample 3D models, thereby improving the training effect of the process feature recognition model and ultimately improving the accuracy of process feature recognition. Furthermore, it can build its own 3D model dataset, eliminating the need for manual extraction and conversion of design data, thus improving efficiency and reducing costs.

[0131] Based on any of the above embodiments, after step 330, the method further includes steps 340 to 390.

[0132] Step 340: Display the feature names of each sample process feature in the sample 3D model, and display the sample 3D model.

[0133] For example, such as Figure 4 As shown, a custom plugin is used for labeling. At this time, the interface displays "Set Feature Color", which shows the feature name corresponding to the feature type of each sample process feature, and displays the sample 3D model on the right.

[0134] Step 350: Determine the name of the first target feature selected by the user based on the first feature selection instruction.

[0135] Here, the first feature selection instruction is triggered by the user in the feature name display area; that is, the user selects the first target feature name from the feature library (including various feature names). The first target feature name creates a feature for the target selected by the user. For example, such as... Figure 4 As shown, the user selects "flare," meaning the first target feature name is "flare."

[0136] Step 360: Based on the feature type-color mapping relationship, determine the annotation color corresponding to the first target feature name.

[0137] The feature type-color mapping relationship is used to indicate the one-to-one correspondence between the feature type corresponding to each feature name and the corresponding color. It should be understood that the feature type-color mapping relationship stores the standard color corresponding to each feature type in advance.

[0138] It should be understood that after the user selects the first target feature name, the standard color (label color) of the first target feature name should be displayed immediately as visual feedback to improve the user experience.

[0139] In one embodiment, a CSV configuration file is created in advance, defining the preset parameter ranges (parameter constraints) for each preset process feature, as well as the feature type-color mapping relationship (i.e., the mapping relationship between feature semantics and color). Then, a three-tier architecture system is used to dynamically load and parse the feature configuration: the data access layer reads the CSV configuration file and converts it into a two-dimensional structure table; the business logic layer converts the data into structured objects and establishes a bidirectional mapping between feature IDs (feature types) and colors (specifically color values); the user interaction layer dynamically loads these parameters and presets the corresponding colors (such as annotation colors and background colors). This adopts an architecture design that separates layered configuration from business logic.

[0140] Step 370: Render the annotation color on each sample constituent surface of the sample process feature corresponding to the first target feature name in the sample 3D model, and render a color different from the annotation color on the sample bottom surface of the sample process feature corresponding to the first target feature name in the sample 3D model.

[0141] Here, the color different from the labeled color is the dedicated background color. In one embodiment, this dedicated background color can be a preset background color, such as red.

[0142] For example, such as Figure 4 As shown, the label color corresponding to the straight tube is yellow, so the sample surfaces of the straight tube are rendered in yellow; the label color corresponding to the curved tube is green, so the sample surfaces of the curved tube are rendered in green.

[0143] In one specific embodiment, such as Figure 4As shown, the feature faces (component faces) and their corresponding selectors are also displayed, allowing users to select all faces (component faces) of a feature using the main selector, or select the bottom face (bottom face) of a feature using the auxiliary selector. When all faces of a feature are selected, the annotation color is rendered on each sample component face of the sample process feature corresponding to the first target feature name in the sample 3D model, i.e., the preset feature color is automatically applied. When the bottom face of a feature is selected, a color different from the annotation color is rendered on the sample bottom face of the sample process feature corresponding to the first target feature name in the sample 3D model, i.e., the special bottom face color is automatically applied.

[0144] Step 380: After obtaining the color update instruction, update the color of the sample constituent surface indicated by the color update instruction to obtain the color of each sample constituent surface included in the sample process feature corresponding to the first target feature name.

[0145] Here, the color update command is triggered by the user on the interface. In one embodiment, the color update command is used not only to indicate the sample constituent surface that needs to be updated, but also to indicate the updated color.

[0146] For example, such as Figure 4 As shown, the colors of the feature faces (constituent faces) and the feature base faces are also displayed, allowing users to select the sample constituent faces that need updating and the updated colors (represented by color values). For example, Figure 4 The selection is made of the constituent faces and the bottom face of the flared opening. The color of each constituent face is changed to the color corresponding to color value 37, and the color of the bottom face is changed to the color corresponding to color value 48.

[0147] It should be understood that although the above-mentioned annotation colors have been rendered on the sample constituent surfaces of the sample process features corresponding to the first target feature name in the sample 3D model, and the sample bottom surface of the sample process features corresponding to the first target feature name in the sample 3D model has been rendered with a color different from the annotation color, the annotation color may be incorrect. Therefore, the user can choose whether to update the color, thereby improving the accuracy of the label annotation.

[0148] In one specific embodiment, the colors of each sample constituent surface, including the sample process features corresponding to the first target feature name, are saved together with the sample 3D model (.prt file).

[0149] Step 390: Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, determine the recognition result label corresponding to the three-dimensional model of the sample.

[0150] It should be understood that by repeating steps 350 to 380 above, the colors of the surfaces that constitute each sample in the three-dimensional model of the sample can be obtained.

[0151] It should be noted that some samples have both a bottom and a top surface, so their colors include the labeled colors as well as the bottom surface colors.

[0152] Specifically, based on the feature type-color mapping relationship, the feature type corresponding to the labeled color of each sample constituent surface is determined, and the detection result labels of which surfaces in each sample constituent surface are bottom surfaces are determined. Then, based on each feature type, the semantic segmentation result sub-labels of each sample constituent surface are determined. Based on the semantic segmentation result sub-labels of each sample constituent surface, the semantic segmentation result labels of the sample 3D model (including the semantic segmentation result sub-labels of each sample constituent surface) are determined, and the bottom surface detection result labels are determined based on the detection result labels of which surfaces in each sample constituent surface are bottom surfaces. The semantic segmentation result sub-label of any sample constituent surface is used to indicate the feature type of that sample constituent surface. The semantic segmentation result label is used to indicate the feature type of each sample constituent surface in the sample 3D model. The bottom surface detection result label includes the detection result labels of each sample constituent surface in the sample 3D model. The detection result label of any sample constituent surface is used to indicate whether the sample constituent surface is the bottom surface used to calculate the geometric parameters of the sample process features.

[0153] In one specific embodiment, the recognition result label corresponding to the sample 3D model is determined based on the instance segmentation result label of the sample 3D model and the color of each sample constituent face in the sample 3D model. The instance segmentation result label is used to indicate the constituent faces included by each sample process feature in the sample 3D model. For example, the recognition result label includes the aforementioned semantic segmentation result label, instance segmentation result label, and bottom surface detection result label.

