Corrosion simulation system based on neural network

Abstract

The corrosion simulation system based on the neural network comprises a data acquisition unit, a whole vehicle network unit, a sub-network unit, a chain connection unit and a chain screening unit, and the data acquisition unit acquires whole vehicle historical corrosion test data and current data of vehicle driving; the historical corrosion test data of the whole vehicle can be used for training the neural network of the whole vehicle, the neural network of the whole vehicle can be divided into a plurality of sub-networks, the current data can be primarily processed through the neural network of the whole vehicle to obtain corrosion areas, the sub-networks are in one-to-one correspondence with the corrosion areas, and the current data are further processed by the matched sub-networks to obtain the corrosion areas. The method comprises the following steps: determining a corrosion component according to the sub-networks, constructing a corrosion chain according to the relevance between the sub-networks, simulating the corrosion evolution, introducing a sparrow search algorithm to screen out the corrosion chain with the maximum corrosion conduction probability, and obtaining the corrosion change of the vehicle under long-time driving according to the corrosion chain. Therefore, the vehicle can be designed, optimized and improved according to the corrosion change.

Description

Technical Field

[0001] This invention relates to the field of vehicle testing technology, and in particular to a corrosion simulation system based on neural networks. Background Technology

[0002] Automotive corrosion is a key factor affecting the performance stability, driving safety, and service life of vehicle components. Its occurrence is closely related to vehicle structural design, component material properties, environmental conditions in the region of use, and driving conditions. Currently, most automotive corrosion detection and simulation rely on long-term real-vehicle testing, simulating the corrosion process and collecting data under dynamic driving conditions. This method is not only time-consuming and costly, but also difficult to adapt to the fast-paced R&D strategies of current automakers. Although some corrosion simulation solutions combining intelligent algorithms have emerged in the existing technology, these solutions mostly lack global and local collaborative design, fail to construct sub-networks based on the corrosion laws of the whole vehicle, and are difficult to effectively explore the structural relationships and corrosion propagation paths between components. They cannot form a complete corrosion chain and lack a priority screening mechanism for multiple potential corrosion paths, resulting in insufficient accuracy of corrosion simulation and failing to meet the actual needs of protective design optimization in the automotive R&D stage and corrosion early warning in the operation and maintenance stage. Summary of the Invention

[0003] In view of this, the present invention proposes a corrosion simulation system based on neural networks, which can not only perform corrosion simulation, but also form corrosion chains, obtain corrosion evolution laws, and support corrosion protection decisions in automobile R&D and maintenance.

[0004] The technical solution of this invention is implemented as follows: A corrosion simulation system based on neural networks includes: The data acquisition unit is used to acquire historical corrosion test data of the entire vehicle during driving, as well as current data. The vehicle network unit has a built-in vehicle neural network trained based on historical corrosion test data of the vehicle. The vehicle neural network processes the current data and outputs the corrosion area. Sub-network units are used to segment the whole vehicle neural network and obtain several sub-networks. The sub-networks corresponding to the eroded areas preprocess the current data to obtain the eroded parts. The chain connection unit is used to connect other subnetworks, starting from the subnetwork corresponding to the corrosion area, to form a corrosion chain. The chain screening unit has a built-in sparrow search algorithm to screen corrosion chains and obtain the corrosion chains with the highest corrosion conduction probability. The data acquisition unit, vehicle network unit, sub-network unit, chain connection unit, and chain screening unit are sequentially connected in terms of data, and the sub-network unit is connected in terms of data to the data acquisition unit.

[0005] Preferably, the execution steps of the data acquisition unit include: Collect historical corrosion test data of the whole vehicle during actual vehicle testing. The historical corrosion test data of the whole vehicle includes material process data, driving condition data, previous corrosion detection records and environmental characteristic data of the test area. The vehicle under test collects current data through onboard sensors and a cloud platform. This current data includes current mileage, environmental parameters, component working status data, and operating condition data.

