Optimization model training method, structure optimization method and device of nuclear power support hanger

Abstract

The invention discloses an optimization model training method, structure optimization method and device of a nuclear power support hanger, the optimization model comprises a topology model and a trained first machine learning model, the three-dimensional model of the support hanger for training is firstly obtained, the three-dimensional model of the support hanger for training is input to the topology model, and the topology model of the support hanger for training is obtained; obtaining a first support hanger three-dimensional model; secondly, inputting the first support hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first support hanger three-dimensional model; and finally, according to the evaluation result, updating design parameters in the topology model to obtain an updated topology model, and taking the updated topology model and the trained first machine learning model as a trained optimization model, so that the trained optimization model can be called, and the optimization efficiency is improved. And taking the to-be-optimized three-dimensional model of the support hanger as input to obtain an optimized three-dimensional model of the support hanger.

Description

Technical Field

[0001] The present application relates to the technical field of nuclear power structural design, and in particular to an optimization model training method, a structural optimization method and a device for a nuclear power support bracket. Background Art

[0002] In the nuclear power sector, supports and hangers primarily serve pipelines, providing load bearing, shock absorption, and protection. Their design must consider multiple factors, including load, service environment, and material properties, to ensure optimal performance. With the continuous advancement of nuclear power technology, traditional support and hanger structures are no longer able to meet the service requirements of advanced new reactor types, necessitating structural optimization of support and hanger designs.

[0003] In the traditional structural optimization design of supports and hangers, designers rely on personal experience to make the initial configuration, and then use simulation methods such as finite element analysis to perform stress analysis on the initial model. During the optimization process, the details of the structure still require designers to spend a lot of time and energy on repeated comparisons and manual calculations, resulting in a significant increase in the overall design cycle and cost, low design efficiency and a low degree of automation. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to provide an optimization model training method, a structural optimization method and a device for nuclear power supports and hangers, aiming to solve the problems of low efficiency and low automation level in the structural optimization process of supports and hangers in existing methods.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] An embodiment of the present application provides a method for training an optimization model of a nuclear power support and hanger, wherein the optimization model includes a topology model and a trained first machine learning model. The method includes:

[0007] Obtain a three-dimensional model of a support and hanger for training;

[0008] Inputting the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model;

[0009] Inputting the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model;

[0010] updating the design parameters in the topology model according to the evaluation result to obtain an updated topology model;

[0011] The updated topology model and the trained first machine learning model are used as the trained optimization model.

[0012] In the above solution, updating the topology model according to the evaluation result to obtain an updated topology model includes:

[0013] When the evaluation result indicates that the design conditions are not met, the design parameters in the topology model are updated based on the parameter adjustment instructions included in the evaluation result to obtain an updated topology model.

[0014] In the above solution, the method further includes:

[0015] Obtain a mechanical test dataset for training; wherein the mechanical test dataset represents a dataset obtained by performing one or more mechanical property tests on a solid three-dimensional model; the solid three-dimensional model represents a solid support and hanger obtained by 3D printing (also known as additive manufacturing) a three-dimensional model of the second support and hanger; and the three-dimensional model of the second support and hanger represents the three-dimensional model of the first support and hanger corresponding to the evaluation result when the evaluation result indicates that the design conditions are met;

[0016] Based on the mechanical test data set, the design parameters in the topology model are updated to obtain an updated topology model.

[0017] In the above solution, the method further includes:

[0018] Acquire topology optimization training data, input the topology optimization training data into a first machine learning model before training, perform training for evaluating topology optimization results, and obtain the first machine learning model after training; the topology optimization training data includes design parameters for training, topology optimization results for training, and a label indicating whether the design requirements are met.

[0019] In the above solution, the optimization model also includes a trained second machine learning model, and the method further includes:

[0020] Obtain a three-dimensional model database, input the three-dimensional model database into a second machine learning model before training, perform grid division training, and obtain a trained second machine learning model.

[0021] In the above solution, the optimization model further includes a trained third machine learning model, and the method further includes:

[0022] Acquire specific feature training data, input the specific feature training data into the third machine learning model before training, perform recognition training of the filamentous structure and isolated structure of the support and hanger, and obtain the third machine learning model after training; wherein,

[0023] The specific feature training data includes edge feature data and / or corner feature data and / or connection feature data of the supports and hangers.

[0024] In the above solution, the design parameters include one or more of the following: load, boundary conditions, material properties, and design domain.

[0025] In the above solution, after obtaining the three-dimensional model of the training support and hanger, the method further includes:

[0026] By accessing the automated script of the finite element analysis software, a static analysis is performed on the three-dimensional model of the training support and hanger to obtain the initial setting information of the design domain; the automated script is provided with one or more of the following information: initial load, initial boundary conditions, initial material properties, and static analysis steps.

[0027] The present application also provides a method for optimizing the structure of a nuclear power support bracket, the method comprising:

[0028] Obtaining a three-dimensional model of the support and hanger to be optimized;

[0029] The trained optimization model is called, and the three-dimensional model of the support and hanger to be optimized is used as input to obtain the optimized three-dimensional model of the support and hanger; wherein,

[0030] The trained optimization model includes an updated topology model and a trained first machine learning model; the trained optimization model is obtained using any of the above-mentioned training methods.

[0031] The present application also provides a training device for an optimization model of a support and hanger structure for nuclear power plants, wherein the optimization model includes a topology model and a trained first machine learning model, and the device includes:

[0032] The first acquisition module is used to obtain a three-dimensional model of a support and hanger for training;

[0033] A first calculation module is used to input the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model;

[0034] a second computing module, configured to input the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model;

[0035] The first updating module is used to update the design parameters in the topology model according to the evaluation results to obtain an updated topology model; and use the updated topology model and the trained first machine learning model as a trained optimization model.