[0154] For example, the identification result labels can be represented using JSON (JavaScript Object Notation) data, as shown below:

[0155] "{"seg":{"0":5,"1":1,"2":1,"3":1,"4":2,"5":2,"6":1,"7":2,"8":2,"9":1,"10":1," 11":2,"12":2,"13":1,"14":1,"15":2,"16":2,"17":1,"18":1,"19":5,"20":5,"21":5," 22":0,"23":0,"24":0,"25":4,"26":4,"27":4,"28":4,"29":0,"30":3,"31":3,"32":3," 33":0,"34":3,"35":3,"36":3,"37":0,"38":3,"39":3,"40":3,"41":0,"42":0,"43":0},

[0156] "inst":[[1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,1,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,1,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]],

[0157] "bottom":{"0":0,"1":1,"2":1,"3":0,"4":1,"5":0,"6":0,"7":1,"8":0,"9":1,"10":0, "11":1,"12":0,"13":1,"14":0,"15":1,"16":0,"17":0,"18":1,"19":0,"20":0,"21":1," 22":1,"23":1,"24":1,"25":0."26":0,"27":1,"28":0,"29":1,"30":0,"31":1,"32":0," 33":1,"34":0,"35":1,"36":0,"37':1,"38":0,"39":1,"40":0,"41":1,"42":1,"43":1}}"

[0158] This JSON data is structured and labeled. The JSON data adopts a three-layer architecture. The seg field records the mapping from the surface index to the feature ID (feature type), which is the semantic segmentation result label. Among them, feature ID 0 represents a blank, feature ID 1 represents a straight pipe, feature ID 2 represents a bent pipe, feature ID 3 represents a flanged hole, feature ID 4 represents a flared end, and feature ID 5 represents a narrowed end. The inst field stores an N×N relationship matrix, which groups the surfaces belonging to the same processing feature instance together. inst[i][j]=1 indicates that surface i and surface j belong to the same processing feature instance (process feature). The 0th row (i=0) is the connection relationship of surface 0, and the 1st row (i=1) is the connection relationship of surface 1. The bottom field identifies the processing contact surface (bottom surface), providing complete semantic information for processing feature recognition and parameter extraction tasks. Among them, 1 indicates that the corresponding feature surface (constituent surface) is the bottom surface, and 0 indicates that the corresponding feature surface (constituent surface) is not the bottom surface. Based on this, the multimodal data organization format, a pairing design of a three-layer JSON structure (seg / inst / bottom) and standardized STEP files, not only preserves geometric accuracy but also provides rich semantic information, offering a brand-new data paradigm for manufacturing feature recognition.

[0159] It should be understood that embodiments of the present invention require labeling each sample's process characteristics, such as... Figure 4 As shown, process features of the same type are marked with the same color, and their bottom surfaces are marked with different colors to facilitate subsequent geometric parameter calculations.

[0160] The process feature recognition method provided in this invention displays the feature names of each sample process feature in the sample 3D model, allowing users to select the target feature to be labeled, thereby improving the user experience. It also displays the sample 3D model for user viewing, further enhancing the user experience. Based on the feature type-color mapping relationship, it determines the labeling color corresponding to the first target feature name. After the user selects the first target feature name, the standard color of the first target feature name is immediately displayed as visual feedback, improving the user experience and allowing the user to easily check for color errors. If an error is found, the user can trigger a color update command to update the color of the sample constituent surface indicated by the color update command, thereby improving the accuracy of labeling and thus improving the training effect of the process feature recognition model, ultimately improving the accuracy of process feature recognition.

[0161] Based on any of the above embodiments, after step 350, the method further includes:

[0162] After receiving the face selection instruction, the sample process features corresponding to the first target feature name are determined based on the face selection instruction, including each sample constituting face.

[0163] Here, the face selection instruction is triggered by the user on the interface. In one embodiment, the face selection instruction is used to indicate that the selected sample constitutes a face.

[0164] For example, such as Figure 4 As shown, the user selects the flare, that is, the first target feature name is flare, and the user selects all the faces of the feature (the faces of each sample) through the main selector. That is, the number of the faces of each sample included in the flare is 4. For example, the 4 selected face of each sample are two conical surfaces and two cylindrical surfaces, and the bottom surface is the inner cylindrical surface.

[0165] Accordingly, step 390 above includes:

[0166] Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, and the constituent surfaces of each sample included in the process features of each sample in the three-dimensional model of the sample, the identification result label corresponding to the three-dimensional model of the sample is determined.

[0167] It should be understood that by repeating the above steps, the constituent surfaces of each sample's process feature in the sample 3D model can be obtained. Then, based on the constituent surfaces of each sample's process feature in the sample 3D model, the instance segmentation result label of the sample 3D model can be determined. This instance segmentation result label is used to indicate the constituent surfaces of each sample's process feature in the sample 3D model.

[0168] Specifically, based on the color of each sample's constituent face in the sample 3D model, the semantic segmentation result label of the sample 3D model can be determined, as well as the bottom surface detection result label of the sample 3D model. Thus, the identification result labels include the aforementioned semantic segmentation result label, instance segmentation result label, and bottom surface detection result label.

[0169] The process feature recognition method provided in this embodiment of the invention displays the feature names of each sample process feature in the sample 3D model in the above manner, and displays the sample 3D model; determines the first target feature name selected by the user based on the first feature selection instruction, and determines the sample constituent faces included in the sample process feature corresponding to the first target feature name based on the face selection instruction, so that the user can select the sample constituent faces belonging to the same sample process feature, thereby accurately determining the instance segmentation result of the sample 3D model, and solving the instance segmentation problem that cannot distinguish between similar adjacent sample process features (such as two adjacent holes) by color alone, ultimately improving the training effect of the process feature recognition model and improving the recognition accuracy of process features.

[0170] Based on any of the above embodiments, after obtaining the surface selection instruction and determining the sample constituent surfaces included in the sample process feature corresponding to the first target feature name based on the surface selection instruction, the method further includes:

[0171] Upon receiving the annotation application instruction, construct the unique identifier information of the sample process feature corresponding to the first target feature name;

[0172] The unique identifier information of the sample process feature corresponding to the first target feature name is bound to each sample constituent surface included in the sample process feature corresponding to the first target feature name.

[0173] Here, the annotation application command is triggered by the user on the interface, for example, such as Figure 4 As shown, the user clicks on the application.

[0174] For example, such as Figure 4 As shown, after the user selects the flare, that is, the first target feature name is flare, and after the user marks the color and selects the sample constituent face and bottom face included in the flare, the user can click Apply to construct the unique identification information of the flare.

[0175] Here, the unique identifier of any sample process feature is used to uniquely identify that sample process feature. Furthermore, since each sample process feature corresponds to an instance segmentation result label, the unique identifier of any sample process feature can also be indirectly used to uniquely identify the instance segmentation result of that sample process feature; that is, the unique identifier of a sample process feature can also identify the instance segmentation result label of that sample process feature. In other words, all faces in the same feature group are bound to the same unique identifier. This unique identifier can be a UUID (Universally Unique Identifier).

[0176] In one embodiment, unique identification information is constructed based on a real-time generated timestamp, thereby determining high-precision unique identification information based on a high-precision timestamp to ensure uniqueness.

[0177] It should be understood that the unique identification information of the sample process feature corresponding to the first target feature name is bound to each sample constituent surface included in the sample process feature corresponding to the first target feature name, that is, the unique identification information is assigned to each sample constituent surface included in the sample process feature.

