A structural configuration generation method, device, equipment and readable storage medium

Abstract

The application discloses a structure configuration generation method and device, equipment and a readable storage medium, applied to the computer technology field, comprising: decoupling processing a global grid model to obtain a reduced order model and a structure optimization model; applying external load and boundary conditions to the reduced order model and solving to obtain displacement load; inputting the displacement load, design constraints and manufacturing constraints into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration. The application greatly reduces the calculation scale of the structure configuration optimization model, improves the calculation efficiency, improves the grid resolution in the target structure area to be optimized, and generates a high-fidelity structure configuration; unnecessary materials can be reduced without affecting the safety of the train; under the premise of maintaining similar geometric characteristics of the target structure area to be optimized, a high-fidelity structure configuration can be generated for any size, external load and boundary condition, and universality under a global parameter space is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a structure configuration generation method, device and equipment and readable storage medium. BACKGROUND

[0002] In order to obtain high-fidelity and clear structure configuration, the traditional structure configuration generation method often relies on improving the grid resolution in the finite element model, which undoubtedly leads to a sharp expansion of the calculation scale and a significant increase in resource consumption. In recent years, with the wide application of machine learning methods, a method has appeared which uses machine learning methods to replace the complex calculation process in the traditional iterative optimization to achieve rapid generation of structure configuration. However, this method has certain defects, such as when dealing with complex external load and boundary conditions, large grid scale, and the requirement of clear and smooth result boundary in actual engineering problems, the training sample generation process itself still has very high calculation cost; and this method is usually strongly dependent on specific problem settings, and its actual application is also limited to the model type and boundary condition range used for training.

[0003] Therefore, how to improve the calculation efficiency and generalization ability of structure configuration generation and reduce the calculation cost is a technical problem that needs to be solved at present. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a structure configuration generation method, device, equipment and readable storage medium, which solves the problems of low calculation efficiency and poor generalization ability of structure configuration generation, and high calculation cost in the prior art.

[0005] To solve the above technical problems, the present application provides a structure configuration generation method, comprising:

[0006] decoupling the global grid model to obtain a reduced-order model and a structure optimization model;

[0007] applying external load and boundary conditions to the reduced-order model and solving to obtain displacement load;

[0008] inputting the displacement load, design constraints and manufacturing constraints into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration;

[0009] The global grid model is a finite element model of the whole vehicle body structure; the reduced-order model includes a non-design space area, a load application area and a boundary constraint area; the structure optimization model includes a target structure area to be optimized and a grid transition area; and the optimal structure configuration is the optimal structure configuration of the target structure area to be optimized.

[0010] Optionally, applying external load and boundary conditions to the reduced-order model and solving to obtain displacement load comprises:

[0011] applying the external load and the boundary condition to the reduced order model to obtain a total stiffness matrix of the reduced order model and equivalent nodal forces applied on each node in the reduced order model;

[0012] calculating displacement vectors of all nodes in the reduced order model based on the total stiffness matrix and the equivalent nodal forces;

[0013] selectively screening the displacement vectors of all nodes to obtain the displacement vectors at the interface between the reduced order model and the structural optimization model as the displacement load.

[0014] Optionally, before decoupling the global mesh model to obtain the reduced order model and the structural optimization model, the method further comprises:

[0015] obtaining a training data set containing various training samples; the training samples include a reference optimal structure configuration and displacement loads for training, design constraints for training and manufacturing constraints for training corresponding to the reference optimal structure configuration;

[0016] normalizing the displacement loads for training, the design constraints for training and the manufacturing constraints for training to obtain normalized input data;

[0017] inputting the input data into an initial structure configuration optimization model to output a predicted optimal structure configuration;

[0018] calculating a loss value according to the predicted optimal structure configuration and the reference optimal structure configuration;

[0019] optimizing the initial structure configuration optimization model based on the loss value until the loss value meets a preset value or the number of iterations of training meets a maximum number of iterations, and then stopping training to obtain the trained structure configuration optimization model.

[0020] Optionally, calculating a loss value according to the predicted optimal structure configuration and the reference optimal structure configuration comprises:

[0021] calculating a density field loss based on the predicted optimal structure configuration and the reference optimal structure configuration;

[0022] calculating a physical constraint loss based on the predicted optimal structure configuration and the reference optimal structure configuration;

[0023] calculating the loss value based on the density field loss and the physical constraint loss.

[0024] Optionally, a training dataset containing various training samples is obtained, including:

[0025] According to the design space of the target structure to be optimized, the external load for training, the boundary condition for training, the design constraint for training, and the manufacturing constraint for training are determined;

[0026] Based on the external load for training, the boundary condition for training, the design constraint for training, and the manufacturing constraint for training, a set of parameter combinations covering the parameter space is generated by using a Latin hypercube method; the set of parameter combinations contains N parameter combinations;

[0027] Based on each set of parameter combinations, the reduced-order model is subjected to the external load for training and the boundary condition for training, and is solved to obtain the displacement load for training;

[0028] Based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training, the optimal structure configuration of the reference is obtained;

[0029] Based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training, and the corresponding optimal structure configuration of the reference, the training dataset is constructed.

[0030] Optionally, obtaining the optimal structure configuration of the reference based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training includes:

[0031] Applying the displacement load for training to the boundary nodes of the structure optimization model to maximize the flexibility of the structure optimization model as the optimization objective;

[0032] Taking the material density of the elements in the target structure to be optimized as a variable;

[0033] Taking the design constraint for training and the manufacturing constraint for training as constraint conditions;

[0034] Based on the optimization objective, the variable, and the constraint conditions, the optimal material density of each element in the target structure to be optimized is obtained by using a gradient optimization algorithm to iteratively solve, as the optimal structure configuration of the reference.

