GAN-based framework-support building structure design method, system and computer

Abstract

The application discloses a framework-support building structure design method and system based on a generative adversarial network and a computer. The design method comprises four steps of pre-training a generative adversarial network, inputting and identifying a building drawing, sequentially arranging columns, supports and a structure drawing based on the building drawing and outputting the structure drawing. Evaluation indexes of the generative adversarial network comprise not only picture quality but also subjective and objective evaluations of the building structure. According to a conditional probability formula and characteristics of the generative adversarial network, the design method decomposes a design process into two steps, improves accuracy of a design result, realizes component arrangement design of the building structure in a short time, is very accurate, has little difference in performance from a design result of a senior structure designer, is simple to use, can obtain a corresponding structure drawing by inputting a processed building drawing, can assist a building designer to complete component design of the building structure and has strong popularization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and more particularly, to a framework-support building structure design method and system based on a generative adversarial network and a computer. BACKGROUND

[0002] The structural member arrangement, as the first step and the most important step of structural design, often determines the safety performance and economic performance of the structure. The existing structural member arrangement is arranged by a structural engineer after the building drawing is preliminarily designed by an architectural designer, so that the structural engineer needs to have rich design experience and coordinate with the architectural designer to complete the arrangement, thereby consuming a large amount of manpower and financial resources. Meanwhile, with the economic development and the improvement of people's living standards, prefabricated buildings have been favored by governments and building engineers all over the world due to the characteristics of fast construction speed and low energy consumption. Among them, prefabricated steel structure buildings, as an important member of prefabricated buildings, have the characteristics of light weight, high strength, good toughness and seismic performance, and have been widely used in industrial buildings and residences. However, due to the complexity and large number of components of prefabricated steel structure buildings, the specific arrangement positions of the components need to be repeatedly negotiated during the design stage, thereby consuming a large amount of time.

[0003] In recent years, with the development of artificial intelligence, especially the proposal of generative adversarial network (GAN), it provides the possibility for automatic structural design. As an innovative algorithm of deep learning, GAN can generate high-quality pictures due to its special network structure. Therefore, GAN can be used to process building drawings to generate corresponding structural drawings. However, there are still the following problems in applying GAN to intelligent design of frame-support building structures: (1) the existing GAN model is not suitable for intelligent design of steel frame-support structures; (2) for the pictures generated by GAN, a suitable evaluation index needs to be selected to evaluate the quality of the pictures.

[0004] Therefore, it is an urgent technical problem to develop a generative adversarial network model suitable for frame-support building structures to realize intelligent design of frame-support building structure components. SUMMARY

[0005] In view of the above defects in the prior art, the present application provides a framework-support building structure design method and system based on a generative adversarial network and a computer to solve the problem of how to realize intelligent design of frame-support building structure components.

[0006] To achieve the above-mentioned purpose, on the one hand, the present application provides a framework-support building structure design method based on a generative adversarial network, characterized in that it comprises the following steps:

[0007] S1, respectively pre-training the column part and the support part of the generative adversarial network for column design and support design;

[0008] S2, semantic processing of the architectural drawing: setting different colors to mark and fill the key components in the architectural drawing, including walls, doors and windows, columns and supports;

[0009] S3, intelligent design of columns: the pre-trained generative adversarial network column part recognizes the architectural drawing obtained in step S2, and draws the arrangement position of the column on the architectural drawing;

[0010] S4, intelligent design of supports: the pre-trained generative adversarial network support part recognizes the arrangement position of the key components and columns in the architectural drawing output in step S3, further optimizes the arrangement of the columns according to the arrangement rules of the supports, so that the arrangement position of the supports is more reasonable, and the arrangement position of the supports is drawn on this basis, thereby generating an intelligently designed architectural structure drawing;

[0011] The types of architectural structure drawings used in the pre-training of steps S3 and S4 are the same as the types of architectural structure drawings to be generated;

[0012] The evaluation index of the architectural structure drawing generated in steps S3 and S4 includes not only the quality of the generated picture, but also the subjective and objective evaluation of the architectural structure.

