Transmission device design simulation key feature extraction and automatic mapping system and method thereof

Abstract

The invention relates to the technical field of intelligent feature mapping, in particular to a transmission device design simulation key feature extraction and automatic mapping system and a method thereof.According to the method, coordinate positioning and space connection parameter calculation are introduced in the component relation extraction stage, a topological index containing a three-dimensional structure relation and a transmission direction is constructed, and the three-dimensional structure relation and the transmission direction are extracted; the integrity of connection information between structural members is effectively improved, a structural path length control and edge coupling direction judgment mechanism is introduced, nested expression and high-dimensional tensor splicing of path features are realized in combination with node sizes and material fields, and the attention extraction capability of a fusion graph neural network on the coupling relationship between nodes and edges is utilized to improve the accuracy of the fusion graph neural network. And the expression integrity of the path vector on the structure dependence intensity is enhanced. In semantic association, a path-label binding relation is established through combination of a size difference value, a material code and a performance label, and semantic synchronization precision between label data and a structure path is improved.

Description

Transmission device design simulation key feature extraction and automatic mapping system and method Technical Field

[0001] This invention relates to the field of intelligent feature mapping technology, and in particular to a system and method for extracting and automatically mapping key features in transmission device design simulation. Background Technology

[0002] The field of intelligent feature mapping technology aims to use artificial intelligence methods to automatically identify, abstract, and match key features in complex data objects, and establish a correspondence between features and target data, parameters, or models. Through machine learning algorithms, deep neural networks, and pattern recognition, design data, simulation data, or operational data can be quickly mapped to structured databases or design models, improving the accuracy and consistency of feature extraction and mapping, and achieving efficient data integration and intelligent processing in engineering design, computer-aided simulation, and automated decision-making.

[0003] The purpose of the key feature extraction and automatic mapping system for transmission device design simulation is to process multiple simulation data and design elements during the transmission device design simulation process. Through artificial intelligence algorithms, it realizes the automatic identification and extraction of key features and maps them to the design database or parameter set. It aims to solve the problems of low feature extraction efficiency and poor result consistency, thereby shortening the design cycle, improving the accuracy and stability of the matching between design simulation and database parameters, and realizing the efficiency and intelligence of the design process.

[0004] In existing technologies, component relationships are often handled by directly comparing scattered parameters during the structural identification stage. This lacks complete spatial coordinate indexes and component connection direction information, resulting in missing connection boundary information. The path extraction stage generally relies on static rule filtering and lacks a mechanism for modeling structural similarity between paths and fusing semantic dimensions, leading to unstable classification of similar structures. Furthermore, existing semantic label generation methods often rely on rule mapping or fixed field nesting, failing to consider the differences and connection coupling strength of nodes within the structural path. This results in semantic labels failing to reflect the dependencies in structural features. Existing optimization logic often adopts a single-objective control standard, failing to control the coordination between error, latency, and computational load, which can easily lead to local convergence of structural configurations. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a system and method for extracting key features and automatically mapping transmission device design simulations.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A key feature extraction and automatic mapping system for transmission device design simulation includes: a structure construction module: based on transmission device design drawings, extracting components such as the drive shaft, gear pairs, and couplings; locating component coordinates; calculating three-dimensional distances and contact parameters; establishing connection paths and direction indices according to the assembly sequence; and generating a node-connection topology table; a nested modeling module: based on the node-connection topology table, selecting gear transmission paths of length 3 to 5; using a graph neural network; organizing component dimensions and material properties; extracting the force direction and coupling relationship between connection pairs; merging the combination structure of nodes and edges; and generating a path nesting structure representation set; a semantic fusion module: based on... The nested path structure representation set summarizes size differences and material codes, compares the similarity between paths, filters out close groups and associates them with lifespan, load, and temperature rise levels, and fuses the semantics and labels of each group to generate a set of structural semantic combination pairs; Model configuration module: Based on the set of structural semantic combination pairs, it constructs multiple sets of structural parameter configurations, uses a non-dominated sorting genetic algorithm to measure the error, response time, and computational cost of each group, selects non-dominated configurations based on the results, updates the structural records, and generates a set of structural performance configuration solutions; State mapping module: Based on the set of structural performance configuration solutions, it filters out resource occupancy and response overrun combinations, analyzes semantic output differences, compares historical structural labels, extracts and fuses matching index items, and obtains the structural index mapping inference results.

[0007] As a further aspect of the present invention, the node-connection topology table includes component number information, connection boundary matrix, and load transmission direction vector; the path nesting structure representation set includes node size vector, connection stiffness sequence, and structural path encoding index; the structural semantic combination set includes path feature vector, functional semantic label, and performance level identifier value; the structural performance configuration solution set includes structural layer number parameter, response delay value, and computational resource consumption item; and the structural index mapping inference result includes recommended load threshold, semantic matching label, and structural output vector group.