[0035] Optionally, after the displacement load, the design constraint, and the manufacturing constraint are input into the pre-trained structure configuration optimization model to obtain the generated optimal structure configuration, the following steps are further included:

[0036] When the optimal structure configuration does not meet the preset requirement, the external load, the boundary condition, the design constraint and the manufacturing constraint are taken as new samples based on the external load, the boundary condition, the design constraint and the manufacturing constraint;

[0037] The trained structure configuration optimization model is trained by using the new samples until a training termination condition is met, and the trained structure configuration optimization model is updated.

[0038] The application further provides a structure configuration generation device, which comprises:

[0039] A decoupling module is configured to perform decoupling processing on the global grid model to obtain a reduced-order model and a structure optimization model;

[0040] A solving module is configured to apply the external load and the boundary condition to the reduced-order model and perform solving to obtain a displacement load;

[0041] A generating module is configured to input the displacement load, the design constraint and the manufacturing constraint into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration;

[0042] The global grid model is a finite element model of a whole structure of a vehicle body; the reduced-order model comprises a non-design space area, a load application area and a boundary constraint area; the structure optimization model comprises a target structure area to be optimized and a grid transition area; and the optimal structure configuration is an optimal structure configuration of the target structure area to be optimized.

[0043] The application further provides a structure configuration generation device, which comprises:

[0044] A memory is configured to store a computer program;

[0045] A processor is configured to implement the structure configuration generation method as described above when the computer program is executed.

[0046] The application further provides a computer readable storage medium, wherein the computer readable storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to implement the structure configuration generation method as described above.

[0047] It can be seen that the application obtains a reduced-order model and a structure optimization model by decoupling a global grid model; applies external load and boundary conditions to the reduced-order model and solves to obtain displacement load; inputs the displacement load, design constraints and manufacturing constraints to a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration; wherein the global grid model is a finite element model of a whole structure of a vehicle body; the reduced-order model includes a non-design space area, a load application area and a boundary constraint area; and the structure optimization model includes a target structure area to be optimized and a grid transition area. The application has the beneficial effects that the complex global grid model is decoupled into two parts of the reduced-order model and the structure optimization model, and the complex external load and boundary conditions are converted into an intermediate variable of displacement load, which greatly reduces the calculation scale of the structure configuration optimization model and improves the calculation efficiency, so as to also improve the grid resolution in the target structure area to be optimized, generate a high-fidelity and smooth structure configuration, and reduce unnecessary materials without affecting the safety of the train; and under the premise of maintaining similar geometric characteristics of the target structure area to be optimized, a high-fidelity and smooth structure configuration can be generated for any size (referring to adjustable size parameters of the target structure limited by design constraints and manufacturing constraints), external load and boundary conditions, and universality in the global parameter space is achieved.

[0048] In addition, the application also provides a structure configuration generation device, equipment and readable storage medium, which also have the beneficial effects. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0050] Figure 1 A flow chart of a structure configuration generation method provided by the embodiment of the application;

[0051] Figure 2 A global grid model division schematic diagram provided by the embodiment of the application;

[0052] Figure 3 A reduced-order model schematic diagram provided by the embodiment of the application;

[0053] Figure 4 A structure optimization model schematic diagram provided by the embodiment of the application;

[0054] Figure 5 A structure configuration generation device structure schematic diagram provided by the embodiment of the application;

[0055] Figure 6 A structural schematic diagram of a structure configuration generation device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0057] Traditional structure configuration generation methods often rely on improving grid resolution to obtain high-fidelity clear structure configurations, which undoubtedly leads to a sharp expansion of the calculation scale and a significant increase in resource consumption. In recent years, many studies have attempted to establish an end-to-end mapping relationship from optimization parameters to structure configurations by means of machine learning methods to replace the complex calculation process in traditional iterative optimization, thereby realizing fast configuration prediction. However, such end-to-end machine learning methods still need to obtain a large number of optimization results covering different external loads, boundary conditions, design constraints and manufacturing constraints as training samples in the model training stage. Especially in the case of dealing with actual engineering problems with complex external loads and boundary conditions, large grid scale and the requirement of clear and smooth boundary of optimization results, the sample generation process itself still has a very high calculation cost. In addition, existing machine learning assisted optimization methods are usually strongly dependent on specific problem settings, and their actual applications are also limited to the model type and boundary condition range used for training. Once the input conditions (such as external loads / boundary conditions) exceed the parameter distribution during training, the structure configuration generated by the model is likely to deviate significantly from the physical reality, and therefore its generalization ability is obviously limited.

[0058] To solve the above problems, the present application provides a physically intelligent driven end-to-end structure configuration generation method, which breaks through the algorithm power bottleneck of high-precision structure configuration optimization by decoupling the global grid model and the displacement load generalization mapping, and significantly improves the optimization efficiency of the structure configuration. Based on the trained structure configuration optimization model, high-fidelity and clear smooth structure configurations can be generated for any size, external load and boundary condition (including non-sample working conditions) under the premise of maintaining similar geometric features in the design domain (referring to the target structure region to be optimized), thereby realizing universal prediction in the global parameter space.

[0059] For details, please refer to Figure 1 , Figure 1 A flowchart of a structure configuration generation method is provided for an embodiment of the present application. The method can include:

[0060] S101: decouple the global grid model to obtain a reduced order model and a structure optimization model.

