A design method for topology optimization of stiffened steel plate shear walls based on generative adversarial networks

Abstract

The present invention discloses a design method for topology optimization of a reinforced steel plate shear wall based on a generative adversarial network. The method comprises the following steps: S1, given a steel plate shear wall, constructing a corresponding steel plate shear wall structure, optimizing the structure using a bidirectional progressive structural optimization method, and obtaining an optimized reinforced steel plate shear wall structure; S2, based on the original steel plate shear wall structure in S1, obtaining its corresponding steel plate shear wall S, Mises diagram and ESEDEN diagram, and based on the optimized reinforced steel plate shear wall structure in S1, obtaining its corresponding steel plate shear wall stiffening rib arrangement diagram to construct a reinforced steel plate shear wall stiffening rib arrangement sample database; S3, constructing a new modified and optimized pix2pixHD generative adversarial network model; S4, based on the improved and optimized generative adversarial network model in S3, training the generative adversarial network model through the reinforced steel plate shear wall stiffening rib arrangement sample database in S3; S5, using a comprehensive evaluation index Score IoU Test the trained model and put it into use after passing the evaluation.

Description

Technical Field

[0001] The present invention belongs to the field of civil engineering and computer deep learning application technology, and in particular to a design method for topological optimization of reinforced steel plate shear walls based on a generative adversarial network. Background Art

[0002] Steel plate shear walls are lateral-load-resisting components with excellent ductility and energy dissipation capacity. However, unreinforced steel plate shear walls have numerous drawbacks, such as significant pinching in their hysteretic curves and a tendency to experience out-of-plane buckling during loading, which can generate noise. Stiffened steel plate shear walls, on the other hand, improve their performance by arranging certain stiffening ribs. Common stiffening methods include horizontal and vertical stiffeners. However, the optimal or preferable arrangement of these stiffening ribs has not been verified through optimization, and design and verification still rely on engineers' experience. With the increasing advancement of intelligent building and civil engineering, it is necessary and urgent to propose an efficient and universal topology optimization method for stiffened steel plate shear walls to promote intelligent building structural design. With the development of computer vision, artificial intelligence and other methods have provided new means and approaches for solving various technical problems in civil engineering. However, AI-driven and assisted applications are currently only widely used in engineering construction and operation and maintenance, and their integration into structural design is relatively rare. Summary of the Invention

[0003] Purpose of the invention: The technical problem to be solved by the present invention is to overcome the current method of arranging stiffening ribs relying on the experience of engineers and propose a topology optimization design method for stiffened steel plate shear walls based on a generative adversarial network. This method can perform the optimal arrangement of stiffening ribs for a given steel plate shear wall. Compared with relying on the experience of engineers and ordinary topology optimization design methods, this method is more efficient and has a wider range of applications, and can realize the rapid and intelligent design of stiffened steel plate shear wall structures.

[0004] Technical solution, in order to solve the above technical problems, the present invention proposes a design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks, the method comprising the following steps:

[0005] Step S1: obtaining a structural diagram of a steel plate shear wall that requires stiffening rib arrangement, constructing a corresponding steel plate shear wall "sandwich" structure, optimizing the structure using a bidirectional progressive structural optimization method, and obtaining an optimized stiffened steel plate shear wall structural diagram;

[0006] Step S2: Generate the corresponding S.Mises diagram and ESEDEN diagram of the steel plate shear wall using the steel plate shear wall structural diagram, generate the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall using the stiffened steel plate shear wall structural diagram, and construct a training data set and a test data set;

[0007] Step S3: Input the steel plate shear wall S.Mises diagram and ESEDEN diagram obtained in step S2 into a generative adversarial network model as trainA for training. The output of the generative adversarial network model is a stiffening rib arrangement diagram of the stiffened steel plate shear wall. The training obtains a generative adversarial network model that takes the steel plate shear wall S.Mises diagram and ESEDEN diagram as input and outputs the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall.

[0008] Step S4: Use the test set to test the generative adversarial network model of step S3, and calculate the comprehensive evaluation index Score of the stiffening rib layout diagram of the reinforced steel plate shear wall output by the generative adversarial network model. IoU , determine whether the comprehensive evaluation index value is greater than a preset threshold. If it is, the generated adversarial neural network model meets the requirements. If not, the generated adversarial neural network model does not meet the requirements and retraining is performed through steps S1-S3;

[0009] Step S5: The preset S.Mises diagram of the steel plate shear wall and the ESEDEN diagram are fused and input into the generative adversarial network model trained in step S3 to obtain the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall.

[0010] Furthermore, in step S2, the steps for obtaining the S.Mises diagram and ESEDEN diagram of the steel plate shear wall and the stiffening rib arrangement diagram of the stiffened steel plate shear wall are as follows:

[0011] (2.1) The steps to obtain the S.Mises graph and ESEDEN graph are as follows:

[0012] (2.1.1) Based on the steel plate shear wall structure diagram, a corresponding steel plate shear wall model was established in the finite element analysis software ABAQUS. The connection plate was not considered in the finite element modeling, and all connections were simulated using tie connections. The columns and column stiffeners were merged into the same component through the merge method. The bottom of the side columns of the steel plate shear wall was fixed to the ground, and a fully fixed constraint was applied to limit all degrees of freedom to zero. UZ = 0 was set for the top beam to simulate lateral support. For loading, a reference point was selected at the right column of the model, and a horizontal displacement was applied to UX in the positive direction through coupling. The inter-story drift angle applied during the elastic analysis was 1 / 500. S4R elements were selected for all components of the model, and the mesh size was set to 40.