[0008] As a further embodiment of the present invention, the structure construction module includes: a component identification submodule: based on the design drawing information of the transmission device, extracting components such as the drive shaft, gear pair, and coupling, extracting the number field in the component model and establishing a mapping table, reading the three-dimensional center coordinates of the component and parsing the axial displacement vector, extracting the boundary coordinate range and identifying the axial overlap relationship, and generating a component spatial coordinate set; and a connection construction submodule: based on the component spatial coordinate set, calculating the Euclidean distance between the center points of the components and determining the critical contact state, extracting the boundary contact surface area ratio and screening effective structural pairs, extracting the assembly sequence code and constructing the connection path sequence, marking the load transmission direction vector and completing the connection information, and generating a node-connection topology table.

[0009] As a further embodiment of the present invention, the nested modeling module includes: a path filtering submodule: based on the node-connection topology table, filtering gear transmission paths with lengths of 3 to 5, extracting the number field in the path and sorting it to generate an index sequence, filtering out discontinuous path segments by traversing the connection edge set, and calculating the consistency between the path node arrangement direction and the connection edge order to generate a path number sequence set; a feature construction submodule: based on the path number sequence set, using a graph neural network, extracting the geometric dimension vector and material number corresponding to the components in each path segment, marking the index position of nodes and edges in the path order, extracting the force direction coordinate difference and connection stiffness value through the connection segment, splicing the node attribute field and the edge attribute field to form a combined vector, and generating a structural feature combination set; a structure generation submodule: based on the structural feature combination set, constructing a structural path vector and adding path index encoding, performing a uniform length padding operation on the node vector and edge vector, extracting the sequence number identifier and corresponding combined structure position index of each path, labeling the nested relationship mapping matrix and archiving all path data, and generating a path nested structure representation set.

[0010] As a further aspect of the present invention, the graph neural network takes a path number sequence set as input, extracts the geometric dimension vector and material number of the nodes in the path as node features, and extracts the force direction coordinate difference and connection stiffness value of the connecting edges as edge features, constructs a graph structure input tensor, the node features are propagated through adjacent edges and weighted and fused with the edge features to generate an updated node representation, and the node representation is aggregated layer by layer through multi-layer convolution iteration to form an overall path feature representation, and the node set representation is mapped to a path-level vector through a readout function to generate a structural feature combination set.

[0011] As a further embodiment of the present invention, the semantic fusion module includes: a feature extraction submodule: based on the path nesting structure representation set, extracting the size features and material coding parameters in each path, constructing a path vector sequence and performing a normalization operation, constructing a similarity matrix between path pairs and filtering highly similar path groups, labeling the grouping index and outputting a set of combined structures to generate a set of similar structure paths; and a label fusion submodule: based on the set of similar structure paths, extracting the lifespan level, load level, and temperature rise level labels corresponding to each group of path combinations, constructing a label ranking list and synchronizing the path number field, establishing an index mapping relationship between labels and path structures, organizing the structure path and label combination fields, and generating a set of structural semantic combination pairs.

[0012] As a further embodiment of the present invention, the model configuration module includes: a parameter construction submodule: based on the set of structural semantic combinations, constructing structural parameter combinations and assigning configuration numbers, mapping each group of structures to fields such as the number of nodes, the number of layers, and the discard ratio to form a configuration table, arranging field combinations in numerical order and binding structural label indexes, constructing all structural configuration combination records, and generating a parameter configuration combination set; a performance evaluation submodule: based on the set of parameter configuration combinations, using a non-dominated sorting genetic algorithm, inputting structural configuration combinations and synchronizing semantic label vectors, calculating the response offset through the error difference between the structural output and the label fields, recording the response time and computational resource value of a unit combination, extracting the combination number and performance evaluation field to form a performance index, and generating a configuration performance index set; and a configuration filtering submodule: based on the set of configuration performance indicators, comparing the error value of combination items, computational resource quantity, and response time, extracting any configuration group number that is not dominated by other combinations, deleting records in the indicator values ​​that are inferior to combination items, updating the structural configuration index list and sorting by number, and generating a structural performance configuration solution set.

[0013] As a further aspect of the present invention, the non-dominated sorting genetic algorithm uses a set of parameter configuration combinations as the initial population, calculates the fitness vector of each combination in terms of error value, response time, and computational resource quantity, sorts the population in layers according to the Pareto dominance relationship, divides individuals not dominated by combinations into the first non-dominated layer, and the remaining individuals form the second to the next layers in sequence, calculates the crowding distance in each layer and determines the selection probability accordingly, performs selection, crossover, and mutation operations to generate a new generation of population, repeats fitness calculation and non-dominated layering until the number of iterations or convergence conditions are met, and outputs a set of structural performance configuration solutions containing multiple sets of fit solutions.