[0061] The execution subject of the embodiment is a terminal. The embodiment does not limit the type of the terminal, as long as the operation of the structure configuration generation method can be completed. The method can be applied to a train structure configuration optimization scene. It needs to be noted that a global grid model is built according to a structure performance requirement, and the global grid model is a finite element model of the whole train body structure. In simple terms, first, the performance requirements that the train body needs to meet are determined, and then a global grid model covering the key structure of the whole train is built to simulate and calculate whether these performance requirements are met. The performance requirement determines which components the global grid model needs to include and how fine the grid needs to be, to ensure that the subsequent structure can accurately calculate whether the safety, stiffness and other requirements are met. The global grid model is decoupled into a reduced order model and a structure optimization model, and the reduced order model and the structure optimization model also refer to finite element models. For ease of understanding, the reduced order model can also be referred to as a reduced order region, and the structure optimization model can also be referred to as a structure optimization region. For details, refer to Figure 2 , Figure 3 and Figure 4 , Figure 2 a global grid model division schematic diagram provided by the embodiment of the application; Figure 3 a reduced order model schematic diagram provided by the embodiment of the application; Figure 4 a structure optimization model schematic diagram provided by the embodiment of the application. The model decoupling division needs to avoid the junction of the two regions, and the grid node density at the junction is large, so the division is implemented at the junction of the low-resolution grid (referring to the reduced order model) and the transition grid. The reduced order model covers the non-design space area, the load application area and the boundary constraint area, and uses a low-resolution grid; the structure optimization region includes a high-resolution grid target structure area to be optimized and a grid transition area. That is, the target structure area to be optimized, the grid transition area and the reduced order region, and the grid resolution of the three areas decreases in turn.

[0062] S102: apply external loads and boundary conditions to the reduced order model and solve to obtain displacement loads.

[0063] It needs to be noted that the external loads and boundary conditions in the step are external loads and boundary conditions applied to the global grid model; the displacement loads solved in the step refer to displacement loads at the junction of the reduced order model and the structure optimization model, and the displacement loads are in the form of vectors.

[0064] Further, the above-mentioned application of external loads and boundary conditions to the reduced order model and the solving to obtain displacement loads can specifically include:

[0065] Step 11: Apply external loads and boundary conditions to the reduced order model to obtain the total stiffness matrix of the reduced order model and equivalent nodal forces applied on each node in the reduced order model;

[0066] Step 12: Based on the total stiffness matrix and the equivalent nodal forces, calculate the displacement vector of all nodes in the reduced order model;

[0067] Step 13: Selectively screen from the displacement vectors of all nodes to obtain the displacement vector at the interface between the reduced order model and the structural optimization model as the displacement load.

[0068] Specifically, the reduced order region is independently calculated, and the displacement results at the interface with the structural optimization model are extracted. The specific steps are as follows: apply the external loads (which can include concentrated forces, distributed forces, body forces, and moments, etc.) and boundary conditions applied by the global grid model to the reduced order model; solve and analyze the reduced order model, and solve U S according to the following balance equation. Extract the displacement results of all nodes at the interface, and the extracted displacement results can be represented by the vector {Uinterface} as the displacement load.

[0069] The balance equation is:

[0070] [K S ]×{U S}={F S}。

[0071] Where K S is the total stiffness matrix of the reduced order model; U S is the displacement vector of all nodes in the reduced order model; and F S is the equivalent nodal force vector applied on the nodes of the reduced order model.

[0072] The vector representation of the displacement load {Uinterface} is:

[0073] {Uinterface}=[T]×{U S}。

[0074] Where T is a selection matrix, that is, a matrix composed of all nodes at the interface.

[0075] S103: Input the displacement load, design constraint, and manufacturing constraint into the pre-trained structural configuration optimization model to obtain the generated optimal structural configuration.

[0076] The displacement load obtained in step 102, and the obtained design constraint and manufacturing constraint are taken as the input of the structural configuration optimization model, and a high-resolution structural configuration with clear and smooth boundary is directly output by the structural configuration optimization model. It should be noted that the method is for structural configuration optimization of the target structure region to be optimized, and therefore, the high-resolution structural configuration corresponds to the high-resolution structural configuration of the target structure region to be optimized, that is, the optimal structural configuration is the optimal structural configuration of the target structure region to be optimized. In this step, the displacement load, the design constraint and the manufacturing constraint are input into the pre-trained structural configuration optimization model, the model generates the optimal material density field of the target structure region to be optimized, represents the optimal structural configuration of the target structure region to be optimized, and realizes the structural configuration optimization of the target structure region to be optimized. The target structure region to be optimized in this embodiment can be a train body component region such as a car frame beam, a car body covering member, and a column.

[0077] Before being input into the structural configuration optimization model, the design constraint, the manufacturing constraint and the displacement load can also be normalized and then input into the structural configuration optimization model. The structural configuration optimization model in this embodiment is not specifically limited, and the structural optimization model refers to a machine learning model. For example, it can be a neural network model, or it can also be a convolutional neural network model.

[0078] Further, the training process of the structural configuration optimization model is described below, which can specifically include:

[0079] Step 21: Obtain a training data set containing various training samples; the training samples include a reference optimal structural configuration and a displacement load for training, a design constraint for training and a manufacturing constraint for training corresponding to the reference optimal structural configuration.