[0013] The framework-support architectural structure design method based on the generative adversarial network provided by the application can quickly determine the optimal arrangement position of the columns and supports according to the position information of the key components in the architectural drawing, thereby assisting the architectural designer to quickly complete the design of the structural components and saving a large amount of design time.

[0014] Preferably, the design process is decoupled according to the conditional probability formula according to the framework-support structure design steps: log p(A col , A br ) = log p(A col ) + log p(A br |A col ), wherein p(A col , A br ) represents the probability distribution function of the design process of the steel framework-support structure, p(A col ) represents the distribution of arranging only columns, and p(A br |A col ) represents the distribution of arranging supports after arranging columns.

[0015] Preferably, in the generative adversarial network in steps S3 and S4, the generator comprises three parts of a down-sampling layer, a residual block and an up-sampling layer, and the discriminator is composed of two PatchGAN discriminators.

[0016] Preferably, the generative adversarial network in steps S3 and S4 is implemented by using a deep learning framework Pytorch.

[0017] Preferably, the loss function applied by the generative adversarial network in steps S3 and S4 is represented as:

[0018]

[0019] where G represents the generator, and D represents the discriminator. After decoupling, the formula is consistent with the form of the loss function of the generative adversarial network, indicating that the compatibility between the two is good.

[0020] Preferably, in step S2, the architectural drawing unifies the sizes of walls, doors and windows, columns and supports to the same thickness.

[0021] Preferably, the architectural drawings used for pre-training in steps S3 and S4 and the architectural drawings to be generated are steel frame-support or concrete frame-support buildings.

[0022] Preferably, the subjective evaluation of the building structure in the evaluation index includes the score of experts on the rationality of the details and the whole of the building structure, and the objective evaluation includes the amount of building materials and the mechanical properties.

[0023] In another aspect, the present application provides a framework-support building structure design system based on a generative adversarial network, characterized in that it comprises:

[0024] An image acquisition module is configured to acquire a preset training data set for image generation.

[0025] A loss function module is configured to acquire a preset loss function for training of the image generation model.

[0026] A decoupling module is configured to decouple the samples according to steps.

[0027] A first image generation module of the generative adversarial network is configured to design columns for the architectural drawing and generate an image.

[0028] A second image generation module of the generative adversarial network is configured to design supports for the architectural drawing with column design and generate an image.

[0029] In still another aspect, the present application provides a computer, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the building structure design method based on the generative adversarial network according to any one of the above aspects when executing the computer program.

[0030] Compared with the prior art, the above-mentioned application has the following advantages or beneficial effects:

[0031] (1) The design method of the present application divides the design process into two steps according to the conditional probability formula and the characteristics of the generative adversarial network, thereby improving the accuracy of the design result;

[0032] (2) The present application can quickly determine the optimal arrangement position of columns and supports according to the position information of key components in the architectural drawing, thereby quickly completing the design of structural components and saving a large amount of design time;

[0033] (3) The design result of the present application is very accurate, and the performance is close to that of a senior structural designer; the use method is simple, and only needs to input the processed architectural drawing to obtain the corresponding structural drawing;

[0034] (4) It can assist the architectural designer to complete the component design of frame-support building structure, and has strong popularization. BRIEF DESCRIPTION OF DRAWINGS

[0035] The present application and its features, shapes and advantages will become more apparent through reading the detailed description of the non-limiting embodiments with reference to the following drawings. The same reference signs indicate the same parts throughout the drawings. The drawings are not necessarily drawn to scale, and the emphasis is on illustrating the main idea of the present application.