[0014] As a further aspect of the present invention, the state mapping module includes: a resource filtering submodule: based on the structural performance configuration solution set, extracting the floating-point computation amount and response time of each group of structural configurations, reading the system-allocated computational resource limits and constructing a comparison list, filtering out configuration combinations that do not meet the resource conditions and recording the remaining configurations, arranging the structural configuration sequence and completing resource adaptation matching, and generating a resource-available configuration set; and a tag matching submodule: based on the resource-available configuration set, extracting the structural output semantic vector and comparing it with historical tag vectors, calculating the vector difference and extracting the tag field with the highest similarity, matching the design indicators associated with the tag field, organizing all structural indicators and constructing a mapping set, and obtaining the structural indicator mapping inference results.

[0015] A method for extracting and automatically mapping key features in transmission device design simulation, based on the aforementioned system for extracting and automatically mapping key features in transmission device design simulation, includes the following steps: S1: Based on the transmission device design drawings, extract the component numbers of the drive shaft, gear pair, and coupling; read the coordinates to construct node indices; calculate geometric distances and contact areas to generate edge connection data; and generate a node-connection topology table; S2: Based on the node relationship topology data, select 3 to 5 path structures; use a graph neural network to fuse node dimensions, material numbers, and edge coupling information; and combine the tensors of nodes and edges to construct... S3: Based on the nested path vector group, calculate the semantic vector difference to extract the path group, associate the lifespan, load, and temperature rise level label fields with the path index, and generate a set of structural semantic combination pairs; S4: Based on the set of structural semantic binding labels, construct structural parameter combinations, use a non-dominated sorting genetic algorithm to generate a population, calculate the error, response time, and computational cost, and select the optimal configuration to generate a set of structural performance configuration solutions; S5: Based on the set of non-dominated structural configurations, filter out out-of-limit combinations, extract the semantic vector and historical label matching results, output the structural index fields, and generate structural index mapping inference results.

[0016] Compared with existing technologies, the advantages and positive effects of this invention are as follows: This invention introduces coordinate positioning and spatial connection parameter calculation in the component relationship extraction stage to construct a topological index containing three-dimensional structural relationships and transmission directions, effectively improving the integrity of connection information between structural components. It also introduces a structural path length control and edge coupling direction judgment mechanism, combining node dimensions and material fields to achieve nested expression of path features and high-dimensional tensor splicing. Furthermore, by fusing graph neural networks to extract attention to the coupling relationships between nodes and edges, the invention enhances the completeness of the path vector's expression of structural dependence strength. In semantic association, a path-label binding relationship is established through a combination of size difference, material encoding, and performance labels, improving the semantic synchronization accuracy between label data and structural paths. Finally, this invention employs a non-dominated sorting genetic algorithm for multi-objective optimization of three-dimensional performance in terms of error, computational resources, and response time, selecting multiple sets of balanced configuration results to enhance the adaptability of parameter combinations. In structural semantic output comparison, a resource threshold filtering mechanism and a historical path label similarity comparison method are introduced to improve the robustness of structural semantic matching results and the accuracy of parameter inference. Attached Figure Description

[0017] Figure 1 is a system flowchart of the present invention; Figure 2 is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1 (Referring to Figure 1): This invention provides a technical solution: a transmission device design simulation key feature extraction and automatic mapping system, comprising: a structure construction module: based on the transmission device design drawing information, extracting components such as the drive shaft, gear pairs, and couplings, locating component coordinates, calculating three-dimensional distances and contact parameters, establishing connection paths and direction indices according to the assembly sequence, and generating a node-connection topology table; a nested modeling module: based on the node-connection topology table, selecting gear transmission paths of length 3 to 5, using a graph neural network, organizing component dimensions and material properties, extracting the force direction and coupling relationship between connection pairs, assembling the combination structure of nodes and edges, and generating a path nesting structure representation set; and a semantic fusion module: Based on the path nesting structure representation set, the module summarizes size differences and material codes, compares the similarity between paths, filters out close groups and associates them with lifetime, load, and temperature rise levels, and fuses the semantics and labels of each group to generate a set of structural semantic combination pairs. The model configuration module constructs multiple sets of structural parameter configurations based on the set of structural semantic combination pairs. It uses a non-dominated sorting genetic algorithm to measure the error, response time, and computational cost of each group, selects non-dominated configurations based on the results, updates the structural records, and generates a set of structural performance configuration solutions. The state mapping module filters out resource occupancy and response exceedance combinations based on the set of structural performance configurations, analyzes semantic output differences, compares historical structural labels, extracts and fuses matching index items, and obtains the structural index mapping inference results.

[0020] The node-connection topology table includes component number information, connection boundary matrix, and load transmission direction vector. The path nesting structure representation set includes node size vector, connection stiffness sequence, and structural path encoding index. The structural semantic combination set includes path feature vector, functional semantic label, and performance level identifier value. The structural performance configuration solution set includes structural layer number parameter, response delay value, and computational resource consumption item. The structural index mapping inference result includes recommended load threshold, semantic matching label, and structural output vector group.