[0080] The displacement load, the design constraint and the manufacturing constraint obtained in this step are all used for model training, and the optimal structural configuration is used for reference. The model training in this embodiment is supervised training.

[0081] Specifically, the above obtaining of the training data set containing various training samples can include:

[0082] Step 211: According to the design space of the target structure to be optimized, determine the external load for training, the boundary condition for training, the design constraint for training and the manufacturing constraint for training.

[0083] In this step, based on the design space of the target structure to be optimized (such as design requirements, spatial structure, etc.), the range of variation and combination of the external loads F (such as the magnitude and location of concentrated force, pressure, torque, etc.), boundary conditions B (such as the location and magnitude of boundary conditions such as displacement, torsion, etc.), design constraints D (such as the upper and lower limits of constraints such as volume constraints, modal constraints, mass constraints, etc.) and manufacturing constraints M (such as draft constraints, symmetry constraints, extrusion constraints, etc.) acting on the global mesh model are clarified.

[0084] Step 212: Based on the external load, boundary conditions, design constraints, and manufacturing constraints used for training, the Latin hypersol method is used to generate a set of parameter combinations covering the parameter space; the set of parameter combinations contains N parameter combinations.

[0085] This step is based on the training data determined in step 211. The Latin hypersol method is used to generate N sets of parameter combinations covering the parameter space. The formula is as follows:

[0086] .

[0087] in, For parameter combination set; For the first A combination of parameters; The first The external load F used for training, the first... B, the boundary condition used for training The design constraints D used for training and the first A manufacturing constraint M is used for training.

[0088] Step 213: Based on each set of parameter combinations, apply external loads and boundary conditions for training to the reduced-order model, and solve for the displacement loads used for training.

[0089] Specifically, based on each set of parameters To generate data pairs, first apply a reduction model. and And solve for it to obtain the first... Displacement loads used for training For the detailed solution process, please refer to steps 11 to 13 in step 102.

[0090] Step 214: Based on the design constraints, manufacturing constraints, and displacement loads used for training, obtain the optimal structural configuration for reference.

[0091] Specifically, , , The input is fed into a physics-constrained configuration optimization framework to obtain the first... The optimal structural configuration of a reference .

[0092] Furthermore, the optimal structural configuration obtained above, based on the design constraints, manufacturing constraints, and displacement loads used for training, can specifically include:

[0093] Step 2141: Apply the displacement loads used for training to the boundary nodes of the structural optimization model, with the optimization objective being to maximize the flexibility of the structural optimization model;

[0094] Step 2142: Use the unit material density within the target structural region to be optimized as a variable;

[0095] Step 2143: Use the design constraints and manufacturing constraints for training as the constraint conditions;

[0096] Step 2144: Based on the optimization objective, variables and constraints, use the gradient optimization algorithm to iteratively solve the problem and obtain the optimal material density of each unit in the target structural region to be optimized, which serves as the optimal structural configuration for reference.

[0097] Specifically, the displacement load solution method provided in steps 11 to 13 of step S102 will be used to calculate the displacement load for training based on the external load and boundary conditions used for training. This displacement load will then be applied as a forced displacement boundary condition to the boundary nodes of the structural optimization model, with the optimization objective being to maximize the compliance. The compliance C calculation formula is as follows:

[0098] .

[0099] in, Let T be the stiffness submatrix of the target structural region to be optimized; T represents the vector transpose; in this formula... This refers to the displacement load used for training.

[0100] Simultaneously apply design constraints (such as volume fraction, mass, modal, stress, or displacement constraints) and manufacturing constraints (such as draft, member dimensions, symmetry, or periodicity constraints) for training; these are the constraint conditions. The relative density of the element materials within the target structural region to be optimized is then considered. [0, 1] (e is the element index) is used as a continuous design variable, and the relationship between density and element stiffness is established through a material property interpolation model. A global stiffness matrix is ​​assembled based on the element stiffness matrix. The gradient optimization algorithm is used to iteratively solve this nonlinear constrained optimization problem: under the condition of satisfying the optimization objective and the constraints, the density field of the target structural region to be optimized is updated. To maximize flexibility The process continues until convergence yields the optimal material density field that approximates a 0-1 distribution. (i.e., reference density, i.e., the optimal structural configuration of the reference), characterizing the optimal structural configuration of the target structural region to be optimized.

[0101] Material property interpolation models include: .

[0102] in, The element stiffness matrix; To meet The penalty function; This is the stiffness matrix of the solid element.

[0103] Step 215: Based on the design constraints, manufacturing constraints, and displacement loads used for training, as well as the corresponding optimal structural configuration for reference, a training dataset is constructed.

[0104] Based on the aforementioned design constraints, manufacturing constraints, and displacement loads used for training, as well as the corresponding optimal structural configuration for reference, a training dataset is constructed. Furthermore, the training dataset can be divided into a training set and a validation set in a certain ratio (e.g., 8:2) for model parameter learning and generalization ability evaluation, respectively.

[0105] Step 22: Normalize the displacement load, design constraints, and manufacturing constraints used for training to obtain normalized input data.

[0106] Specifically, the model input data in the training dataset can be normalized. , , Map to the [-1,1] interval to eliminate the impact of differences in parameter magnitudes on model training.

[0107] Step 23: Input the input data into the initial structural configuration optimization model and output the predicted optimal structural configuration.

[0108] Specifically, a convolutional neural network model framework is built as a structural configuration optimization model. The network parameters are initialized, and the normalized displacement load, design constraints, and manufacturing constraints are input. The output is the predicted optimal structural configuration.