[0013] (2.1.2) Create a field output in the analysis step module, with the scope being the middle steel plate section of the steel plate shear wall, and the output variables set to MISES, Mises equivalent stress, and ELEDEN, all energy density components;

[0014] (2.1.3) Create and submit the Job file. After the job is completed, open the Job.odb file. In the visualization module, draw a cloud diagram on the undeformed image to obtain the S.Mises diagram and ESEDEN diagram of the middle steel plate.

[0015] (2.2) The steps for obtaining the stiffening rib layout diagram of the reinforced steel plate shear wall are as follows:

[0016] (2.2.1) Establishment of mathematical optimization model: Selecting a mathematical model for linear structural stiffness optimization, with maximum stiffness as the optimization goal;

[0017]

[0018] Where C is the minimum strain energy of the steel plate shear wall structure, F and U are the load vector and displacement vector of the steel plate shear wall respectively, and V * represents the volume ratio of the preset steel plate shear wall stiffener, M is the number of elements in the optimization area, Ve represents the volume fraction of the e-th element, x e Indicates that the relative density value of the e-th unit is 1 or x min , 1 represents the real unit, x min Take 0.001 to represent a virtual unit;

[0019] (2.2.2) Improvements to the bidirectional progressive structural optimization method: Improvement 1: Change the solid element to a shell element; Improvement 2: Change the material property of the solid element to an elastic modulus of 2.06*10 5 N / mm 2 and Poisson's ratio 0.3, and change the virtual element material properties to elastic modulus 2.06*10 -4 N / mm 2 and Poisson's ratio 0.3; Improvement 3, change one optimization component into two optimization components, and the corresponding related variables should also be increased accordingly, including component part, unit elmts, unit node nds, design variable xe, sensitivity ae, previous generation sensitivity oae and dictionary fm for saving weight factors.

[0020] (2.2.3) Establish the finite element optimization model. The stiffeners and the intermediate steel plate are constrained by Ties to create a steel plate wall. The material is the S4R shell structure. The stiffeners are distributed on both sides of the steel plate to form a "sandwich" structure. The optimized area is not connected to the surrounding frame, but only to the steel plate. The optimization model does not assign material properties to the optimized area. The rest of the modeling method is the same as 2.1.1.

[0021] (2.4) The steps for optimizing the design of the reinforced steel plate shear wall structure are as follows:

[0022] (2.4.1) Establish a steel plate shear wall model in ABAQUS, set the stiffener optimization regions to Part-1 and Part-2 respectively, do not assign material properties to Part-1 and Part-2, and use meshing technology to discretize the optimization regions;

[0023] (2.4.2) Set the parameters required for optimization, including volume retention ratio, filter radius, and evolution rate;

[0024] (2.4.3) Assign initial element material properties to Part-1 and Part-2;

[0025] (2.4.4) Calculate the weighting factor of each unit sensitivity;

[0026] (2.4.5) Perform finite element analysis on the steel plate shear wall model;

[0027] (2.4.6) Extract the unit sensitivity of Part-1 and Part-2, output the objective function and save it;

[0028] (2.4.7) The unit sensitivity is added to the unit sensitivity of the previous generation and averaged to improve the sensitivity stability, and saved for subsequent iterations;

[0029] (2.4.8) Calculate the volume retention ratio of the next iteration of the stiffener;

[0030] (2.4.9) The global sensitivity is obtained by averaging the corresponding unit sensitivities in Part-1 and Part-2. The threshold of the stiffener unit sensitivity is determined using the dichotomy method. The sensitivity of each unit is judged, and stiffener units are added or deleted based on the threshold.

[0031] (2.4.10) Repeat steps (2.4.5) to (2.4.9) until the volume reaches the preset volume retention ratio and meets the convergence criteria;

[0032] (2.5) Obtain the stiffener layout diagram of the reinforced steel plate shear wall. Run the script of the bidirectional progressive structural optimization method based on the finite element optimization model to obtain the Final_design.cae file. In this file, open the structure mesh and display the structure in the previous view state in the assembly interface. Then hide the other parts and only display the part-1-1 part. Select the set label in the color coding task bar, display the solid element part in red, and the virtual element part in green to obtain the final stiffener layout diagram.

[0033] Furthermore, in step S2, the perspective of the S.Mises diagram and ESEDEN diagram of the steel plate shear wall and the stiffener arrangement diagram of the stiffened steel plate shear wall is selected as the front view. The scaling ratio and position of the middle steel plate in the three diagrams are kept consistent. The diagram is saved in "png" format through the print function in ABAQUS, the viewport decoration is turned off, the viewport background is set to white and the viewport background is printed, the color reduction to 256 is unchecked, and the saved image size is set to 4096*1706p.