[0036] The present application also provides a structural optimization device for a nuclear power support bracket, the device comprising:

[0037] The second acquisition module is used to obtain the three-dimensional model of the support and hanger to be optimized;

[0038] The third calculation module is used to call the trained optimization model and take the three-dimensional model of the support and hanger to be optimized as input to obtain the optimized three-dimensional model of the support and hanger; wherein,

[0039] The trained optimization model includes an updated topology model and a trained first machine learning model; the trained optimization model is obtained using any of the above-mentioned training methods.

[0040] The present application proposes an optimization model training method, a structural optimization method and a device for a support bracket for nuclear power, wherein the optimization model includes a topological model and a trained first machine learning model. In the training method of the optimization model, first, a training support bracket three-dimensional model is obtained, and the training support bracket three-dimensional model is input into the topological model to obtain a first support bracket three-dimensional model; secondly, the first support bracket three-dimensional model is input into the trained first machine learning model to obtain an evaluation result of the first support bracket three-dimensional model; finally, according to the evaluation result, the design parameters in the topological model are updated to obtain an updated topological model, and the updated topological model and the trained first machine learning model are used as the trained optimization model, that is, a training process of the optimization model is completed, so that the design parameters of the topological model of the optimization model are more reasonable, so that the trained optimization model can be called, and the support bracket three-dimensional model to be optimized is used as input to obtain the optimized support bracket three-dimensional model, thereby improving the efficiency and automation of the structural optimization process of the support bracket, while also avoiding excessive safety margin. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a training method for an optimization model of a nuclear power support and hanger provided in an embodiment of the present application;

[0042] Figure 2 This is a flow chart of a structural optimization method for a nuclear power support and hanger provided in an embodiment of the present application;

[0043] Figure 3 This is a flow chart of another method for training an optimization model and a method for optimizing a structure of a support and hanger for nuclear power provided in an embodiment of the present application;

[0044] Figure 4 This is a structural schematic diagram of a training device for an optimization model of a nuclear power support and hanger provided in an embodiment of the present application;

[0045] Figure 5 This is a structural schematic diagram of a structural optimization device for a nuclear power support bracket provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0047] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0048] With the continuous development of nuclear power technology, the support and hanger structures used to provide load-bearing, shock absorption, and protection for pipelines in the nuclear power field can no longer meet current design requirements. The design of supports and hangers in the nuclear power field must take into account multiple factors, including load, service environment, and material properties. Therefore, the design and optimization of these support and hanger structures are demanding and difficult for designers. In related structural design methods, designers rely on personal experience to perform initial configurations, then use simulation methods such as finite element methods to perform stress analysis on the initial model. Finally, based on the analysis results, the model is modified to meet the corresponding mechanical requirements. This method requires designers to spend a lot of time and energy on repeated comparisons and manual calculations, resulting in a significant increase in the overall design cycle and cost, low efficiency and automation, and excessive safety margins.

[0049] The present application proposes an optimization model training method, a structural optimization method and a device for a support bracket for nuclear power, wherein the optimization model includes a topological model and a trained first machine learning model. In the training method of the optimization model, first, a training support bracket three-dimensional model is obtained, and the training support bracket three-dimensional model is input into the topological model to obtain a first support bracket three-dimensional model; secondly, the first support bracket three-dimensional model is input into the trained first machine learning model to obtain an evaluation result of the first support bracket three-dimensional model; finally, according to the evaluation result, the design parameters in the topological model are updated to obtain an updated topological model, and the updated topological model and the trained first machine learning model are used as the trained optimization model, that is, a training process of the optimization model is completed, so that the design parameters of the topological model of the optimization model are more reasonable, so that the trained optimization model can be called, and the support bracket three-dimensional model to be optimized is used as input to obtain the optimized support bracket three-dimensional model, thereby improving the efficiency and automation of the structural optimization process of the support bracket, while also avoiding excessive safety margin.

[0050] The optimization model training method, structural optimization method and device of the nuclear power support bracket provided in the embodiment of the present application are described in detail below with reference to the accompanying drawings through specific embodiments and their application scenarios.

[0051] The embodiment of the present application provides a method for training an optimization model of a nuclear power support and hanger, wherein the optimization model includes a topology model and a trained first machine learning model, such as Figure 1 As shown, the method may include the following steps:

[0052] Step 101: Obtain a three-dimensional model of a training support and hanger.

[0053] Step 102: Input the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model.

[0054] Step 103: Input the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model.

[0055] Step 104: Update the design parameters in the topology model according to the evaluation result to obtain an updated topology model; and use the updated topology model and the trained first machine learning model as a trained optimization model.

[0056] In the embodiment of the present application, in step 101, the training support and hanger 3D model may be one or more 3D models that have been meshed using finite element methods. In other words, the input to the topological model may be a meshed 3D support and hanger model. To better perform calculations in the topological model, a training support and hanger 3D model with good meshing quality may be obtained.

[0057] In one embodiment, the optimization model further includes a trained second machine learning model, and the method further includes:

[0058] Obtain a three-dimensional model database, input the three-dimensional model database into a second machine learning model before training, perform grid division training, and obtain a trained second machine learning model.

[0059] In an embodiment of the present application, the three-dimensional model database may include one or more three-dimensional models of supports and hangers created using three-dimensional modeling software, and the second machine learning model is trained using the three-dimensional model database so that the trained second machine learning model can be used to perform better meshing operations. In actual application, a suitable deep learning architecture can be selected to design a network structure, including the number of layers, number of neurons, and activation functions, etc., and the loss function and optimization algorithm can be determined to continuously adjust the parameters in the second machine learning model to optimize the meshing performance of the second machine learning model. After calling the trained second machine learning model, a three-dimensional model of supports and hangers for training with better meshing can be obtained, so that subsequent calculations in the topology model can be better performed, making the calculations in the topology model more accurate.