[0109] Step 24: Calculate the loss based on the predicted optimal structural configuration and the reference optimal structural configuration to obtain the loss value.

[0110] Specifically, the loss calculation process is not limited in the embodiment. Further, the loss value is obtained by calculating the loss according to the predicted optimal structure configuration and the reference optimal structure configuration, which can specifically include: calculating the density field loss based on the predicted optimal structure configuration and the reference optimal structure configuration; calculating the physical constraint loss based on the predicted optimal structure configuration and the reference optimal structure configuration; and calculating the loss value based on the density field loss and the physical constraint loss.

[0111] Specifically, the model training process can refer to the following process:

[0112] The total loss function formula is as follows:

[0113] Loss = a x L data1 + b x L data2 .

[0114] Wherein, a and b are weight coefficients, used to balance the influence of two parts of loss (weight coefficients are dynamically adjusted according to the training stage).

[0115] L data1 is the density field loss, and the calculation method is as follows:

[0116] .

[0117] Wherein, M is the total number of all target structure region units to be optimized in the batch; is the predicted density, i.e. the predicted optimal structure configuration; is the reference density, i.e. the reference optimal structure configuration.

[0118] It is known that the finite element discretization is obtained . Wherein, is the structure stiffness matrix, is the input displacement load, is the corresponding node force vector, and the physical constraint loss L data2 can be expressed as:

[0119] .

[0120] Wherein, is the structure stiffness of the structure configuration optimization model, which is directly related to the material density field predicted by the structure configuration optimization model, and the density value affects the element stiffness through the interpolation function; is the node force vector obtained by solving and calculating the structure optimization model.

[0121] Step 25: based on the loss value, the initial structure configuration optimization model is optimized until the loss value meets the preset value or the iteration number of training meets the maximum iteration number, the training is stopped, and the trained structure configuration optimization model is obtained.

[0122] After a training iteration is completed, the performance of the structure configuration optimization model can be evaluated by a validation set in the training data set, and the number of iterations is adjusted. Until the training loss function meets the preset value or meets the maximum number of iterations, the structure configuration optimization model capable of generating the structure end to end is output.

[0123] In order to solve the problem that the model lacks continuous evolution mechanism and cannot dynamically adjust and improve performance when encountering new samples or new constraints, which limits its applicability and flexibility in actual engineering. Further, after the displacement load, design constraint and manufacturing constraint are input into the pre-trained structure configuration optimization model to obtain the generated optimal structure configuration, the following can be included: when the optimal structure configuration does not meet the preset requirements, the external load, boundary condition, design constraint and manufacturing constraint are used as new samples; the trained structure configuration optimization model is trained until the training termination condition is met, and the trained structure configuration optimization model is updated.

[0124] Specifically, the structure configuration optimization model in the embodiment can drive the structure configuration optimization model to dynamically optimize through incremental learning. For example, when it is found that the optimal structure configuration predicted by the structure configuration optimization model does not meet the requirements, such as the difference with the result solved by the physical constraint driven configuration optimization framework is large, the manufacturing path is not available or a completely new optimization constraint combination is encountered, the system automatically takes this case as a new sample to dynamically optimize the structure configuration optimization model, so that it can continuously improve and adapt to new requirements.

[0125] The application embodiment provided by the application decouples the global grid model to obtain a reduced order model and a structure optimization model; external loads and boundary conditions are applied to the reduced order model, and displacement loads are obtained by solving; the displacement loads, design constraints and manufacturing constraints are input into the pre-trained structure configuration optimization model to obtain a generated optimal structure configuration; wherein the global grid model is a finite element model of the whole structure of the vehicle body; the reduced order model includes a non-design space area, a load application area and a boundary constraint area; and the structure optimization model includes a target structure area to be optimized and a grid transition area. The method decouples the complex global grid model into two parts of the reduced order model and the structure optimization model, and converts the complex external loads and boundary conditions into an intermediate variable of displacement loads, which greatly reduces the calculation scale of the structure configuration optimization model and improves the calculation efficiency, so as to improve the grid resolution in the target structure area to be optimized, generate a high-fidelity and smooth structure configuration, reduce unnecessary materials and make the train body lighter without affecting the safety of the train; and under the premise of maintaining similar geometric characteristics of the target structure area to be optimized, a high-fidelity and smooth structure configuration can be generated for any size (referring to adjustable size parameters of the target structure limited by design constraints and manufacturing constraints), external loads and boundary conditions, realizing universality in the global parameter space; and the method also integrates an incremental learning mechanism, which can autonomously optimize the prediction results for new constraint combinations to continuously improve the model adaptability and evolution ability.

[0126] The structure configuration generation device provided by the application embodiment is described below, and the structure configuration generation device described below can be correspondingly referred to the structure configuration generation method described above.

[0127] For details, please refer to Figure 5 , Figure 5 A structure configuration generation device provided by the application embodiment can include:

[0128] The decoupling module 100 is configured to decouple the global grid model to obtain a reduced order model and a structure optimization model;

[0129] The solving module 200 is configured to apply external loads and boundary conditions to the reduced order model and solve to obtain displacement loads;

[0130] The generation module 300 is configured to input the displacement loads, design constraints and manufacturing constraints into the pre-trained structure configuration optimization model to obtain a generated optimal structure configuration;

[0131] The global grid model is a finite element model of the whole body structure; the reduced order model includes a non-design space area, a load application area and a boundary constraint area; the structure optimization model includes a target structure area to be optimized and a grid transition area; and the optimal structure configuration is an optimal structure configuration of the target structure area to be optimized.