[0034] Furthermore, in step S2, the steps for creating the training set and test set for training the generative adversarial network model are as follows:

[0035] (1) For a group of steel plate shear walls with different structures, the S.Mises diagram and ESEDEN diagram of the steel plate shear wall are obtained through finite element modeling analysis. The corresponding stiffening rib arrangement diagram of the steel plate shear wall is obtained through the bidirectional progressive structural optimization method. Moreover, the perspective of the S.Mises diagram, ESEDEN diagram and stiffening rib arrangement diagram of the steel plate shear wall is selected as the front view, and the scaling ratio and position of the middle steel plate in the three diagrams are consistent. The diagram is saved in "png" format through the print function in ABAQUS, and the saved image size is set to 4096*1706p;

[0036] (2) Data enhancement is performed on the S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall obtained in (1) to increase the amount of training sample data in the sample library. The enhancement method is offline enhancement. The S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall are rotated, translated and noise is added to increase the data volume of the training library. The enhanced images are saved in ".png" format, and the saved image size is set to 2048*1024p;

[0037] (3) The enhanced steel plate shear wall S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram obtained in (2) are divided into a training set and a test set in proportion. The training set is used for model training, and the test set is used to evaluate the model effect. Each element in the training set and the test set is a steel plate shear wall S.Mises diagram, ESEDEN diagram and the corresponding stiffener arrangement diagram.

[0038] Furthermore, in step S3, the initial learning rate of the adversarial network model during training is 1×10 -4 , the learning rate remains unchanged for the first 50 training rounds of model learning; the learning rate decays linearly for the next 50 training rounds until it decays to 0.

[0039] Furthermore, in step S4, the comprehensive evaluation index is calculated as follows:

[0040] (1) The total amount of stiffeners in the steel plate shear wall, SRratio, is calculated as follows:

[0041]

[0042] Among them, A se Represents the total area of ​​the solid unit, A ve represents the total area of ​​the virtual unit;

[0043] (2) Based on the evaluation of the confusion matrix, the weighted intersection-over-union (WIoU) is calculated as follows:

[0044]

[0045] Among them, k is 3 (representing the category, category 0 is background, category 1 is entity unit, category 2 is virtual unit, and category 3 is black grid line), i represents the true value, and j represents the predicted value; P ij Indicates that i is predicted as j, which is a false negative FN; P ji Indicates that j is predicted as i, which is a false positive FP; P ii It means that i is predicted as i, which is the true TP, w0 = 0, w1 = 0.5, w2 = 0.4, w3 = 0.1;

[0046] (3) Structural evaluation based on IoU, structural intersection over union (SIoU), the calculation formula of this indicator is:

[0047]

[0048] Among them, A inter A represents the intersection area of ​​the entity elements in the generated layout diagram and the real layout diagram. union A represents the union area of ​​the solid elements in the generated layout diagram and the real layout diagram, union =A target +A GAN -A inter , A target Represents the area of ​​the solid unit in the actual layout diagram, A GAN Indicates the area of ​​the solid element in the generated layout;

[0049] (4) The calculation formula of comprehensive evaluation index is as follows:

[0050] Score IoU =η SRratio ×(η SIoU ×SIoU+η WIoU ×WIoU)

[0051] Where SRratio, WIoU, and SIoU are the total amount of stiffeners in the steel plate shear wall in (1), the weighted intersection-over-union ratio in (2), and the structural intersection-over-union ratio in (3), respectively; η SIoU and η WIoU Represent the weight coefficients of SIoU and WIoU respectively, both are 0.5; SRratio GAN Indicates the total amount of steel plate shear wall stiffeners in the generated layout drawing, SRratio target Indicates the total amount of steel plate shear wall stiffeners in the actual layout diagram;

[0052] All comprehensive evaluation indicators Score in the test set IoU The sum is added to get the average value. If the average value is greater than the preset threshold, the generative adversarial network model is selected as the final generative adversarial network model.

[0053] Beneficial effects: Compared with the existing technology, the present invention adopts the above technical solution to achieve the following beneficial technical effects:

[0054] (1) The present invention proposes a design method for topology optimization of stiffened steel plate shear walls based on generative adversarial networks. This method applies deep learning methods in artificial intelligence to structural design, solving the problems of relying on engineers' experience in designing stiffened steel plate shear walls, which have not been proven to be optimal or better through optimization, and the low efficiency of ordinary topology optimization design.