[0060] In one embodiment, design parameters include one or more of loads, boundary conditions, material properties, and a design domain. These design parameters refer to parameters within the topology model, and the design domain refers to the region within the three-dimensional support and hanger model where material density variations are permitted during topology model calculations. By setting these design parameters within the topology model, topology model calculations can be more closely aligned with actual structural optimization requirements.

[0061] In one embodiment, after obtaining the three-dimensional model of the training support and hanger, the method further includes: performing a static analysis on the three-dimensional model of the training support and hanger by accessing an automated script of finite element analysis software to obtain initial setting information of the design domain; one or more of the following information is set in the automated script: initial load, initial boundary conditions, initial material properties, and static analysis steps.

[0062] In an embodiment of the present application, the above-mentioned initial load, initial boundary conditions, initial material properties, and static analysis steps are all used to perform static analysis, so as to determine the initial setting information of the design domain in the topological model calculation. After obtaining the initial setting information of the design domain, the design domain in the design parameters can be obtained, so as to facilitate subsequent calculations in the topological model. In actual application, by running the automation script, results such as displacement, stress, and strain can also be extracted to identify stress concentration areas and areas with excessive displacement. By accessing the above-mentioned automation script, on the one hand, there is no need to manually set the initial design parameters of the topological model, thereby improving the efficiency of optimization model training or calculation. On the other hand, by identifying stress concentration areas and areas with excessive displacement, a reference can be provided for the further application and optimization of subsequent loads and boundary conditions.

[0063] In an embodiment of the present application, in step 102, the training support and hanger three-dimensional model is input into the topological model to obtain a first support and hanger three-dimensional model. The first support and hanger three-dimensional model may include the training support and hanger three-dimensional model after a portion of the grid cells are removed. The calculation principle of the topological model may be to set an optimization objective function (e.g., minimizing mass) and calculate based on design parameters and constraints to obtain one or more grid cells with a density of 1 or 0. A cell density of 1 indicates that material needs to be placed at that location, and a cell density of 0 indicates that material can be removed at that location, thereby forming different topological configurations.

[0064] In this embodiment of the present application, in step 103, the first hanger 3D model is input into the trained first machine learning model to obtain an evaluation result of the first hanger 3D model. The trained first machine learning model is used to evaluate whether the topological calculation result of the first hanger 3D model meets the design requirements.

[0065] In one embodiment, the method further includes: obtaining topology optimization training data, inputting the topology optimization training data into a first machine learning model before training, performing training for evaluating topology optimization results, and obtaining the first machine learning model after training; the topology optimization training data includes design parameters for training, topology optimization results for training, and a label indicating whether the design requirements are met.

[0066] In an embodiment of the present application, topology optimization related cases can be collected, including design parameters, optimization results, and labels indicating whether the design requirements are met. The collected data can be cleaned, standardized, and feature extracted to extract key features in preparation for training the first machine learning model. The first machine learning model trained with these data can be a supervised learning model and / or a reinforcement model. In other words, the qualification of the topology calculation results can be judged by the supervised learning model, and the parameters can be updated based on the reinforcement learning action, and the topology calculation and result qualification verification can be re-executed until the convergence conditions are met. The first machine learning model trained by the above method can obtain the evaluation results of the three-dimensional model of the first hanger more quickly and accurately.

[0067] In one embodiment, in step 104, updating the topology model according to the evaluation result to obtain an updated topology model includes:

[0068] When the evaluation result indicates that the design conditions are not met, the design parameters in the topology model are updated based on the parameter adjustment instructions included in the evaluation result to obtain an updated topology model.

[0069] In an embodiment of the present application, the evaluation result indicating that the design conditions are not met may be that the calculation result of the topology model is unqualified. In this case, the first machine learning model includes a reinforcement model, and the reinforcement model can output parameter adjustment instructions. The parameter adjustment instructions can be instructions included in the evaluation results, or instructions generated by the reinforcement model after the evaluation results are generated. By updating the design parameters in the topology model based on the parameter adjustment instructions included in the evaluation results when the evaluation results indicate that the design conditions are not met, the topology model can be continuously iteratively updated, and the updated topology model and the trained first machine learning model are used as a trained optimization model, so that the performance of the optimization model is better and the accuracy is higher.

[0070] In one embodiment, the method further includes: obtaining a mechanical test data set for training; wherein, the mechanical test data set represents a data set obtained by performing one or more mechanical property tests on a solid three-dimensional model; the solid three-dimensional model represents a solid support and hanger obtained by 3D printing a second support and hanger three-dimensional model; the second support and hanger three-dimensional model represents the first support and hanger three-dimensional model corresponding to the evaluation result when the evaluation result indicates that the design conditions are met; based on the mechanical test data set, the design parameters in the topological model are updated to obtain an updated topological model.

[0071] In an embodiment of the present application, the three-dimensional model of the second support and hanger can be 3D printed to obtain a physical support and hanger. In actual application, the three-dimensional model of the second support and hanger can be automatically converted into the data format required by the 3D printing software through a specific interface, and the corresponding process and printing parameters can be selected for 3D printing. Afterwards, a typical mechanical property test is performed on the physical support and hanger obtained after 3D printing. The mechanical performance test may include but is not limited to tests or experiments such as tension and compression, seismic resistance, and fatigue resistance. When conducting the test, the test data can be automatically collected, including key performance indicators such as stress, strain, and displacement, and thresholds can be set for each performance indicator. If the test performance does not meet the requirements, that is, exceeds the set threshold, the test results are used as a mechanical test data set for training. The mechanical test data set for training may include: obtaining the failure mode, dynamic response data and measured material properties of the support and hanger finished product through mechanical performance tests (such as tension and compression, seismic resistance, and fatigue resistance), combined with 3D printing process parameters (layer thickness, filling rate, and post-processing conditions) to form a multi-dimensional feedback data set.

