An intelligent design and topological optimization method and system for key nodes of an aerial building machine

Abstract

The application belongs to the technical field of structural optimization design, and discloses a kind of intelligent design and topological optimization method and system of key node of air building machine, comprising: S1, topological optimization of key node of air building machine: high-quality big data learning library of key node of air building machine is established using topological optimization technology;S2, boundary balance generative adversarial network: introduce the generative design method based on deep learning, fully learn the topological characteristics of existing design scheme, on the basis of inheriting these characteristics, global exploration is carried out on unknown latent space, thereby deriving a large number of global optimization design scheme;S3, post-processing of intelligent generated scheme: edge regularization processing is carried out on the generated rib plate layout scheme by Python third-party OpenCV2 library;S4, reconstruction and evaluation of intelligent generated scheme: to realize the batch automatic three-dimensional reconstruction of generated rib plate layout scheme, macro tool is used and secondary development is carried out.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of structural optimization design, and particularly relates to an intelligent design and topology optimization method and system for key nodes of an air building machine. BACKGROUND

[0002] With the acceleration of global urbanization and population growth, high-rise buildings have begun to be built on a large scale, and air building machines have also emerged as the times require. The air building machine can transform high-risk and poor high-altitude operations into safe and comfortable ground construction by alternating climbing and working on the built concrete structure, and has the characteristics of high bearing capacity, strong adaptability, convenient operation, intelligent control and comfortable environment. The nodes, such as wall support, hydraulic cylinder support, support column top beam, etc., are the key connection and force transmission parts of the air building machine, and have a direct impact on the material consumption, processing and installation, and bearing capacity of the whole structure. Therefore, how to design an economical and reasonable, safe and reliable node is crucial.

[0003] However, the existing design method of key nodes of the air building machine mainly adopts the traditional design method based on knowledge and experience, that is, the designer first constructs an initial model according to experience, and then iteratively optimizes and improves the model through human cognition and observation until an optimal design scheme is obtained. This method often has problems such as low efficiency, complicated process, resource waste, and the obtained scheme is single in form and lacks layout optimization, which is difficult to meet the research and development needs of advanced design methods in the current intelligent manufacturing era. Therefore, it is necessary to carry out innovative research on the traditional design method of key nodes of the air building machine, establish a new design method that meets the needs of the theme of the times and engineering practice, and improve the intelligent degree of node design.

[0004] With the passage of time, big data and deep learning technology have made significant progress and play an increasingly important role in science and engineering. Among them, the development of computer vision is particularly eye-catching. As a major technological breakthrough in the field of computer vision, the generative adversarial network (GAN) has become an outstanding representative in this field, and its emergence provides a new research idea for the new configuration design problem of key nodes of the air building machine. GAN is composed of a generator and a discriminator, and through adversarial training, it can generate high-quality images, videos or other types of data. Since the original GAN is prone to gradient disappearance and is not easy to train, many other algorithms have been derived based on it, such as DCGAN, LSGAN, BEGAN, WGAN, etc., and these generative models all need a large amount of data for training. A high-quality big data learning library is the premise of training and using GAN and the guarantee of training effect, so there is an urgent need for a method that can generate multiple models and these models also have good mechanical performance effect.

[0005] In view of the above problems, the topological optimization technology provides a relatively effective solution. Topological optimization is a method of integrating engineering design problems and optimization mathematical theory, which aims to explore the best design in the design area that meets various constraint conditions by optimizing material distribution under given boundary conditions and load action. The use of topological optimization to establish a data set not only can inherit the characteristics of topological optimization and improve the material utilization, but also can enhance the robustness of the deep learning model. In addition, by integrating the manufacturing process constraints into the definition of each working condition optimization problem, the types of basic models can be enriched, and therefore, it is a very suitable choice to use topological optimization technology to construct an excellent big data learning library. In summary, it is a very forward-looking and reasonable direction to combine topological optimization technology with generative adversarial network to optimize the key nodes of the air building machine.

[0006] In addition, when dealing with a large number of design schemes, comprehensive evaluation is an indispensable key step, because it plays a crucial role in the scientific scheme selection decision-making work. Single attribute evaluation is relatively easy to implement, but when the mutual influence and importance between multiple attributes need to be considered, the problem becomes more complex. Among the commonly used multi-attribute evaluation methods are linear programming (LINMAP), analytic hierarchy process (AHP), and TOPSIS, but there are still few studies on multi-attribute evaluation methods in the field of generative design. Therefore, it is also a problem to be solved to consider a multi-attribute decision-making method to achieve accurate and efficient evaluation of a large number of schemes.