[0178] In one specific embodiment, the unique identifier information of the sample process feature corresponding to the first target feature name is saved along with the sample 3D model (.prt file). Based on this, a robust association is established between the sample 3D model and the recognition result label, so that subsequent labeling does not need to be repeated.

[0179] Accordingly, after determining the recognition result label corresponding to the three-dimensional model of the sample based on the color of each sample constituent surface in the sample three-dimensional model and the sample constituent surfaces included in each sample process feature in the sample three-dimensional model, the method further includes:

[0180] Based on the preset parameter ranges corresponding to the process features of each sample in the three-dimensional model of the sample, the geometric parameters of the process features of each sample in the three-dimensional model of the sample are updated to obtain a new three-dimensional model of the sample.

[0181] Based on the unique identifier information of each sample process feature in the new sample 3D model, and the color of each sample constituent surface in the new sample 3D model, the identification result label of the new sample 3D model is determined.

[0182] This preset parameter range is used to limit the geometric parameters of process features, i.e., geometric parameter constraints. Therefore, the updated geometric parameters are within the preset parameter range.

[0183] For example, after completing the above labeling, a batch dataset generation process is initiated via a command-line tool. Specifically, the system reads the labeled .prt files in batches, parses the B-rep data (data representing the 3D models of the samples), fully preserves the geometric topological accuracy, and automatically executes a loop of model cloning, parameter randomization, and geometric reconstruction, combined with a predefined range of expression parameters (preset parameter range), thereby obtaining new 3D models of the samples. Based on this, while maintaining the semantic correlation of features, a large number of geometric variants can be generated efficiently, significantly improving the diversity and engineering representativeness of the dataset, i.e., achieving data augmentation.

[0184] Specifically, when labeling a new sample 3D model, the unique identifier information of each sample process feature in the new sample 3D model can be viewed. Based on this unique identifier information, the instance segmentation result sub-labels corresponding to each sample process feature in the new sample 3D model can be determined (the instance segmentation result sub-label of any sample process feature is used to indicate the constituent surfaces of that sample process feature). Then, based on these instance segmentation result sub-labels, the instance segmentation result labels of the new sample 3D model (including the instance segmentation result sub-labels of each sample process feature) are determined. Furthermore, based on the color of each constituent surface in the new sample 3D model and the feature type-color mapping relationship, the feature type of each constituent surface in the new sample 3D model is determined, thereby determining the semantic segmentation result label of the new sample 3D model. Based on these labels, the recognition result label of the new sample 3D model is determined. In other words, the recognition result label corresponding to the sample 3D model is determined as the recognition result label of the new sample 3D model.

[0185] It should be noted that since each sample process feature in the sample 3D model is identified by a unique identifier, and the construction of a new sample 3D model does not change the sample process features, only the geometric parameters are changed, each sample process feature in the new sample 3D model constructed based on the sample 3D model is also identified by the same unique identifier.

[0186] For example, within the range of an expression (which is customizable and requires no damage to the model), a parametric modeling expression is randomly driven to generate multiple variants. For instance, if the first straight pipe segment changes from 85mm to 100mm, this generates a new 3D model of the sample. Simultaneously, a STEP file (geometric model) and a JSON label file (recognition result labels) are exported for each variant (new sample 3D model). Based on this, during the geometric parameter perturbation process, only the geometric parameters change; the categories and number of process features included in the model remain unchanged. Therefore, unique identifiers maintain the topological relationships of feature groups, ensuring that geometric changes do not disrupt established label annotations (such as semantic annotations, instance annotations, and bottom surface annotations). This allows for data augmentation without the need for repeated label annotation, thereby improving the efficiency of training dataset construction and reducing its cost.

[0187] It should be understood that this invention provides a method for efficiently generating high-quality labeled data (new sample 3D models and their corresponding recognition result labels), and constructs a complete automatic generation system for training datasets. Its core lies in building a data pipeline that combines configuration-driven, interactive annotation, and batch generation. Through a color-coding and unique identifier binding mechanism, it achieves accurate semantic mapping and large-scale annotation of manufacturing features. In other words, by employing an expression-driven parametric modeling method, it can automatically generate large-scale, high-quality labeled datasets based on basic 3D models (sample 3D models). For example, it has achieved the automatic construction of datasets in two major industrial fields: piping and sheet metal, generating over 30,000 precisely annotated sample 3D models respectively. The piping dataset covers 12 categories of standard manufacturing features, and the sheet metal dataset contains 16 categories of standard manufacturing features, comprehensively covering the typical processing elements of related manufactured products. In particular, by identifying the unique identifiers of each sample's process features in the sample 3D models, it ensures the accuracy and consistency of data throughout the entire process from R&D to manufacturing, providing reliable data support for intelligent manufacturing systems.

[0188] To facilitate understanding of the above embodiments, a specific embodiment is described here. This invention provides a method for constructing a training dataset that deeply integrates interactive design and automated processing technologies, offering a method for constructing a sample 3D model dataset for process feature recognition. Specifically, the configuration-driven architecture employs a modular architecture driven by configuration, decoupling data definition and processing logic through independent feature configuration files. This supports dynamic expansion and flexible adjustment of the manufacturing feature library (preset process feature set), effectively solving the technical bottleneck of fixed feature types and difficulty in updating in traditional systems. The feature labeling method uses a color-coding and unique identifier binding mechanism, combining visual color identifiers with logically unique identifiers, providing intuitive visual feedback while ensuring accurate maintenance of the feature group topology. Parametric batch generation automatically generates geometric variants through parameter-driven expressions, achieving large-scale expansion of the dataset while maintaining feature semantics. This one-time labeling, multiple-generation mode significantly improves data construction efficiency.

[0189] The process feature recognition method provided in this invention updates the geometric parameters of each process feature in the sample 3D model based on the preset parameter range corresponding to each sample process feature, thus obtaining a new sample 3D model. This allows for data augmentation on the constructed sample 3D model without repeating the entire sample 3D model construction process, thereby improving the efficiency of training dataset construction and reducing construction costs. Furthermore, the sample 3D model can be a real 3D model, and the model obtained by data augmentation on it can also be considered a real 3D model, thus improving model training performance and increasing the accuracy of process feature recognition. Based on the unique identifier information of each sample process feature in the new sample 3D model and the color of each sample's constituent surface in the new sample 3D model, the recognition result label of the new sample 3D model is determined. Therefore, by only updating the geometric parameters of each sample process feature in the sample 3D model, repeated labeling is unnecessary, further improving the efficiency of training dataset construction and reducing training dataset construction costs.

[0190] Based on any of the above embodiments, after step 330, the method further includes:

[0191] If the topological rationality verification of the sample 3D model is passed, the sample 3D model is added to the training dataset.

[0192] Here, topological rationality verification is used to verify whether the topological relationships of the sample 3D model are reasonable. In one embodiment, topological rationality verification can ensure that the model is valid, closed, manifold, and has unique common edges. A valid model means that the model does not have defects such as self-intersection, gaps, or geometric degradation; a closed model means that there are no free edges or missing faces; a manifold model means that there are no non-manifold shells or faces shared by two or more shells; and unique common edges mean that each common edge strictly belongs to a single wire loop.