[0132] Based on the above embodiment, the solving module 200 can include:

[0133] An acquisition unit is configured to apply the external load and the boundary condition to the reduced order model to obtain a total stiffness matrix of the reduced order model and equivalent node forces applied to each node in the reduced order model;

[0134] A displacement vector calculation unit is configured to calculate displacement vectors of all nodes in the reduced order model based on the total stiffness matrix and the equivalent node forces;

[0135] A selection unit is configured to selectively filter the displacement vectors of all nodes to obtain the displacement vectors at the junction of the reduced order model and the structure optimization model as the displacement load.

[0136] Based on the above embodiment, the structure configuration generation device can further include:

[0137] A training data set acquisition module is configured to acquire a training data set containing various training samples; the training samples include a reference optimal structure configuration and a displacement load for training, a design constraint for training and a manufacturing constraint for training corresponding to the reference optimal structure configuration;

[0138] A normalization module is configured to normalize the displacement load for training, the design constraint for training and the manufacturing constraint for training to obtain normalized input data;

[0139] An input and output module is configured to input the input data into an initial structure configuration optimization model and output a predicted optimal structure configuration;

[0140] A loss calculation module is configured to calculate a loss value according to the predicted optimal structure configuration and the reference optimal structure configuration;

[0141] An optimization module is configured to optimize the initial structure configuration optimization model based on the loss value until the loss value meets a preset value or the number of iterations meets a maximum number of iterations, and then stop training to obtain a trained structure configuration optimization model.

[0142] Based on the above embodiment, the training data set acquisition module can include:

[0143] determining, according to a design space of a target structure to be optimized, an external load for training, a boundary condition for training, a design constraint for training, and a manufacturing constraint for training;

[0144] a parameter set generation unit configured to generate a parameter combination set covering a parameter space based on the external load for training, the boundary condition for training, the design constraint for training, and the manufacturing constraint for training, by using a Latin hypercube method, wherein the parameter combination set includes N parameter combinations;

[0145] a displacement load calculation unit configured to apply the external load for training and the boundary condition for training to the reduced-order model based on each of the parameter combinations, and to solve the reduced-order model to obtain a displacement load for training;

[0146] an optimal structure configuration generation unit configured to obtain a reference optimal structure configuration based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training;

[0147] a training data set construction unit configured to construct the training data set based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training, and the corresponding reference optimal structure configuration.

[0148] Based on the above embodiment, the optimal structure configuration generation unit can include:

[0149] an optimization target determination subunit configured to apply the displacement load for training to a boundary node of the structure optimization model to maximize the flexibility of the structure optimization model as an optimization target;

[0150] a variable determination subunit configured to take the material density of a cell in the target structure to be optimized as a variable;

[0151] a constraint condition determination subunit configured to take the design constraint for training and the manufacturing constraint for training as constraint conditions;

[0152] a solving subunit configured to obtain the optimal material density of each cell in the target structure to be optimized as the reference optimal structure configuration by using a gradient optimization algorithm to iteratively solve based on the optimization target, the variable, and the constraint conditions.

[0153] Based on the above embodiment, the loss calculation module can include:

[0154] a density field loss calculation unit configured to calculate a density field loss based on the predicted optimal structure configuration and the reference optimal structure configuration;

[0155] a physical constraint loss calculation unit configured to calculate a physical constraint loss based on the predicted optimal structure configuration and the reference optimal structure configuration;

[0156] a loss value calculation unit configured to calculate the loss value based on the density field loss and the physical constraint loss.

[0157] Based on the above embodiment, the structure configuration generation apparatus can further comprise:

[0158] a new sample generation module configured to generate a new sample based on the external load, the boundary condition, the design constraint and the manufacturing constraint when the optimal structure configuration does not meet the preset requirement;

[0159] a model optimization module configured to train the trained structure configuration optimization model using the new sample until a training termination condition is met, and update the trained structure configuration optimization model.

[0160] It should be noted that the order of the modules and units in the above structure configuration generation apparatus can be changed without affecting the logic.

[0161] The structure configuration generation apparatus provided by the embodiment of the present application comprises a decoupling module 100 configured to decouple a global grid model to obtain a reduced-order model and a structure optimization model; a solving module 200 configured to apply an external load and a boundary condition to the reduced-order model and solve to obtain a displacement load; and a generation module 300 configured to input the displacement load, a design constraint and a manufacturing constraint into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration. The apparatus decouples a complex global grid model into two parts of a reduced-order model and a structure optimization model, and converts a complex external load and a boundary condition into an intermediate variable of a displacement load, thereby greatly reducing the calculation scale of the structure configuration optimization model and improving the calculation efficiency. In this way, the grid resolution in the target structure region to be optimized can also be improved, a high-fidelity and smooth structure configuration is generated, unnecessary materials are reduced to make the train body lighter without affecting the safety of the train, and a high-fidelity and smooth structure configuration is generated for any size (referring to adjustable size parameters of the target structure, limited by the design constraint and the manufacturing constraint), external load and boundary condition under the premise of keeping similar geometric characteristics of the target structure region to be optimized, achieving universality in the global parameter space. Furthermore, the apparatus also integrates an incremental learning mechanism, which can autonomously optimize the prediction results for new constraint combinations and continuously improve the model adaptability and evolution ability.

[0162] The structural configuration generation device provided by the embodiment of the present application is described below. The structural configuration generation device described below can be referred to in correspondence with the structural configuration generation method described above.