[0021] The structural construction module includes: a component identification submodule: based on the transmission device design drawings, it extracts components such as the drive shaft, gear pairs, and couplings; extracts the number field from the component model and establishes a mapping table; reads the 3D center coordinates of the components and parses the axial displacement vector; extracts the boundary coordinate range and identifies axial overlap relationships; and generates a component spatial coordinate set. The connection construction submodule: based on the component spatial coordinate set, it calculates the Euclidean distance between the center points of the components and determines the critical contact state; extracts the boundary contact surface area ratio and filters effective structural pairs; extracts the assembly sequence code and constructs the connection path sequence; marks the load transfer direction vector and completes the connection information; and generates a node-connection topology table. The process involves extracting components such as the drive shaft, gear pairs, and couplings. The component model's identification number field is extracted and a mapping table is established. The 3D center coordinates of the components are read, and the axial displacement vectors are analyzed. Boundary coordinate ranges are extracted, and axial overlap relationships are identified. A spatial coordinate set for the components is generated. Using data processing tools, the component model data is first structured. The identification number field, 3D coordinates, and boundary coordinates of each component are extracted separately. The spatial coordinates of each component are compared, the 3D distance between the center points of each component is calculated, and the existence of axial overlap relationships is identified. Based on this, a spatial coordinate set for the components is generated. Furthermore, the identification numbers and coordinate information of all components are merged, and the accuracy of the data is ensured through mapping relationships, ultimately generating the component spatial coordinate set.

[0022] The connection construction submodule calculates the Euclidean distance between the center points of components and determines the critical contact state based on the component spatial coordinate set. It extracts the boundary contact surface area ratio and filters valid structural pairs, extracts the assembly sequence code and constructs the connection path sequence, marks the load transmission direction vector and completes the connection information, and generates a node-connection topology table. First, by calculating the Euclidean distance between the center points of each component, it determines whether they are in contact. If the distance between the center points is less than a preset threshold, the two components are considered to be in contact. Then, it calculates the area ratio between the boundary contact surfaces, compares and filters valid structural pairs. Subsequently, it extracts the code according to the assembly sequence, constructs the connection path sequence in sequence, marks the load transmission direction during this process, completes the connection relationship based on the obtained information, and finally generates a node-connection topology table.

[0023] The nested modeling module includes: a path filtering submodule: based on a node-connection topology table, it filters gear transmission paths of length 3 to 5, extracts the number field from the path and sorts it to generate an index sequence, filters out discontinuous path segments by traversing the connection edge set, and calculates the consistency between the path node arrangement direction and the connection edge order to generate a path number sequence set; a feature construction submodule: based on the path number sequence set, it uses a graph neural network to extract the geometric dimension vector and material number corresponding to the components in each path segment, marks the index position of nodes and edges in the path order, extracts the force direction coordinate difference and connection stiffness value through the connection segment, and concatenates the node attribute field and edge attribute field to form a combined vector to generate a structural feature combination set; and a structure generation submodule: based on the structural feature combination set, it constructs a structural path vector and adds it to the path. The path index encoding process involves padding node and edge vectors to the same length, extracting the sequence number and corresponding combined structure position index for each path, labeling the nested relationship mapping matrix, and archiving all path data to generate a path nested structure representation set. The path filtering submodule, based on the node-connection topology table, filters the node numbers of each path, extracting gear transmission paths with lengths of 3 to 5. It uses Python's `sort` function to sort the number fields in the paths to generate an index sequence, uses a `for` loop to iterate through the connection edge set, filters out discontinuous path segments, and calculates the consistency between the path node arrangement direction and the connection edge order by executing NumPy's `dot` function, generating a path number sequence set.

[0024] Feature construction submodule: Based on the path number sequence set, a graph neural network is used to extract the geometric dimension vector and material number of each component in the path as node features, mark the index position of nodes and edges in the path sequence, calculate the difference of force direction coordinates and connection stiffness values ​​through the adjacent edge set, and use the concatenate function in NumPy to concatenate the node attribute fields and edge attribute fields to form a combined vector, generating a structural feature combination set.

[0025] The structure generation submodule constructs structural path vectors based on the structural feature combination set and embeds the path index encoding. It executes the pad function in NumPy to perform uniform length padding on the node vectors and edge vectors, extracts the sequence number identifier of each path and the corresponding combined structure position index, labels the nested relationship mapping matrix, archives all path data, and generates a path nested structure representation set.