[0193] Furthermore, if the topological rationality verification of the sample 3D model fails, the sample 3D model will not be added to the training dataset; that is, the sample 3D model will not be used as a training sample.

[0194] like Figure 5 and Figure 6 As shown, Figure 5 The sample 3D model that has passed the topological rationality verification. Figure 6 For sample 3D models that fail topological rationality verification (damaged models), such as Figure 6 As shown in the red box, there is an incorrect topology within the red box.

[0195] The process feature recognition method provided in this embodiment of the invention can ensure the accuracy of the three-dimensional model of the sample used for training by means of the above method, thereby improving the model training effect and thus improving the recognition accuracy of process features.

[0196] Based on any of the above embodiments, in this method, the training dataset is optimized based on at least one of the following methods:

[0197] If the number of sample 3D models in the training dataset is different from the number of identification result labels in the training dataset, delete the first abnormal sample 3D model and / or the first abnormal identification result label in the training dataset; the first abnormal sample 3D model is a sample 3D model without a corresponding identification result label, and the first abnormal identification result label is an identification result label without a corresponding sample 3D model.

[0198] If a second abnormal sample 3D model and a second abnormal identification result label exist in the training dataset, delete the second abnormal sample 3D model and the second abnormal identification result label from the training dataset; the exported file name of the second abnormal sample 3D model is different from the exported file name of the second abnormal identification result label;

[0199] If a 3D model of a third anomalous sample and a label of a third anomalous identification result exist in the training dataset, delete the 3D model of the third anomalous sample and the label of the third anomalous identification result from the training dataset; the number of constituent faces of the 3D model of the third anomalous sample is different from the number of constituent faces indicated by the label of the third anomalous identification result.

[0200] In one specific embodiment, it is verified whether the number of exported files of the sample 3D model in the training dataset is the same as (matches) the number of exported files of the recognition result labels in the training dataset, that is, it is necessary to ensure that the two numbers are consistent.

[0201] Here, the second anomaly identification result label is the identification result label corresponding to the 3D model of the second anomaly sample. In a specific embodiment, it is verified whether the exported file name of the sample 3D model in the training dataset is the same as (consistent with) the exported file name of the identification result label corresponding to the sample 3D model. That is, it is necessary to ensure that the exported file names of the two are consistent. If they are inconsistent, it means that there is no corresponding sample 3D model or identification result label.

[0202] Here, the third anomaly identification result label is the identification result label corresponding to the 3D model of the third anomaly sample. In a specific embodiment, it is verified whether the number of constituent faces indicated by the 3D model of the sample in the training dataset is the same as the number of constituent faces indicated by the identification result label corresponding to the 3D model of the sample; that is, it is necessary to ensure that the number of constituent faces indicated by the sample and its label is consistent.

[0203] It should be understood that the embodiments of the present invention establish a quality verification system and construct a full-process quality verification system. Through an automatic verification and assurance mechanism, the quality of each link from labeling to data generation is controllable.

[0204] The process feature recognition method provided in this embodiment of the invention optimizes the training dataset in the above manner. If any defect is detected, the sample 3D model or recognition result label in the training dataset can be discarded, thereby improving the data quality of the training dataset, thereby improving the model training effect, and ultimately improving the recognition accuracy of process features.

[0205] Based on any of the above embodiments, in this method, step 110 includes steps 111 to 113.

[0206] Step 111: Convert the 3D model into a face adjacency graph.

[0207] In this context, the nodes of the face adjacency graph represent constituting faces, and the edges of the face adjacency graph represent the connecting edges between two constituting faces. That is, the model's topological information (3D model) is transformed into a face adjacency graph (FAG) as the "skeleton" of the model. The constituting faces (feature faces) of the 3D model serve as nodes in the FAG, and the edges of the 3D model represent the connections between adjacent nodes (constituting faces).

[0208] Step 112: Construct an attribute adjacency graph based on the geometric information of the 3D model, the attribute information of the 3D model, and the face adjacency graph.

[0209] The geometric information includes face geometric information and edge geometric information, and the attribute information includes face attribute information and edge attribute information. The face geometric information is used to be attached to the nodes of the face adjacency graph, the edge geometric information is used to be attached to the edges of the face adjacency graph, the face attribute information is used to be attached to the nodes of the face adjacency graph, and the edge attribute information is used to be attached to the edges of the face adjacency graph.

[0210] Here, surface geometry information may include, but is not limited to: 3D coordinates and surface normal vectors, etc. Edge geometry information may include, but is not limited to: 3D coordinates and tangent vectors, etc. Surface attribute information may include, but is not limited to: surface type, area, centroid, and number of boundary loops, etc. Edge attribute information may include, but is not limited to: curve type, length, convexity, and dihedral angle between adjacent surfaces, etc.

[0211] In one specific embodiment, face geometry information is extracted from the face mesh obtained through geometric processing using UV-Net (Unified Vertex-Normal Network), and face attribute information is extracted from the face attributes. Similarly, edge geometry information is extracted from the edge mesh obtained through geometric processing, and edge attribute information is extracted from the edge attributes. This UV-Net is a deep learning network for 3D shape analysis and processing that enhances geometric feature representation by uniformly encoding vertex and normal information.

[0212] It should be understood that further transforming the face adjacency graph into an attribute adjacency graph (AAG) allows the graph structure to contain both topological connections (through the face adjacency graph) and rich geometric and attribute features, which facilitates subsequent graph neural network processing, i.e., retains more original model information, thereby improving the accuracy of process feature recognition.

[0213] Step 113: Input the attribute adjacency graph into the process feature recognition model to obtain the recognition result output by the process feature recognition model.

[0214] It should be understood that, since the three-dimensional model is transformed into an attribute adjacency graph as described above, the process feature recognition model can treat each input three-dimensional model as a graph structure, thereby improving the accuracy of process feature recognition.

[0215] In one embodiment, the process feature recognition model can extract local and global features from the attribute adjacency graph, and can fuse the local and global features to obtain the recognition result based on the fused features. Specifically, local and global features are extracted by combining graph neural networks (GNNs) and a transformer layer.