[0072] In embodiments of the present application, the specific process for updating the design parameters in the topology model based on the mechanical test dataset may include, but is not limited to, adding local stress / fatigue constraints based on failure locations, calibrating finite element parameters based on measured material properties, and updating the dynamic load spectrum. By updating the design parameters in the topology model based on the mechanical test dataset to obtain an updated topology model, the topology model can be further updated to obtain a more accurate and efficient optimization model.

[0073] In one embodiment, the optimization model further includes a trained third machine learning model, and the method further includes:

[0074] Acquire specific feature training data, input the specific feature training data into a third machine learning model before training, perform recognition training for the filamentous structure and isolated structure of the support and hanger, and obtain a third machine learning model after training; wherein the specific feature training data includes edge feature data and / or corner feature data and / or connection feature data of the support and hanger.

[0075] In an embodiment of the present application, the third machine learning model can be used to reconstruct the three-dimensional model of the second support hanger to obtain an updated three-dimensional model of the second support hanger. Even if the evaluation results show that the design conditions are met, the three-dimensional model of the second support hanger may still include filaments and isolated areas. The trained third machine learning model is obtained by inputting the edge feature data and / or corner feature data and / or connection feature data of the support hanger into the third machine learning model for training. Furthermore, the trained third machine learning model can be called, and the three-dimensional model of the second support hanger is used as input to obtain an updated three-dimensional model of the second support hanger, automatically identifying and removing unnecessary features in the output results of the topological model, and ensuring the feasibility of manufacturing and processing the three-dimensional model of the second support hanger.

[0076] The embodiment of the present application provides a method for optimizing the structure of a support bracket for nuclear power plants, such as Figure 2 As shown, the method may include the following steps:

[0077] Step 201: obtaining a three-dimensional model of the support and hanger to be optimized;

[0078] Step 202: call the trained optimization model, take the three-dimensional model of the support and hanger to be optimized as input, and obtain the optimized three-dimensional model of the support and hanger.

[0079] In an embodiment of the present application, the trained optimization model includes an updated topological model and a trained first machine learning model; the trained optimization model is obtained by any of the above-mentioned training methods. By calling the trained optimization model, based on the topological optimization process and guided by the target performance, the design process of structural optimization can be simplified. At the same time, by using machine learning related technical means, it is possible to reduce repetitive modeling and other work, and realize the automatic iterative verification process of support and hanger structure stress analysis-design parameter definition-structural optimization, thereby reducing the requirements for designers to get started, and facilitating the convenient and efficient implementation of subsequent design activities. Furthermore, 3D printing technology can also be used at the same time to realize the rapid preparation of support and hanger finished products, and through the typical performance verification test of the support and hanger, the design accuracy of the support and hanger structure can be further improved to form a design closed loop.

[0080] The present application will be described in further detail below in conjunction with application examples.

[0081] In the application examples of this application, Figure 3 As shown, this application describes in detail the support and hanger structure optimization method of the application through steps S1 to S13.

[0082] S1: Create a three-dimensional model database of supports and hangers.

[0083] The above steps can be specifically implemented in the following way: according to the "Pipe Support and Hanger Manual" for nuclear power, one or more three-dimensional models of supports and hangers can be created using three-dimensional modeling software. Since the "Pipe Support and Hanger Manual" for nuclear power covers and summarizes the forms, classifications, structures of its components, specific dimensional parameters, etc. of commonly used supports and hangers in the nuclear power field, the three-dimensional models of one or more supports and hangers created using three-dimensional modeling software can form a support and hanger model database. The three-dimensional models of supports and hangers in the support and hanger model database can include different geometric shapes (such as L-type, T-type, I-type, etc.), sizes (such as length, width, height, thickness, etc.) and material properties (such as carbon steel, stainless steel, alloy steel, etc.). Furthermore, not only can all models be stored in a centralized database, but each three-dimensional model of a support and hanger can also be accompanied by complete metadata, which includes but is not limited to design-related parameters and material properties.

[0084] S2: Automatically import the support and hanger model into the finite element simulation software.

[0085] The above steps can be specifically implemented in the following ways: initialize the script environment of the finite element simulation software, create a new programming script file, and use the script file to import the three-dimensional model of the support and hanger; use the application programming interface (API, Application Programming Interface) of the finite element simulation software to import a specific format (such as STEP / IGES format) file in the support and hanger model database of the three-dimensional modeling software, read the metadata of the support and hanger model during the import process, and assign material properties to the imported support and hanger model based on the read metadata.

[0086] S3: Perform automated finite element static analysis.

[0087] The above steps can be specifically implemented through stages (1), (2) and (3):

[0088] (1) Data preparation and model training stage: Collect previous successful or high-quality support and hanger finite element model meshing cases or results, including support and hanger three-dimensional model data, corresponding meshing results and corresponding parameter settings, and use these cases or results as training data. Furthermore, the collected training data can be annotated to determine the degree of fineness of the meshing corresponding to different areas; the training data can also be standardized to determine the key geometric features and material properties that affect the meshing. Then, the machine learning model is trained. For example, a suitable deep learning architecture can be selected, such as convolutional neural networks (CNN), generative adversarial networks (GAN) or recurrent neural networks (RNN), and then the network structure is designed, including the number of layers, number of neurons and activation functions, and the loss function and optimization algorithm are determined. Specifically, the above training data can also be divided into training sets and test sets to train and verify the performance of the convolutional neural network model. During the training process, the model structure and parameters can be continuously adjusted to optimize the model performance.

[0089] (2) Meshing stage: The trained convolutional neural network model is called to import the three-dimensional model of the support and hanger to be evaluated into the finite element analysis software. Based on the output of the convolutional neural network model, the mesh is automatically generated in the finite element analysis software. To improve the degree of automation, the three-dimensional finite element model of the support and hanger can also be imported through the API of the finite element simulation software in S2, and automatic predictive meshing is performed. The mesh is then generated based on the output of the convolutional neural network model.

