Fabricated culvert structure parameter deep learning optimization system and method thereof

Abstract

The invention relates to the technical field of fabricated culverts, in particular to a fabricated culvert structure parameter deep learning optimization system and a method thereof, which integrate multi-module technologies such as section parameterization, finite element modeling, neural network training, digital twinning and digital plotting. Classification and parameterization analysis are carried out for different culvert types, a component template is created, a finite element framework is used for establishing a model, and boundary conditions and loads are designed. Through neural network training, establishing a parameterized sample library, realizing extraction and real-time monitoring of geometric structure characteristic parameters, reinforcing learning of physical constraints, and synchronous updating of a physical model; the system is also integrated with Openapi to develop an AutoCAD interface, modularly develops a CAD file, and automatically generates an assembly type BIM model and a construction drawing by using a BIM model family library; the problems of low efficiency, insufficient optimization and link splitting of traditional design are effectively solved, the design efficiency and quality are improved, and a new way is developed for innovative design of the culvert structure.

Description

Technical Field

[0001] The present invention relates to the technical field of prefabricated culverts, in particular to a deep learning optimization system and method for the structural parameters of prefabricated culverts. Background Art

[0002] With the acceleration of the urbanization process and the increasingly extensive utilization of underground space, as an important municipal engineering facility, the prefabricated culvert structure plays an increasingly important role in the fields of urban drainage, underground passages, etc. The traditional design method of culvert structures mainly relies on the experience of engineers and manual calculations. Although this method is reliable, it is inefficient and difficult to fully consider all possible optimization schemes, often resulting in material waste and increased costs.

[0003] In recent years, with the development of computer technology, parametric design and optimization algorithms have been applied to a certain extent in the design of culvert structures. These methods can improve the design efficiency to a certain extent, but there are still some limitations. For example, although conventional parametric design software can perform a certain degree of automated design, its optimization algorithm is relatively simple and difficult to handle complex non-linear problems in the design of culvert structures. In addition, these methods lack learning and adaptation capabilities and are difficult to make full use of historical design experience and data.

[0004] On the other hand, the existing design methods of culvert structures often separate the links of structural analysis, optimization design, CAD drawing, and BIM modeling. There are lags and inconsistencies in data transfer and integration between these links, affecting the overall efficiency and quality of the design. Especially when dealing with complex corner culverts or culvert structures with special shapes, the existing methods are difficult to quickly and accurately perform design optimization and performance evaluation.

[0005] In addition, with the popularization of prefabricated buildings, the design of culvert structures also needs to consider the standardization and modularization of precast components, as well as the convenience of on-site assembly. However, the existing design methods do not consider this aspect sufficiently and are difficult to achieve the overall optimization of prefabricated culvert structures. Summary of the Invention

[0006] In the face of these technical problems, the present invention proposes a deep learning optimization system for the structural parameters of prefabricated culverts. The system aims to solve the problems of low efficiency, insufficient optimization, and fragmentation of each link in the existing culvert structure design, and realize the intelligentization and automation of culvert structure design.

[0007] The present invention proposes a deep learning optimization system for the structural parameters of prefabricated culverts, including:

[0008] A cross-section parameterization module, configured to:

[0009] Perform classification and parametric analysis according to different types of culverts;

[0010] Create component templates using different component parameters;

[0011] A finite element modeling module, connected to the cross-section parameterization module, for:

[0012] Establish a culvert model using a finite element framework and set material parameters;

[0013] Design boundary conditions and load forms according to structural design requirements;

[0014] Generate optimization objectives and constraint conditions;

[0015] A neural network training module, connected to the finite element modeling module, for:

[0016] Establish a cross-section parameterization sample library;

[0017] Train the neural network to obtain a trained neural network;

[0018] A digital twin module, connected to the neural network training module, for:

[0019] Extract the characteristic parameters of the geometric structure and construct a digital twin model;

[0020] Monitor key geometric parameters in real time during the iteration process;

[0021] Perform reinforcement learning of physical constraints and drive the physical model to update synchronously;

[0022] A digital drawing module, connected to the digital twin module, for:

[0023] Complete the API interface development of AutoCAD using Openapi;

[0024] Conduct modular development on CAD files to obtain standardized CAD;

[0025] A BIM model drawing engine module, connected to the digital drawing module, for:

[0026] Automatically generate an assembled BIM model using the family library of the assembled BIM model;

[0027] Generate construction drawings.

[0028] Preferably, it further includes a model intelligent optimization module, and the model intelligent optimization module includes:

[0029] A geometric parameter calculation unit, for:

[0030] Pre-determine the structural form of the current culvert and select parameters according to geometric characteristics;

[0031] Use the digital twin model to calculate the current structural mechanical properties and structural characteristics.