[0216] For example, the process feature recognition model includes a GNN layer, a layer normalization layer, a Transformer layer, an average pooling layer, a multi-layer perceptron layer, a feature concatenation layer, and a multi-task layer (including a semantic segmentation head, an instance segmentation head, and a bottom surface detection head connected in sequence). The GNN layer is used to perform edge information message passing and node message passing in sequence; the layer normalization layer is used to perform layer normalization processing; the Transformer layer is used to perform batch processing, feature filling and feature masking, Transformer encoding and feature inverse filling in sequence; the average pooling layer is used to perform average pooling; the multi-layer perceptron layer is used to perform projection to obtain local features and global features; the feature concatenation layer is used to concatenate local features and global features; and the semantic segmentation head is used to perform semantic segmentation. The semantic segmentation head classifies each node (constituent surface) into one of the predefined feature types, such as venting, flaring, and square hole. This semantic segmentation result indicates the feature type of each constituent surface in the 3D model. The instance segmentation head performs instance segmentation to obtain instance segmentation results, grouping nodes (surfaces) belonging to the same processing instance, thus determining which faces constitute the process feature. This instance segmentation result indicates the constituent surfaces included in each process feature in the 3D model. The bottom surface detection head performs bottom surface detection to obtain bottom surface detection results, which include the detection results of each constituent surface in the 3D model. This bottom surface detection head is a binary classification head used to determine whether a feature surface is used for subsequent geometric parameter calculations. Furthermore, the number of GNN layers can be multiple, i.e., multiple GNN layers are connected sequentially, for example, stacking three GNN layers. Based on this, this embodiment of the invention efficiently processes boundary representation (B-rep) data through unified geometric sampling, attribute embedding, and hierarchical feature fusion.

[0217] The process feature recognition method provided in this invention transforms a 3D model into an attribute adjacency graph. Therefore, the process feature recognition model can treat each input 3D model as a graph structure for processing, thereby improving the accuracy of process feature recognition. Furthermore, the face adjacency graph is further transformed into an attribute adjacency graph, so that the graph structure contains both topological connections and rich geometric and attribute features, which facilitates subsequent graph neural network processing, i.e., retains more original model information, and further improves the accuracy of process feature recognition.

[0218] Based on any of the above embodiments, in this method, when at least one bottom surface of the process feature is located or spans the bending region of the three-dimensional model, the geometric parameters of the process feature are calculated based on the following method:

[0219] The process features are projected to obtain a two-dimensional contour graphic;

[0220] Based on the two-dimensional contour graphic, the geometric parameters of the process feature are determined.

[0221] It should be noted that, considering many process features are often located near or across bends—meaning at least one bottom surface of a process feature lies within or crosses a bend in the 3D model—the geometric parameters calculated using conventional algorithms are inaccurate. Therefore, the process features are first projected to obtain a 2D contour graphic, and then the geometric parameters of the process features are determined based on this 2D contour graphic. For example, on sheet metal parts, many features (such as ribs, square holes near bend edges, etc.) are often located near bends or cross bend areas (e.g.,... Figure 7 As shown in the figure, this leads to inaccurate calculation of feature dimensions (such as the length and width of the hole).

[0222] In one specific embodiment, the geometric parameters are calculated using a "normal projection-bounding box decomposition" algorithm. This algorithm consists of two steps. Step one: normal projection. First, the bent surfaces (such as cylindrical surfaces) associated with the process feature and the two adjacent planar regions are identified. Then, the projection direction is determined, typically perpendicular to the bend line. The projection operation projects the entire three-dimensional geometry of the feature onto a two-dimensional plane to obtain a two-dimensional contour graphic. Step Two: Bounding Box Decomposition. First, analyze the 2D contour graphic obtained after projection and identify the decomposition points. The key point is to identify the points in the projected contour that correspond to the intersection of the curved surface and the planar region in the actual 3D geometry. On the projection map, these points are usually the endpoints of the "straight line segments" of the curved area after projection. Then, using these identified decomposition points, the entire projected contour is divided into several continuous line segments / sub-contours. The parts located on both sides of the projected line segments in the curved area correspond to the feature boundary projections on two adjacent planar regions in the original 3D model. Calculate the 2D axial bounding box for each sub-contour obtained after segmentation. For the sub-contour representing the planar region, calculating its bounding box can obtain the true length and width of the feature boundary on that planar region. The length of the line segment representing the projection of the curved area or its relationship with the planar bounding box can be used to infer information such as the bending angle.

[0223] For example, the above method can effectively separate the dimensional contributions of the bending area and the planar area in sheet metal features.

[0224] The process feature identification method provided in this embodiment of the invention projects the process feature to obtain a two-dimensional contour graphic when at least one bottom surface of the process feature is located in or spans the bending area of ​​a three-dimensional model. Based on the two-dimensional contour graphic, the geometric parameters of the process feature are accurately determined, thereby improving the accuracy of geometric parameter calculation.

[0225] Based on any of the above embodiments, after step 110, the method further includes:

[0226] Based on the feature type of each process feature, the parameter calculation tool corresponding to each process feature is scheduled.

[0227] The parameter calculation tool corresponding to any of the aforementioned process features is used to calculate the geometric parameters of the process feature. Different feature types correspond to different parameter calculation tools; that is, one parameter calculation tool corresponds to one feature type. In one embodiment, a built-in dynamic feature type binding mechanism supports flexible expansion of parameter calculation tools for new feature types, and also allows for updating and deleting parameter calculation tools.

[0228] In one specific embodiment, a FeatureClassFactory module is designed to integrate and dispatch computational logic. This module maintains a dynamic mapping table between feature types and corresponding parameter calculation tools. During the parameter calculation phase, the identified feature types are used for scheduling, and the FeatureClassFactory instantiates the corresponding parameter calculation tool (such as a calculator). This ensures the system's high modularity and scalability; when adding new process features, only a new calculation class needs to be registered, without modifying the core framework.

[0229] The process feature recognition method provided in this invention schedules the parameter calculation tools corresponding to each process feature based on the feature type of each process feature, thereby determining the corresponding parameter calculation tools for different feature types to more accurately calculate the geometric parameters of the process features, and thus improving the accuracy of geometric parameter calculation.

[0230] Based on any of the above embodiments, in this method, the recognition result includes semantic segmentation result, instance segmentation result, and bottom surface detection result; the semantic segmentation result is used to indicate the feature type of each constituent surface in the three-dimensional model, the instance segmentation result is used to indicate each constituent surface included by each process feature in the three-dimensional model, and the bottom surface detection result includes the detection result of each constituent surface in the three-dimensional model.

[0231] Based on this, the above three tasks can accurately identify and locate process features and their related constituent surfaces, providing crucial support for downstream operations and thus improving the accuracy of process feature identification. Furthermore, this invention proposes an end-to-end solution that integrates process feature identification (semantic segmentation, instance segmentation) and key geometric attribute (bottom surface) detection within a unified deep learning model. It directly outputs structured feature instances (identification results) suitable for geometric parameter calculation, overcoming the bottleneck of separating process feature identification from geometric parameter calculation in traditional solutions. This improves the computational efficiency of geometric parameters and reduces their computational cost.

[0232] Accordingly, the feature types of each process feature in the three-dimensional model are determined based on the following method:

[0233] Based on the semantic segmentation results and the instance segmentation results, the feature types of each process feature in the three-dimensional model are determined.

[0234] Since semantic segmentation results are used to indicate the feature types of each constituent surface in the 3D model, and instance segmentation results are used to indicate the constituent surfaces included by each process feature in the 3D model, the feature types of each process feature in the 3D model can be determined based on the two.

