A bolt identification method, system and electronic device

Abstract

The application provides a bolt recognition method and system and an electronic device, and relates to the technical field of computer-aided engineering. The application determines the connection relationship features between the candidate sub-regions and adjacent virtual components by acquiring the geometric features of the candidate regions in the assembly model, combining the connection topological information of the candidate sub-regions in the assembly model, and inputting the connection relationship features into a multi-modal learning model for bolt recognition, so that efficient and accurate recognition of the bolt is realized, and the fault tolerance and recognition accuracy are higher in complex assembly structures.

Description

Technical Field

[0001] This application relates to the field of computer-aided engineering technology, and in particular to a bolt identification method, system, and electronic device. Background Technology

[0002] In recent years, with the widespread application of structural simulation technology in fields such as automobiles, aerospace, and engineering machinery, the efficiency and accuracy of finite element analysis preprocessing have increasingly become key factors restricting the improvement of simulation efficiency. Among them, bolted connections, as a common and critical structural connection form in complex assemblies, directly affect the reliability of simulation results and the accuracy of engineering decisions due to their modeling quality.

[0003] Currently, bolt modeling typically relies on engineers manually identifying the geometric features of bolts in assembly models and defining connection relationships based on experience. However, when dealing with large assembly models containing numerous bolts, the bolt identification process is not only labor-intensive and inefficient, but also prone to misidentification or omission, further affecting modeling consistency and subsequent simulation accuracy. Furthermore, most existing tools rely on simple geometric rules for bolt identification, making it difficult to accurately distinguish bolts from similar structures, thus limiting identification accuracy.

[0004] Therefore, there is an urgent need for a bolt identification method with high recognition accuracy. Summary of the Invention

[0005] In view of the above problems, embodiments of this application provide a bolt identification method, system, and electronic device to overcome or at least partially solve the above problems.

[0006] In a first aspect, this application provides a method for identifying bolts, the method comprising: Obtain the geometric features of any candidate region in the assembly model; The geometric features of the candidate region are matched with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, and each candidate sub-region corresponds to at least one geometric feature. Obtain the connection topology information of each of the candidate sub-regions in the assembly model; Based on the connection topology information of each candidate sub-region, determine the connection relationship characteristics between each candidate sub-region and its adjacent virtual components; The geometric features and connectivity features of each candidate sub-region are input into a pre-trained multimodal learning model to identify bolts in each candidate sub-region.

[0007] Optionally, matching the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region includes: Obtain the assembly type information corresponding to the assembly model; Based on the assembly type information, and according to a preset dynamic weight allocation mechanism, corresponding weights are assigned to each geometric feature of the candidate region. Based on the weights of each of the geometric features, the similarity score between the candidate region and the standard bolt features in the standardized bolt feature library is calculated. Based on the similarity score, multiple candidate sub-regions that meet the similarity score requirements are selected from the candidate regions.

[0008] Optionally, determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region includes: Obtain the mating hole group structure between the target candidate sub-region and its adjacent virtual components in the assembly model, wherein the target candidate sub-region is any one of the multiple candidate sub-regions; Based on the mating hole group structure formed by the target candidate sub-region and its adjacent virtual components in the assembly model, the hole group logic characteristics between the target candidate sub-region and its adjacent virtual components are determined. Based on the spatial position of the target candidate sub-region in the assembly model, and the relative position between the target candidate sub-region and its adjacent virtual components, the assembly hierarchy features between the target candidate sub-region and its adjacent virtual components are determined. Based on the logical features of the hole group and the assembly hierarchy features, the connection relationship features are obtained.

[0009] Optionally, the multimodal learning model is trained according to the following steps: The initial machine learning model is initially trained using the first training dataset, which includes geometric point cloud data information, material property information, and semantic label information of standard bolt features stored in the standardized bolt feature library. The geometric point cloud data information, material property information, and semantic label information of the standard bolt features are pre-stored in the standardized bolt feature library. The parameters of the initial machine learning model, which has been initially trained, are adjusted using a second training dataset, which includes at least sample data from real-world engineering scenarios.

[0010] Optionally, the training process of the multimodal learning model further includes: For the occluded standard bolt features in the first training dataset, the occluded standard bolt features are reconstructed using a preset occlusion compensation algorithm. The rationality of the standard bolt features after feature reconstruction is verified according to the predefined assembly symmetry rules.

[0011] Optionally, the training process of the multimodal learning model further includes: If there are false positives or false negatives in the bolt identification results, obtain correction data for the false positives or false negatives. The corrected data is used as an incremental training sample and added to the second training dataset; The step of tuning the parameters of the initially trained machine learning model using the second training dataset includes: The parameters of the fully connected layers of the initial machine learning model, which has been initially trained, are updated using the incremental training samples.

[0012] Optionally, the method further includes: Obtain the stress condition information of the identified bolts; Based on the stress condition information, the connection simulation parameters corresponding to the identified bolts are determined according to the preset connection parameter calculation rules. The connection simulation parameters include preload parameters and connection stiffness parameters. The rationality of the connection simulation parameters corresponding to the identified bolts is verified. For bolts whose verification results are unreasonable, parameter correction suggestions are generated by matching them with similar modeling cases in the preset enterprise knowledge base.

[0013] Optionally, the method further includes: Obtain the type and version information of the target solver; Based on the type and version information of the target solver, a preset semantic conversion rule is applied to convert the connection simulation parameters corresponding to the identified bolts into connection simulation parameters compatible with the format of the target solver. Based on the identified bolts and their corresponding connection simulation parameters compatible with the format of the target solver, a visualization report is generated using the target solver.