[0032] Preferably, it further includes a CAD automatic generation module, and the CAD automatic generation module includes:

[0033] A BIM model construction unit for:

[0034] Screen components based on the BIM model according to the input basic structure information;

[0035] Construct a BIM model;

[0036] A data extraction unit, connected to the BIM model construction unit, for:

[0037] Quickly identify model nodes using Openapi;

[0038] Extract data in the model based on the Openapi module;

[0039] A structure optimization unit, connected to the data extraction unit, for:

[0040] Return the obtained target data information to the structure simulation module;

[0041] Drive the automatic optimization of structural parameters to complete the iterative optimization process.

[0042] Preferably, it further includes a model performance evaluation and optimization iteration module, and the model performance evaluation and optimization iteration module includes:

[0043] A model verification unit for:

[0044] Based on the digital twin model, establish a model verification scheme;

[0045] Automatically complete the comparison and analysis of the model simulation results before and after optimization;

[0046] A design iteration unit, connected to the model verification unit, for:

[0047] Based on the model performance evaluation, perform design iteration updates on the intelligent optimization module;

[0048] A BIM model driving unit, connected to the design iteration unit, for:

[0049] Based on the obtained model parameters, drive the BIM model to automatically generate drawings and analyze;

[0050] Machine according to the optimized CAD drawings and complete the assembly of the model.

[0051] Preferably, in the cross-section parameterization module, classification and parametric analysis are carried out according to different types of culverts, including straight culvert parameterization, corner culvert parameterization, and rectangular culvert parameterization.

[0052] Preferably, in the finite element modeling module, the finite element framework is used to establish the culvert model and set the material parameters. The specific steps include:

[0053] Based on the cross-section parameterization module, a parametric finite element cross-section is created for parametric modeling of the culvert structure;

[0054] After the model is established, set the model cross-section, the number of steel bars, the reinforcement ratio, etc.;

[0055] Setting of material parameters, the concrete and reinforcement materials of the culvert are set based on parameters;

[0056] Setting of boundary conditions, which are set using the finite element cell attributes;

[0057] Setting of loads, establishing load cases, and performing loading calculations using the finite element loading process.

[0058] Preferably, the neural network training module includes:

[0059] A sample database establishment unit for:

[0060] The sample data set obtained by parameterizing the finite element cells;

[0061] A neural network training unit, connected to the sample database establishment unit, for:

[0062] Training the neural network using the sample database;

[0063] Constructing a digital twin model;

[0064] Using the reinforcement learning algorithm to obtain the optimal neural network constructed.

[0065] Preferably, in the finite element modeling module, a parametric unit sample library is established based on the parameterization module, and the parametric unit sample library is imported into the finite element for stress, deformation, and mass analysis to obtain a sample parameter library.

[0066] Preferably, it further includes a calculation result visualization module for visually displaying the calculation results.

[0067] The deep learning optimization method for the structure parameters of prefabricated culverts includes the following steps:

[0068] Step 1, parameter initialization:

[0069] Preset design parameters, including the dimensions of components, the design scope of the culvert, the threshold of the number of culvert calculation times, the number of components, and the optimization target value;

[0070] Step 2: Obtain the optimal finite element model using the genetic algorithm. The specific steps are as follows:

[0071] Perform mechanical simulation using the finite element method to solve for the stress distribution of the culvert;

[0072] Select models with good fitness to enter the next round of mechanical simulation. The selection method uses the roulette wheel method;

[0073] Optimize the structural parameters through the genetic algorithm. The optimized parameters include: the type of components, the cross-sectional shape of components, the reinforcement ratio, the steel bar spacing, the stirrup spacing, the number of longitudinal steel bars, and the number of stirrups;

[0074] Compare the calculated stress value with the target to determine whether the design requirements are met;

[0075] Step 3: Determine whether the number of mechanical simulation times exceeds the threshold;

[0076] If it has exceeded the threshold, end the mechanical simulation;

[0077] If it has not exceeded the threshold, enter Step 4;

[0078] Step 4: Determine whether the convergence condition is satisfied:

[0079] If it is satisfied, end the mechanical simulation;

[0080] If it is not satisfied, re-enter Step 1, and use the adaptive moment estimation algorithm to construct the objective function and constraint conditions according to the optimization objective and constraint conditions, and search for and solve the optimal parameters.

[0081] The system of the present invention constructs a complete intelligent design ecosystem for culvert structures by integrating multiple functional modules such as cross-section parameterization, finite element modeling, neural network training, digital twin, digital drawing, and BIM model drawing. This holistic solution brings beneficial effects in many aspects:

[0082] Firstly, from a macroscopic perspective, the system of the present invention realizes the intelligence and automation of the entire process of culvert structure design. Through deep learning and digital twin technologies, the system can quickly and accurately capture the complex non-linear relationship between culvert structure parameters and performance, so as to find the optimal solution among a large number of possible solutions. This not only greatly improves the design efficiency, but also can obtain a better design solution, effectively reducing the project cost.