[0026] A graph neural network (GNN) takes a path number sequence set as input, extracts the geometric dimension vectors and material numbers of nodes in the path as node features, and extracts the force direction coordinate differences and connection stiffness values ​​of connecting edges as edge features. It constructs a graph structure input tensor. Node features are propagated through adjacent edges and weighted and fused with edge features to generate updated node representations. Through multi-layer convolutional iterations, the node representations are aggregated layer by layer to form the overall path feature representation. A readout function maps the node set representation to path-level vectors, generating a structural feature combination set. The graph neural network follows the formula:

[0027] in: Represents a node In the Layer feature representation, This represents a constant coefficient, typically used to adjust the weights during node feature updates. Represents a node The set of neighboring nodes, where ReLU represents the activation function, is used to increase the nonlinearity of the network. This represents the edge weight matrix, which is responsible for processing the feature information of the edges. Represents a node With nodes Edge features between them Representing an edge The correction coefficient is used to adjust the impact of edge features on node propagation. This represents the adjustment weight coefficient for the edge features between nodes, used to adjust the contribution of edge features to the propagation of node information. Represents a node In the Layer feature representation, updated node feature values; Execution process: First, node... initial features The geometric dimensions, material properties, and path sequence of each key component in the transmission device are set as node feature vectors. Then, the feature representation of each node is iteratively calculated. At each layer, the node features are updated using a formula. In each iteration, the node... Features Based on the characteristics of neighboring nodes and the edge features between them A weighted update is performed, and a correction factor is introduced. and weighting coefficients The propagation of node features is optimized based on the physical properties of edges and connection stiffness. The ReLU activation function ensures the nonlinear mapping of the network, making information exchange between nodes more accurate. After multiple iterations, the nodes... Features It has been updated to reflect the comprehensive characteristics of the interaction between the various components of the transmission device, forming an efficient set of structural features, which can be further used in the automatic mapping system of the transmission device optimization design and simulation process.

[0028] The semantic fusion module includes: a feature extraction submodule: based on the path nesting structure representation set, it extracts the size features and material coding parameters of each path, constructs a path vector sequence and performs normalization, builds a similarity matrix between path pairs and filters highly similar path groups, labels the grouping index and outputs a set of combined structures, generating a set of similar structure paths; a label fusion submodule: based on the similar structure path set, it extracts the lifespan level, load level, and temperature rise level labels corresponding to each path combination, constructs a label ranking list and synchronizes the path number field, establishes the index mapping relationship between labels and path structures, organizes the structure path and label combination fields, and generates a set of structural semantic combination pairs; the feature extraction submodule: based on the path nesting structure representation set, it extracts the size features and material coding parameters of each path, constructs a path vector sequence, performs normalization, and uses the StandardScaler function in scikit-learn to normalize the size features and material coding parameters in the path. The first module standardizes the dimensions of each feature, calculates the similarity matrix between path pairs using the `dot` function in NumPy, filters highly similar path groups, sets threshold conditions using the `where` function in NumPy to extract path pairs that meet the conditions, and adds path index labels. The `groupby` function in Pandas is then used to group the paths, outputting a set of combined structures and generating a set of similar structured paths. The second module, the tag fusion submodule, extracts lifespan, load, and temperature rise level labels corresponding to each path combination based on the similar structured path set, constructs a tag ranking list, associates the tags with the path structure using the `merge` function in Pandas, synchronizes the path number field according to the path order, establishes the index mapping relationship between tags and path structures using the `set_index` function in Pandas, organizes the structured path and tag combination fields, and merges the corresponding fields of tags and paths using the `concat` function in Pandas to generate a set of structural semantic combination pairs.

[0029] The model configuration module includes: a parameter construction submodule: based on the set of structural semantic combinations, it constructs structural parameter combinations and assigns configuration numbers. It maps each structure to fields such as the number of nodes, number of layers, and discard ratio to form a configuration table. The field combinations are arranged in numerical order and bound to structural label indexes. This builds a record of all structural configuration combinations, generating a parameter configuration combination set. The performance evaluation submodule: based on the parameter configuration combination set, it uses a non-dominated sorting genetic algorithm. It inputs structural configuration combinations and synchronizes semantic label vectors. It calculates the response offset based on the error difference between the structural output and the label fields. It records the response time and computational resource value for each combination, extracts the combination number and performance evaluation fields to form a performance index, and generates a configuration. Performance metric set; Configuration filtering submodule: Based on the configuration performance metric set, compare the error value of the combination item, the amount of computing resources, and the response time; extract the configuration group number that is not dominated by other combinations; delete records with metric values ​​that are inferior to the combination item; update the structure configuration index list and sort the numbers; generate a structure performance configuration solution set; Parameter construction submodule: Based on the structure semantic combination pair set, construct structure parameter combinations and assign configuration numbers; construct a configuration table by mapping each structure with fields such as the number of nodes, the number of layers, and the discard ratio; organize the structure data using the DataFrame function in pandas; associate the structure with related fields; and use the merge function in pandas. The first module integrates the number of nodes, layers, and dropout ratio with other fields in the configuration table, arranges the field combinations in numerical order, sorts the field combinations using the `sort_values` function, binds the structure label index, constructs all structure configuration combination records, and generates a parameter configuration combination set. The second module, the performance evaluation module, uses a non-dominated sorting genetic algorithm based on the parameter configuration combination set. It inputs the structure configuration combinations and synchronizes semantic label vectors, calculates the error difference between the structure output and the label field using the `mean_squared_error` function in scikit-learn, obtains the response offset using the `abs` function in NumPy, and records the response of each combination. The process involves: 1) Extracting the combination number and performance evaluation field based on duration and computational resource values; 2) Grouping the data using the `groupby` function in pandas; 3) Building a performance index; and 4) Generating a configuration performance metric set. The configuration filtering submodule compares the combination item error value, computational resource quantity, and response time based on the configuration performance metric set. It uses the `where` function in NumPy for conditional filtering, extracts any configuration group number not dominated by other combinations, deletes records with metric values ​​inferior to the combination item using the `drop_duplicates` function in pandas, updates the structure configuration index list, and sorts the numbers using the `sort_index` function to generate a structure performance configuration solution set.