[0014] A second aspect of this application provides a bolt identification system, the system comprising: The first acquisition module is used to acquire the geometric features of any candidate region in the assembly model; The matching module is used to match the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, and each candidate sub-region corresponds to at least one geometric feature. The second acquisition module is used to acquire the connection topology information of each of the candidate sub-regions in the assembly model; The determination module is used to determine the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region; The identification module is used to input the geometric features and connection relationship features of each candidate sub-region into a pre-trained multimodal learning model, so as to identify bolts in each candidate sub-region through the multimodal learning model.

[0015] Optionally, the matching module, which matches the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, includes: The first acquisition submodule is used to acquire the assembly type information corresponding to the assembly model; The weight assignment submodule is used to assign corresponding weights to various geometric features of the candidate region based on the assembly type information and according to a preset dynamic weight allocation mechanism. The calculation submodule is used to calculate the similarity score between the candidate region and the standard bolt features in the standardized bolt feature library according to the weights of each of the geometric features. The filtering submodule is used to filter out multiple candidate subregions that meet the similarity score requirements from the candidate regions based on the similarity score.

[0016] Optionally, the step of determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region, the determining module includes: The second acquisition submodule is used to acquire the mating hole group structure between the target candidate sub-region and its adjacent virtual components in the assembly model. The first determining submodule is used to determine the hole group logic features between the target candidate subregion and its adjacent virtual components based on the mating hole group structure formed by the target candidate subregion and its adjacent virtual components in the assembly model. The second determining submodule is used to determine the assembly hierarchy features between the target candidate subregion and its adjacent virtual components based on the spatial position of the target candidate subregion in the assembly model and the relative position between the target candidate subregion and its adjacent virtual components. The third determining submodule is used to obtain the connection relationship features based on the hole group logic features and the assembly level features.

[0017] Optionally, the system further includes: The model training submodule is used to perform initial training on the initial machine learning model using the first training dataset. The first training dataset includes geometric point cloud data information, material property information and semantic label information of standard bolt features stored in the standardized bolt feature library. The geometric point cloud data information, material property information and semantic label information of the standard bolt features are pre-stored in the standardized bolt feature library. The parameter tuning submodule is used to tune the parameters of the initially trained machine learning model using a second training dataset, which includes at least sample data from real-world engineering scenarios.

[0018] Optionally, the system further includes: The feature reconstruction submodule is used to reconstruct the occluded standard bolt features in the first training dataset using a preset occlusion compensation algorithm. The feature verification submodule is used to verify the rationality of standard bolt features after feature reconstruction according to predefined assembly symmetry rules.

[0019] Optionally, the system further includes: The third acquisition submodule is used to acquire correction data for the false detection results or the missed detection results when there are false detection results or missed detection results in the bolt identification results. Add a submodule to add the corrected data as an incremental training sample to the second training dataset; The parameter tuning submodule for the initial machine learning model trained using the second training dataset includes: The parameter update subunit is used to update the parameters of the fully connected layer of the initially trained machine learning model using the incremental training samples.

[0020] Optionally, the system further includes: The fourth acquisition submodule is used to acquire the stress condition information of the identified bolts; The fourth determination submodule is used to determine the connection simulation parameters corresponding to the identified bolts based on the stress condition information and according to the preset connection parameter calculation rules. The connection simulation parameters include preload parameters and connection stiffness parameters. The generation submodule is used to verify the rationality of the connection simulation parameters corresponding to the identified bolts, and for bolts whose verification results are unreasonable, parameter correction suggestions are generated by matching similar modeling cases in the preset enterprise knowledge base.

[0021] A third aspect of this application provides an electronic device, including a memory, a processor, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the bolt identification method as described in the first aspect of this application.

[0022] The beneficial effects of this application are: This application proposes a bolt identification method, comprising: acquiring the geometric features of any candidate region in an assembly model; matching the geometric features of the candidate region with a pre-defined standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, each candidate sub-region corresponding to at least one geometric feature; acquiring the connection topology information of each candidate sub-region in the assembly model; determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region; and inputting the geometric features and connection relationship features of each candidate sub-region into a pre-trained multimodal learning model to identify bolts in each candidate sub-region through the multimodal learning model. This application achieves efficient and accurate bolt identification by acquiring the geometric features of candidate regions in the assembly model, combining the connection topology information of candidate sub-regions in the assembly model, determining the connection relationship features between candidate sub-regions and their adjacent virtual components, and inputting these features into a multimodal learning model for bolt identification. This results in higher fault tolerance and recognition accuracy in complex assembly structures. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic flowchart illustrating the steps of a bolt identification method provided in an embodiment of this application; Figure 2 This is a flowchart of a multi-level feature-driven bolt recognition priority matching process provided in an embodiment of this application; Figure 3 This application provides a multimodal machine learning framework and a flowchart of a bolt recognition enhancement process. Figure 4 This is a flowchart illustrating an automated parameter generation process provided in an embodiment of this application; Figure 5 This is a flowchart of a visual report generation process provided in an embodiment of this application; Figure 6 This is a schematic diagram of a bolt identification system provided in an embodiment of this application; Figure 7 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0026] In a first aspect, this application provides a method for identifying bolts, such as... Figure 1 As shown, the method includes: Step S101: Obtain the geometric features of any candidate region in the assembly model.

[0027] In step S101, during the preprocessing for bolt identification of the assembly model, a local area with potential bolt structures (i.e., a candidate area) needs to be selected in the assembly model. Geometric feature information for identification is then extracted from this candidate area. The geometric features described in this application are categorized as follows: core features: such as the diameter of the main cylinder, the size of the head flange, the thread length, and the head shape; auxiliary features: such as the relief groove, chamfer, guide cone surface, and gasket groove; contextual features: such as the hole distribution pattern and constraint alignment direction between the candidate area and surrounding virtual components. In this application, the assembly model can be a CAD model or a mesh model. The hole distribution pattern refers to the arrangement of holes in the candidate area and adjacent virtual components. For example, in some cases, multiple bolts may be distributed around a common hole group. This application can assist in bolt identification by analyzing the geometric relationship of the hole group. The constraint alignment direction refers to the installation direction and alignment method that the bolt may be associated with surrounding virtual components during assembly. This application can determine whether the candidate area conforms to the standard structure of bolted connections by analyzing these assembly constraints.