[0055] (2) This paper utilizes an improved and optimized generative adversarial network model to overcome the limitations of current stiffened steel plate shear wall design methods and proposes a universal design method for stiffened steel plate shear wall topology optimization. If sufficient training data is available, the proposed method can be applied to the topology optimization design of various types of stiffened steel plate shear walls, thus possessing strong practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0057] Figure 1 A schematic diagram of a flow chart of a design method for topology optimization of a stiffened steel plate shear wall based on a generative adversarial network provided by an embodiment of the present invention;

[0058] Figure 2 "Sandwich" plate structure for finite element optimization model;

[0059] Figure 3 A schematic diagram for generating the S, Mises diagram and ESEDEN diagram of the steel plate shear wall in the method of the present invention;

[0060] Figure 4generating a schematic diagram for the stiffening rib arrangement diagram in the method of the present invention;

[0061] Figure 5 The specific steps for optimizing the design of the reinforced steel plate shear wall structure in the method of the present invention;

[0062] Figure 6 A schematic diagram of a sample database for the topology optimization design method of stiffened steel plate shear walls in the method of the present invention is provided;

[0063] Figure 7 A schematic diagram of the adversarial generation network structure of the present invention;

[0064] Figure 8 A schematic diagram of the use of a grid structure in the topology optimization design method for a stiffened steel plate shear wall after comprehensive evaluation is passed;

[0065] Figure 9 This is a dataset for the topology optimization design of a typical stiffened steel plate shear wall involved in this invention;

[0066] Figure 10 A diagram showing the generation of topological optimization design for the stiffened steel plate shear wall at each stage of neural network model training in the present invention. Specific implementation methods

[0067] In order to make the technical solutions, advantages and ultimate objectives of the present invention clearer, the present invention will be further explained and illustrated below with reference to the accompanying drawings and specific implementation methods.

[0068] The present invention proposes a design method for topology optimization of reinforced steel plate shear walls based on a generative adversarial network, which includes the following steps:

[0069] Step S1: obtaining a structural diagram of a steel plate shear wall that requires stiffening rib arrangement, constructing a corresponding steel plate shear wall "sandwich" structure, optimizing the structure using a bidirectional progressive structural optimization method, and obtaining an optimized stiffened steel plate shear wall structural diagram;

[0070] Step S2: Generate the corresponding S.Mises diagram and ESEDEN diagram of the steel plate shear wall using the steel plate shear wall structural diagram, generate the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall using the stiffened steel plate shear wall structural diagram, and construct a training data set and a test data set;

[0071] Step S3: Input the steel plate shear wall S.Mises diagram and ESEDEN diagram obtained in step S2 into a generative adversarial network model as trainA for training. The output of the generative adversarial network model is a stiffening rib arrangement diagram of the stiffened steel plate shear wall. The training obtains a generative adversarial network model that takes the steel plate shear wall S.Mises diagram and ESEDEN diagram as input and outputs the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall.

[0072] Step S4: Use the test set to test the generative adversarial network model of step S3, and calculate the comprehensive evaluation index Score of the stiffening rib layout diagram of the reinforced steel plate shear wall output by the generative adversarial network model. IoU , determine whether the comprehensive evaluation index value is greater than a preset threshold. If it is, the generated adversarial neural network model meets the requirements. If not, the generated adversarial neural network model does not meet the requirements and retraining is performed through steps S1-S3;

[0073] Step S5: The preset S.Mises diagram of the steel plate shear wall and the ESEDEN diagram are fused and input into the generative adversarial network model trained in step S3 to obtain the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall.

[0074] Furthermore, in step S2, the steps for obtaining the S,Mises diagram and ESEDEN diagram of the steel plate shear wall and the stiffening rib arrangement diagram of the stiffened steel plate shear wall are as follows:

[0075] (2.1) The steps to obtain the S.Mises graph and ESEDEN graph are as follows:

[0076] (2.1.1) Based on the steel plate shear wall structure diagram, a corresponding steel plate shear wall model was established in the finite element analysis software ABAQUS. The connection plate was not considered in the finite element modeling, and all connections were simulated using tie connections. The columns and column stiffeners were merged into the same component through the merge method. The bottom of the side columns of the steel plate shear wall was fixed to the ground, and a fully fixed constraint was applied to limit all degrees of freedom to zero. UZ = 0 was set for the top beam to simulate lateral support. For loading, a reference point was selected at the right column of the model, and a horizontal displacement was applied to UX in the positive direction through coupling. The inter-story drift angle applied during the elastic analysis was 1 / 500. S4R elements were selected for all components of the model, and the mesh size was set to 40.

[0077] (2.1.2) Create a field output in the analysis step module, with the scope being the middle steel plate section of the steel plate shear wall, and the output variables set to MISES, Mises equivalent stress, and ELEDEN, all energy density components;

[0078] (2.1.3) Create and submit the Job file. After the job is completed, open the Job.odb file. In the visualization module, draw a cloud diagram on the undeformed diagram to obtain the S.Mises diagram and ESEDEN diagram of the middle steel plate.