[0090] (3) Static analysis stage: import the automation script through the API interface of the finite element analysis software, automatically set the boundary conditions, loads, material properties, static analysis steps, etc. of the support and hanger finite element model, and automatically submit the finite element analysis job. The automation script can be an automation script written in a programming language (such as MATLAB / Python); after the finite element analysis job is completed, extract the displacement, stress, strain and other results; identify the stress concentration area and the area with excessive displacement as the focus of subsequent structural optimization, that is, determine the area where the subsequent support and hanger needs to be structurally optimized based on the analysis results.

[0091] S4: Define the design problem.

[0092] Defining a design problem may include defining design variables and establishing a design rule base including constraints and objective functions.

[0093] Defining design variables can include defining the density of elements within the design domain as a variable. In practical applications, the density of all elements within the design domain can be initially set to 1, indicating that these elements with a density of 1 are solid. In practical applications, design variables can be defined within the design domain. These design variables can be the density values ​​of each element after the design domain is discretized. The design domain refers to the area within the support and hanger model where material density variations are allowed. The design domain can also be a constraint. When performing topology optimization, the design domain can be defined first.

[0094] Establish a design rule library containing constraints and objective functions to store commonly used constraints and objective functions for quick subsequent use. This can include clarifying the mathematical expression and physical meaning of each design rule, encoding each design rule into executable code or script, storing it in a database or file system, and using templates or functions to enable quick definition and use.

[0095] Exemplarily, the above-mentioned design rule library can be established or called through Python code, which may specifically include: storing and managing design rules through the DesignRuleLibrary class to establish the design rule library; adding the objective function and constraints to the library through the add_rule method, and obtaining specific design rules through the get_rule method; the constraints and objective functions in the design rule library can be called, and the constraints are selected as the maximum stress value and volume ratio; the volume ratio is set to 0.7, indicating that the maximum volume of the support and hanger structure after topology optimization is 70% of the volume of the support and hanger before optimization, and the minimum flexibility is used as the optimization objective function.

[0096] In the design rule library including constraint conditions and objective functions, the constraint conditions may also include but are not limited to volume constraint, stress constraint, displacement constraint, and frequency constraint.

[0097] For example, the mathematical expression of the volume constraint is: V≤V allow , where V allow Is the maximum volume allowed. The physical meaning of this expression is to limit the volume of the optimized structure to not exceed the predetermined allowable value V allow to ensure the feasibility of the design in manufacturing or functionality.

[0098] For example, the mathematical expression of stress constraint is: σ max ≤σ allow , where σ max is the maximum stress of the structure under maximum load, σ allow is the allowable stress of the material. The physical meaning of this expression is to ensure that the maximum stress of the structure under the worst load conditions does not exceed the allowable stress σ of the material allow , to avoid structural failure.

[0099] For example, the mathematical expression of the displacement constraint is: max ≤u allow , where u max is the maximum displacement of the structure in a specific direction, u allow is the maximum displacement allowed. The physical meaning of this expression is to limit the maximum displacement of the structure in a specific direction to no more than the allowed value u allow to ensure the stability and functionality of the structure.

[0100] For example, the mathematical expression of the frequency constraint is: min ≤f≤f max , where f min is the minimum natural frequency of the structure, f max is the maximum natural frequency of the structure. The physical meaning of this expression is that for a dynamic system, the natural frequency f of the structure is limited to a certain range to meet the dynamic performance requirements.

[0101] In the design rule library including constraint conditions and objective functions, the objective functions may include but are not limited to minimizing mass, optimizing strength-to-weight ratio, minimizing maximum displacement, minimizing flexibility, and minimizing strain energy.

[0102] For example, the mathematical expression for minimizing mass is: Minimize M = ∫ Ω ρ(x)dV, where ρ(x) is the material density at each point in the design domain and M is the mass. The physical meaning of this expression is that the objective function aims to reduce the total mass of the structure to reduce the weight of the structure.

[0103] For example, the mathematical expression for the optimal strength-to-weight ratio is:

[0104]

[0105] Among them, σ yield is the yield strength of the material, V is the total volume of the structure, and ρ(x) is the material density at each point in the design domain. The physical meaning of this expression is that the objective function aims to find a material distribution that maximizes the ratio of the strength (stiffness) to the weight (volume) of the structure, thereby achieving the most efficient use of materials.

[0106] For example, the mathematical expression for minimizing the maximum displacement is:

[0107] Minimize u max =max x∈Ω {u(x)}

[0108] Where u(x) is the displacement at position x. The physical meaning of this expression is that this objective function ensures that the maximum displacement of the structure under load does not exceed the maximum value allowed by the design to ensure the functionality and safety of the structure.

[0109] For example, the mathematical expression of minimum flexibility is:

[0110]

[0111] Among them, C is the structural flexibility, U is the overall displacement of the structure, K is the total stiffness matrix, and x e is the density of the e-th unit, p is the penalty factor (usually p ≥ 3), u e is the local displacement vector of the e-th element, and k0 is the element stiffness matrix. The physical meaning of this expression is that the objective function aims to find a material distribution that minimizes the deformation (displacement) of the structure under a given external force.

[0112] S5: Build a mathematical model.

[0113] The above steps can be specifically implemented in the following ways: selecting a suitable topology optimization method, such as the variable density method and the level set method, to establish a mathematical model for topology optimization.

[0114] For example, when the variable density method is selected, the established topology optimization mathematical model can be:

[0115] Find:x={x1,x2,...,x e ,...,x n}

[0116] The physical meaning of this expression is to solve the optimal design variable x, that is, to find the x value that meets the objective function under the constraints; where x e is the density of the e-th unit.

[0117]

[0118] The physical meaning of this expression is to define the objective function as minimizing flexibility, where C is the structural flexibility, U is the overall displacement of the structure, K is the total stiffness matrix, and x e is the density of the e-th unit, p is the penalty factor (usually p ≥ 3), u e is the local displacement vector of the e-th unit, and k0 is the unit stiffness matrix.