[0028] Step S102: Match the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, each candidate sub-region corresponding to at least one geometric feature.

[0029] In step S102, the geometric features of the candidate regions selected in the previous steps are matched with a preset standardized bolt feature library. The aim is to identify multiple candidate sub-regions that may be bolts by comparing the geometric shapes in the candidate regions with the typical features of standard bolts.

[0030] Step S103: Obtain the connection topology information of each candidate sub-region in the assembly model.

[0031] In step S103, the connection topology information of each candidate sub-region in the assembly model is obtained. This allows for further exploration of the structural relationships between the candidate sub-regions and their surrounding components, building upon the preliminary geometric feature matching described above. This connection topology information includes, but is not limited to: the spatial positional relationship between the candidate sub-region and adjacent virtual components, hole group mating structures (such as the combination of "through hole + threaded hole"), and assembly hierarchy features between components (such as the connection order between the main surface and the slave surface). Extracting this connection topology information provides contextual semantic support for subsequent identification steps, helping to determine whether the candidate sub-region possesses typical bolt connection features, thereby improving the ability to distinguish between false positives and false negatives in complex structures.

[0032] Step S104: Based on the connection topology information of each candidate sub-region, determine the connection relationship features between each candidate sub-region and its adjacent virtual components.

[0033] In step S104, based on the connection topology information of each candidate sub-region, the connection relationship features between each candidate sub-region and its adjacent virtual components are determined. The aim is to further analyze structural semantic relationships based on geometric recognition, enhancing the accuracy and robustness of the recognition. In some cases, the connection relationship features specifically include two types of key information: firstly, hole group logic features, i.e., identifying whether the candidate sub-region participates in forming a typical "through hole + threaded hole" mating structure, excluding interfering elements such as isolated holes or non-connecting holes; secondly, assembly level features, which determine whether the candidate sub-region conforms to common bolt assembly paths by analyzing its positional hierarchy within the assembly (e.g., master-slave assembly relationships between components). This step enriches the feature dimensions upon which bolt recognition relies through structural logic reasoning, providing more discriminative contextual information for subsequent recognition decisions based on multimodal learning models.

[0034] Step S105: Input the geometric features and connectivity features of each candidate sub-region into a pre-trained multimodal learning model to identify bolts in each candidate sub-region through the multimodal learning model.

[0035] In step S105, the geometric features and connectivity features of each candidate sub-region are input into a pre-trained multimodal learning model. This model comprehensively utilizes multi-source heterogeneous information to identify bolts within the candidate sub-regions. In this application, the multimodal learning model integrates various feature types, such as geometric point cloud data, material properties, and semantic labels, during the training phase and is fine-tuned using sample data from actual engineering scenarios. This enables the model to possess excellent contextual understanding and recognition generalization capabilities. During the recognition process, the multimodal learning model considers not only the local geometry of the candidate sub-regions but also their surrounding semantic environment, achieving accurate identification of typical bolt structures. This improves the accuracy and fault tolerance in complex assembly scenarios and provides high-quality recognition results for automated modeling.

[0036] This application obtains the geometric features of candidate regions in the assembly model, combines them with the connection topology information of the assembly model, extracts connection relationship features such as hole group logic and assembly level, and inputs them into a multimodal learning model for bolt recognition, thereby achieving efficient and accurate bolt recognition, and possessing higher fault tolerance and recognition accuracy in complex assembly structures.

[0037] In one embodiment, matching the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region includes: Obtain the assembly type information corresponding to the assembly model; Based on the assembly type information, and according to a preset dynamic weight allocation mechanism, corresponding weights are assigned to each geometric feature of the candidate region. Based on the weights of each of the geometric features, the similarity score between the candidate region and the standard bolt features in the standardized bolt feature library is calculated. Based on the similarity score, multiple candidate sub-regions that meet the similarity score requirements are selected from the candidate regions.

[0038] In this embodiment, the geometric features of the candidate region are matched with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region. Specifically, this includes the following steps: First, the assembly type information of the assembly model is obtained to determine whether the assembly model belongs to different categories such as vehicle body structure, aerospace thin-walled component, or mechanical base. Based on the assembly type information, a preset dynamic weight allocation mechanism is applied to weight the geometric features extracted from the candidate region (such as cylinder diameter, head shape, undercut groove, chamfer, etc.) to highlight features with greater discriminative value in specific assembly structures.

[0039] Furthermore, a similarity score is calculated between the weighted geometric features and typical bolt models in the standardized bolt feature library to evaluate the degree of matching between the candidate regions and each standard bolt model. Based on these similarity scores, several candidate sub-regions that meet the set thresholds are selected as the key target regions for subsequent analysis.

[0040] In one embodiment, determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region includes: Obtain the mating hole group structure between the candidate sub-region and its adjacent virtual components in the assembly model; Based on the mating hole group structure formed by the candidate sub-region and its adjacent virtual components in the assembly model, the hole group logic characteristics between the candidate sub-region and its adjacent virtual components are determined. Based on the spatial position of the candidate sub-region in the assembly model and the relative position between the candidate sub-region and its adjacent virtual components, the assembly hierarchy features between the candidate sub-region and its adjacent virtual components are determined. The connection relationship features are determined based on the logical features of the hole group and the assembly level features.