[0079] (2.2) The steps for obtaining the stiffening rib layout diagram of the reinforced steel plate shear wall are as follows:

[0080] (2.2.1) Establishment of mathematical optimization model: Selecting a mathematical model for linear structural stiffness optimization, with maximum stiffness as the optimization goal;

[0081]

[0082] Where C is the minimum strain energy of the steel plate shear wall structure, F and U are the load vector and displacement vector of the steel plate shear wall respectively, and V * represents the volume ratio of the preset steel plate shear wall stiffener, M is the number of elements in the optimization area, Ve represents the volume fraction of the e-th element, x e Indicates that the relative density value of the e-th unit is 1 or x min , 1 represents the real unit, x min Take 0.001 to represent a virtual unit;

[0083] (2.2.2) Improvements to the bidirectional progressive structural optimization method include: 1. Changing the solid element to a shell element; 2. Changing the material properties of the solid element to an elastic modulus of 2.06*10 5 N / mm 2 and Poisson's ratio 0.3, and change the virtual element material properties to elastic modulus 2.06*10 -4 N / mm 2 and Poisson's ratio 0.3; 3. Change one optimization component into two optimization components, and the corresponding related variables are also increased accordingly, including component part, unit elmts, unit node nds, design variable xe, sensitivity ae, previous generation sensitivity oae and dictionary fm for saving weight factors;

[0084] (2.2.3) Establish the finite element optimization model. The stiffeners and the intermediate steel plate are constrained by Ties to create a steel plate wall. The material is the S4R shell structure. The stiffeners are distributed on both sides of the steel plate to form a "sandwich" structure. The optimized area is not connected to the surrounding frame, but only to the steel plate. The optimization model does not assign material properties to the optimized area. The rest of the modeling method is the same as 2.1.1.

[0085] (2.4) The steps for optimizing the design of the reinforced steel plate shear wall structure are as follows:

[0086] (2.4.1) Establish a steel plate shear wall model in ABAQUS, set the stiffener optimization regions to Part-1 and Part-2 respectively, do not assign material properties to Part-1 and Part-2, and use meshing technology to discretize the optimization regions;

[0087] (2.4.2) Set the parameters required for optimization, including volume retention ratio, filter radius, and evolution rate;

[0088] (2.4.3) Assign initial element material properties to Part-1 and Part-2;

[0089] (2.4.4) Calculate the weighting factor of each unit sensitivity;

[0090] (2.4.5) Perform finite element analysis on the steel plate shear wall model;

[0091] (2.4.6) Extract the unit sensitivity of Part-1 and Part-2, output the objective function and save it;

[0092] (2.4.7) The unit sensitivity is added to the unit sensitivity of the previous generation and averaged to improve the sensitivity stability, and saved for subsequent iterations;

[0093] (2.4.8) Calculate the volume retention ratio of the next iteration of the stiffener;

[0094] (2.4.9) The global sensitivity is obtained by averaging the corresponding unit sensitivities in Part-1 and Part-2. The threshold of the stiffener unit sensitivity is determined using the dichotomy method. The sensitivity of each unit is judged, and stiffener units are added or deleted based on the threshold.

[0095] (2.4.10) Repeat steps (2.4.5) to (2.4.9) until the volume reaches the preset volume retention ratio and meets the convergence criteria;

[0096] (2.5) Obtain the stiffener layout diagram of the reinforced steel plate shear wall. Run the script of the bidirectional progressive structural optimization method based on the finite element optimization model to obtain the Final_design.cae file. In this file, open the structure mesh and display the structure in the previous view state in the assembly interface. Hide the other parts and only display the part-1-1 part. Select the set label in the color coding task bar, display the solid element part in red, and the virtual element part in green to obtain the final stiffener layout diagram.

[0097] Furthermore, in step S2, the perspective of the S.Mises diagram and ESEDEN diagram of the steel plate shear wall and the stiffener arrangement diagram of the stiffened steel plate shear wall is selected as the front view. The scaling ratio and position of the middle steel plate in the three diagrams are kept consistent. The diagram is saved in "png" format through the print function in ABAQUS, the viewport decoration is turned off, and the viewport background is set to white and printed. The option to reduce to 256 colors is not checked, and the saved image size is set to 4096*1706p.

[0098] Furthermore, in step S2, the steps for creating the training set and test set for training the generative adversarial network model are as follows:

[0099] (1) For a group of steel plate shear walls with different structures, the S.Mises diagram and ESEDEN diagram of the steel plate shear wall are obtained through finite element modeling analysis. The corresponding stiffening rib arrangement diagram of the steel plate shear wall is obtained through the bidirectional progressive structural optimization method. Moreover, the perspective of the S.Mises diagram, ESEDEN diagram and stiffening rib arrangement diagram of the steel plate shear wall is selected as the front view, and the scaling ratio and position of the middle steel plate in the three diagrams are consistent. The diagram is saved in "png" format through the print function in ABAQUS, and the saved image size is set to 4096*1706p;

[0100] (2) Data enhancement is performed on the S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall obtained in (1) to increase the amount of training sample data in the sample library. The enhancement method is offline enhancement. The S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall are rotated, translated and noise is added to increase the data volume of the training library. The enhanced images are saved in ".png" format, and the saved image size is set to 2048*1024p;

[0101] (3) The enhanced steel plate shear wall S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram obtained in (2) are divided into a training set and a test set in proportion. The training set is used for model training, and the test set is used to evaluate the model effect. Each element in the training set and the test set is a steel plate shear wall S.Mises diagram, ESEDEN diagram and the corresponding stiffener arrangement diagram.