[0006] Preferably, it also includes a data processing unit, which is used to preprocess the historical corrosion test data of the whole vehicle and the current data. The preprocessing includes missing value filling, outlier removal and format standardization. The data processing unit is connected to the data acquisition unit, the whole vehicle network unit and the sub-network unit respectively.

[0007] Preferably, the execution steps of the vehicle network unit include: The historical corrosion test data of the whole vehicle were labeled and divided into training set and test set according to the proportion; The vehicle neural network is trained using a training set and tested using a test set to obtain the trained vehicle neural network. The current data is input into the vehicle's neural network, which outputs the confidence level of different regions of the vehicle. The region with the highest confidence level is output as the erosion region.

[0008] Preferably, the execution step of the sub-network unit segmenting the whole vehicle neural network and obtaining several sub-networks is as follows: Retrieve vehicle design manuals and safety standards, and formulate vehicle component division rules, which include dividing the vehicle into several independent areas according to safety priority, structural correlation, and corrosion characteristics; The network hierarchy, feature extraction module, and output node of the vehicle neural network are analyzed in a structured manner to clarify the vehicle feature learning function of each module. Establish a mapping relationship between vehicle component division rules and vehicle neural network modules, and determine the vehicle neural network feature extraction module and output branch corresponding to each vehicle region; Based on the mapping relationship, the whole vehicle neural network is modularly cut to separate independent network architectures that correspond one-to-one with each vehicle region. Each split, independent network architecture is configured with a dedicated input interface and a dedicated output layer for corrosion detection, adapting to the feature data format of the corresponding vehicle region.

[0009] Preferably, the sub-network unit obtains the eroded component through the following steps: Based on the erosion area output by the vehicle network unit, the corresponding target subnetwork is matched; The target sub-network receives the current data and performs feature extraction to obtain core feature data related to corrosion detection; The core feature data is processed by feature filtering, dimensionality reduction and format adaptation to obtain preprocessed data that meets the input requirements of the subnetwork. Corrosion status detection is performed on the preprocessed data to locate corroded components.

[0010] Preferably, the execution steps of the chain connection unit are as follows: The subnetwork corresponding to the corrosion area output by the whole vehicle network unit is used as the starting subnetwork of the corrosion chain; Screen the associated subnetworks that have structural connections or corrosion propagation paths with the corroded components corresponding to the starting subnetwork; Based on the physical rules of corrosion propagation, connection conditions between sub-networks are set, including component spatial distance thresholds and the feasibility of corrosion medium conduction. According to the connection conditions, the starting subnetwork and the associated subnetwork are connected in sequence to form multiple erosion chains.

[0011] Preferably, the specific steps for screening the associated subnetworks that have structural correlation or corrosion propagation paths with the corroded components corresponding to the starting subnetwork are as follows: The cosine similarity algorithm is used to calculate the structural feature matching degree between the eroded component corresponding to the starting sub-network and other components; The Pearson correlation coefficient algorithm is used to analyze the correlation between the corrosion propagation medium and corrosion type of the corrosion component corresponding to the starting sub-network and other components, and obtain the corrosion propagation correlation score. The overall correlation between components is obtained by weighted summation of the structural feature matching degree and the corrosion propagation correlation score. Subnetworks with a comprehensive correlation degree between components that is greater than the comprehensive correlation degree threshold are output as associated subnetworks.

[0012] Preferably, the execution steps of the chain screening unit are as follows: Each corrosion chain is used as an initial population individual, and a chain fitness function is constructed based on corrosion propagation probability, node corrosion confidence, and component safety weight; The Sparrow Search algorithm calculates the fitness value of each individual in the population and selects individuals with high fitness as discoverers based on the fitness values. The discoverer updates its position using a random search strategy, and the remaining individuals in the population act as followers, mimicking the path characteristics of the discoverer to update their own positions. The sparrow search algorithm randomly selects a portion of the population as watchdogs, checks whether they are trapped in a local optimum, and if so, reinitializes the individual. Repeat the optimization process until the preset number of iterations is reached or the fitness value tends to stabilize. Then, output the corrosion chain with the highest fitness value. This chain has the highest corrosion propagation probability.