[0235] The process feature recognition method provided in this invention, through the above three tasks, can accurately identify and locate process features and their related constituent surfaces, thereby improving the accuracy of process feature recognition; and completes process feature recognition (semantic segmentation, instance segmentation) and key geometric attribute (bottom surface) detection in a unified deep learning model, directly outputting structured feature instances (recognition results) that can be used for geometric parameter calculation, breaking through the process bottleneck of separating process feature recognition and geometric parameter calculation in traditional solutions, thereby improving the calculation efficiency of geometric parameters and reducing the calculation cost of geometric parameters.

[0236] Based on any of the above embodiments, after obtaining the geometric parameters of each process feature in the three-dimensional model, the method further includes:

[0237] Display the feature names of each process feature in the 3D model, and display the 3D model;

[0238] The name of the second target feature selected by the user is determined based on the second feature selection instruction;

[0239] Display the geometric parameters of the process feature corresponding to the second target feature name, and render the process feature corresponding to the second target feature name on the three-dimensional model using a preset rendering method; the preset rendering method is used to distinguish the process feature corresponding to the second target feature name from other process features in the three-dimensional model besides the process feature corresponding to the second target feature name.

[0240] Here, the second feature selection instruction is triggered by the user in the feature name display area, that is, the user selects the second target feature name from the feature list (including each feature name), for example, by clicking on the feature name. The second target feature name creates a feature for the target selected by the user.

[0241] Here, the preset rendering method can be set according to actual needs, such as red highlighting.

[0242] For example, such as Figure 8 As shown, this 3D model is a piping model. The interface displays a "feature list," which includes the feature names of all process features in the 3D model, and the 3D model is displayed on the right. If the user selects a bend, i.e., the second target feature name is "bend," the corresponding geometric information (including geometric parameters) is displayed. This geometric information includes multiple attributes (such as bend radius, bending angle, spatial angle, and coordinates), showing the attribute names and their values. Simultaneously, the corresponding process feature of the bend is highlighted in red on the 3D model when the user selects it.

[0243] For example, such as Figure 9As shown, the 3D model is a sheet metal model. The interface displays a "feature list," which includes the feature names of all process features in the 3D model, and the 3D model is displayed on the right. If the user selects a bend, i.e., the second target feature name is "bend," the corresponding geometric information (including geometric parameters) is displayed. This geometric information includes multiple attributes (such as angle, bend radius, height, and coordinates), showing the attribute names and their values. Simultaneously, the corresponding process feature of the bend is highlighted in red on the 3D model when the user selects it.

[0244] The process feature recognition method provided in this embodiment of the invention displays the feature names of each process feature in a three-dimensional model, and displays the three-dimensional model so that users can select the target feature they want to view, thereby improving the user experience. It also displays the geometric parameters of the process feature corresponding to the second target feature name, and renders the process feature corresponding to the second target feature name on the three-dimensional model through a preset rendering method, thereby improving the user experience and making it easier for users to view the geometric parameters of the corresponding process feature, thus enabling rapid process design.

[0245] Based on the above embodiments, it should be noted that the CAPP system for 3D processes has 3D machining process design tools. Therefore, by identifying the process features of the product and automatically capturing the geometric parameters of the process features, intelligent process decisions can be made.

[0246] Furthermore, it should be noted that for industrial applications such as intelligent process planning and processing cost analysis, obtaining the feature types and their precise geometric parameters is crucial. This invention develops an automated feature post-processing framework based on geometric topology analysis, enabling precise parameter calculation for 16 sheet metal features and 12 pipe features.

[0247] Based on the above embodiments, the system demonstrates strong robustness in handling complex spatial relationships of pipeline features such as multi-section bends, providing a reliable technical path for intelligent process planning.

[0248] Furthermore, after submitting the above identification results and the geometric parameters of each process feature to the CAPP system, a list of process feature parameters is formed, thereby improving the efficiency of process design. Moreover, the automatic calculation of geometric parameters eliminates errors caused by human factors, effectively ensuring the accuracy of the data and thus improving the accuracy of process design.

[0249] Furthermore, by automatically extracting the geometric parameters of process features, the CAPP system enables online and structured management of cost analysis models for sheet metal parts and piping fittings. By acquiring model parameters and manufacturing features of sheet metal parts and pipes, real-time linkage between the product's 3D model design and the cost analysis model is achieved, helping R&D engineers make rapid cost decisions. Combining the cost model, the geometric parameters of the 3D model, and the processing technology, the system automatically calculates the processing costs of parts and manages the cost results through archiving.

[0250] Furthermore, this process feature identification method is integrated into the product system. R&D drawings and related firmware data objects are uniformly managed within the PLM system. The PLM system and CAPP system achieve synchronous transmission of component data through an interface. During synchronization, the system automatically invokes the drawing conversion service and triggers the process feature identification process of the 3D model. The resulting data after the above processing is then transmitted back to the CAPP system for centralized storage and data visualization.

[0251] The process feature recognition device provided by the present invention is described below. The process feature recognition device described below and the process feature recognition method described above can be referred to in correspondence.

[0252] Figure 10 This is a schematic diagram of the process feature recognition device provided in an embodiment of the present invention, as shown below. Figure 10 As shown, the process feature identification device includes:

[0253] The feature recognition module 1101 is used to input a three-dimensional model into a process feature recognition model and obtain the recognition result output by the process feature recognition model. The recognition result includes the feature type of each process feature in the three-dimensional model, the constituent surfaces included in each process feature, and the detection result of each constituent surface in the three-dimensional model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the three-dimensional model.

[0254] The type determination module 1102 is used to determine the geometric type of each bottom surface included in any of the process features based on the identification result;

[0255] The method determination module 1103 is used to determine the parameter calculation method for each of the bottom surfaces based on the geometric type of each bottom surface;

[0256] The parameter calculation module 1104 is used to calculate the geometric parameters of the process features based on the parameter calculation method of each of the bottom surfaces;

[0257] The process feature recognition model is trained based on the sample 3D model and the recognition result labels corresponding to the sample 3D model.

[0258] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11 As shown, the electronic device may include: a processor 1210, a communications interface 1220, a memory 1230, and a communications bus 1240, wherein the processor 1210, the communications interface 1220, and the memory 1230 communicate with each other through the communications bus 1240. The processor 1210 can call logic instructions in the memory 1230 to execute a process feature recognition method. This method includes: inputting a three-dimensional model into a process feature recognition model to obtain recognition results output by the process feature recognition model; the recognition results include the feature type of each process feature in the three-dimensional model, the constituent surfaces included in each process feature, and the detection results of each constituent surface in the three-dimensional model; the detection result of any constituent surface is used to indicate whether the constituent surface is a bottom surface used to calculate the geometric parameters of the process feature, the bottom surface being a geometric surface used to define the dimensions of the three-dimensional model; for any process feature, based on the recognition results, determining the geometric type of each bottom surface included in the process feature; based on the geometric type of each bottom surface, determining the parameter calculation method for each bottom surface; based on the parameter calculation method of each bottom surface, calculating the geometric parameters of the process feature; wherein the process feature recognition model is trained based on a sample three-dimensional model and the recognition result labels corresponding to the sample three-dimensional model.