[0083] Secondly, from the perspective of the coordination of each module, the system of the present invention realizes seamless connection of all links in the design process. The cross-section parameterization module provides a flexible design space for subsequent optimization; the finite element modeling module ensures the accuracy of structural analysis; the neural network training module endows the system with powerful learning and prediction capabilities. The digital twin module realizes real-time synchronization of the virtual model and the actual structure, greatly improving the reliability of the optimization process. The digital drawing and BIM model drawing modules directly transform the optimization results into available engineering drawings and models, greatly improving the practicality of the design results.

[0084] From a microscopic level, the system of the present invention shows significant advantages in dealing with complex culvert structures. For example, for corner culverts or culverts with special shapes, the system can quickly generate and evaluate multiple possible solutions through parametric modeling and deep learning to find the optimal design solution. At the same time, the reinforcement learning algorithm of the system can continuously learn and adjust during the optimization process to adapt to different design requirements and constraints.

[0085] In addition, the system of the present invention also fully considers the characteristics of prefabricated buildings. Through the BIM model drawing engine module, the system can automatically generate component models and construction drawings that meet the requirements of prefabrication, providing precise guidance for the production and on-site assembly of precast components.

[0086] Generally speaking, the deep learning optimization system for the parameters of the prefabricated culvert structure of the present invention effectively solves the problems of low efficiency, insufficient optimization, and fragmentation of each link in the traditional method through the coordination of each module. The intelligent and automated characteristics of the system not only improve the design efficiency and quality, but also provide new possibilities for the innovative design of culvert structures. This systematic and intelligent design method is of great significance for improving the overall level of China's municipal engineering and is expected to have a profound impact in a wider field of civil engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 is the main flow chart of the present invention;

[0088] Figure 2 is the diagram of the model intelligent optimization module of the present invention;

[0089] Figure 3 is the diagram of the CAD automatic generation module of the present invention;

[0090] Figure 4 is the diagram of the model performance evaluation and optimization iteration module of the present invention;

[0091] Figure 5 is the detailed flow chart of the neural network training module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0092] Please refer to the appendix Figures 1-5 , the present invention discloses a deep learning optimization system and method for the structural parameters of an assembled culvert. The system includes multiple functional modules, and realizes the intelligent optimization of the culvert structural parameters through deep learning and digital twin technology.

[0093] Specifically, the deep learning optimization system for the structural parameters of the assembled culvert of the present invention includes a cross-section parameterization module 1, a finite element modeling module 2, a neural network training module 3, a digital twin module 4, a digital drawing module 5, and a BIM model drawing engine module 6. These modules are interconnected and work together to jointly complete the parameter optimization task of the culvert structure.

[0094] First, the cross-section parameterization module 1 is used to classify and parametrically analyze different types of culverts, and create component templates using different component parameters. For example, for different types such as straight culverts, corner culverts, and rectangular culverts, this module can establish corresponding parametric models respectively. Preferably, the culvert cross-section parameters may include geometric dimensions such as width, height, wall thickness, and material parameters such as concrete strength.

[0095] Next, the finite element modeling module 2 is connected to the cross-section parameterization module 1, and is used to establish a culvert model using the finite element framework and set material parameters. This module also designs boundary conditions and load forms according to the structural design requirements, and generates optimization objectives and constraint conditions. In one embodiment, the boundary conditions may include fixed constraints, hinge supports, etc.; the load forms may include dead loads, live loads, horizontal loads, etc. The optimization objective may be to minimize the structural weight, and the constraint conditions may include requirements such as strength, stiffness, and stability.

[0096] The neural network training module 3 is connected to the finite element modeling module 2, and is used to establish a cross-section parameterization sample library and train the neural network to obtain a trained neural network. This module uses deep learning algorithms such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs) to model the parameter-performance relationship of the culvert structure. During the training process, the Adam optimizer and the mean squared error loss function can be used, the learning rate can be set to 0.001, and the number of training epochs can be set to 1000 epochs.

[0097] The digital twin module 4 is connected to the neural network training module 3, and is used to extract the characteristic parameters of the geometric structure and build a digital twin model. During the iteration process, this module monitors the key geometric parameters in real time, performs reinforcement learning of physical constraints, and drives the physical model to be updated synchronously. For example, the deep deterministic policy gradient (DDPG) algorithm can be used for reinforcement learning, and the reward function can be designed as:

[0098] R = -w1·stress - w2·displacement + w3·material_saving,

[0099] Among them, w1, w2, and w3 are weight coefficients, stress is the maximum stress, displacement is the maximum displacement, and material_saving is the material saving amount.

[0100] The digital drawing module 5 is connected to the digital twin module 4. By using Openapi to complete the development of the API interface of AutoCAD, modular development of CAD files is carried out to obtain standardized CAD. This enables the system to automatically generate CAD drawings that meet engineering standards, greatly improving the design efficiency.