[0013] Compared with the prior art, the beneficial effects of the present invention are: The data acquisition unit can acquire historical corrosion test data and current data of the vehicle during driving. The historical corrosion test data can be used by the vehicle network unit to train the vehicle neural network. The vehicle neural network can initially identify the corrosion area of ​​the vehicle based on the current data to achieve coarse localization. After determining the corrosion area, the corresponding sub-network can be matched from the sub-networks segmented by the vehicle neural network. Then, the sub-network can further refine the localization of the current data and obtain the corroded parts, which are the most likely locations for corrosion. Then, corrosion chains are constructed through the correlation between sub-networks, and a sparrow search algorithm is introduced to filter the corrosion chains to obtain the corrosion chains with the highest corrosion propagation probability. This realizes the whole process of corrosion simulation from global corrosion pattern mining to accurate local corrosion localization, and then to priority screening of cross-component corrosion propagation paths. It avoids the defects of long cycle and high cost of real vehicle testing, and solves the problems of existing intelligent algorithms lacking global and local coordination and difficulty in forming complete corrosion chains and path screening. It significantly improves the accuracy, efficiency and practicality of corrosion simulation and can effectively support protection optimization in the automotive R&D stage and corrosion early warning in the operation and maintenance stage. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a corrosion simulation system based on a neural network according to the present invention. In the diagram, 1 is the data acquisition unit; 2 is the vehicle network unit; 3 is the sub-network unit; 4 is the chain connection unit; 5 is the chain screening unit; and 6 is the data processing unit. Detailed Implementation

[0016] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0017] See Figure 1 The present invention provides a corrosion simulation system based on a neural network, comprising: Data acquisition unit 1 is used to acquire historical corrosion test data of the whole vehicle during vehicle operation and current data; The vehicle network unit 2 has a built-in vehicle neural network trained based on historical corrosion test data of the whole vehicle. The vehicle neural network processes the current data and outputs the corrosion area. Sub-network unit 3 is used to segment the whole vehicle neural network and obtain several sub-networks. The sub-network corresponding to the eroded area preprocesses the current data to obtain the eroded parts. Chain connection unit 4 is used to connect other sub-networks from the sub-network corresponding to the corrosion area to form a corrosion chain; Chain screening unit 5 has a built-in sparrow search algorithm to screen corrosion chains and obtain the corrosion chains with the highest corrosion conduction probability; The data acquisition unit 1, the vehicle network unit 2, the sub-network unit 3, the chain connection unit 4, and the chain screening unit 5 are connected in sequence, and the sub-network unit 3 is connected to the data acquisition unit 1.

[0018] This invention discloses a neural network-based corrosion simulation system for simulating automotive corrosion. Unlike traditional long-term real-vehicle testing, this system employs a neural network for intelligent and efficient simulation. Before corrosion simulation, data acquisition unit 1 collects historical corrosion test data of the entire vehicle, including extensive test data from previous real-vehicle corrosion tests. This test data can be used to train the vehicle neural network. After inputting the historical corrosion test data into vehicle network unit 2, vehicle network unit 2 can label and classify the data, which can then be used to construct the vehicle neural network. The input consists of relevant data from vehicle operation, and the output is the corrosion area, such as different areas like the chassis, sheet metal, and sensors. The whole-vehicle neural network can perform a coarse assessment of corrosion location. After inputting the current data into the whole-vehicle neural network, the corresponding corrosion area of ​​the vehicle can be obtained. Fine assessment is performed through sub-networks. Sub-network unit 3 is used to segment the whole-vehicle neural network to obtain several sub-networks, each corresponding to a different corrosion area of ​​the vehicle. After inputting the current data into a sub-network, the sub-network can perform a fine assessment to identify the corroded component. The corroded component is the component with the highest probability of corrosion, allowing for reasonable improvements or replacements of that component and surrounding components, thus optimizing the vehicle design.