[0030] The non-dominated sorting genetic algorithm uses a set of parameter configuration combinations as the initial population. It calculates the fitness vector for each combination based on three metrics: error value, response time, and computational resources. The population is then hierarchically sorted according to Pareto dominance, with individuals not dominated by any combination forming the first non-dominated layer, and the remaining individuals forming the second and subsequent layers. Within each layer, crowding distance is calculated to determine the selection probability. Selection, crossover, and mutation operations are performed to generate a new generation of population. Fitness calculation and non-dominated layering are repeated until the required number of iterations or convergence conditions are met. The output is a set of structural performance configuration solutions containing multiple fits. The non-dominated sorting genetic algorithm follows the formula:

[0031] in: Indicates configuration combination The performance indicators These represent the target weighting coefficients, used to determine the importance of different performance objectives in the calculation. Indicates the first Configuration combinations and tags The error between them This represents the error correction factor, used to adjust the impact of the error on the final performance index based on its characteristics. This represents the error deviation value, used to compensate for systematic deviations caused by errors. This represents the standardized coefficient of the response error. Represents the target adjustment coefficient; Execution process: First, input each configuration combination and its corresponding semantic label vector, and then calculate the actual response. With predicted response Error between The model's adaptability and accuracy to the transmission device design are evaluated, and then the error value is corrected using an error correction factor. and error deviation value Adjustments were made to correct for the impact of errors, ensuring that the performance evaluation better reflects actual physical characteristics, and then the response error normalization coefficient was adjusted. The error impact between different performance targets is standardized to make the error impact of each target relatively consistent, and finally the target adjustment coefficient is determined. In multi-objective optimization, the relative importance of each objective is adjusted to ensure that the performance objectives are balanced during the evaluation process, through weighting coefficients. By weighting the errors of each objective, a comprehensive performance index is generated for each configuration combination. It is used to optimize the key feature extraction and automatic mapping system in the design process of transmission devices, thereby achieving precise design optimization.

[0032] The state mapping module includes: a resource filtering submodule: based on the structural performance configuration solution set, it extracts the floating-point computational cost and response time of each structural configuration, reads the system-allocated computational resource limits and constructs a comparison list, filters out configuration combinations that do not meet the resource conditions and records the remaining configurations, arranges the structural configuration sequence and completes resource adaptation matching, generating a set of available resource configurations; a label matching submodule: based on the available resource configuration set, it extracts the structural output semantic vector and compares it with historical label vectors, calculates the vector difference and extracts the label field with the highest similarity, matches the design indicators associated with the label field, organizes all structural indicators and constructs a mapping set, and obtains the structural indicator mapping inference results; the resource filtering submodule... Based on the structural performance configuration deset, the floating-point computation cost and response time of each structural configuration are extracted. The floating-point computation cost and response time data of each structural configuration are extracted using the DataFrame function in pandas. The less_equal function in numpy is used to compare with the computing resource limits allocated by the system to filter out configuration combinations that do not meet the resource conditions. The drop function in pandas is used to delete configuration records that do not meet the resource conditions, and the remaining configurations that meet the conditions are recorded. The structural configuration sequence is arranged and sorted using the argsort function in numpy to complete the resource adaptation and matching, and generate a resource-available configuration set.

[0033] The tag matching submodule extracts the structural output semantic vector based on the available resource configuration set and compares it with the historical tag vector. It uses the spatial.distance.cosine function in SciPy to calculate the vector difference, extracts the tag field with the highest similarity using the argsort function in NumPy, matches the design indicators associated with the tag field, uses the merge function in Pandas to associate the matched tag field with the design indicator field, organizes all structural indicators and constructs a mapping set, and obtains the structural indicator mapping inference results.