[0119] st:V=0.7·V0

[0120] The physical meaning of this expression is to define a volume constraint so that the optimized material volume is 70% of the initial volume; where V is the optimized material volume and V0 is the initial design domain volume (i.e., the volume before optimization).

[0121] σ max ≤σ allow

[0122] The physical meaning of this expression is to define the stress constraint so that the maximum stress in the structure does not exceed the maximum stress allowed by the material itself; max is the maximum equivalent stress in the structure, σ allow The maximum stress allowed by the material (determined by the material strength).

[0123] 0 <x min ≤x i ≤1

[0124] The physical meaning of this expression is to define the design variable constraints so that the design variables are located in an interval; where x i is the density of the i-th unit, x min is the minimum value; when x i =1, indicating solid material, when x i =x min , indicating that there is approximately no material (i.e., hollow element).

[0125] S6: Set optimization parameters.

[0126] The above steps can be specifically implemented in the following ways: select an optimization algorithm, define algorithm parameters, such as the number of iterations and convergence criteria, and call the optimization algorithm for calculation. Among them, the optimization algorithm can be any one of gradient descent method, genetic algorithm, particle swarm optimization, simulated annealing, and evolutionary structure optimization. In actual application, the optimization algorithm can be encapsulated into a module or function to achieve reuse in different design problems and improve design efficiency. For example, one or more optimization algorithms can be selected, an optimization algorithm library can be established, a modular structure can be designed through python code, a class can be used to encapsulate each algorithm, the optimization parameters can be defined for each group of optimization algorithms, and the specific optimization algorithm logic can be implemented in each subclass. After the algorithm library is established, an optimization algorithm can be called for calculation.

[0127] S7: Apply further boundary conditions and loads.

[0128] The above steps can be specifically implemented as follows: Based on the results of the previous finite element analysis, further appropriate boundary conditions and external loads are added. Because the application of boundaries and loads in the nuclear power field is not fixed, and the actual working conditions faced by the supports and hangers are relatively complex, it is necessary to further add appropriate boundary conditions and external loads during the calculation process. For example, based on the results of the previous finite element analysis, the bolted connection of the support and hanger can be set as the boundary, and an external load can be added to the contact portion between the support and hanger and the pipeline, with the load size being half of the allowable load of the support and hanger.

[0129] S8: Perform topology optimization calculation and analysis.

[0130] S9: Automated result evaluation and post-processing.

[0131] Step S9 can be implemented as follows: collecting topology optimization cases, including design parameters, optimization results, and labels indicating whether design requirements are met; cleaning, standardizing, and feature extracting the collected data; and extracting key features for use in training the machine learning model. Design parameters may include, but are not limited to, loads, boundary conditions, material properties, and the initial design domain; optimization results may include, but are not limited to, density distribution and stress / displacement fields; labels may include labels indicating whether evaluation criteria are met; and key features may include, but are not limited to, load direction and design domain aspect ratio. The machine learning model trained using this data can be a supervised learning model, such as a neural network.

[0132] When evaluating the results of topology optimization, the input of the trained machine learning model can be the design parameters and current design variables, and the output is the qualified probability of the optimization result or the predicted value of key performance indicators (such as maximum stress and volume) for rapid pre-screening.

[0133] When the evaluation result is qualified, the next step can be processed; when the evaluation result is unqualified, the reinforcement learning model can be triggered to generate parameter adjustment instructions and update the parameters based on the reinforcement learning action. Among them, when training the reinforcement model, the reinforcement model can adopt reinforcement learning such as the Deep Deterministic Policy Gradient algorithm (DDPG). The input of the reinforcement model is the optimization state (such as design variables, performance indicators, and number of iterations), and the output is the design parameter adjustment instruction (such as the penalty factor increment Δp, the filter radius increment Δr). Its reward function is dynamically calculated according to the evaluation indicators (maximum allowable stress, volume threshold) to drive the adaptive optimization of parameters.

[0134] When using the above-mentioned machine learning models (such as supervised learning models) and reinforcement models for optimizing result evaluation, the specific automated iterative process can be: initialize the design domain and set constraints. In each round of iteration, first use the supervised learning model to predict the eligibility of the optimization results: if qualified, proceed to the next step; otherwise, trigger the reinforcement learning to generate parameter adjustment instructions and return to S4-S8; update the parameters based on the reinforcement learning actions, and re-execute the topology optimization and finite element verification until the convergence conditions are met (continuous predictions are qualified and the finite element verification passes, or the maximum number of iterations is reached).

[0135] S10: Reconstruction of the support and hanger model.

[0136] The above steps can be implemented by combining image processing and machine learning techniques to automatically identify and remove unnecessary features from the optimization results, such as filaments and isolated areas, to ensure manufacturing feasibility. In practical applications, a machine learning model can be trained and applied to the topology-optimized support and hanger structure image to automatically identify and remove filaments and isolated areas, resulting in a reconstructed support and hanger structure model. During the training process, image processing techniques can be used to extract key features of the pre-optimized support and hanger structure, such as edges, corners, and regions, for use in machine learning training and analysis.

[0137] S11: 3D printing to prepare the finished support bracket.

[0138] The above steps can be specifically implemented in the following ways: through the API interface, the optimized support and hanger three-dimensional model is automatically converted into the data format required by the 3D printing software. For example, the data format can be 3D Slicer or Meshmixer; at the same time, the three-dimensional model is pre-processed using artificial intelligence (AI) algorithms, such as slicing, support generation, and printing path optimization; according to the support and hanger material (carbon steel or stainless steel), the directed energy deposition process or the selective laser melting process is selected, and the pre-processing result file is automatically imported into the 3D printing equipment to prepare the finished product of the support and hanger.

[0139] S12: Typical performance test verification.