[0163] Please refer to Figure 6 , Figure 6 The structural configuration generation device provided by the embodiment of the present application can include:

[0164] The memory 10 is configured to store a computer program.

[0165] The processor 20 is configured to execute the computer program to implement the structural configuration generation method described above.

[0166] The memory 10, the processor 20 and the communication interface 31 can communicate with each other through the communication bus 32.

[0167] In the embodiment of the present application, the memory 10 stores one or more programs. The program can include program code, and the program code includes computer operation instructions. In the embodiment of the present application, the memory 10 can store programs for implementing the following functions:

[0168] Decoupling the global grid model to obtain a reduced model and a structure optimization model;

[0169] Applying external loads and boundary conditions to the reduced model and solving to obtain displacement loads;

[0170] Inputting the displacement loads, design constraints and manufacturing constraints into the pre-trained structure configuration optimization model to obtain a generated optimal structure configuration;

[0171] The global grid model is a finite element model of the whole vehicle body structure. The reduced model includes a non-design space area, a load application area and a boundary constraint area. The structure optimization model includes a target structure area to be optimized and a grid transition area. The optimal structure configuration is the optimal structure configuration of the target structure area to be optimized.

[0172] In a possible implementation, the memory 10 can include a program storage area and a data storage area. The program storage area can store an operating system and at least one application program required by a function, etc. The data storage area can store data created during use.

[0173] Further, the memory 10 can include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory can also include a NVRAM. The memory stores an operating system and operation instructions, executable modules or data structures, or a subset or extension thereof, wherein the operation instructions can include various operation instructions for implementing various operations. The operating system can include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0174] The processor 20 can be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic device, and can be a microprocessor or any conventional processor, etc. The processor 20 can invoke programs stored in the memory 10.

[0175] The communication interface 31 can be an interface of a communication module, for connecting with other devices or systems.

[0176] Of course, it needs to be explained that, Figure 6 The structure shown does not constitute a limitation on the structure configuration generation device in the embodiments of the present application, and in actual applications, the structure configuration generation device can include more or less components than Figure 6 those shown, or combine certain components.

[0177] The computer readable storage medium provided by the embodiments of the present application is described below, and the computer readable storage medium described below can be referred to each other corresponding to the structure configuration generation method described above.

[0178] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the structure configuration generation method described above.

[0179] The computer readable storage medium can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0180] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0181] Those skilled in the art will further realize that the mere conception of the examples described herein is not inducing any patentable instrument, and that each example presents only one illustrative aspect of the present application. The present application is thus deemed to cover any and all adaptations or variations of preferred examples. It is intended to embrace each and every possible modification or variation of this application that can come within the scope of the present application. It is intended to include also any and all comebacks or equivalents of the moieties eluded herein.

[0182] Finally, it is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component" includes reference to one or more components.

[0183] The above detailed description merely describes a specific implementation of the application. The application is not limited to the specific examples described herein, but also covers any and all alternatives, modifications, changes, and equivalents of its examples falling within the scope of the application. Accordingly, the description is to be regarded as illustrative in nature and is not to be considered limiting to the scope of the application.

Claims

1. A method of generating a structural configuration, characterized by, The application relates to a method for generating an optimal structure configuration of a vehicle body, comprising the following steps: decoupling a global grid model to obtain a reduced-order model and a structure optimization model; applying external loads and boundary conditions to the reduced-order model and solving to obtain displacement loads; inputting the displacement loads, design constraints and manufacturing constraints into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration; the design constraints include upper and lower limits of volume constraints, modal constraints and mass constraints, and the manufacturing constraints include draft constraints, symmetry constraints and extrusion constraints; wherein the global grid model is a finite element model of a whole vehicle body structure; the reduced-order model comprises a non-design space area, a load application area and a boundary constraint area; the structure optimization model comprises a target structure area to be optimized and a grid transition area; the optimal structure configuration is an optimal structure configuration of the target structure area to be optimized; the grid resolutions of the target structure area to be optimized, the grid transition area and the reduced-order model are successively reduced; applying external loads and boundary conditions to the reduced-order model and solving to obtain displacement loads, comprising the following steps: applying the external loads and the boundary conditions to the reduced-order model to obtain a total stiffness matrix of the reduced-order model and equivalent node forces applied to each node in the reduced-order model; based on the total stiffness matrix and the equivalent node forces, calculating displacement vectors of all nodes in the reduced-order model; selectively screening the displacement vectors of all nodes to obtain the displacement vectors at the junction of the reduced-order model and the structure optimization model as the displacement loads; after inputting the displacement loads, design constraints and manufacturing constraints into the pre-trained structure configuration optimization model to obtain the generated optimal structure configuration, further comprising the following steps: when the optimal structure configuration does not meet preset requirements, taking the external loads, the boundary conditions, the design constraints and the manufacturing constraints as new samples; training the trained structure configuration optimization model by using the new samples until a training termination condition is met, and updating the trained structure configuration optimization model; the total loss function of the pre-trained structure configuration optimization model comprises: ； a and b are weight coefficients, respectively; is a density field loss; is a physical constraint loss; ； ； Wherein, M is the total number of all target structure region units in the batch to be optimized; is the predicted density, i.e., the predicted optimal structure configuration; is the reference density, i.e., the reference optimal structure configuration; is the structural stiffness of the structure configuration optimization model, which is directly related to the material density field predicted by the structure configuration optimization model, and the density value affects the unit stiffness through an interpolation function; refers to the displacement load used for training; is the node force vector obtained by solving and calculating the structure configuration optimization model.