[0041] In this embodiment, firstly, the component boundary information corresponding to the candidate sub-region in geometric space is extracted from the assembly model, and adjacent virtual components that are in direct contact with or connected through hole structures are identified. Based on this, the mating hole group structure formed between the candidate sub-region and the adjacent virtual components is further obtained. This mating hole group structure may consist of combinations of different types of holes, such as through holes, threaded holes, countersunk holes, and positioning holes. Its distribution and hole type combination are important bases for judging the authenticity of the bolt connection.

[0042] Subsequently, based on the geometric properties of the mating hole group (including hole diameter, hole depth, axial alignment, axial spacing, central symmetry, etc.) and the contact method between the components containing the hole, the logical features of the hole group in the candidate sub-region are extracted and established. For example, if a candidate sub-region has a set of aligned through holes and threaded holes with two adjacent virtual components, and satisfies conditions such as axial collinearity, hole diameter matching, and reasonable mating clearance, it can be determined that it has typical bolt mating hole group characteristics; conversely, if the region only connects a single component or the hole group is arranged in a disorderly manner, its logical features can be determined as non-standard connection.

[0043] Based on the logical features of the hole group, the assembly semantic position of the candidate sub-region in the overall structure is determined by further combining its three-dimensional spatial position in the assembly model (e.g., located at the model boundary, center, or a specific structural level) and its relative assembly relationship with adjacent virtual components (e.g., stacking relationship, insertion direction, contact surface orientation, etc.). This information constitutes the assembly hierarchy feature, which can reflect whether the candidate sub-region is located on the surface of the main structural component or at a controlled connection node of a subordinate structure. For example, in a car body model, the master and slave surfaces of bolted connections can be determined by identifying its assembly sequence among components such as A-pillars, crossbeams, and reinforcing plates, thereby enhancing the reliability of the judgment.

[0044] By comprehensively extracting the logical features of the hole group and the assembly level features, the connection relationship features between each candidate sub-region and its adjacent virtual components are determined. This not only accurately reflects the connection pattern of the candidate sub-region in the geometric structure, but also captures its role in assembly semantics and mechanical function, providing richer contextual information input for the subsequent bolt recognition model.

[0045] Figure 2 A flowchart of a multi-level feature-driven bolt recognition priority matching process is provided in this application, as follows: Figure 2 As shown: The entire process is based on candidate regions in the assembly model, and sequentially completes steps such as geometric feature extraction, feature weighting, semantic reasoning and candidate region scoring, providing high-quality input for subsequent bolt recognition.

[0046] First, the system performs multi-level feature classification on the candidate regions, dividing their geometric features into three categories: Key features, such as the diameter of the main cylinder and the size of the bolt head flange, directly reflect the basic structure of the bolt. Auxiliary features, such as thread relief grooves and gasket grooves, are used to improve the completeness of identification and the ability to cover non-standard structures. Contextual features, such as hole group distribution and assembly constraints, reflect the spatial and logical relationships between the candidate area and surrounding components.

[0047] The aforementioned features are input into a dynamic weight allocation process, which automatically adjusts the weights of different feature dimensions based on the assembly type to adapt to the discrimination requirements of different application scenarios. The weight adjustment follows a priority rule system to guide the subsequent matching calculation process.

[0048] Based on weighted configuration, a fuzzy matching algorithm is used to compare the similarity of candidate regions with standard bolts in a standardized bolt feature library. This is further refined using the Levenshtein distance algorithm, which outputs a matching score for each candidate region. Based on the score, several sub-regions exceeding a set threshold are identified as key candidate sub-regions for subsequent identification steps. The score of each candidate sub-region serves as a preliminary criterion for determining whether it is a bolt. The Levenshtein distance algorithm measures the structural differences between candidate regions and standard bolts in the standard bolt feature library, refining the differences in geometric features by calculating minimal editing operations (such as insertion, deletion, and replacement). Building upon fuzzy matching, the Levenshtein distance algorithm further optimizes the matching results, making the matching between candidate regions and standard bolts more accurate. Finally, combining the matching scores from this algorithm, several candidate sub-regions with similarity scores exceeding a set threshold are selected as key analysis targets in subsequent identification steps.

[0049] Simultaneously, a topological connectivity reasoning engine is introduced in parallel to analyze the connection topology information, i.e., contextual features, between candidate regions and their adjacent virtual components, thereby completing two key reasoning tasks: The hole group logic is verified by determining whether there is a typical hole group fit structure such as "through hole + threaded hole"; Based on the master-slave structure analysis of the assembly, the semantic hierarchy in the connection path is determined, including the division of master and slave surfaces, thereby enabling reasoning analysis of the assembly hierarchy. Specifically, by identifying key components in the assembly and their interrelationships, the components in the connection path are classified according to their primary and secondary relationships within the assembly structure. This allows for accurate analysis of the assembly hierarchy and connection relationships between components during subsequent bolt identification. This analysis helps improve the accuracy of identification, especially in complex assemblies, effectively distinguishing components at different levels and providing clearer structural information for bolt identification.

[0050] In one embodiment, the multimodal learning model is trained according to the following steps: The initial machine learning model is initially trained using the first training dataset, which includes geometric point cloud data, material property information, and semantic label information of standard bolt features stored in the standardized bolt feature library. The parameters of the initial machine learning model, which has been initially trained, are adjusted using a second training dataset, which includes at least sample data from real-world engineering scenarios.

[0051] In this embodiment, the training process of the multimodal learning model adopts a phased strategy, aiming to integrate standard data and real engineering data to improve the model's generalization ability and practical adaptability. First, the initial machine learning model is pre-trained using a first training dataset. The first training dataset comes from a standardized bolt feature library and contains a large number of standard bolt samples with standardized modeling, including geometric point cloud data information (such as surface morphology, boundary contours, local details, etc.), material property information (such as the density and reflectivity of materials such as steel and aluminum), and semantic label information attached to the assembly model, so that the model has basic shape understanding and type recognition capabilities.