[0036] Figure 1 Theoretical basis diagram of the present application;

[0037] Figure 2 Neural network structure diagram used in the embodiment of the present application;

[0038] Figure 3 Generator in the generative adversarial network used in the embodiment of the present application;

[0039] Figure 4 Discriminator in the generative adversarial network used in the embodiment of the present application;

[0040] Figure 5 Mechanical performance comparison diagram of the design result of the embodiment of the present application and the artificial design result;

[0041] Figure 6 Flowchart of the building structure design method of the embodiment of the present application. DETAILED DESCRIPTION

[0042] The structure of the present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the present invention.

[0043] Example

[0044] like Figure 1 As shown, in an embodiment of a steel frame-braced structure design, the columns are typically arranged first, followed by the bracing. Assume its distribution function is p(A col A br Considering this is a two-stage process, it can be decomposed into two sub-distributions: log p(A col A br ) = log p(A col )+log p(A br |A col ), where p(A) col ) represents the distribution of only columns, p(A) br |A col The ) represents the distribution of supports after the columns are arranged. During GAN training, if the real data distribution can be decoupled and decomposed into multiple low-dimensional sub-distributions, the learning difficulty of the GAN can be effectively reduced, making the final result closer to the real data distribution.

[0045] Based on the above theory, this invention proposes a novel generative adversarial network, such as... Figure 2 As shown, the neural network consists of two parts, each with a generator G and a discriminator D. G1 and D1 are used to determine the positions of columns based on the architectural drawings, while G2 and D2 further determine the positions of supports. The specific network structure of the generator and discriminator is as follows. Figure 3 and Figure 4 As shown.

[0046] To ensure the design results generated by the GAN are realistic and effective, this invention trains the GAN by collecting a large number of steel frame-bracing structure drawings used in actual engineering projects. To facilitate computer recognition and processing, this embodiment performs semantic processing on the collected drawings, that is, key components in the drawings, including walls, doors, windows, columns, and supports, are marked and filled with different colors. The dimensions of the components in the drawings are uniformly set to the same thickness. During training, the first part of the GAN is first trained using architectural drawings and structural drawings with only columns. Then, the second part of the GAN is trained using the structural drawings with only columns generated from the first part and the complete structural drawings.

[0047] For the pictures generated by GAN, it is very important to select the appropriate evaluation index. Considering the physical meaning of structural drawings, the evaluation index is not only limited to the quality of the generated pictures, but also needs to include professional knowledge and objective evaluation. Therefore, two evaluation indexes are adopted in this embodiment, which are expert scoring and objective comparison. Among them, the former invites experts from Tongji Design Institute to score the drawings as the evaluation standard, and the latter compares the material consumption and mechanical properties.

[0048] According to the expert scoring results, the scores of GAN and the scores of structural engineers are averaged, as shown in Table 1. After calculation, the average score of GAN is only 6.4% lower than the score of structural engineering design, which shows that the design of GAN is not much different from the design of structural engineers in subjectivity, and can fully meet the design needs. At the same time, the score of the key node arrangement of GAN is only 4.9% lower than that of the structural engineer, and the overall rationality score is 8.1% lower, which shows that GAN is closer to the design of the engineer in the arrangement of the key node. The reason for this phenomenon is that the convolution layer in GAN mainly convolves a small part of the picture, so as to pay more attention to details.

[0049] Table 1 Expert scoring results

[0050]

[0051] Secondly, in order to compare the difference between the design consumption of GAN and the structural engineer, this embodiment respectively counts the number of columns and the length of supports in 10 structural drawings, and calculates the average difference, as shown in Table 2. The difference in steel consumption between the two is not large, which is acceptable for structural design. In addition to the comparison of material consumption, this embodiment uses the method of finite element analysis to compare the mechanical properties of the structures designed by the structural engineer and GAN. Through the analysis results of the finite element model, the inter-story drift angle of each layer of the structure is obtained, as shown in Table 3. Figure 5 From the figure, it can be seen that the maximum difference of the inter-story drift angle of the structure designed by the structural engineer and GAN is 0.015%, and is less than the limit value of 0.4% in Chinese seismic code. Therefore, the design of GAN is not much different from the structural engineer in mechanical properties, and its safety can meet the requirements of daily use.