[0102] Furthermore, in step S3, the initial learning rate of the adversarial network model during training is 1×10 -4 , the learning rate remains unchanged for the first 50 training rounds of model learning; the learning rate decays linearly for the next 50 training rounds until it decays to 0.

[0103] Furthermore, in step S4, the comprehensive evaluation index is calculated as follows:

[0104] (1) The total amount of stiffeners in the steel plate shear wall, SRratio, is calculated as follows:

[0105]

[0106] Among them, A se Represents the total area of ​​the solid unit, A ve Represents the total area of ​​the virtual cell.

[0107] (2) Based on the evaluation of the confusion matrix, the weighted intersection-over-union (WIoU) is calculated as follows:

[0108]

[0109] Among them, k is 3 (representing the category, category 0 is background, category 1 is entity unit, category 2 is virtual unit, and category 3 is black grid line), i represents the true value, and j represents the predicted value; P ij Indicates that i is predicted as j, which is a false negative (FN); P ji Indicates that j is predicted as i, which is a false positive (FP); P ii It means that i is predicted to be i, which is true (TP). w0=0, w1=0.5, w2=0.4, w3=0.1.

[0110] (3) Structural evaluation based on IoU, structural intersection over union (SIoU), the calculation formula of this indicator is:

[0111]

[0112] Among them, A inter A represents the intersection area of ​​the entity elements in the generated layout diagram and the real layout diagram. union A represents the union area of ​​the solid elements in the generated layout diagram and the real layout diagram, union =A target +A GAN -A inter , A target Represents the area of ​​the solid unit in the actual layout diagram, A GAN Indicates the area of ​​the solid element in the generated layout diagram.

[0113] (4) The calculation formula of comprehensive evaluation index is as follows:

[0114] Score IoU =η SRratio ×(η SIoU ×SIoU+η WIoU ×WIoU)

[0115] Where SRratio, WIoU, and SIoU are the total amount of stiffeners in the steel plate shear wall in (1), the weighted intersection-over-union ratio in (2), and the structural intersection-over-union ratio in (3), respectively; η SIoU and η WIoU Represent the weight coefficients of SIoU and WIoU respectively, both are 0.5; SRratio GAN Indicates the total amount of steel plate shear wall stiffeners in the generated layout drawing, SRratio target Indicates the total amount of steel plate shear wall stiffeners in the actual layout drawing.

[0116] All comprehensive evaluation indicators Score in the test set IoU The sum is added to get the average value. If the average value is greater than the preset threshold, the generative adversarial network model is selected as the final generative adversarial network model.

[0117] In order to verify the accuracy of the present invention, a topology optimization design of a typical stiffened steel plate shear wall was carried out. Figure 9 This is the dataset for the topology optimization design of a typical reinforced steel plate shear wall involved in this invention. Figure 10 Figure 1 shows the topology optimization design of the reinforced steel plate shear wall at each stage of neural network model training. Table 1 shows the evaluation index for the total number of stiffeners in the steel plate shear wall, the SRratio; Table 2 shows the weighted intersection-over-union ratio (WIoU), the structural intersection-over-union ratio (SIoU), and the comprehensive evaluation index (ScoreIoU). Combining Tables 1 and 2, the comprehensive evaluation index for the topology optimization design of the reinforced steel plate shear wall for the typical dataset was calculated to be 0.9146, exceeding the preset threshold of 0.8. Figure 9 、 Figure 10 Tables 1 and 2 illustrate the reliability and accuracy of this method from qualitative and quantitative perspectives, respectively, thereby proving the effectiveness of the design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks proposed in this method.

[0118] Table 1 Evaluation index of the total amount of stiffeners in steel plate shear wall SRratio

[0119] <![CDATA[SRratio GAN ]]> <![CDATA[SRratio target ]]> <![CDATA[η Srratio ]]> Typical data sets 0.1437 0.1456 0.9868

[0120] Table 2 Weighted Intersection-over-Union (WIoU), Structural Intersection-over-Union (SIoU) and Comprehensive Evaluation Index Score IoU

[0121]

[0122]

[0123] The specific embodiment described above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any replacement or equivalent transformation method that can be easily thought of by any person skilled in the art within the technical scope disclosed by the present invention should be included in the scope of protection of the present invention.

Claims

1. A design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks, characterized in that: The method comprises the following steps: Step S1: Obtain a structural diagram of a steel plate shear wall that requires stiffening rib arrangement, construct a corresponding steel plate shear wall "sandwich" structure, and optimize the structure using a bidirectional progressive structural optimization method to obtain an optimized stiffened steel plate shear wall structural diagram; Step S2: Generate the corresponding S.Mises diagram and ESEDEN diagram of the steel plate shear wall using the steel plate shear wall structural diagram, generate the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall using the stiffened steel plate shear wall structural diagram, and construct a training data set and a test data set; Step S3: Input the steel plate shear wall S.Mises diagram and ESEDEN diagram obtained in step S2 into a generative adversarial network model as trainA for training. The output of the generative adversarial network model is a stiffening rib arrangement diagram of the stiffened steel plate shear wall. The training obtains a generative adversarial network model that takes the steel plate shear wall S.Mises diagram and ESEDEN diagram as input and outputs the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall. Step S4: Use the test set to test the generative adversarial network model of step S3, and calculate the comprehensive evaluation index Score of the stiffening rib layout diagram of the reinforced steel plate shear wall output by the generative adversarial network model. IoU , determine whether the comprehensive evaluation index value is greater than a preset threshold. If it is, the generated adversarial neural network model meets the requirements. If not, the generated adversarial neural network model does not meet the requirements and retraining is performed through steps S1-S3; Step S5: The preset S.Mises diagram of the steel plate shear wall and the ESEDEN diagram are fused and input into the generative adversarial network model trained in step S3 to obtain the corresponding stiffening rib arrangement diagram of the stiffened steel plate shear wall.