[0052] After the initial model was built, a second training dataset was introduced to fine-tune the model parameters. This second training dataset came from real-world engineering scenarios, covering real samples under various complex assembly environments, including situations where bolts were occluded, deformed, had missing features, or used non-standard assembly methods. By introducing these real samples, the model can gradually adapt to imperfect inputs in engineering and learn more connection patterns and boundary judgment capabilities with engineering semantics through parameter adjustments, thereby improving recognition accuracy and robustness in practical applications. This training process achieves a transfer from a standard ideal state to real-world engineering applications, constructing a highly adaptive recognition model with multimodal perception capabilities.

[0053] In one embodiment, the training process of the multimodal learning model further includes: For the occluded standard bolt features in the first training dataset, the occluded standard bolt features are reconstructed using a preset occlusion compensation algorithm. The rationality of the standard bolt features after feature reconstruction is verified according to the predefined assembly symmetry rules.

[0054] In this embodiment, to enhance the model's ability to identify incomplete data during the training process of the multimodal learning model, an occlusion compensation and symmetry verification mechanism is introduced, specifically for handling standard bolt features that are occluded or partially missing in the first training dataset. Specifically, for these occluded standard bolt samples, a preset occlusion compensation algorithm is used to reconstruct their features. This compensation algorithm, based on known geometric patterns and prior information about the bolt structure, combines locally visible areas (such as partially exposed threaded sections, flange edges, etc.) to infer and reconstruct the complete bolt shape, generating reconstructed features consistent with the geometric shape before occlusion.

[0055] To ensure the rationality and physical feasibility of the compensation results, the reconstructed features are further verified based on predefined assembly symmetry rules. In this application, these symmetry rules include, but are not limited to, the axial symmetry of hole arrangement, the circumferential uniformity of bolt distribution, and the equivalence of mating relationships between adjacent bolts. By comparing these rules with the actual distribution in the assembly model, it is determined whether the reconstructed results conform to common engineering layout logic. The combined application of occlusion compensation and symmetry verification not only improves the completeness of the training samples but also significantly enhances the model's robustness and discriminative ability in complex scenarios such as occlusion and defects.

[0056] In one embodiment, the training process of the multimodal learning model further includes: If there are false positives or false negatives in the bolt identification results, obtain correction data for the false positives or false negatives. The corrected data is used as an incremental training sample and added to the second training dataset; The step of adjusting the parameters of the initially trained machine learning model using the second training dataset includes: The parameters of the fully connected layers of the initial machine learning model, which has been initially trained, are updated using the incremental training samples.

[0057] In this embodiment, to further improve the recognition accuracy and continuous learning capability of the multimodal learning model in practical applications, an online incremental learning mechanism is introduced during model training to handle false detections and missed detections that occur during the recognition process. Specifically, when recognizing bolts in the application stage, if the detection results show false recognition (identifying non-bolts as bolts) or missed recognition (failing to identify actual bolts), relevant correction data will be collected as feedback input. This correction data can come from user manual annotation, engineering review, or verification simulation comparison, and has clear judgment labels and real geometric / topological information.

[0058] Subsequently, these corrected data are used to construct incremental training samples and dynamically added to the second training dataset, ensuring the training data is updatable and adaptable to various scenarios. During the parameter update phase, the entire model is not retrained; instead, a lightweight strategy is employed, fine-tuning only the parameters of the fully connected layers in the initially trained machine learning model. This approach preserves the original model's recognition ability on standard samples while improving its adaptability to special structures, atypical distributions, or new types of bolts through new sample updates, ensuring continuous evolution and iterative optimization of the model in engineering applications. This embodiment enables rapid adaptation and self-correction to complex working conditions, improving recognition stability and accuracy over long-term operation.

[0059] Figure 3This application provides a multimodal machine learning framework and a flowchart of a bolt recognition enhancement process, such as... Figure 3 As shown: This flowchart integrates multiple data sources, including geometric point cloud information, material properties, and engineering semantics. Through a unified fusion and inference mechanism, it enhances the model's ability to identify complex bolt structures, compensate for occlusion, and learn online adaptively.

[0060] The entire process begins with multi-source data input, including three main input types: Geometric point cloud data: Extract the spatial shape of the bolt candidate region, and perform three-dimensional convolutional encoding through a 3D-CNN network (three-dimensional convolutional neural network) to extract local and global geometric features; Material property data, such as material type and density, are numerically characterized using a feature encoder. Engineering semantic data: semantic tags from the assembly model are converted into semantic vectors through the tag embedding module.

[0061] The aforementioned multimodal features are fused in the multimodal fusion layer to form a unified representation, which is then input into the GNN (Graph Neural Network) context inference module to achieve a comprehensive understanding and recognition of the candidate sub-regions and their contextual connections (such as hole distribution, assembly level, etc.).

[0062] To enhance the model's adaptability to occluded scenes, an occlusion compensation mechanism is introduced. Internally, it focuses on the visible area through an attention mechanism and drives the geometric completion network to reconstruct the occluded area by combining symmetry constraint rules. The reconstruction result is then verified for structural rationality through a completion verification mechanism.

[0063] Meanwhile, the GNN context inference results also undergo confidence evaluation processing, identifying candidate regions with low recognition confidence. For these low-confidence regions, user feedback correction data is obtained through a user correction interface and input into the incremental learning engine to achieve lightweight fine-tuning of the model. This incremental learning engine mainly updates the parameters of the fully connected layers at the back end of the model, avoiding large-scale retraining while improving the model's adaptability to novel bolt structures and non-standard geometries.