[0034] A method for extracting and automatically mapping key features in transmission device design simulation is presented. This method, based on the aforementioned system for extracting and automatically mapping key features in transmission device design simulation, includes the following steps: S1: Based on the transmission device design drawings, extract the component numbers of the drive shaft, gear pairs, and couplings; read the coordinates to construct node indices; calculate geometric distances and contact areas to generate edge connection data; and generate a node-connection topology table. S2: Based on the node relationship topology data, select 3 to 5 path structures; use a graph neural network to fuse node dimensions, material numbers, and edge coupling information; and combine the tensors of nodes and edges. S3: Based on the nested path vector group, calculate the semantic vector difference to extract the path group, associate the lifetime, load, and temperature rise level label fields with the path index, and generate a set of structural semantic combination pairs; S4: Based on the structural semantic binding label set, construct the structural parameter combination, use a non-dominated sorting genetic algorithm to generate a population, calculate the error, response time, and computational cost, and select the optimal configuration to generate a structural performance configuration solution set; S5: Based on the non-dominated structural configuration set, filter out out-of-limit combinations, extract the semantic vector and historical label matching results, output the structural index field, and generate the structural index mapping inference result.

[0035] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A system for extracting key features and automatically mapping them in the design simulation of transmission devices, characterized in that: The system includes: a structure construction module: based on the design drawings of the transmission device, extracting components such as the drive shaft, gear pairs, and couplings, locating component coordinates, calculating three-dimensional distances and contact parameters, establishing connection paths and direction indexes according to the assembly sequence, and generating a node-connection topology table; a nested modeling module: based on the node-connection topology table, selecting gear transmission paths of length 3 to 5, using a graph neural network, organizing component dimensions and material properties, extracting the force direction and coupling relationship between connection pairs, assembling the combination structure of nodes and edges, and generating a path nested structure representation set; and a semantic fusion module: based on the path nested structure representation set, summarizing dimensional differences and... Material coding involves comparing path similarity to identify closely related groups and associating them with lifetime, load, and temperature rise levels. Semantics and labels from each group are then fused to generate a set of structural semantic combinations. The model configuration module, based on this set of structural semantic combinations, constructs multiple sets of structural parameter configurations. A non-dominated sorting genetic algorithm is used to measure the error, response time, and computational cost of each group. Based on the results, non-dominated configurations are selected, structural records are updated, and a set of structural performance configuration solutions is generated. The state mapping module, based on this set of structural performance configuration solutions, filters out resource occupancy and response exceedance combinations, analyzes semantic output differences, compares historical structural labels, extracts and fuses matching index items, and obtains the structural index mapping inference results.

2. The transmission device design simulation key feature extraction and automatic mapping system according to claim 1, characterized in that, The node-connection topology table includes component number information, connection boundary matrix, and load transmission direction vector. The path nesting structure representation set includes node size vector, connection stiffness sequence, and structural path encoding index. The structural semantic combination set includes path feature vector, functional semantic label, and performance level identifier value. The structural performance configuration solution set includes structural layer number parameter, response delay value, and computational resource consumption item. The structural index mapping inference result includes recommended load threshold, semantic matching label, and structural output vector group.

3. The key feature extraction and automatic mapping system for transmission device design simulation according to claim 1, characterized in that, The structural construction module includes: a component identification submodule: based on the transmission device design drawings, it extracts components such as the drive shaft, gear pairs, and couplings; extracts the number field from the component model and establishes a mapping table; reads the three-dimensional center coordinates of the components and parses the axial displacement vector; extracts the boundary coordinate range and identifies the axial overlap relationship; and generates a component spatial coordinate set. A connection construction submodule: based on the component spatial coordinate set, it calculates the Euclidean distance between the center points of the components and determines the critical contact state; extracts the boundary contact surface area ratio and filters effective structural pairs; extracts the assembly sequence code and constructs the connection path sequence; marks the load transmission direction vector and completes the connection information; and generates a node-connection topology table.

4. The transmission device design simulation key feature extraction and automatic mapping system according to claim 1, characterized in that, The nested modeling module includes: a path filtering submodule: based on the node-connection topology table, filtering gear transmission paths with lengths of 3 to 5, extracting the number field from the path and sorting it to generate an index sequence, filtering out discontinuous path segments by traversing the connection edge set, and calculating the consistency between the path node arrangement direction and the connection edge order to generate a path number sequence set; a feature construction submodule: based on the path number sequence set, using a graph neural network, extracting the geometric dimension vector and material number corresponding to the components in each path segment, marking the index position of nodes and edges in the path order, extracting the force direction coordinate difference and connection stiffness value through the connection segment, splicing the node attribute field and edge attribute field to form a combined vector, and generating a structural feature combination set; a structure generation submodule: based on the structural feature combination set, constructing a structural path vector and adding path index encoding, performing a uniform length padding operation on the node vector and edge vector, extracting the sequence number identifier and corresponding combined structure position index of each path, labeling the nested relationship mapping matrix and archiving all path data, and generating a path nested structure representation set.

5. The transmission device design simulation key feature extraction and automatic mapping system according to claim 4, characterized in that, The graph neural network takes a path number sequence set as input, extracts the geometric dimension vector and material number of the nodes in the path as node features, and extracts the force direction coordinate difference and connection stiffness value of the connecting edges as edge features to construct a graph structure input tensor. The node features are propagated through adjacent edges and weighted and fused with the edge features to generate an updated node representation. Through multi-layer convolution iteration, the node representation is aggregated layer by layer to form the overall feature representation of the path. The node set representation is mapped to a path-level vector through a readout function to generate a structural feature combination set.