2. The design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks according to claim 1 is characterized in that: In step S2, the steps for obtaining the S.Mises diagram and ESEDEN diagram of the steel plate shear wall and the stiffening rib arrangement diagram of the stiffened steel plate shear wall are as follows: (2.1) The steps to obtain the S.Mises graph and ESEDEN graph are as follows: (2.1.1) Based on the steel plate shear wall structure diagram, a corresponding steel plate shear wall model was established in the finite element analysis software ABAQUS. The connection plate was not considered in the finite element modeling, and all connections were simulated using tie connections. The columns and column stiffeners were merged into the same component through the merge method. The bottom of the side columns of the steel plate shear wall was fixed to the ground, and a fully fixed constraint was applied to limit all degrees of freedom to zero. UZ = 0 was set for the top beam to simulate lateral support. For loading, a reference point was selected at the right column of the model, and a horizontal displacement was applied to UX in the positive direction through coupling. The inter-story drift angle applied during the elastic analysis was 1 / 500. S4R elements were selected for all components of the model, and the mesh size was set to 40. (2.1.2) Create a field output in the analysis step module, with the scope being the middle steel plate section of the steel plate shear wall, and the output variables set to MISES, Mises equivalent stress, and ELEDEN, all energy density components; (2.1.3) Create and submit the Job file. After the job is completed, open the Job.odb file. In the visualization module, draw a cloud diagram on the undeformed diagram to obtain the S.Mises diagram and ESEDEN diagram of the middle steel plate. (2.2) The steps for obtaining the stiffening rib layout diagram of the reinforced steel plate shear wall are as follows: (2.2.1) Establishment of mathematical optimization model: Selecting a mathematical model for linear structural stiffness optimization, with maximum stiffness as the optimization goal; Where C is the minimum strain energy of the steel plate shear wall structure, F and U are the load vector and displacement vector of the steel plate shear wall respectively, and V * represents the volume ratio of the preset steel plate shear wall stiffener, M is the number of elements in the optimization area, Ve represents the volume fraction of the e-th element, x e Indicates that the relative density value of the e-th unit is 1 or x min , 1 represents the real unit, x min Take 0.001 to represent a virtual unit; (2.2.2) Improvements to the bidirectional progressive structural optimization method include:

1. Changing the solid element to a shell element; 2. Changing the material properties of the solid element to an elastic modulus of 2.06*10 5 N / mm 2 and Poisson's ratio 0.3, and change the virtual element material properties to elastic modulus 2.06*10 -4 N / mm 2 and Poisson's ratio 0.3; 3. Change one optimization component into two optimization components, and the corresponding related variables are also increased accordingly, including component part, unit elmts, unit node nds, design variable xe, sensitivity ae, previous generation sensitivity oae and dictionary fm for saving weight factors; (2.2.3) Establish the finite element optimization model. Tie constraints are used to create a steel plate wall with stiffeners and intermediate steel plates. The material is an S4R shell structure. Stiffeners are distributed on both sides of the steel plate to form a "sandwich" structure. The optimized area is not connected to the surrounding frame, but only to the steel plate. The optimization model does not assign material properties to the optimized area. The rest of the modeling is the same as in 2.1.

1. (2.4) The steps for optimizing the design of the reinforced steel plate shear wall structure are as follows: (2.4.1) Establish a steel plate shear wall model in ABAQUS, set the stiffener optimization regions to Part-1 and Part-2 respectively, do not assign material properties to Part-1 and Part-2, and use meshing technology to discretize the optimization regions; (2.4.2) Set the parameters required for optimization, including volume retention ratio, filter radius, and evolution rate; (2.4.3) Assign initial element material properties to Part-1 and Part-2; (2.4.4) Calculate the weighting factor of each unit sensitivity; (2.4.5) Perform finite element analysis on the steel plate shear wall model; (2.4.6) Extract the unit sensitivity of Part-1 and Part-2, output the objective function and save it; (2.4.7) The unit sensitivity is added to the unit sensitivity of the previous generation and averaged to improve the sensitivity stability, and saved for subsequent iterations; (2.4.8) Calculate the volume retention ratio of the next iteration of the stiffener; (2.4.9) The global sensitivity is obtained by averaging the corresponding unit sensitivities in Part-1 and Part-2. The threshold of the stiffener unit sensitivity is determined using the dichotomy method. The sensitivity of each unit is judged, and stiffener units are added or deleted based on the threshold. (2.4.10) Repeat steps (2.4.5) to (2.4.9) until the volume reaches the preset volume retention ratio and meets the convergence criteria; (2.5) Obtain the stiffener layout diagram of the reinforced steel plate shear wall. Run the script of the bidirectional progressive structural optimization method based on the finite element optimization model to obtain the Final_design.cae file. In this file, open the structure mesh and display the structure in the previous view state in the assembly interface. Hide the other parts and only display the part-1-1 part. Select the set label in the color coding task bar, display the solid element part in red, and the virtual element part in green to obtain the final stiffener layout diagram.