[0064] In addition, it supports the generation of synthetic data, which means adding the corrected data as incremental training samples to the second training dataset, which can be used to expand the training dataset and improve the model's recognition ability.

[0065] In one embodiment, the stress condition information of the identified bolts is obtained; based on the stress condition information, the connection simulation parameters corresponding to the identified bolts are determined according to the preset connection parameter calculation rules, the connection simulation parameters including preload parameters and connection stiffness parameters; the rationality of the connection simulation parameters corresponding to the identified bolts is verified, and for bolts whose verification results are unreasonable, parameter correction suggestions are generated by matching similar modeling cases in the preset enterprise knowledge base.

[0066] In this embodiment, to achieve automatic connection from bolt geometry identification to simulation modeling, after bolt identification is completed, it is necessary to further obtain the stress condition information of each identified bolt. The stress condition information may include the load direction at the bolt location, the material properties of the connecting components, and the type of external working condition (such as tension, shear, vibration, etc.). After obtaining this information, based on preset connection parameter calculation rules, the connection simulation parameters corresponding to the bolt are automatically derived, mainly including but not limited to: preload parameters calculated based on bolt specifications, material strength, friction coefficient of the assembly surface, etc., and connection stiffness parameters determined based on connection interface characteristics, material stiffness, and stress mode.

[0067] Subsequently, the generated simulation parameters are validated for rationality through methods such as unit analysis, boundary condition checks, simulation convergence prediction, and comparison with historical models. If certain bolt parameter combinations are found to be physically unreasonable, have abnormal value ranges, or deviate significantly from actual working conditions, the model will be matched with a large number of pre-stored modeling cases in the enterprise's knowledge base to identify historical bolt models with similar structural locations, dimensions, and stress characteristics. Based on their validated simulation parameters, parameter correction suggestions or recommended values ​​will be generated.

[0068] This embodiment automatically derives key simulation parameters such as preload and connection stiffness by combining bolt identification results with their stress conditions in the assembly model. A rationality verification mechanism is introduced to ensure the parameters conform to engineering physics. For bolts identified as abnormal, correction suggestions can be generated based on similar modeling cases in the enterprise knowledge base, thereby achieving intelligent parameter correction and optimization. This method not only improves the automation and efficiency of bolt simulation modeling but also enhances the reliability and engineering adaptability of parameter settings, effectively reducing manual configuration errors and ensuring the accuracy and stability of subsequent simulation analysis.

[0069] Figure 4 A flowchart for the automated parameter generation process provided in this application, such as Figure 4 As shown: The system retrieves bolt specification information from a standardized bolt feature library. Based on this information (such as bolt stress conditions), including the load on the bolt's location, the material properties of the connecting components, and the type of external working conditions, the system automatically derives the corresponding connection simulation parameters for each bolt according to pre-defined connection parameter calculation rules. The generated simulation parameters are then validated for reasonableness. If any bolt parameter combinations are found to be physically unreasonable, have abnormal value ranges, or deviate significantly from actual working conditions, the system matches them with pre-stored modeling cases in a pre-defined enterprise knowledge base to identify historical bolt models with similar structural locations, dimensions, and stress characteristics. Based on these validated simulation parameters, the system generates parameter correction suggestions or recommended values.

[0070] In one embodiment, the type and version information of the target solver are obtained; based on the type and version information of the target solver, a preset semantic conversion rule is applied to convert the connection simulation parameters corresponding to the identified bolts into connection simulation parameters compatible with the format of the target solver; based on the identified bolts and their corresponding connection simulation parameters compatible with the format of the target solver, a visualization report is generated through the target solver.

[0071] In this embodiment, to achieve efficient integration of bolt identification results with multiple simulation platforms, the type and version information of the target solver are first obtained, such as whether it is a NASTRAN (NASA Structural Analysis, finite element analysis for structural analysis) solver, ABAQUS solver, ANSYS solver, etc., and the supported connection definition syntax, keyword format, and parameter unit system are identified. Subsequently, based on preset semantic conversion rules, the connection simulation parameters (such as preload, connection stiffness, contact properties, etc.) generated during bolt identification are converted into an input format compatible with the target solver, ensuring that the syntax structure, parameter naming, and unit system strictly match the target environment, avoiding errors or result failures due to format conflicts during simulation. After format conversion, the target solver can be further invoked, and a visual report can be automatically generated using the identified bolt geometry and connection parameters. This report intuitively displays the modeling status, boundary condition arrangement, and parameter distribution of the connection area, assisting engineers in quickly completing modeling confirmation and simulation preparation verification.

[0072] In some cases, visualization reports can include: bolt distribution heatmaps, preload histograms, and connection stiffness matrix summaries. Furthermore, visualization reports can be linked to CAE models; clicking on bolt markers in the report will highlight the corresponding locations in the CAE model. It also supports multi-user online annotation of suspicious bolts, generating a list of bolts awaiting confirmation, recording user correction history, and enabling team-level knowledge accumulation of bolt identification rules.

[0073] This embodiment obtains the type and version information of the target solver and, combined with preset semantic conversion rules, automatically converts the connection simulation parameters in the bolt identification results into a format compatible with the target solver. This achieves seamless integration and automatic configuration of the simulation model, avoiding format errors and parameter loss problems that are prone to occur during manual conversion. Simultaneously, the target solver can be directly invoked to generate visual reports, making the simulation modeling results more intuitive and verifiable. This enhances the intelligence and cross-platform adaptability of the pre-processing workflow, effectively improving engineering efficiency and the accuracy of simulation preparation.