[0052] Table 2 Material consumption difference percentage

[0053] Column Support Difference percentage 7.3％ 16.2％

[0054] Referring to Figure 6 , a framework-support building structure intelligent design method based on a generative adversarial network, comprising the following steps:

[0055] S1, pre-training the column part and the support part of the generative adversarial network respectively for column design and support design;

[0056] S2, semantic processing of the architectural drawing: different colors are marked and filled in the key components in the architectural drawing, including walls, doors and windows, columns and supports;

[0057] S3, intelligent design of columns: the pre-trained generative adversarial network column part recognizes the architectural drawing obtained in step S2, and draws the arrangement position of the column on the architectural drawing;

[0058] S4, intelligent design of supports: the pre-trained generative adversarial network support part recognizes the arrangement position of the key components and the column in the architectural drawing output by step S3, further optimizes the arrangement of the column according to the arrangement rules of the support, so that the position of the support arrangement is more reasonable, and the arrangement position of the support is drawn on this basis, thereby generating the intelligent designed structural drawing.

[0059] In another embodiment, the generative adversarial network is pre-trained by a reinforced concrete frame-support building structural drawing, and is better used for intelligent design of key components of the reinforced concrete frame-support building structural drawing.

[0060] In another embodiment, steps S1 and S2 are performed synchronously.

[0061] Another embodiment of the present application also provides a system for implementing the above-mentioned building structure design method based on the generative adversarial network, comprising:

[0062] An image acquisition module is configured to acquire a preset training data set for image generation;

[0063] A loss function module is configured to acquire a preset loss function for training of the image generation model;

[0064] A decoupling module is configured to decouple the samples according to steps;

[0065] A first image generation module of the generative adversarial network is configured to design columns for the architectural drawing and generate an image;

[0066] A second image generation module of the generative adversarial network is configured to design supports for the column-designed architectural drawing and generate an image.

[0067] Another embodiment of the present application also provides a computer, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor. The processor invokes the computer program stored in the memory to execute the steps in the building structure design method provided by the embodiments of the present application. Please refer to the above description of the building structure design method for details. Figure 6, specifically comprising:

[0068] S1, pre-training the column part and the support part of the generative adversarial network respectively for column design and support design;

[0069] S2, performing semantic processing on the architectural drawings;

[0070] S3, the pre-trained generative adversarial network column part identifies the architectural drawings obtained in step S2, and draws the arrangement position of the column on the architectural drawings;

[0071] S4, the pre-trained generative adversarial network support part identifies the arrangement position of the key components and the column in the architectural drawings output in step S3, further optimizes the arrangement of the column according to the arrangement rules of the support, so that the position of the support arrangement is more reasonable, and draws the arrangement position of the support on this basis, thereby generating the intelligently designed structural drawings.

[0072] In summary, the present application provides a framework-support building structure design method, system and computer based on generative adversarial network. The design method includes four steps of pre-training generative adversarial network, inputting and identifying architectural drawings, arranging columns and supports on the basis of the architectural drawings, and outputting structural drawings. The evaluation index of the generative adversarial network includes not only the quality of the generated pictures, but also the subjective and objective evaluation of the building structure. According to the conditional probability formula and the characteristics of the generative adversarial network, the design process is divided into two steps, which improves the accuracy of the design result; the component arrangement design of the building structure can be realized in a short time; the design result is very accurate, and the performance is close to that of experienced structural designers; the method is simple, only the processed architectural drawings need to be inputted, and the corresponding structural drawings can be obtained; it can assist the building designers to complete the component design of the building structure, and has strong popularization.

[0073] Those skilled in the art should understand that those skilled in the art can realize variations in combination with the prior art and the above embodiments, which are not described here. Such variations do not affect the essential content of the present application, and are not described here.