2. The structural configuration generation method according to claim 1, characterized by, before decoupling the global grid model to obtain the reduced-order model and the structure optimization model, further comprising the following steps: obtaining a training data set containing various training samples; the training samples include a reference optimal structure configuration and displacement loads for training, design constraints for training and manufacturing constraints for training corresponding to the reference optimal structure configuration; performing normalization processing on the displacement loads for training, the design constraints for training and the manufacturing constraints for training to obtain normalized input data; inputting the input data into an initial structure configuration optimization model to output a predicted optimal structure configuration; calculating a loss value according to the predicted optimal structure configuration and the reference optimal structure configuration; optimizing the initial structure configuration optimization model based on the loss value until the loss value meets a preset value or the number of iterations of training meets a maximum number of iterations, and stopping training to obtain the trained structure configuration optimization model.

3. The structural configuration generation method according to claim 2, characterized by, loss calculation based on the predicted optimal structure configuration and the reference optimal structure configuration to obtain a loss value, including: calculating a density field loss based on the predicted optimal structure configuration and the reference optimal structure configuration; calculating a physical constraint loss based on the predicted optimal structure configuration and the reference optimal structure configuration; calculating the loss value based on the density field loss and the physical constraint loss.

4. The structural configuration generation method according to claim 2, characterized by, obtaining a training data set containing various training samples, including: determining an external load for training, a boundary condition for training, a design constraint for training, and a manufacturing constraint for training according to a design space of a target structure to be optimized; generating a parameter combination set covering a parameter space by using a Latin hypercube method based on the external load for training, the boundary condition for training, the design constraint for training, and the manufacturing constraint for training; the parameter combination set contains N parameter combinations; applying the external load for training and the boundary condition for training to the reduced-order model based on each parameter combination, and solving to obtain a displacement load for training; obtaining a reference optimal structure configuration based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training; constructing the training data set based on the design constraint for training, the manufacturing constraint for training, the displacement load for training, and the corresponding reference optimal structure configuration.

5. The structural configuration generation method according to claim 4, characterized by, obtaining a reference optimal structure configuration based on the design constraint for training, the manufacturing constraint for training, and the displacement load for training, including: applying the displacement load for training to the boundary nodes of the structure optimization model to maximize the flexibility of the structure optimization model as an optimization objective; taking the element material density in the target structure to be optimized as a variable; taking the design constraint for training and the manufacturing constraint for training as constraint conditions; obtaining the optimal material density of each element in the target structure to be optimized as the reference optimal structure configuration by using a gradient optimization algorithm to iteratively solve based on the optimization objective, the variable, and the constraint conditions.

6. A structure configuration generating apparatus characterized by comprising: including: a decoupling module for decoupling the global grid model to obtain a reduced-order model and a structure optimization model; a solving module for applying an external load and a boundary condition to the reduced-order model and solving to obtain a displacement load; a generating module for inputting the displacement load, a design constraint, and a manufacturing constraint into a pre-trained structure configuration optimization model to obtain a generated optimal structure configuration; the design constraint includes upper and lower limits of a volume constraint, a modal constraint, and a mass constraint, and the manufacturing constraint includes a draw constraint, a symmetry constraint, and an extrusion constraint; The global grid model is a finite element model of the whole structure of the vehicle body; the reduced order model includes a non-design space area, a load application area, and a boundary constraint area; the structure optimization model includes a target structure area to be optimized and a grid transition area; the optimal structure configuration is an optimal structure configuration of the target structure area to be optimized; the grid resolution of the target structure area to be optimized, the grid transition area, and the reduced order model decreases in turn; The solving module comprises: The obtaining unit is configured to apply the external load and the boundary condition to the reduced order model to obtain a total stiffness matrix of the reduced order model and equivalent node forces applied to each node in the reduced order model; The displacement vector calculation unit is configured to calculate displacement vectors of all nodes in the reduced order model based on the total stiffness matrix and the equivalent node forces; The selection unit is configured to selectively filter the displacement vectors of all nodes to obtain the displacement vectors at the junction of the reduced order model and the structure optimization model as the displacement load; Further comprising: The new sample generation module is configured to use the external load, the boundary condition, the design constraint, and the manufacturing constraint as a new sample when the optimal structure configuration does not meet the preset requirement; The model optimization module is configured to train the trained structure configuration optimization model using the new sample until a training termination condition is met, and update the trained structure configuration optimization model; The total loss function of the pre-trained structure configuration optimization model comprises: ； a and b are weight coefficients, respectively; is a density field loss; is a physical constraint loss; ； ； Wherein, M is the total number of all target structure region units in the batch to be optimized; is the predicted density, i.e., the predicted optimal structure configuration; is the reference density, i.e., the reference optimal structure configuration; is the structural stiffness of the structure configuration optimization model, which is directly related to the material density field predicted by the structure configuration optimization model, and the density value affects the unit stiffness through an interpolation function; refers to the displacement load for training; is the node force vector obtained by solving and calculating the structure configuration optimization model.

7. A structure configuration generating apparatus characterized by comprising: The memory is configured to store a computer program; The processor is configured to implement the steps of the structure configuration generation method according to any one of claims 1 to 5 when executing the computer program. The computer executable instructions are stored in the readable storage medium, and when loaded and executed by the processor, the steps of the structure configuration generation method according to any one of claims 1 to 5 are implemented.

8. A readable storage medium, characterized by, ​

Images