[0007] Through the above analysis, the problems and defects of the prior art are that the method of combining topological optimization with generative adversarial network has great potential advantages in structural intelligent design. Although this method has achieved remarkable results in many fields, it has not been applied in the field of key node design of air building machine. SUMMARY

[0008] In view of the problems existing in the prior art, the present application provides a key node intelligent design and topological optimization method and system for air building machine

[0009] The present application is implemented in the following way: a key node intelligent design and topological optimization method for air building machine,

[0010] S1, topological optimization of key nodes of air building machine: using topological optimization technology to establish a high-quality big data learning library of key nodes of air building machine;

[0011] S2, boundary balance generative adversarial network: introducing a deep learning-based generative design method, fully learning the topological characteristics of existing design schemes, and on the basis of inheriting these characteristics, globally exploring the unknown potential space, thereby deriving a large number of globally optimized design schemes;

[0012] S3, post-processing of the intelligent generated scheme: edge regularization processing of the generated rib plate layout scheme through the Python third-party OpenCV2 library;

[0013] S4, reconstruction and evaluation of the intelligent generated scheme: to realize the batch automatic three-dimensional reconstruction of the generated rib plate layout scheme, macro tools are used and they are secondarily developed.

[0014] Further, S1 specifically comprises: firstly, determining the design area of the initial topology optimization node according to the layout and structural features of the joist node, and performing discretization processing on it, then optimizing the design of the node by applying corresponding loads and constraints, and finally extracting the two-dimensional rib plate layout features of the obtained topology optimization node.

[0015] Further, S1 also comprises: the stiffness of the key node of the air building machine is an important factor affecting its carrying capacity, so the maximum stiffness (minimum flexibility) is taken as the preferred objective function; in order to realize the lightweight design of the node, the limit value of the volume fraction of the structure is taken as the constraint condition; the relative density of the element is taken as the design variable, and its mathematical model is as follows:

[0016]

[0017]

[0018]

[0019]

[0020] Among them minimizing a certain physical quantity; subject to certain constraint conditions; is the relative density of the element, ; is the flexibility function of the structure; is the external load matrix of the structure; is the overall stiffness matrix of the structure; is the displacement matrix of the structure; is the design variable the optimized volume of the element corresponding to the design variable; is the sum of the optimized volumes of all elements; is the initial volume of the entire structure; is the limit value of the volume fraction; is the total number of elements after the structure is discretized.

[0021] Further, S2 specifically comprises: after the establishment of the big data learning library, a suitable generative adversarial network architecture needs to be selected and trained using the data set, and a BEGAN deep learning model is used to intelligently generate a new rib plate layout scheme, and the objective function of BEGAN is as follows:

[0022]

[0023] for

[0024] for

[0025]

[0026] wherein represents the loss of training pixel-level autoencoder; represents the autoencoder function; represents the real / fake sample; represents the target standard; represents the discriminator loss function; represents the generator loss function; represents the real image; represents a random sample taken from a uniform distribution [-1, 1]; and represents a sample from ; and represent the parameters of the discriminator and the generator; the variable is used to control the degree of emphasis during gradient descent , which is constantly adjusted by to maintain a balance , the initial value of is 0, represents the number of iteration steps; is the proportional gain (i.e. learning rate) of ; the hyperparameter represents the diversity ratio, which is used to balance the performance between the generator and the discriminator and to control the quality and diversity of the generated samples, and the global convergence measure is:

[0027] .

[0028] Further, S3 specifically includes: first, the rib plate layout image is binarized by a threshold segmentation function in OpenCV2; for the generated rib plate layout scheme, black represents the rib plate design area, and white represents the background area; by converting the image into a binary form, the shape contour of the rib plate design area can be highlighted, so that it is easier to be processed and analyzed by a computer vision algorithm; then, small connected domains in the image are filtered to remove impurities by a measurement function of scikit-image (skimage); specific details are as follows: each independent black area in the image is numbered, and the area of the corresponding black area under each number is calculated, then a judgment statement is used to retain the black area with an area greater than a set threshold, and the corresponding number of the retained black area is used for region integration; finally, the binary open operation of OpenCV2 is used to remove noise again to improve the image quality, and the rib plate layout image processed by the binary open operation becomes clearer and more regular.

[0029] Further, S4 is roughly divided into three parts: first, the filtered rib plate layout scheme is scaled and feature contour extraction is performed; then, the following process is recorded as a macro command, that is, the feature contour of a single scheme is stretched by SolidWorks (SW), and the rib plate is exported together with the top plate and the bottom plate as a.STEP entity file; finally, the SW macro command is developed again to realize the batch automatic three-dimensional reconstruction of the rib plate layout scheme.

[0030] Another object of the application is to provide a key node intelligent design and topology optimization system of an aerial building machine, which applies the key node intelligent design and topology optimization method of the aerial building machine.