[0259] Furthermore, the logical instructions in the aforementioned memory 1230 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0260] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the process feature recognition method provided by the above methods. The method includes: inputting a three-dimensional model into a process feature recognition model to obtain a recognition result output by the process feature recognition model; the recognition result includes the feature type of each process feature in the three-dimensional model, and each constituent surface included by each process feature, and the detection result of each constituent surface in the three-dimensional model; the detection result of any constituent surface is used to indicate whether the constituent surface is a bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is a geometric surface used to define the size of the three-dimensional model; for any process feature, based on the recognition result, determining the geometric type of each bottom surface included by the process feature; based on the geometric type of each bottom surface, determining the parameter calculation method of each bottom surface; based on the parameter calculation method of each bottom surface, calculating the geometric parameters of the process feature; wherein, the process feature recognition model is trained based on a sample three-dimensional model and the recognition result label corresponding to the sample three-dimensional model.

[0261] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the process feature recognition method provided by the above methods. The method includes: inputting a three-dimensional model into a process feature recognition model to obtain a recognition result output by the process feature recognition model; the recognition result includes the feature type of each process feature in the three-dimensional model, and each constituent surface included in each process feature, and the detection result of each constituent surface in the three-dimensional model; the detection result of any constituent surface is used to indicate whether the constituent surface is a bottom surface used to calculate the geometric parameters of the process feature, the bottom surface being a geometric surface used to define the size of the three-dimensional model; for any process feature, based on the recognition result, determining the geometric type of each bottom surface included in the process feature; based on the geometric type of each bottom surface, determining the parameter calculation method of each bottom surface; based on the parameter calculation method of each bottom surface, calculating the geometric parameters of the process feature; wherein the process feature recognition model is trained based on a sample three-dimensional model and the recognition result labels corresponding to the sample three-dimensional model.

[0262] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0263] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0264] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0265] The above embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Although the invention has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of the invention do not depart from the spirit and scope of the invention and should be covered within the protection scope of the invention.

Claims

1. A process feature recognition method, characterized by, include: The three-dimensional model is input into the process feature recognition model to obtain the recognition result output by the process feature recognition model. The recognition result includes the feature type of each process feature in the three-dimensional model, the constituent surfaces of each process feature, and the detection result of each constituent surface in the three-dimensional model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the three-dimensional model. For any of the aforementioned process features, based on the identification results, the geometric type of each bottom surface included in the process feature is determined; Based on the geometry of each bottom surface, the parameter calculation method for each bottom surface is determined. When the geometry of the bottom surface is planar, the parallel surfaces of the bottom surface are found by traversing the topological relationship of the 3D model and the normal distance is calculated to obtain the length and width, or the adjacent surfaces are found by finding the shared edge of the bottom surface and the angle between the normal vectors is calculated to obtain the angle. When the geometry of the bottom surface is curved, the radius of the cylindrical surface and the large radius of the torus are directly read through the geometry adapter, and the diameter, length, bending angle and spatial angle are obtained by calculating the span of the surface parameter domain or the geometric relationship of the endpoints. For circular hole features, the cylindrical surface radius method is implemented: locate the cylindrical surface, extract the radius value, and calculate the diameter. For tubular features, the parameter domain difference method is used: obtain the UV domain of the curved surface, calculate the V-direction span, and output the length. The characteristics of the bend are calculated using a composite method: the bending angle is calculated using the endpoint geometry method, and the acute angles of the axes of adjacent bends are also compared. When the geometry of the bottom surface is a mixed type, a dynamic local coordinate system is constructed: a temporary local coordinate system is automatically determined and constructed based on the main direction of the process features; Coordinate transformation and bounding box analysis: Transform all relevant surfaces of the feature to a temporary local coordinate system and calculate the axial bounding box of the bottom surface to extract the length, width, and height dimensions; Multi-method fusion: Combine geometric relationship analysis and direct parameter extraction methods to calculate other geometric parameters; Based on the parameter calculation method of each of the bottom surfaces, the geometric parameters of the process features are calculated; The process feature recognition model is trained based on the sample 3D model and the recognition result labels corresponding to the sample 3D model; When at least one bottom surface of the process feature is located in or spans the bending area of ​​the three-dimensional model, the process feature is first projected to obtain a two-dimensional contour graphic, and then the geometric parameters of the process feature are determined based on the two-dimensional contour graphic.

2. The process feature identification method of claim 1, wherein, The process feature recognition model is trained based on a training dataset, which includes multiple sample 3D models and recognition result labels corresponding to each sample 3D model. The three-dimensional model of any of the samples is constructed based on the following method: Based on the process scenario to be constructed, a preset set of process features is determined; Several preset process features are determined from the preset process feature set, and the geometric parameters of the several preset process features are randomly determined based on the preset parameter ranges corresponding to the several preset process features. Based on the aforementioned preset process features and their geometric parameters, a three-dimensional model of the sample is constructed.

3. The process feature identification method according to claim 2, characterized in that, After constructing the sample 3D model based on the aforementioned preset process features and their geometric parameters, the method further includes: Display the feature names of each sample process feature in the sample 3D model, and display the sample 3D model; The name of the first target feature selected by the user is determined based on the first feature selection instruction; Based on the feature type-color mapping relationship, the label color corresponding to the first target feature name is determined; the feature type-color mapping relationship is used to indicate the one-to-one correspondence between the feature type corresponding to each feature name and each color. The annotation color is rendered on each sample constituent surface of the sample process feature corresponding to the first target feature name in the sample 3D model, and a color different from the annotation color is rendered on the sample bottom surface of the sample process feature corresponding to the first target feature name in the sample 3D model. After receiving the color update instruction, the color of the sample constituent surface indicated by the color update instruction is updated to obtain the color of each sample constituent surface included in the sample process feature corresponding to the first target feature name. Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, the recognition result label corresponding to the three-dimensional model of the sample is determined.

4. The process feature identification method according to claim 3, characterized in that, After determining the name of the first target feature selected by the user based on the first feature selection instruction, the method further includes: After receiving the face selection instruction, the sample process features corresponding to the first target feature name are determined based on the face selection instruction, including each sample constituent face. Accordingly, determining the recognition result label corresponding to the three-dimensional model of the sample based on the color of each sample constituent face in the sample three-dimensional model includes: Based on the colors of the constituent surfaces of each sample in the three-dimensional model of the sample, and the constituent surfaces of each sample included in the process features of each sample in the three-dimensional model of the sample, the identification result label corresponding to the three-dimensional model of the sample is determined.