[0101] Finally, the BIM model drawing engine module 6 is connected to the digital drawing module 5. By using the family library of the prefabricated BIM model, a prefabricated BIM model is automatically generated and construction drawings are generated. This function realizes the seamless connection from parameter optimization to actual construction drawings, providing directly usable results for engineering practice.

[0102] The system of the present invention further includes a model intelligent optimization module 7. This module includes a geometric parameter calculation unit 71, which is used to pre-determine the structural form of the current culvert and select parameters according to geometric characteristics, and use the digital twin model to calculate the current structural mechanical properties and structural characteristics. For example, for a straight culvert, its cross-section can be pre-determined to be circular or rectangular, and then a suitable range of geometric parameters can be selected. The structural mechanical properties may include stress distribution, deformation and other indicators, and the structural characteristics may include stiffness, natural vibration frequency, etc.

[0103] In addition, the present invention further includes a CAD automatic generation module 8, which includes a BIM model construction unit 81, a data extraction unit 82, and a structure optimization unit 83. The BIM model construction unit 81 screens and constructs a BIM model based on the input basic structure information based on the BIM model. The data extraction unit 82 uses Openapi to quickly identify model nodes and extracts data in the model based on the Openapi module. The structure optimization unit 83 returns the obtained target data information to the structure simulation module to drive the automatic optimization of structural parameters and complete the iterative optimization process.

[0104] During the optimization process, intelligent optimization methods such as genetic algorithms or particle swarm algorithms can be used. For example, when using a genetic algorithm, the population size can be set to 100, the crossover probability can be 0.8, the mutation probability can be 0.1, and the number of iterations can be 200. These parameters can be adjusted according to the complexity of the specific problem to obtain the best optimization effect.

[0105] Through the collaborative work of the above modules, the system of the present invention can achieve the intelligent optimization of the structural parameters of the prefabricated culvert, improving the design efficiency and quality. The system makes full use of the advantages of deep learning and digital twin technologies, realizing the intelligentization and automation of the culvert structure design, and providing strong technical support for engineering practice.

[0106] The system of the present invention further includes a model performance evaluation and optimization iteration module 9, which further improves the intelligent level of the system. The model performance evaluation and optimization iteration module 9 includes a model verification unit 91, a design iteration unit 92, and a BIM model driving unit 93.

[0107] Based on the digital twin model, the model verification unit 91 establishes a model verification scheme and automatically completes the comparison and analysis of the simulation results of the model before and after optimization. In an embodiment of the present invention, the model verification scheme may include multiple dimensions such as stress distribution comparison, deformation comparison, and material usage comparison. For example, the root mean square error (RMSE) can be used to quantify the difference between the models before and after optimization:

[0108]

[0109] where y i is the output of the model before optimization, is the output of the model after optimization, and n is the number of samples. Preferably, an RMSE value less than 0.05 can be regarded as a significant optimization effect.

[0110] The design iteration unit 92 is connected to the model verification unit 91 and performs design iteration updates on the intelligent optimization module based on the model performance evaluation. In a preferred embodiment of the present invention, the design iteration uses the gradient descent method with an adaptive step size, and the iteration formula is as follows:

[0111]

[0112] where x k is the design parameter of the k-th iteration, α k is the adaptive step size, is the gradient of the objective function. The adaptive step size can be determined by the Armijo criterion to ensure convergence and efficiency.

[0113] The BIM model driving unit 93 is connected to the design iteration unit 92 and drives the BIM model to automatically generate drawings and analyze based on the obtained model parameters. Preferably, the BIM model driving unit 93 can also process according to the optimized CAD drawings and complete the assembly of the model. This function realizes the seamless connection from parameter optimization to the physical model, greatly improving the efficiency of project implementation.

[0114] In an embodiment of the present invention, in the cross-section parameterization module 1, classification and parametric analysis are performed according to different types of culverts, including straight culvert parameterization, corner culvert parameterization, and rectangular culvert parameterization. For straight culverts, circular or rectangular cross-sections can be adopted; for corner culverts, variable cross-section design can be used; for rectangular culverts, the thicknesses of the top slab, bottom slab, and side walls can be optimized. Orthogonal experimental method can be used for parametric analysis to determine the influence degree of each parameter on the structural performance.

[0115] In the finite element modeling module 2 of the present invention, the finite element framework is used to establish the culvert model and set the material parameters. The specific steps include:

[0116] First, based on the cross-section parameterization module 1, a parametric finite element cross-section is created for parametric modeling of the culvert structure. After the model is established, parameters such as the model cross-section, the number of steel bars, and the reinforcement ratio are set. Preferably, commercial finite element software such as ANSYS can be used, or a custom finite element program can be developed.