[0031] The topology optimization module is used to establish a high-quality big data learning library of key nodes of the aerial building machine by using a topology optimization technology;

[0032] The adversarial network generation module is used to introduce a generative design method based on deep learning, fully learn the topology features of existing design schemes, globally explore the unknown potential space on the basis of inheriting these features, and derive a large number of globally optimized design schemes;

[0033] The intelligent generated scheme post-processing module is used to perform edge regularization processing on the generated rib plate layout scheme by using a Python third-party OpenCV2 library.

[0034] The intelligent generated scheme reconstruction and evaluation module is used to realize the batch automatic three-dimensional reconstruction of the generated rib plate layout scheme by using a macro tool and developing it again.

[0035] Another object of the present application is to provide a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the intelligent design and topology optimization method for key nodes of an aerial building machine.

[0036] Another object of the present application is to provide a computer readable storage medium storing a computer program, the computer program being executed by a processor to cause the processor to perform the steps of the intelligent design and topology optimization method for key nodes of an aerial building machine.

[0037] Another object of the present application is to provide an information data processing terminal for implementing the intelligent design and topology optimization system for key nodes of an aerial building machine.

[0038] In combination with the above technical solutions and the technical problems solved, the technical solutions to be protected by the present application have the following advantages and positive effects:

[0039] First, in view of the technical problems existing in the prior art and the difficulty in solving the problems, some creative technical effects are brought about after solving the problems. The specific description is as follows:

[0040] Since the aerial building machine faces three complex working conditions of construction, jacking and shutdown in its high-altitude working environment. This special working environment is full of uncertainty, so that the influence of various working environments must be fully considered in the optimization design of key nodes of the aerial building machine. Therefore, the present application proposes an intelligent design and optimization method and system for key nodes of an aerial building machine based on topology optimization and generative adversarial network, aiming to design the nodes under various working conditions and evaluate the combined influence of various working conditions on the designed nodes. By combining topology optimization technology, generative adversarial network and multi-attribute decision method, and considering the uncertainty of high-altitude working environment, the rapid generation and flexible selection of the key node design scheme of the aerial building machine under various working conditions are realized.

[0041] The application utilizes a large number of topological optimization structure models integrated with multi-working condition factors to establish a deep learning database; secondly, a generator and a discriminator of the BEGAN algorithm are built based on a TensorFlow deep learning framework, and a novel and diverse node design scheme is generated by using the antagonistic training mechanism of the two; then, the generated scheme is post-processed to obtain a high-quality design scheme that is more in line with actual engineering requirements; finally, the optimal design scheme under multiple working conditions is selected by the MADM method. The static performance of the optimal design scheme obtained based on the MADM method under each working condition is greatly improved compared with the initial design scheme, and the overall optimization degree under the construction working condition is 43.45%, the overall optimization degree under the jacking working condition is 43.67%, and the overall optimization degree under the shutdown working condition is 42.89%, which has lighter weight and better mechanical properties. The results show that the intelligent design and optimization technology of the key node of the air building machine integrating topological optimization and generative adversarial network proposed in the application can not only continuously generate novel design schemes with rich styles and reasonable stress, but also accurately select the optimal scheme under the premise of considering the comprehensive effect of each working condition.

[0042] Secondly, in view of the problems of low efficiency, complicated process, resource waste, and the obtained scheme being single in form and insufficient in layout optimization, the application proposes a generative design method combining topological optimization technology, generative adversarial network and multi-attribute decision-making method. The method can automatically and quickly explore in a large design space under the consideration of the uncertainty of high-altitude operation environment, that is, the key node of the air building machine is optimized and designed from a global perspective to obtain a design scheme with lighter weight and better performance, thereby making up for the defects of the traditional design method which is mainly based on experience and locally optimized, and realizing the full optimization design of the key node of the air building machine.

[0043] The application proposes an intelligent design method of the key node of the air building machine based on generative adversarial network, aiming to optimize and design the key node of the air building machine from a global perspective and improve the intelligent degree of node design. The data set is constructed by using topological optimization technology and integrated with multi-working condition factors, novel schemes are intelligently generated based on the BEGAN algorithm and post-processed, multi-attribute decision-making evaluation of the design schemes is performed by MADM, and finally the intelligent design and evaluation of the key node of the air building machine under multiple working conditions are realized. The intelligent design and topological optimization method can not only retain the basic characteristics of topological optimization, but also continuously generate novel design schemes with rich styles, realize the global optimization of the key node of the air building machine, and further improve the intelligent degree of node design. In addition, the MADM method can obtain the comprehensive score ranking of each alternative scheme, and can meet the actual requirements by setting appropriate weights, and it cooperates with the intelligent design method to construct an integrated solution of the design and evaluation of the key node of the air building machine.