5. The process feature identification method according to claim 4, characterized in that, After obtaining the surface selection instruction, and determining the sample constituent surfaces included in the sample process feature corresponding to the first target feature name based on the surface selection instruction, the method further includes: Upon receiving the annotation application instruction, construct the unique identifier information of the sample process feature corresponding to the first target feature name; The unique identifier information of the sample process feature corresponding to the first target feature name is bound to each sample constituent surface included in the sample process feature corresponding to the first target feature name. After determining the recognition result label corresponding to the three-dimensional model of the sample based on the color of each sample constituent surface in the sample three-dimensional model and the sample constituent surfaces included in each sample process feature in the sample three-dimensional model, the method further includes: Based on the preset parameter ranges corresponding to the process features of each sample in the three-dimensional model of the sample, the geometric parameters of the process features of each sample in the three-dimensional model of the sample are updated to obtain a new three-dimensional model of the sample. Based on the unique identifier information of each sample process feature in the new sample 3D model, and the color of each sample constituent surface in the new sample 3D model, the identification result label of the new sample 3D model is determined.

6. The process feature identification method according to claim 2, characterized in that, After constructing the sample 3D model based on the aforementioned preset process features and their geometric parameters, the method further includes: If the topological rationality verification of the sample 3D model is passed, the sample 3D model is added to the training dataset; The training dataset is optimized based on at least one of the following methods: If the number of sample 3D models in the training dataset is different from the number of identification result labels in the training dataset, delete the first abnormal sample 3D model and / or the first abnormal identification result label in the training dataset; the first abnormal sample 3D model is a sample 3D model without a corresponding identification result label, and the first abnormal identification result label is an identification result label without a corresponding sample 3D model. If a second abnormal sample 3D model and a second abnormal identification result label exist in the training dataset, delete the second abnormal sample 3D model and the second abnormal identification result label from the training dataset; the exported file name of the second abnormal sample 3D model is different from the exported file name of the second abnormal identification result label; If a 3D model of a third anomalous sample and a label of a third anomalous identification result exist in the training dataset, delete the 3D model of the third anomalous sample and the label of the third anomalous identification result from the training dataset; the number of constituent faces of the 3D model of the third anomalous sample is different from the number of constituent faces indicated by the label of the third anomalous identification result.

7. The process feature identification method according to claim 1, characterized in that, The step of inputting the 3D model into the process feature recognition model and obtaining the recognition result output by the process feature recognition model includes: The 3D model is converted into a face adjacency graph; the nodes of the face adjacency graph represent the constituent faces, and the edges of the face adjacency graph represent the connecting edges between two constituent faces. Based on the geometric information of the 3D model, the attribute information of the 3D model, and the face adjacency graph, an attribute adjacency graph is constructed; the geometric information includes face geometric information and edge geometric information, the attribute information includes face attribute information and edge attribute information, the face geometric information is used to be attached to the nodes of the face adjacency graph, the edge geometric information is used to be attached to the edges of the face adjacency graph, the face attribute information is used to be attached to the nodes of the face adjacency graph, and the edge attribute information is used to be attached to the edges of the face adjacency graph; The attribute adjacency graph is input into the process feature recognition model to obtain the recognition result output by the process feature recognition model.

8. The process feature identification method according to any one of claims 1 to 7, characterized in that, After inputting the 3D model into the process feature recognition model and obtaining the recognition result output by the process feature recognition model, the method further includes: Based on the feature type of each process feature, the parameter calculation tool corresponding to each process feature is scheduled respectively; The parameter calculation tool corresponding to any of the process features is used to calculate the geometric parameters of the process feature.

9. The process feature identification method according to any one of claims 1 to 7, characterized in that, The recognition results include semantic segmentation results, instance segmentation results, and bottom surface detection results; The semantic segmentation result is used to indicate the feature type of each constituent surface in the three-dimensional model, the instance segmentation result is used to indicate the constituent surfaces included by each process feature in the three-dimensional model, and the bottom surface detection result includes the detection results of each constituent surface in the three-dimensional model; The feature types of each process feature in the three-dimensional model are determined based on the following method: Based on the semantic segmentation results and the instance segmentation results, the feature types of each process feature in the three-dimensional model are determined.

10. The process feature identification method according to any one of claims 1 to 7, characterized in that, After obtaining the geometric parameters of each process feature in the three-dimensional model, the method further includes: Display the feature names of each process feature in the 3D model, and display the 3D model; The name of the second target feature selected by the user is determined based on the second feature selection instruction; Display the geometric parameters of the process feature corresponding to the second target feature name, and render the process feature corresponding to the second target feature name on the three-dimensional model using a preset rendering method; the preset rendering method is used to distinguish the process feature corresponding to the second target feature name from other process features in the three-dimensional model besides the process feature corresponding to the second target feature name.

11. A process feature identification device, characterized in that, include: The feature recognition module is used to input a 3D model into a process feature recognition model and obtain the recognition result output by the process feature recognition model. The recognition result includes the feature type of each process feature in the 3D model, the constituent surfaces included in each process feature, and the detection result of each constituent surface in the 3D model. The detection result of any constituent surface is used to indicate whether the constituent surface is the bottom surface used to calculate the geometric parameters of the process feature, and the bottom surface is the geometric surface used to define the size of the 3D model. A type determination module is used to determine the geometric type of each bottom surface included in any of the process features based on the identification result; The method determination module is used to determine the parameter calculation method for each of the bottom surfaces based on their geometric type. When the bottom surface's geometric type is planar, it finds parallel surfaces of the bottom surface by traversing the topological relationships of the 3D model and calculates the normal distance to obtain the length and width, or finds adjacent surfaces through the shared edges of the bottom surface and calculates the angle between the normal vectors to obtain the angle. When the bottom surface's geometric type is curved, it directly reads the radius of the cylindrical surface and the large radius of the torus through the geometric adapter, and obtains the diameter, length, bending angle, and spatial angle by calculating the span of the surface parameter domain or the geometric relationship of the endpoints. For circular hole features, the cylindrical surface radius method is implemented: locating the cylindrical surface, extracting the radius value, and calculating the diameter. For tubular features, the parameter domain difference method is used: obtaining the UV domain of the curved surface, calculating the V-direction span, and outputting the length. The characteristics of the bend are calculated using a composite method: the bending angle is calculated using the endpoint geometry method, and the acute angles of the axes of adjacent bends are also compared. When the geometry of the bottom surface is a mixed type, a dynamic local coordinate system is constructed: a temporary local coordinate system is automatically determined and constructed based on the main direction of the process features; Coordinate transformation and bounding box analysis: Transform all relevant surfaces of the feature to a temporary local coordinate system and calculate the axial bounding box of the bottom surface to extract the length, width, and height dimensions; Multi-method fusion: Combine geometric relationship analysis and direct parameter extraction methods to calculate other geometric parameters; The parameter calculation module is used to calculate the geometric parameters of the process features based on the parameter calculation method of each of the bottom surfaces; The process feature recognition model is trained based on the sample 3D model and the recognition result labels corresponding to the sample 3D model; When at least one bottom surface of the process feature is located in or spans the bending area of ​​the three-dimensional model, the process feature is first projected to obtain a two-dimensional contour graphic, and then the geometric parameters of the process feature are determined based on the two-dimensional contour graphic.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the process feature recognition method as described in any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the process feature identification method as described in any one of claims 1 to 10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the process feature identification method as described in any one of claims 1 to 10.

Images