[0117] Second, the material parameters are set. The concrete and reinforcement materials of the culvert are both set based on parameters. For example, parameters such as the elastic modulus E, Poisson's ratio ν, and density ρ of the concrete can be adjusted according to the actual engineering requirements. The yield strength f y and ultimate strength f u of the steel bars can also be selected according to the design specifications. Specifically:

[0118] For the concrete material, the elastic modulus E reflects the ability of the material to resist deformation and is usually determined according to the concrete strength grade. Poisson's ratio ν represents the ratio of the lateral strain to the longitudinal strain when the material is stressed, and generally ranges from 0.1 to 0.2. The density ρ is used to calculate the self-weight of the structure, and the density of ordinary concrete is about 2400 kg / m 3 .

[0119] For the steel bar material, the yield strength f y is the stress value when the steel bar begins to undergo plastic deformation and needs to meet the requirements of relevant design specifications. The ultimate strength f u represents the maximum tensile stress that the steel bar can withstand and is usually greater than the yield strength.

[0120] The selection of these parameters should be combined with specific engineering conditions, design standards, and construction technology requirements to ensure the safety and economy of the structure. In practical applications, the material parameters can also be checked and optimized through test data.

[0121] Then, the boundary conditions are set using the finite element element attributes. For example, a fixed constraint can be set at the bottom of the culvert to simulate the foundation support, and a uniform load can be applied at the top of the culvert to simulate the filling pressure.

[0122] Finally, load settings are carried out to establish load cases, and the loading calculation is performed using the finite element loading process. The load cases can include dead loads, live loads, horizontal loads, etc., and can be combined according to the actual engineering requirements.

[0123] The neural network training module 3 of the present invention includes a sample database establishment unit 31 and a neural network training unit 32. The sample database establishment unit 31 sorts and stores the sample data sets obtained by parameterizing the finite element units. Preferably, the sample data can include input parameters (such as geometric dimensions, material parameters, etc.) and output parameters (such as maximum stress, maximum displacement, etc.).

[0124] The neural network training unit 32 is connected to the sample database establishment unit 31, and uses the sample database to train the neural network to construct a digital twin model. In an embodiment of the present invention, a multi-layer perceptron (MLP) can be used as the basic structure of the neural network. The number of hidden layers can be set to 3 - 5 layers, and the number of neurons in each layer can be adjusted according to the complexity of the problem, usually between 50 - 200. The activation function can select the ReLU function:

[0125] f(x) = max(0, x),

[0126] The neural network training unit 32 also uses a reinforcement learning algorithm to obtain the optimal neural network constructed. For example, the deep Q - network (DQN) algorithm can be used, and its Q - value update formula is:

[0127]

[0128] where s t is the current state, a t is the current action, r t is the immediate reward, γ is the discount factor, and α is the learning rate.

[0129] Through the collaborative work of these modules, the system of the present invention can realize the intelligent optimization of the structural parameters of the prefabricated culvert, improve the design efficiency and quality. The system makes full use of the advantages of deep learning and digital twin technology, realizes the intelligentization and automation of the culvert structure design, and provides strong technical support for engineering practice.

[0130] In the finite element modeling module 2 of the present invention, a parameterized unit sample library is established based on the parameterization module, and the parameterized unit sample library is imported into the finite element to perform stress, deformation, and mass analysis to obtain a sample parameter library. This process provides a rich data basis for subsequent deep learning and optimization.

[0131] In a preferred embodiment of the present invention, the parametric unit sample library may include culvert units of different types, different sizes, and different material parameters. For example, for a straight culvert, parameters such as cross-sectional shape (circular, rectangular), size (width, height, wall thickness), and material strength can be set; for a corner culvert, parameters such as corner angle and transition section length can be added. These parameters can be discretized within a certain range to form a multi-dimensional parameter space.

[0132] After importing the parametric unit sample library into the finite element software, the system will automatically perform stress, deformation, and mass analysis. Preferably, a linear static analysis method can be used to calculate the stress distribution and deformation of the culvert structure under various load conditions. Stress analysis can obtain indicators such as von Mises equivalent stress and principal stress; deformation analysis can obtain indicators such as maximum displacement and average displacement; mass analysis can calculate the total weight of the structure. These analysis results constitute the output part of the sample parameter library.

[0133] The establishment of the sample parameter library provides the necessary data support for the training of the neural network. Through the analysis of a large number of samples, the system can learn the complex relationship between the culvert structure parameters and their mechanical properties, laying a foundation for subsequent optimization design.

[0134] The system of the present invention further includes a calculation result visualization module 10 for visually displaying the calculation results. The introduction of this module greatly improves the usability and intuitiveness of the system, enabling designers to better understand and evaluate the optimization results.