[0074] Figure 5 This application provides a flowchart of a visual report generation process, such as... Figure 5 As shown: The process begins by obtaining the solver type and version information of the target simulation platform, such as mainstream finite element solvers like NASTRAN, ABAQUS, and ANSYS, and identifying their corresponding supported connection syntax, parameter unit systems, and format constraints. Subsequently, preset semantic conversion rules are invoked to restructure and semantically map the identified bolt connection simulation parameters, ensuring that parameter naming, unit systems, keyword syntax, and other content are fully compatible with the target solver, thus avoiding simulation errors or model invalidation due to format incompatibility.

[0075] After formatting, the system can automatically call the target solver interface to build a simulation model and generate a visualization report. The report includes multi-dimensional index views such as bolt distribution heatmap, preload histogram, and connection stiffness matrix summary, which intuitively show the layout status and parameter distribution of each connection area.

[0076] This application provides an automated integrated processing flow from bolt recognition to simulation. First, candidate regions with potential bolt features are extracted from the assembly model. Based on core geometric features, auxiliary structural features, and contextual spatial relationships, these regions are classified and dynamically weighted at multiple levels. Similarity matching is performed using a standardized bolt feature library to filter out multiple candidate sub-regions to be identified. Subsequently, the connection topology information between each candidate sub-region and its adjacent virtual components is obtained, and the logical features of hole groups and assembly level features are extracted to further construct connection semantics. Then, these geometric and connection features are input into a pre-trained multimodal learning model, which integrates multi-source data such as geometric point clouds, material properties, and semantic labels for recognition, resulting in high-precision bolt recognition results. After identification, based on the stress conditions of the bolts (such as load direction, component material, and working condition type), and combined with the connection parameter calculation rules, simulation parameters, including preload and connection stiffness, are automatically derived. The physical validity is verified through a rationality verification mechanism. If anomalies are found, parameter correction suggestions are automatically generated based on similar modeling cases in the enterprise knowledge base. In the simulation output stage, according to the type and version information of the selected solver (such as NASTRAN, ABAQUS, ANSYS), semantic conversion rules are applied to convert the connection parameters into a compatible format. The model is then built, and a visual report including bolt distribution heatmap, preload histogram, and connection stiffness matrix is ​​generated. The model supports linked highlighting and multi-user online annotation, realizing a closed-loop process of identification-modeling-verification-feedback. This significantly improves the automation, engineering adaptability, and simulation reliability of bolt identification and modeling in complex assembly structures.

[0077] Based on the same inventive concept, a second aspect of the embodiments of this application provides a bolt identification system, such as... Figure 6 As shown, the system includes: The first acquisition module 201 is used to acquire the geometric features of any candidate region in the assembly model; The matching module 202 is used to match the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, and each candidate sub-region corresponds to at least one geometric feature. The second acquisition module 203 is used to acquire the connection topology information of each of the candidate sub-regions in the assembly model; The determining module 204 is used to determine the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region. The connection relationship features include: hole group logic features and assembly level features. The identification module 205 is used to input the geometric features and connection relationship features of each candidate sub-region into a pre-trained multimodal learning model, so as to identify bolts in each candidate sub-region through the multimodal learning model.

[0078] Optionally, the matching module 202, which matches the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, includes: The first acquisition submodule is used to acquire the assembly type information corresponding to the assembly model; The weight assignment submodule is used to assign corresponding weights to various geometric features of the candidate region based on the assembly type information and according to a preset dynamic weight allocation mechanism. The calculation submodule is used to calculate the similarity score between the candidate region and the standard bolt features in the standardized bolt feature library according to the weights of each of the geometric features. The filtering submodule is used to filter out multiple candidate subregions that meet the similarity score requirements from the candidate regions based on the similarity score.

[0079] Optionally, the step of determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region, the determining module 204, includes: The second acquisition submodule is used to acquire the mating hole group structure between the candidate subregion and its adjacent virtual components in the assembly model; The first determining submodule is used to determine the hole group logic features between the candidate subregion and its adjacent virtual components based on the mating hole group structure formed by the candidate subregion and its adjacent virtual components in the assembly model. The second determining submodule is used to determine the assembly hierarchy features between the candidate subregion and its adjacent virtual components based on the spatial position of the candidate subregion in the assembly model and the relative position between the candidate subregion and its adjacent virtual components. The third determining submodule is used to determine the connection relationship features based on the hole group logic features and the assembly level features.

[0080] Optionally, the system further includes: The model training submodule is used to perform initial training on the initial machine learning model using the first training dataset, which includes geometric point cloud data information, material property information and semantic label information of standard bolt features stored in the standardized bolt feature library. The parameter tuning submodule is used to tune the parameters of the initially trained machine learning model using a second training dataset, which includes at least sample data from real-world engineering scenarios.

[0081] Optionally, the system further includes: The feature reconstruction submodule is used to reconstruct the occluded standard bolt features in the first training dataset using a preset occlusion compensation algorithm. The feature verification submodule is used to verify the rationality of standard bolt features after feature reconstruction according to predefined assembly symmetry rules.

[0082] Optionally, the system further includes: The third acquisition submodule is used to acquire correction data for the false detection results or the missed detection results when there are false detection results or missed detection results in the bolt identification results. Add a submodule to add the corrected data as an incremental training sample to the second training dataset; The parameter tuning submodule for the initially trained machine learning model using the second training dataset includes: The parameter update subunit is used to update the parameters of the fully connected layer of the initially trained machine learning model using the incremental training samples.

[0083] Optionally, the system further includes: The fourth acquisition submodule is used to acquire the stress condition information of the identified bolts; The fourth determination submodule is used to determine the connection simulation parameters corresponding to the identified bolts based on the stress condition information and according to the preset connection parameter calculation rules. The connection simulation parameters include preload parameters and connection stiffness parameters. The generation submodule is used to verify the rationality of the connection simulation parameters corresponding to the identified bolts, and for bolts whose verification results are unreasonable, parameter correction suggestions are generated by matching similar modeling cases in the preset enterprise knowledge base.