[0074] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, also can be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of software product, and the computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a plurality of instructions to make a terminal (may be a mobile phone, computer, server, air conditioner, or network equipment, etc.) execute the method described in various embodiments of the present application.

[0075] The preferred embodiments of the present application are described above. It should be understood that the present application is not limited to the specific embodiments described above, and that the devices and structures not described in detail should be understood as being implemented in the ordinary way in the art; any person skilled in the art can make many possible changes and modifications to the technical solutions of the present application, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present application, which does not affect the essential content of the present application. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical solutions of the present application, still belongs to the scope of protection of the technical solutions of the present application.

Claims

1. A framework-supporting building structure design method based on a generative adversarial network framework, characterized by, The method comprises the following steps: S1, pre-training the column part and the support part of the generative adversarial network respectively for column design and support design; The real data are decoupled according to the conditional probability formula following the frame-support structure design steps: wherein, represents a probability distribution function of the design process of the building structure, represents a probability distribution function when only columns are arranged, represents a probability distribution function when supports are arranged after the column arrangement is completed; S2, semantic processing of the architectural drawing: labeling and filling the key components in the architectural drawing, including walls, doors and windows, columns and supports, by setting different colors; S3, intelligent design of columns: the pre-trained generative adversarial network column part recognizes the architectural drawing obtained in step S2 and draws the arrangement position of the columns on the architectural drawing; S4, intelligent design of supports: the pre-trained generative adversarial network support part recognizes the arrangement position of the key components and columns in the architectural drawing output in step S3, further optimizes the arrangement of the columns according to the arrangement rules of the supports, so that the arrangement position of the supports is more reasonable, and draws the arrangement position of the supports on this basis, thereby generating an intelligently designed architectural structure drawing; The types of the architectural structure drawings used for pre-training in steps S3 and S4 are the same as the types of the architectural structure drawings to be generated; The evaluation indexes of the architectural structure drawings generated in steps S3 and S4 include not only the quality of the generated pictures, but also subjective and objective evaluations of the architectural structure; the subjective evaluation in the evaluation indexes includes the scores of experts on the details and overall rationality of the architectural structure, and the objective evaluation includes the amount of building materials and the mechanical properties. 2.The framework-supporting building structure design method based on a generative adversarial network according to claim 1, wherein, In the generative adversarial network in steps S3 and S4, the generator comprises three parts of a down-sampling layer, a residual block and an up-sampling layer, and the discriminator comprises two PatchGAN discriminators. 3.The framework-supporting building structure design method based on a generative adversarial network according to claim 2, wherein, The generative adversarial network in steps S3 and S4 is implemented by using a deep learning framework Pytorch.

4. The framework-supporting building structure design method based on a generative adversarial network according to claim 2, characterized in that, The loss function applied to the generative adversarial network in steps S3 and S4 is represented as: where G represents the generator and D represents the discriminator. 5.The framework-supporting building structure design method based on a generative adversarial network according to claim 1, wherein, In the architectural drawing in step S2, the sizes of the walls, doors and windows, columns and supports are unified to the same thickness.

6. The framework-supporting building structure design method based on a generative adversarial network according to claim 1, wherein, The types of the architectural structure drawings used for pre-training in steps S3 and S4 are steel frame-support or concrete frame-support buildings. 7.A framework-supporting building structure design system based on a generative adversarial network framework, characterized by The method comprises the following steps: An image acquisition module is configured to acquire a preset training data set for image generation; A loss function module is configured to acquire a preset loss function for training of the image generation module; A decoupling module is configured to decouple samples according to steps; A first image generation module of the generative adversarial network is configured to perform column design on the architectural drawing and generate an image; A second image generation module of the generative adversarial network is configured to perform support design on the column-designed architectural drawing and generate an image.

8. A computer, comprising: The method comprises the following steps: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 6 when executing the computer program.

Images