[0140] The above steps can be specifically achieved in the following ways: conduct typical mechanical performance tests on the finished support and hanger brackets, such as tension and compression, seismic resistance, fatigue resistance and other tests, and automatically collect test data, including key performance indicators such as stress, strain, and displacement; set thresholds for each performance indicator. If the test performance does not meet the requirements, that is, exceeds the set threshold, the test results will be fed back and integrated into the design process S4~S11 to correct the model; if the performance meets the requirements, it is determined that the support and hanger model has been optimized, and the entire design process is output to form a set of optimization plans, including optimization parameters, test results and detailed information on model adjustments, etc., which are stored in the optimization plan database to guide subsequent new design problems.

[0141] Among them, the feedback test results may include: obtaining the failure mode, dynamic response data and measured material properties of the finished support bracket through mechanical performance tests (such as tension, compression, seismic resistance, and fatigue resistance), combined with 3D printing process parameters (layer thickness, filling rate, and post-processing conditions) to form a multi-dimensional feedback data set.

[0142] In practical applications, test results are integrated into the design process S4-S11. Model modifications can be made through the following methods: modifying the topology optimization model, adding local stress / fatigue constraints based on the failure location, calibrating finite element parameters based on measured material properties, and updating the dynamic load spectrum; modifying the 3D printing process, optimizing the printing path and support structure for defective areas, and adjusting interlayer bonding parameters to reduce anisotropy. By building a closed "design-manufacturing-testing" loop, machine learning is used to predict failure risks and recommend correction directions. Reinforcement learning is then used to dynamically balance the weights of multi-objective optimizations (such as weight, lifespan, and cost) until the test results are verified to meet the requirements.

[0143] S13: Optimize model output.

[0144] Finally, the optimized support and hanger model structure is output.

[0145] In the above-mentioned application embodiments of the present application, first, through topology optimization, that is, by defining design variables, constraints and objective functions, complex and abstract design problems can be transformed into mathematical problems for quantitative analysis, thereby simplifying the design process; secondly, by creating a data transfer API interface between the support and hanger three-dimensional model database and various software, repeated modeling and data transfer can be avoided; then, by establishing a design rule library and encapsulating the optimization algorithm, the process parameters of topology optimization such as constraints, objective functions and optimization algorithms can be modularized, thereby achieving rapid call in the design iteration process, which can effectively improve the design efficiency; in addition, by writing automated scripts, developing automated evaluation tools for optimization results, and combining machine learning and intelligent algorithms, the automation of finite element analysis and iterative optimization processes can be achieved; finally, by preparing the finished support and hanger products through 3D printing technology, the processing and manufacturing cycle can be greatly reduced. At the same time, combined with verification tests, the accuracy of the support and hanger optimization model can be further improved, and the feasibility of the technical route can be verified. Through the above-mentioned method, a fast and accurate structural optimization process can be carried out for the support and hanger according to actual performance requirements, further simplifying the design process and improving design efficiency and quality.

[0146] In order to implement the training method of the optimization model of the nuclear power support and hanger of the embodiment of the present application, the embodiment of the present application also provides a training device for the optimization model of the nuclear power support and hanger, wherein the optimization model includes a topology model and a trained first machine learning model, such as Figure 4 As shown, the device includes:

[0147] The first acquisition module 401 is used to acquire a three-dimensional model of a training support and hanger;

[0148] A first calculation module 402 is configured to input the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model;

[0149] a second calculation module 403, configured to input the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model;

[0150] The first updating module 404 is used to update the design parameters in the topology model according to the evaluation results to obtain an updated topology model; and use the updated topology model and the trained first machine learning model as a trained optimization model.

[0151] In one embodiment, the second calculation module 404 is used to update the design parameters in the topology model based on the parameter adjustment instructions included in the evaluation result to obtain an updated topology model when the evaluation result indicates that the design conditions are not met.

[0152] In one embodiment, the apparatus further comprises:

[0153] The third acquisition module is used to obtain a mechanical test data set for training; wherein, the mechanical test data set represents a data set obtained by performing one or more mechanical property tests on the solid three-dimensional model; the solid three-dimensional model represents the solid support and hanger obtained after 3D printing the second support and hanger three-dimensional model; the second support and hanger three-dimensional model represents the first support and hanger three-dimensional model corresponding to the evaluation result when the evaluation result shows that the design conditions are met.

[0154] The second updating module is used to update the design parameters in the topology model based on the mechanical test data set to obtain an updated topology model.

[0155] In one embodiment, the apparatus further comprises:

[0156] The fourth acquisition module is used to obtain topology optimization training data;

[0157] The first training module inputs the topology optimization training data into the first machine learning model before training, performs training for evaluating the topology optimization results, and obtains the first machine learning model after training; the topology optimization training data includes design parameters for training, topology optimization results for training, and labels indicating whether the design requirements are met.

[0158] In one embodiment, the optimization model further includes a trained second machine learning model, and the apparatus further includes:

[0159] A fifth acquisition module is used to acquire a three-dimensional model database;

[0160] The second training module is used to input the three-dimensional model database into the second machine learning model before training, perform grid division training, and obtain the trained second machine learning model.

[0161] In one embodiment, the optimization model further includes a trained third machine learning model, and the apparatus further includes:

[0162] A sixth acquisition module, used to acquire specific feature training data;

[0163] The third training module is used to input the specific feature training data into the third machine learning model before training, perform recognition training of the filamentous structure and isolated structure of the support and hanger, and obtain the trained third machine learning model; wherein, the specific feature training data includes edge feature data and / or corner feature data and / or connection feature data of the support and hanger.

[0164] In one embodiment, the design parameters include one or more of the following: loads, boundary conditions, material properties, and design domain.

[0165] In one embodiment, the apparatus further comprises:

[0166] The fourth calculation module is used to perform static analysis on the three-dimensional model of the training support bracket by accessing the automated script of the finite element analysis software to obtain the initial setting information of the design domain; the automated script is provided with one or more of the following information: initial load, initial boundary conditions, initial material properties, and static analysis steps.