[0019] During long-term vehicle operation, corrosion does not occur only in individual components. Corrosion in some components can gradually spread to other areas. Therefore, after identifying the corroded component, the chain connection unit 4 uses the sub-network corresponding to the corroded component as a starting point to establish connections with other sub-networks, thus forming a corrosion chain. This corrosion chain represents the corrosion propagation path that occurs during long-term vehicle operation. However, different corrosion chains have different propagation probabilities, and the correlation between some sub-networks is poor. Therefore, it is necessary to screen the corrosion chains. The chain screening unit 5 has a pre-installed sparrow search algorithm that can screen corrosion chains to obtain the one with the highest corrosion propagation probability. Based on this corrosion chain, comprehensive optimization and improvement of the vehicle design can be performed, thereby extending the vehicle's service life and supporting corrosion protection decisions in automotive R&D and maintenance.

[0020] Preferably, the execution steps of the data acquisition unit 1 include: Collect historical corrosion test data of the whole vehicle during actual vehicle testing. The historical corrosion test data of the whole vehicle includes material process data, driving condition data, previous corrosion detection records and environmental characteristic data of the test area. The vehicle under test collects current data through onboard sensors and a cloud platform. This current data includes current mileage, environmental parameters, component working status data, and operating condition data.

[0021] The data acquisition unit 1 is used to acquire historical corrosion test data and current data of the whole vehicle. The historical corrosion test data of the whole vehicle are the relevant parameters collected when the vehicle was previously subjected to actual corrosion tests, including the material parameters of the vehicle itself, such as surface treatment, substrate and burrs, etc. It also includes the previous corrosion detection records recorded during the test. Since different environmental characteristics have different effects on corrosion, it is also necessary to record the environmental characteristic data during the test.

[0022] Historical corrosion test data of the whole vehicle is used to train the whole vehicle neural network, while the current data is used to perform corrosion simulation of the current vehicle. The current data of the test vehicle is obtained so that the whole vehicle neural network and the sub-network can identify the corrosion area and corrosion component respectively. The current data includes the mileage of the test vehicle in kilometers, and also collects the environmental parameters of the test vehicle and the working status data of various components such as the engine and battery, for use by sub-network unit 3.

[0023] Preferably, it also includes a data processing unit 6, which is used to preprocess the historical corrosion test data of the whole vehicle and the current data. The preprocessing includes missing value filling, outlier removal and format standardization. The data processing unit 6 is connected to the data acquisition unit 1, the whole vehicle network unit 2 and the sub-network unit 3 respectively.

[0024] The data processing unit 6 is used to preprocess the historical corrosion test data and current data of the whole vehicle, so that the whole vehicle neural network and sub-network can process them.

[0025] Preferably, the execution steps of the vehicle network unit 2 include: The historical corrosion test data of the whole vehicle were labeled and divided into training set and test set according to the proportion; The vehicle neural network is trained using a training set and tested using a test set to obtain the trained vehicle neural network. The current data is input into the vehicle's neural network, which outputs the confidence level of different regions of the vehicle. The region with the highest confidence level is output as the erosion region.

[0026] After acquiring the historical corrosion test data of the whole vehicle, the data is labeled to determine the specific partition where the corrosion location is located. Then, the historical corrosion test data of the whole vehicle is divided into a training set and a test set in a 7:3 ratio. The training set is used to train the whole vehicle neural network. After the training converges, the accuracy is tested through the test set. Once the test accuracy reaches a preset threshold, the training can be stopped, thus obtaining the trained whole vehicle neural network. Then, the current data is used as the input of the whole vehicle neural network, and the whole vehicle neural network will output the confidence level of different areas of the vehicle. The confidence levels of all areas are compared, and the area with the highest confidence level is output as the corrosion area.