6. The transmission device design simulation key feature extraction and automatic mapping system according to claim 1, characterized in that, The semantic fusion module includes: a feature extraction submodule: based on the path nesting structure representation set, extracting the size features and material coding parameters of each path, constructing a path vector sequence and performing normalization operation, constructing a similarity matrix between path pairs and filtering highly similar path groups, labeling the grouping index and outputting a set of combined structures to generate a set of similar structure paths; and a label fusion submodule: based on the set of similar structure paths, extracting the lifespan level, load level, and temperature rise level labels corresponding to each group of path combinations, constructing a label ranking list and synchronizing the path number field, establishing an index mapping relationship between labels and path structures, organizing the structure path and label combination fields, and generating a set of structural semantic combination pairs.

7. The transmission device design simulation key feature extraction and automatic mapping system according to claim 1, characterized in that, The model configuration module includes: a parameter construction submodule: based on the set of structural semantic combinations, constructing structural parameter combinations and assigning configuration numbers, mapping each group of structures to fields such as the number of nodes, number of layers, and discard ratio to form a configuration table, arranging field combinations in numerical order and binding structural label indexes, constructing all structural configuration combination records, and generating a parameter configuration combination set; a performance evaluation submodule: based on the set of parameter configuration combinations, using a non-dominated sorting genetic algorithm, inputting structural configuration combinations and synchronizing semantic label vectors, calculating the response offset through the error difference between the structural output and the label fields, recording the response time and computational resource value of each combination, extracting the combination number and performance evaluation field to form a performance index, and generating a configuration performance index set; and a configuration filtering submodule: based on the set of configuration performance indicators, comparing the error value of combination items, computational resource quantity, and response time, extracting any configuration group number that is not dominated by other combinations, deleting records with indicator values ​​inferior to combination items, updating the structural configuration index list and sorting by number, and generating a structural performance configuration solution set.

8. The transmission device design simulation key feature extraction and automatic mapping system according to claim 7, characterized in that, The non-dominated sorting genetic algorithm uses a set of parameter configuration combinations as the initial population, calculates the fitness vector of each combination in terms of error value, response time, and computational resources, and sorts the population in layers according to the Pareto dominance relationship. Individuals not dominated by combinations are divided into the first non-dominated layer, and the remaining individuals form the second and subsequent layers in sequence. The crowding distance is calculated in each layer and the selection probability is determined accordingly. Selection, crossover, and mutation operations are performed to generate a new generation of population. Fitness calculation and non-dominated layering are repeated until the number of iterations or convergence conditions are met, and the output is a set of structural performance configuration solutions containing multiple sets of fit solutions.

9. The transmission device design simulation key feature extraction and automatic mapping system according to claim 1, characterized in that, The state mapping module includes: a resource filtering submodule: based on the structural performance configuration solution set, extracting the floating-point computation amount and response time of each structural configuration, reading the system-allocated computing resource limits and constructing a comparison list, filtering out configuration combinations that do not meet the resource conditions and recording the remaining configurations, arranging the structural configuration sequence and completing resource adaptation matching, and generating a resource-available configuration set; and a tag matching submodule: based on the resource-available configuration set, extracting the structural output semantic vector and comparing it with historical tag vectors, calculating the vector difference and extracting the tag field with the highest similarity, matching the design indicators associated with the tag field, organizing all structural indicators and constructing a mapping set, and obtaining the structural indicator mapping inference results.

10. A method for extracting key features and automatically mapping them in transmission device design simulation, characterized in that, The transmission device design simulation key feature extraction and automatic mapping system according to any one of claims 1-9 includes the following steps: S1: Based on the transmission device design drawings, extract the component numbers of the drive shaft, gear pair, and coupling, read the coordinates to construct the node index, calculate the geometric distance and contact area to generate edge connection data, and generate a node-connection topology table; S2: Based on the node relationship topology data, select 3 to 5 path structures, use a graph neural network to fuse node size, material number, and edge coupling information, combine the tensors of nodes and edges to form a path vector, and generate a path nesting structure representation set. S3: Based on the nested vector group of the structural path, calculate the semantic vector difference to extract the path group, associate the lifespan, load, and temperature rise level label fields with the path index, and generate a set of structural semantic combination pairs; S4: Based on the set of structural semantic binding labels, construct structural parameter combinations, use a non-dominated sorting genetic algorithm to generate a population, calculate the error, response time, and computational cost, and select the optimal configuration to generate a set of structural performance configuration solutions; S5: Based on the set of non-dominated structural configurations, filter out out-of-limit combinations, extract the semantic vector and historical label matching results, output the structural index fields, and generate structural index mapping inference results.

Images