[0167] In order to realize the structural optimization method of the nuclear power support and hanger in the embodiment of the present application, the embodiment of the present application also provides a structural optimization device for the nuclear power support and hanger, such as Figure 5 As shown, the device includes:

[0168] The second acquisition module 501 is used to obtain a three-dimensional model of the support and hanger to be optimized;

[0169] The third calculation module 502 is used to call the trained optimization model, take the three-dimensional model of the support and hanger to be optimized as input, and obtain the optimized three-dimensional model of the support and hanger; wherein,

[0170] The trained optimization model includes an updated topology model and a trained first machine learning model; the trained optimization model is obtained using any of the above-mentioned training methods.

[0171] In addition, an embodiment of the present application also provides a computer program product, which is stored in a storage medium. The computer program can be executed by a processor of an electronic device to complete the steps of the training method of the optimization model of nuclear power supports and hangers, or the steps of the structural optimization method of nuclear power supports and hangers.

[0172] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0173] The term "one or more" herein refers to any combination of at least two of any one or more of a plurality, for example, including at least one of A, B, and C may refer to including any one or more elements selected from the set consisting of A, B, and C. The term "one or more" herein refers to any combination of at least two of any one of a plurality, for example, including at least one of A, B, and C may refer to including any one or more elements selected from the set consisting of A, B, and C.

[0174] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0175] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A training method for an optimization model of a nuclear power support and hanger, characterized in that: The optimization model includes a topology model and a trained first machine learning model, and the method includes: Obtain a three-dimensional model of a support and hanger for training; Inputting the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model; Inputting the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model; updating the design parameters in the topology model according to the evaluation result to obtain an updated topology model; The updated topology model and the trained first machine learning model are used as the trained optimization model.

2. The training method of the optimization model according to claim 1, characterized in that The updating of the topology model according to the evaluation result to obtain an updated topology model includes: When the evaluation result indicates that the design conditions are not met, the design parameters in the topology model are updated based on the parameter adjustment instructions included in the evaluation result to obtain an updated topology model.

3. The training method of the optimization model according to claim 2, characterized in that: The method further comprises: Acquire a mechanical test dataset for training; wherein the mechanical test dataset represents a dataset obtained by performing one or more mechanical property tests on a solid three-dimensional model; the solid three-dimensional model represents a solid support and hanger obtained by 3D printing a second support and hanger three-dimensional model; the second support and hanger three-dimensional model represents the first support and hanger three-dimensional model corresponding to the evaluation result when the evaluation result indicates that the design conditions are met; Based on the mechanical test data set, the design parameters in the topology model are updated to obtain an updated topology model.

4. The training method of the optimization model according to claim 1, wherein: The method further comprises: Acquire topology optimization training data, input the topology optimization training data into a first machine learning model before training, perform training for evaluating topology optimization results, and obtain the first machine learning model after training; the topology optimization training data includes design parameters for training, topology optimization results for training, and a label indicating whether the design requirements are met.

5. The training method of the optimization model according to claim 1, wherein: The optimization model further includes a trained second machine learning model, and the method further includes: Obtain a three-dimensional model database, input the three-dimensional model database into a second machine learning model before training, perform grid division training, and obtain a trained second machine learning model.

6. The training method of the optimization model according to claim 1, wherein: The optimization model further includes a trained third machine learning model, and the method further includes: Acquire specific feature training data, input the specific feature training data into the third machine learning model before training, perform recognition training of the filamentous structure and isolated structure of the support and hanger, and obtain the third machine learning model after training; wherein, The specific feature training data includes edge feature data and / or corner feature data and / or connection feature data of the supports and hangers.

7. The training method of the optimization model according to any one of claims 1 to 6, characterized in that: The design parameters include one or more of the following: loads, boundary conditions, material properties, and design domain.

8. The training method of the optimization model according to claim 7, characterized in that: After obtaining the three-dimensional model of the training support and hanger, the method further includes: By accessing the automated script of the finite element analysis software, a static analysis is performed on the three-dimensional model of the training support and hanger to obtain the initial setting information of the design domain; the automated script is provided with one or more of the following information: initial load, initial boundary conditions, initial material properties, and static analysis steps.

9. A method for optimizing the structure of a nuclear power support bracket, characterized in that: The method comprises: Obtaining a three-dimensional model of the support and hanger to be optimized; The trained optimization model is called, and the three-dimensional model of the support and hanger to be optimized is used as input to obtain the optimized three-dimensional model of the support and hanger; wherein, The trained optimization model includes an updated topology model and a trained first machine learning model; the trained optimization model is obtained using any one of the training methods of claims 1 to 8.

10. A training device for an optimization model of a nuclear power support and hanger, characterized in that: The optimization model includes a topology model and a trained first machine learning model, and the device includes: The first acquisition module is used to acquire a three-dimensional model of a support and hanger for training; A first calculation module is used to input the training support and hanger three-dimensional model into the topological model to obtain a first support and hanger three-dimensional model; a second computing module, configured to input the first hanger three-dimensional model into the trained first machine learning model to obtain an evaluation result of the first hanger three-dimensional model; The first updating module is used to update the design parameters in the topology model according to the evaluation results to obtain an updated topology model; and use the updated topology model and the trained first machine learning model as a trained optimization model.

11. A structural optimization device for a nuclear power support bracket, characterized in that: The device comprises: The second acquisition module is used to obtain a three-dimensional model of the support and hanger to be optimized; The third calculation module is used to call the trained optimization model and take the three-dimensional model of the support and hanger to be optimized as input to obtain the optimized three-dimensional model of the support and hanger; wherein, The trained optimization model includes an updated topology model and a trained first machine learning model; the trained optimization model is obtained using any one of the training methods of claims 1 to 8.

Images