[0084] Based on the same inventive concept, a third aspect of the embodiments of this application provides a method as follows: Figure 7 The illustrated electronic device 100 includes a memory 110, a processor 120, and a program or instructions stored in the memory 110 and executable on the processor 120, wherein the program or instructions, when executed by the processor 120, implement the steps of the bolt identification method as described in the first aspect of this application.

[0085] Each embodiment in this specification focuses on the differences from other embodiments. For the same or similar parts between the embodiments, please refer to each other.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0091] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0092] The above provides a detailed description of the bolt identification method, system, and electronic device. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying bolts, characterized in that, The method includes: Obtain the geometric features of any candidate region in the assembly model; The geometric features of the candidate region are matched with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, and each candidate sub-region corresponds to at least one geometric feature. Obtain the connection topology information of each of the candidate sub-regions in the assembly model; Based on the connection topology information of each candidate sub-region, determine the connection relationship characteristics between each candidate sub-region and its adjacent virtual components; The geometric features and connectivity features of each candidate sub-region are input into a pre-trained multimodal learning model to identify bolts in each candidate sub-region.

2. The bolt identification method according to claim 1, characterized in that, The step of matching the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region includes: Obtain the assembly type information corresponding to the assembly model; Based on the assembly type information, and according to a preset dynamic weight allocation mechanism, corresponding weights are assigned to each geometric feature of the candidate region. Based on the weights of each of the geometric features, the similarity score between the candidate region and the standard bolt features in the standardized bolt feature library is calculated. Based on the similarity score, multiple candidate sub-regions that meet the similarity score requirements are selected from the candidate regions.

3. The bolt identification method according to claim 1, characterized in that, The step of determining the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region includes: Obtain the mating hole group structure between the target candidate sub-region and its adjacent virtual components in the assembly model, wherein the target candidate sub-region is any one of the multiple candidate sub-regions; Based on the mating hole group structure formed by the target candidate sub-region and its adjacent virtual components in the assembly model, the hole group logic characteristics between the target candidate sub-region and its adjacent virtual components are determined. Based on the spatial position of the target candidate sub-region in the assembly model, and the relative position between the target candidate sub-region and its adjacent virtual components, the assembly hierarchy features between the target candidate sub-region and its adjacent virtual components are determined. Based on the logical features of the hole group and the assembly hierarchy features, the connection relationship features are obtained.

4. The bolt identification method according to claim 1, characterized in that, The multimodal learning model is trained according to the following steps: The initial machine learning model is initially trained using the first training dataset, which includes geometric point cloud data information, material property information, and semantic label information of standard bolt features in the standardized bolt feature library. The geometric point cloud data information, material property information, and semantic label information of the standard bolt features are pre-stored in the standardized bolt feature library. The parameters of the initial machine learning model, which has been initially trained, are adjusted using a second training dataset, which includes at least sample data from real-world engineering scenarios.

5. The bolt identification method according to claim 4, characterized in that, The training process of the multimodal learning model also includes: For the occluded standard bolt features in the first training dataset, the occluded standard bolt features are reconstructed using a preset occlusion compensation algorithm. The rationality of the standard bolt features after feature reconstruction is verified according to the predefined assembly symmetry rules.

6. The bolt identification method according to claim 4, characterized in that, The training process of the multimodal learning model also includes: If there are false positives or false negatives in the bolt identification results, obtain correction data for the false positives or false negatives. The corrected data is used as an incremental training sample and added to the second training dataset; The step of tuning the parameters of the initially trained machine learning model using the second training dataset includes: The parameters of the fully connected layers of the initial machine learning model, which has been initially trained, are updated using the incremental training samples.

7. The bolt identification method according to claim 1, characterized in that, The method further includes: Obtain the stress condition information of the identified bolts; Based on the stress condition information, the connection simulation parameters corresponding to the identified bolts are determined according to the preset connection parameter calculation rules. The connection simulation parameters include preload parameters and connection stiffness parameters. The rationality of the connection simulation parameters corresponding to the identified bolts is verified. For bolts whose verification results are unreasonable, parameter correction suggestions are generated by matching them with similar modeling cases in the preset enterprise knowledge base.

8. The bolt identification method according to claim 7, characterized in that, The method further includes: Obtain the type and version information of the target solver; Based on the type and version information of the target solver, a preset semantic conversion rule is applied to convert the connection simulation parameters corresponding to the identified bolts into connection simulation parameters compatible with the format of the target solver. Based on the identified bolts and their corresponding connection simulation parameters compatible with the format of the target solver, a visualization report is generated using the target solver.

9. A bolt identification system, characterized in that, The system includes: The first acquisition module is used to acquire the geometric features of any candidate region in the assembly model; The matching module is used to match the geometric features of the candidate region with a preset standardized bolt feature library to obtain multiple candidate sub-regions to be identified within the candidate region, and each candidate sub-region corresponds to at least one geometric feature. The second acquisition module is used to acquire the connection topology information of each of the candidate sub-regions in the assembly model; The determination module is used to determine the connection relationship features between each candidate sub-region and its adjacent virtual components based on the connection topology information of each candidate sub-region. The connection relationship features include: hole group logic features and assembly level features. The identification module is used to input the geometric features and connection relationship features of each candidate sub-region into a pre-trained multimodal learning model, so as to identify bolts in each candidate sub-region through the multimodal learning model.

10. An electronic device, characterized in that, It includes a memory, a processor, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the bolt identification method as described in any one of claims 1-8.

Images