[0135] In an embodiment of the present invention, the calculation result visualization module 10 can generate various types of charts and graphs. For example, for the stress analysis results, a stress contour map can be generated to visually display the stress distribution in the structure; for the deformation analysis results, a deformation contour map or deformation animation can be generated to show the deformation process of the structure under the action of the load; for the optimization process, an iteration curve can be generated to show the change trend of the objective function with the number of iterations.

[0136] Preferably, the calculation result visualization module 10 can also provide an interactive 3D model display function. Designers can perform operations such as rotation, scaling, and cross-section viewing to comprehensively check the optimized culvert structure. This function is particularly useful for complex corner culverts or culvert structures with special shapes, and can help designers quickly discover potential problems.

[0137] The present invention also proposes a deep learning optimization method for the parameters of prefabricated culvert structures. This method is used in conjunction with the above system to achieve intelligent design and optimization of culvert structures. The method includes the following steps:

[0138] First, perform parameter initialization. In this step, preset design parameters, including the dimensions of components, the design range of the culvert, the threshold of the number of culvert calculation times, the number of components, and the optimization target value. Preferably, these initial parameters can be determined according to engineering experience and design specifications. For example, the design range of the culvert can be determined according to the geological conditions and usage requirements of the project site; the threshold of the number of culvert calculation times can be set to 1000 - 5000 times to balance calculation accuracy and efficiency.

[0139] Secondly, use the genetic algorithm to obtain the optimal finite element model. This step includes the following specific operations:

[0140] (1) Use the finite element method for mechanical simulation to solve for the stress distribution of the culvert. In this process, the system will automatically generate a finite element mesh, set boundary conditions and loads, and then solve. Preferably, the adaptive mesh technology can be used to refine the mesh in the stress concentration area to improve the calculation accuracy.

[0141] (2) Select the models with good fitness to enter the next round of mechanical simulation, and the roulette wheel method is used for the selection method. The fitness function can be designed as:

[0142]

[0143] where σ max is the maximum stress, u max is the maximum displacement, m is the structural mass, and w1, w2, w3 are weight coefficients.

[0144] (3) Optimize the structural parameters through the genetic algorithm. The optimized parameters include: the type of components, the cross-sectional shape of components, the reinforcement ratio, the spacing of steel bars, the spacing of stirrups, the number of longitudinal steel bars, and the number of stirrups. In the genetic algorithm, the real number coding method can be used, the arithmetic crossover can be used for the crossover operation, and the uniform mutation can be used for the mutation operation.

[0145] (4) Compare the calculated stress value with the target to determine whether the design requirements are met. If the requirements are met, proceed to the next step; if not, return to step (1) for further optimization.

[0146] Next, determine whether the number of mechanical simulation times exceeds the threshold. If it has exceeded the threshold, end the mechanical simulation; if not, proceed to the next step. This step aims to control the calculation cost and prevent infinite loops.

[0147] Finally, determine whether the convergence condition is met. If the convergence condition is satisfied, end the mechanical simulation; if not, re-enter the first step, use the adaptive moment estimation algorithm, construct the objective function and constraint conditions according to the optimization target and constraint conditions, and search for and solve the optimal parameters.

[0148] The objective function of the adaptive moment estimation algorithm can be designed as follows:

[0149] f(x) = w1·σ max (x) + w2·u max (x) + w3·m(x),

[0150] Where x is the design variable vector, including various geometric parameters and material parameters. The constraint conditions can include stress constraints, displacement constraints, construction requirements, etc.

[0151] Through the above steps, the method of the present invention can realize the intelligent optimization of the prefabricated culvert structure parameters, greatly improving the design efficiency and quality. This method makes full use of the advantages of deep learning and genetic algorithms, realizing the intelligentization and automation of the culvert structure design, and providing strong technical support for engineering practice.

[0152] To verify the actual effect of the deep learning optimization system and method for the prefabricated culvert structure parameters of the present invention, the present invention selects a typical municipal engineering case for simulation analysis. This case is an urban underground passage project, and a straight culvert structure with a length of 200 meters needs to be designed. The design requirements of the culvert include: the internal cross-section is rectangular, with a net width of 6 meters and a net height of 4.5 meters; the upper cover soil depth is 2 meters; the design service life is 100 years; the anti-seepage grade is P6; the structure uses reinforced concrete materials.

[0153] The present invention selects three different design schemes for comparison: In Example 1, the deep learning optimization system for the prefabricated culvert structure parameters of the present invention is used for design; in Comparative Example 1, the traditional manual experience design method is used; in Comparative Example 2, the conventional parametric design software is used for optimization design. The three schemes are all based on the same initial conditions and design objectives to ensure the fairness of comparison.

[0154] Example 1: The deep learning optimization system of the present invention first constructs a parametric model of the culvert structure based on the initial parameters. Then, the system uses the finite element analysis module to generate a large number of sample data and trains the deep neural network using these data. During the optimization process, the system uses the reinforcement learning algorithm to continuously adjust the design parameters, and finally obtains the optimized culvert structure scheme.