[0027] Preferably, the execution steps of the sub-network unit 3 segmenting the whole vehicle neural network and obtaining several sub-networks are as follows: Retrieve vehicle design manuals and safety standards, and formulate vehicle component division rules, which include dividing the vehicle into several independent areas according to safety priority, structural correlation, and corrosion characteristics; The network hierarchy, feature extraction module, and output node of the vehicle neural network are analyzed in a structured manner to clarify the vehicle feature learning function of each module. Establish a mapping relationship between vehicle component division rules and vehicle neural network modules, and determine the vehicle neural network feature extraction module and output branch corresponding to each vehicle region; Based on the mapping relationship, the whole vehicle neural network is modularly cut to separate independent network architectures that correspond one-to-one with each vehicle region. Each split, independent network architecture is configured with a dedicated input interface and a dedicated output layer for corrosion detection, adapting to the feature data format of the corresponding vehicle region.

[0028] The first function of sub-network unit 3 is to segment the vehicle neural network. The segmented sub-network corresponds to a region of the vehicle. Therefore, the segmentation of the vehicle neural network needs to be based on the region division of the vehicle. First, according to the vehicle connectivity design manual and safety standards, the segmentation rules are preset to divide the vehicle into several independent regions. The segmentation rules include various types, such as the safety priority of different components, the correlation between components, and different corrosion characteristics. After determining the several independent regions of the vehicle, the vehicle neural network can be segmented. First, the vehicle neural network is structured and analyzed to determine the vehicle feature learning function corresponding to each module. Then, the mapping relationship between the vehicle component segmentation rules and the vehicle neural network modules is established. Finally, the vehicle neural network is segmented according to the mapping relationship to obtain an independent network architecture corresponding to each independent region. Then, a dedicated input and output terminal is configured for each independent network architecture to obtain a sub-network adapted to each independent region.

[0029] Preferably, the sub-network unit 3 obtains the eroded component through the following steps: Based on the erosion area output by the vehicle network unit 2, the corresponding target subnetwork is matched; The target sub-network receives the current data and performs feature extraction to obtain core feature data related to corrosion detection; The core feature data is processed by feature filtering, dimensionality reduction and format adaptation to obtain preprocessed data that meets the input requirements of the subnetwork. Corrosion status detection is performed on the preprocessed data to locate corroded components.

[0030] After the whole vehicle neural network processes the eroded area and divides it into sub-networks, it finds the corresponding sub-network based on the obtained eroded area, and then uses the current data as the input of the sub-network. The sub-network can perform feature extraction to obtain core feature data. After preprocessing the core feature data, it performs erosion state detection and locates the eroded parts.

[0031] Preferably, the execution steps of the chain connection unit 4 are as follows: The sub-network corresponding to the corrosion area output by the vehicle network unit 2 is used as the starting sub-network of the corrosion chain; Screen the associated subnetworks that have structural connections or corrosion propagation paths with the corroded components corresponding to the starting subnetwork; Based on the physical rules of corrosion propagation, connection conditions between sub-networks are set, including component spatial distance thresholds and the feasibility of corrosion medium conduction. According to the connection conditions, the starting subnetwork and the associated subnetwork are connected in sequence to form multiple erosion chains.

[0032] After identifying the corroded component, a corrosion chain needs to be established. Since the corrosion of the component occurs fastest, the sub-network corresponding to the corroded component can be used as the starting point, and connections can be established with other sub-networks to form a corrosion chain. However, not all sub-networks meet the conditions, so they need to be screened. The screening rule is to select other sub-networks that have a structural relationship with the corroded component or a corrosion propagation path, and output the corresponding sub-networks as associated sub-networks. Finally, the sub-networks are connected according to the physical rules of corrosion propagation to form a corrosion chain.

[0033] Preferably, the specific steps for screening the associated subnetworks that have structural correlation or corrosion propagation paths with the corroded components corresponding to the starting subnetwork are as follows: The cosine similarity algorithm is used to calculate the structural feature matching degree between the eroded component corresponding to the starting sub-network and other components; The Pearson correlation coefficient algorithm is used to analyze the correlation between the corrosion propagation medium and corrosion type of the corrosion component corresponding to the starting sub-network and other components, and obtain the corrosion propagation correlation score. The overall correlation between components is obtained by weighted summation of the structural feature matching degree and the corrosion propagation correlation score. Subnetworks with a comprehensive correlation degree between components that is greater than the comprehensive correlation degree threshold are output as associated subnetworks.