3. The design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks according to claim 1 is characterized in that: In step S2, the front view is selected for the S.Mises and ESEDEN diagrams of the steel plate shear wall and the stiffener arrangement diagram of the stiffened steel plate shear wall. The scale and position of the middle steel plate in the three diagrams are consistent. The diagrams are saved in "png" format using the print function in ABAQUS. Viewport decoration is turned off, and the viewport background is set to white and printed. The option to reduce to 256 colors is unchecked, and the saved image size is set to 4096*1706p.

4. A design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks according to claim 1 or 3, characterized in that: In step S2, the steps for creating the training set and test set for training the generative adversarial network model are as follows: (1) For a group of steel plate shear walls with different structures, the S.Mises diagram and ESEDEN diagram of the steel plate shear wall are obtained through finite element modeling analysis. The corresponding stiffening rib arrangement diagram of the steel plate shear wall is obtained through the bidirectional progressive structural optimization method. Moreover, the perspective of the S.Mises diagram, ESEDEN diagram and stiffening rib arrangement diagram of the steel plate shear wall is selected as the front view, and the scaling ratio and position of the middle steel plate in the three diagrams are consistent. The diagram is saved in "png" format through the print function in ABAQUS, and the saved image size is set to 4096*1706p; (2) Data enhancement is performed on the S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall obtained in (1) to increase the amount of training sample data in the sample library. The enhancement method is offline enhancement. The S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram of the steel plate shear wall are rotated, translated and noise is added to increase the data volume of the training library. The enhanced images are saved in ".png" format, and the saved image size is set to 2048*1024p; (3) The enhanced steel plate shear wall S.Mises diagram, ESEDEN diagram and stiffener arrangement diagram obtained in (2) are divided into a training set and a test set in proportion. The training set is used for model training, and the test set is used to evaluate the model effect. Each element in the training set and the test set is a steel plate shear wall S.Mises diagram, ESEDEN diagram and the corresponding stiffener arrangement diagram.

5. The design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks according to claim 1 is characterized in that: In step S3, the initial learning rate of the adversarial network model during training is 1×10 -4 , the learning rate remains unchanged for the first 50 training rounds of model learning; the learning rate decays linearly for the next 50 training rounds until it decays to 0.

6. The design method for topology optimization of reinforced steel plate shear walls based on generative adversarial networks according to claim 1 is characterized in that: In step S4, the comprehensive evaluation index is calculated as follows: (1) The total amount of stiffeners in the steel plate shear wall, SRratio, is calculated as follows: Among them, A se Represents the total area of ​​the solid unit, A ve represents the total area of ​​the virtual unit; (2) Based on the evaluation of the confusion matrix, the weighted intersection-over-union (WIoU) is calculated as follows: Among them, k is 3, indicating the category, category 0 is background, category 1 is entity unit, category 2 is virtual unit, category 3 is black grid line, i represents the true value, and j represents the predicted value; P ij Indicates that i is predicted as j, which is a false negative FN; P ji Indicates that j is predicted as i, which is a false positive FP; P ii It means that i is predicted as i, which is the true TP, w0 = 0, w1 = 0.5, w2 = 0.4, w3 = 0.1; (3) Structural evaluation based on IoU, structural intersection over union (SIoU), the calculation formula of this indicator is: Among them, A inter A represents the intersection area of ​​the entity elements in the generated layout diagram and the real layout diagram. union A represents the union area of ​​the solid elements in the generated layout diagram and the real layout diagram, union =A target +A GAN -A inter , A target Represents the area of ​​the solid unit in the actual layout diagram, A GAN Indicates the area of ​​the solid element in the generated layout; (4) The calculation formula of comprehensive evaluation index is as follows: Score IoU =the SRratio ×(the SIoU ×SIoU+η WIoU ×WIoU) Where SRratio, WIoU, and SIoU are the total amount of stiffeners in the steel plate shear wall in (1), the weighted intersection-over-union ratio in (2), and the structural intersection-over-union ratio in (3), respectively; η SIoU and η WIoU Represent the weight coefficients of SIoU and WIoU respectively, both are 0.5; SRratio GAN Indicates the total amount of steel plate shear wall stiffeners in the generated layout drawing, SRratio target Indicates the total amount of steel plate shear wall stiffeners in the actual layout diagram; All comprehensive evaluation indicators Score in the test set IoU The sum is added to get the average value. If the average value is greater than the preset threshold, the generative adversarial network model is selected as the final generative adversarial network model.

Images