[0155] Comparative Example 1: The traditional manual experience design method mainly relies on the experience and manuals of designers, and determines the various parameters of the culvert through repeated manual calculations and adjustments. Although this method is reliable, its efficiency is low, and it is difficult to fully consider all possible optimization schemes.

[0156] Comparative Example 2: Although conventional parametric design software can perform a certain degree of automated design, its optimization algorithm is relatively simple, difficult to handle complex non-linear problems, and lacks learning and adaptation capabilities.

[0157] To comprehensively evaluate the performance of the three schemes, the present invention selects the following key indicators for testing:

[0158] 1. Structural weight (tons / meter): Reflects the material utilization efficiency;

[0159] 2. Maximum stress (MPa): Reflects the structural safety;

[0160] 3. Maximum displacement (mm): Reflects the structural stiffness;

[0161] 4. Design time (hours): Reflects the design efficiency;

[0162] 5. Cost (ten thousand yuan / meter): Reflects the economic performance;

[0163] The detection methods for these indicators are as follows:

[0164] Structural weight: Automatically calculated through the BIM model;

[0165] Maximum stress and maximum displacement: Finite element analysis is performed using ANSYS software;

[0166] Design time: Record the actual time consumed from the start of design to the determination of the scheme;

[0167] Cost: Estimated according to the local material price and labor cost standards;

[0168] The test results are shown in Table 1 below:

[0169] Table 1. Comparison table of experimental results of Example 1 and Comparative Example 1 and Comparative Example 2:

[0170]

[0171]

[0172] Analyzing the test results in Table 1, the following conclusions can be drawn:

[0173] 1. Material utilization efficiency: The scheme of the present invention (Example 1) is significantly lower in structural weight than the other two schemes, 17.8% lighter than Comparative Example 1 and 9.4% lighter than Comparative Example 2. This indicates that the deep learning optimization system of the present invention can utilize materials more effectively and reduce waste.

[0174] 2. Structural performance: In terms of the maximum stress and maximum displacement indicators, the solution of the present invention performs best. The maximum stress is 16.6% smaller than that of Comparative Example 1 and 7.5% smaller than that of Comparative Example 2; the maximum displacement is 28.7% smaller than that of Comparative Example 1 and 15.5% smaller than that of Comparative Example 2. This shows that the optimized structure of the present invention has higher safety and stiffness.

[0175] 3. Design efficiency: The solution of the present invention has an overwhelming advantage in terms of design time, and it only takes 4 hours to complete the design, while the traditional manual method takes 72 hours and the conventional parametric software also takes 24 hours. This reflects the high efficiency of the deep learning system in dealing with complex non-linear problems.

[0176] 4. Economy: Benefiting from the efficient use of materials and the significant reduction in design time, the solution of the present invention also shows the best performance in terms of cost, being 15.8% lower than that of Comparative Example 1 and 8.6% lower than that of Comparative Example 2.

[0177] These results fully demonstrate the superiority of the deep learning optimization system for the structural parameters of the prefabricated culvert of the present invention. This system can not only design a structure with better performance, but also greatly improve the design efficiency and reduce the project cost. This is mainly due to the following aspects:

[0178] 1. The deep learning model can capture the complex non-linear relationship between the culvert structural parameters and performance, so as to find a better design solution.

[0179] 2. The digital twin technology realizes the real-time synchronization of the virtual model and the actual structure, making the optimization process more accurate and reliable.

[0180] 3. The reinforcement learning algorithm can quickly find the optimal solution among a large number of possible solutions, greatly improving the search efficiency.

[0181] 4. The system has a high degree of automation, with seamless connection from parameter optimization to CAD drawing and BIM modeling, greatly improving the design efficiency.

[0182] In summary, the deep learning optimization system for the structural parameters of the prefabricated culvert of the present invention shows obvious advantages in terms of material utilization, structural performance, design efficiency and economy, providing an efficient and reliable new method for the intelligent design of culvert structures. This method is not only applicable to the straight culverts in this case, but also can be extended to the design of other types of culvert structures, having broad application prospects.

[0183] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An optimized system for deep learning of the structural parameters of prefabricated culverts, characterized in that, It includes: A cross-section parameterization module, which is used for: Classifying and performing parametric analysis according to different types of culverts; Creating component templates using different component parameters; A finite element modeling module, connected to the cross-section parameterization module, which is used for: Establishing a culvert model using a finite element framework and setting material parameters; Designing boundary conditions and load forms according to structural design requirements; Generating optimization objectives and constraint conditions; A neural network training module, connected to the finite element modeling module, which is used for: Establishing a cross-section parameterization sample library; Training the neural network to obtain a trained neural network; A digital twin module, connected to the neural network training module, which is used for: Extracting characteristic parameters of geometric structures to construct a digital twin model; Real-time monitoring of key geometric parameters during the iteration process; Performing reinforcement learning of physical constraints and driving the physical model to be updated synchronously; A digital drawing module, connected to the digital twin module, which is used for: Completing the development of the api interface of AutoCAD using Openapi; Performing modular development on CAD files to obtain a standardized CAD; A BIM model drawing engine module, connected to the digital drawing module, which is used for: Automatically generating an assembled BIM model using the family library of the assembled BIM model; Generating construction drawings.

2. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, wherein It further includes a model intelligent optimization module, and the model intelligent optimization module includes: A geometric parameter calculation unit, which is used for: Pre-determining the structural form of the current culvert and selecting parameters according to geometric characteristics; Calculating the current structural mechanical properties and structural characteristics using the digital twin model.

3. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, wherein It further includes a CAD automatic generation module, and the CAD automatic generation module includes: A BIM model construction unit, which is used for: Screening components based on the BIM model according to the input basic structural information; Constructing a BIM model; A data extraction unit, connected to the BIM model construction unit, which is used for: Quickly identifying model nodes using Openapi; Extracting data in the model based on the Openapi module; A structure optimization unit, connected to the data extraction unit, which is used for: Returning the obtained target data information to the structure simulation module; Driving the automatic optimization of structural parameters to complete the iterative optimization process.

4. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, wherein It further includes a model performance evaluation and optimization iteration module, and the model performance evaluation and optimization iteration module includes: A model verification unit, which is used for: Establishing a model verification scheme based on the digital twin model; Automatically completing the comparison and analysis of the simulation results of the model before and after optimization; A design iteration unit, connected to the model verification unit, which is used for: Performing design iteration and update on the intelligent optimization module based on model performance evaluation; A BIM model driving unit, connected to the design iteration unit, which is used for: Driving the BIM model to perform automatic drawing and analysis based on the obtained model parameters; Processing according to the optimized CAD drawings and completing the assembly of the model.

5. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, characterized in that In the cross-section parameterization module, classifying and performing parametric analysis according to different types of culverts includes straight pipe culvert parameterization, corner culvert parameterization, and rectangular culvert parameterization.

6. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, wherein, In the finite element modeling module, a finite element framework is used to establish the culvert model and set material parameters. The specific steps are as follows: Based on the cross-section parameterization module, a parametric finite element cross-section is created for parametric modeling of the culvert structure; After the model is established, set the model cross-section, number of steel bars, reinforcement ratio, etc.; Set the material parameters. The concrete and reinforcement materials of the culvert are set based on parameters; Set the boundary conditions using the finite element element properties; Set the loads, establish the load cases, and perform the loading calculation using the finite element loading process.

7. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, characterized in that, The neural network training module includes: A sample database establishment unit for: Obtaining the sample data set parameterized from the finite element elements; A neural network training unit, connected to the sample database establishment unit, for: Training the neural network using the sample database; Constructing a digital twin model; Using the reinforcement learning algorithm to obtain the optimal neural network constructed.

8. The deep learning optimization system for the structural parameters of the prefabricated culvert according to claim 1, wherein, In the finite element modeling module, a parametric unit sample library is established based on the parameterization module, and the parametric unit sample library is imported into the finite element for stress, deformation, and mass analysis to obtain the sample parameter library.

9. The deep learning optimization system for the structural parameters of the prefabricated culvert according to any one of claims 1-8, characterized in that It also includes a calculation result visualization module for visually displaying the calculation results.

10. Deep learning optimization method for structural parameters of assembled culverts, characterized in that, The steps are as follows: Step 1, parameter initialization: Preset the design parameters, including the dimensions of the components, the design range of the culvert, the threshold of the number of culvert calculation times, the number of components, and the optimization target value; Step 2, obtaining the optimal finite element model using the genetic algorithm. The specific steps are as follows: Perform mechanical simulation using the finite element method to solve for the stress distribution of the culvert; Select the models with good fitness to enter the next round of mechanical simulation, and the selection method is the roulette wheel method; Optimize the structural parameters through the genetic algorithm. The optimized parameters include: type of components, cross-sectional shape of components, reinforcement ratio, steel bar spacing, stirrup spacing, number of longitudinal steel bars, and number of stirrups; Compare the calculated stress value with the target to determine whether the design requirements are met; Step 3, determine whether the number of mechanical simulation times exceeds the threshold; If it has exceeded the threshold, end the mechanical simulation; If it has not exceeded the threshold, enter Step 4; Step 4, determine whether the convergence condition is established: If it is satisfied, end the mechanical simulation; If it is not satisfied, re-enter Step 1, and use the adaptive moment estimation algorithm to construct the objective function and constraint conditions according to the optimization target and constraint conditions, and search for and solve the optimal parameters.

Images