[0034] When screening related subnetworks, two conditions need to be met: the existence of structural correlation or corrosion propagation path. The cosine similarity algorithm is used to calculate the structural feature matching degree between the corroded component corresponding to the starting subnetwork and other components. The Pearson correlation coefficient algorithm is used to quantify the corrosion propagation path. By analyzing the correlation between the corroded component corresponding to the starting subnetwork and other components in terms of corrosion propagation medium and corrosion type, a corrosion propagation correlation score can be obtained. Finally, the structural feature matching degree and the corrosion propagation correlation score are weighted and summed to obtain the comprehensive correlation degree between components. Subnetworks with a comprehensive correlation degree between components greater than the comprehensive correlation degree threshold are identified as related subnetworks, thus enabling the connection of corrosion chains.

[0035] Preferably, the execution steps of the chain screening unit 5 are as follows: Each corrosion chain is used as an initial population individual, and a chain fitness function is constructed based on corrosion propagation probability, node corrosion confidence, and component safety weight; The Sparrow Search algorithm calculates the fitness value of each individual in the population and selects individuals with high fitness as discoverers based on the fitness values. The discoverer updates its position using a random search strategy, and the remaining individuals in the population act as followers, mimicking the path characteristics of the discoverer to update their own positions. The sparrow search algorithm randomly selects a portion of the population as watchdogs, checks whether they are trapped in a local optimum, and if so, reinitializes the individual. Repeat the optimization process until the preset number of iterations is reached or the fitness value tends to stabilize. Then, output the corrosion chain with the highest fitness value. This chain has the highest corrosion propagation probability.

[0036] Using an effective corrosion chain as the initial population ensures the effectiveness of the optimization starting point. Then, after constructing a reasonable fitness function, the optimization speed is improved through the division of labor and cooperation of discoverers exploring new paths, followers optimizing and iterating, and vigilants avoiding local optima, thus ensuring the global optimality of the result. Finally, the output is stabilized by setting a preset iteration termination condition. This method can quickly and accurately locate the core chain with the highest corrosion transmission probability, providing a clear priority basis for vehicle corrosion protection design and operation and maintenance, and greatly enhancing the decision support value of corrosion simulation.

[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A corrosion simulation system based on neural networks, characterized in that, include: The data acquisition unit is used to acquire historical corrosion test data of the entire vehicle during driving, as well as current data. The vehicle network unit has a built-in vehicle neural network trained based on historical corrosion test data of the vehicle. The vehicle neural network processes the current data and outputs the corrosion area. Sub-network units are used to segment the whole vehicle neural network and obtain several sub-networks. The sub-networks corresponding to the eroded areas preprocess the current data to obtain the eroded parts. The chain connection unit is used to connect other subnetworks, starting from the subnetwork corresponding to the corrosion area, to form a corrosion chain. The chain screening unit has a built-in sparrow search algorithm to screen corrosion chains and obtain the corrosion chains with the highest corrosion conduction probability. The data acquisition unit, vehicle network unit, sub-network unit, chain connection unit, and chain screening unit are sequentially connected in terms of data, and the sub-network unit is connected in terms of data to the data acquisition unit.

2. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The execution steps of the data acquisition unit include: Collect historical corrosion test data of the whole vehicle during actual vehicle testing. The historical corrosion test data of the whole vehicle includes material process data, driving condition data, previous corrosion detection records and environmental characteristic data of the test area. The vehicle under test collects current data through onboard sensors and a cloud platform. This current data includes current mileage, environmental parameters, component working status data, and operating condition data.

3. The corrosion simulation system based on a neural network according to claim 1, characterized in that, It also includes a data processing unit, which is used to preprocess the historical corrosion test data of the whole vehicle and the current data. The preprocessing includes missing value filling, outlier removal and format standardization. The data processing unit is connected to the data acquisition unit, the whole vehicle network unit and the sub-network unit respectively.

4. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The execution steps of the vehicle network unit include: The historical corrosion test data of the whole vehicle were labeled and divided into training set and test set according to the proportion; The vehicle neural network is trained using a training set and tested using a test set to obtain the trained vehicle neural network. The current data is input into the vehicle's neural network, which outputs the confidence level of different regions of the vehicle. The region with the highest confidence level is output as the erosion region.

5. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The execution steps for the sub-network unit to segment the whole vehicle neural network and obtain several sub-networks are as follows: Retrieve vehicle design manuals and safety standards, and formulate vehicle component division rules, which include dividing the vehicle into several independent areas according to safety priority, structural correlation, and corrosion characteristics; The network hierarchy, feature extraction module, and output node of the vehicle neural network are analyzed in a structured manner to clarify the vehicle feature learning function of each module. Establish a mapping relationship between vehicle component division rules and vehicle neural network modules, and determine the vehicle neural network feature extraction module and output branch corresponding to each vehicle region; Based on the mapping relationship, the whole vehicle neural network is modularly cut to separate independent network architectures that correspond one-to-one with each vehicle region. Each split, independent network architecture is configured with a dedicated input interface and a dedicated output layer for corrosion detection, adapting to the feature data format of the corresponding vehicle region.

6. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The sub-network unit obtains the eroded component through the following steps: Based on the erosion area output by the vehicle network unit, the corresponding target subnetwork is matched; The target sub-network receives the current data and performs feature extraction to obtain core feature data related to corrosion detection; The core feature data is processed by feature filtering, dimensionality reduction and format adaptation to obtain preprocessed data that meets the input requirements of the subnetwork. Corrosion status detection is performed on the preprocessed data to locate corroded components.

7. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The execution steps of the chain connection unit are as follows: The subnetwork corresponding to the corrosion area output by the whole vehicle network unit is used as the starting subnetwork of the corrosion chain; Screen the associated subnetworks that have structural connections or corrosion propagation paths with the corroded components corresponding to the starting subnetwork; Based on the physical rules of corrosion propagation, connection conditions between sub-networks are set, including component spatial distance thresholds and the feasibility of corrosion medium conduction. According to the connection conditions, the starting subnetwork and the associated subnetwork are connected in sequence to form multiple erosion chains.

8. The corrosion simulation system based on a neural network according to claim 7, characterized in that, The specific steps for screening the associated subnetworks that have structural correlations or corrosion propagation paths with the corroded components corresponding to the starting subnetwork are as follows: The cosine similarity algorithm is used to calculate the structural feature matching degree between the eroded component corresponding to the starting sub-network and other components; The Pearson correlation coefficient algorithm is used to analyze the correlation between the corrosion propagation medium and corrosion type of the corrosion component corresponding to the starting sub-network and other components, and obtain the corrosion propagation correlation score. The overall correlation between components is obtained by weighted summation of the structural feature matching degree and the corrosion propagation correlation score. Subnetworks with a comprehensive correlation degree between components that is greater than the comprehensive correlation degree threshold are output as associated subnetworks.

9. The corrosion simulation system based on a neural network according to claim 1, characterized in that, The execution steps of the chain screening unit are as follows: Each corrosion chain is used as an initial population individual, and a chain fitness function is constructed based on corrosion propagation probability, node corrosion confidence, and component safety weight; The Sparrow Search algorithm calculates the fitness value of each individual in the population and selects individuals with high fitness as discoverers based on the fitness values. The discoverer updates its position using a random search strategy, and the remaining individuals in the population act as followers, mimicking the path characteristics of the discoverer to update their own positions. The sparrow search algorithm randomly selects a portion of the population as watchdogs, checks whether they are trapped in a local optimum, and if so, reinitializes the individual. Repeat the optimization process until the preset number of iterations is reached or the fitness value tends to stabilize. Then, output the corrosion chain with the highest fitness value. This chain is the corrosion chain with the highest corrosion propagation probability.

Images