[0044] Thirdly, the technical solution of the present application fills the technical gap in the industry at home and abroad: Through the prospect of existing literature and patents, we found that the combination of topology optimization and generative adversarial network has achieved remarkable success in many fields such as spatial structure, thermal engineering, optical materials, product form and mechanical engineering. This integration has the unique ability to solve the problems of experience dominance and local optimization in traditional design methods. However, despite the success in these fields, existing research is still relatively insufficient in dealing with multi-condition problems, especially difficult to be directly applied to aerial building machines and other fields that need to consider multi-condition operation.

[0045] In this context, the present application uses the TOPSIS method to evaluate the multi-attribute of the generated scheme, and considers the effect of multi-condition in the scheme generation and evaluation stage, and proposes an integrated solution of intelligent design and evaluation of key nodes of aerial building machines combining topology optimization, generative adversarial network and MADM. This innovative method not only fills the gap of existing technology in multi-condition problems, but also brings advanced and comprehensive technical vision to the industry at home and abroad. Through the technical solution of the present application, we successfully solve the shortcomings of traditional design methods in multi-condition problems, and provide a new intelligent solution for the design of key nodes of aerial building machines and other complex systems. This comprehensive and innovative method has a significant promoting effect in the industry, and contributes a new direction and method to improve the technical level of related fields in China.

[0046] Does the technical solution of the present application solve the technical problems: Currently, the design of key nodes of aerial building machines still mainly uses traditional methods based on knowledge and experience, but this method faces a series of problems such as low efficiency, complicated process, resource waste, single scheme, and insufficient layout optimization, which is difficult to meet the continuous pursuit of advanced design methods in today's intelligent manufacturing era. Therefore, academia has been committed to innovative research on traditional design methods for key nodes of aerial building machines, although great efforts have been made, but so far no great breakthrough has been made.

[0047] Unlike the present application, the technical solution brings significant improvements, successfully solving the various problems of traditional design methods, including low efficiency, time-consuming, resource waste, single design, and insufficient layout optimization. By introducing advanced topology optimization technology, generative adversarial network and multi-attribute decision method, the present application realizes the global optimization of key nodes of aerial building machines considering the uncertainty of high-altitude operation environment, and automatically obtains a lighter and more optimal design scheme. This innovative design method successfully makes up for the shortcomings of traditional design methods in local optimization, and realizes the overall optimization design of key nodes of aerial building machines.

[0048] In summary, the technical scheme of the present application has made a significant breakthrough in solving the technical problems that people have been eager to solve but have failed to succeed. Its advancement and practicality not only meet the current research and development needs of the intelligent manufacturing era, but also inject new vitality into the key node design field of air building. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical scheme of the embodiments of the present application, the drawings needed in the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained from these drawings without creative labor for those skilled in the art.

[0050] Figure 1 is a flow chart of the intelligent design and topology optimization method of the key node of the air building machine provided by the embodiments of the present application;

[0051] Figure 2 is a system structure diagram of the intelligent design and topology optimization of the key node of the air building machine provided by the embodiments of the present application;

[0052] Figure 3 is a topology optimization flow chart of the key node of the air building machine provided by the embodiments of the present application; wherein, (a) the key node of the air building machine; (b) the model before node topology optimization; (c) the model after node topology optimization; (d) the two-dimensional rib plate layout obtained from the topology result;

[0053] Figure 4 is a network structure diagram of the generator and discriminator provided by the embodiments of the present application; wherein, (a) the network structure of the generator; (b) the network structure of the discriminator;

[0054] Figure 5 is a three-dimensional reconstruction process diagram of the rib plate layout scheme provided by the embodiments of the present application; wherein, (a) the intelligently generated rib plate layout scheme (PNG file); (b) the feature contour (DXF file); (c) the initial model after importing the contour (CAD model); (d) the reconstructed three-dimensional beam model (STEP model);

[0055] Figure 6 is a basic step diagram of the TOPSIS method provided by the embodiments of the present application; wherein, (a) the step diagram of the TOPSIS method; (b) the program diagram of the TOPSIS method;

[0056] Figure 7 is a flow chart of the intelligent design method of the key node of the air building machine provided by the embodiments of the present application;

[0057] Figure 8is a flow chart of the determination of the planar design area provided by the embodiment of the present application; wherein, (a) hole area of the top plate and the bottom plate; (b) expansion of the hole area; (c) connection of the hole area; (d) regularization of the hole area;

[0058] Figure 9 is a diagram of ten layout types of the rib plate provided by the embodiment of the present application; wherein, (a) type 1; (b) type 2; (c) type 3; (d) type 4; (e) type 5; (f) type 6; (g) type 7; (h) type 8; (i) type 9; (j) type 10

[0059] Figure 10 is an initial model and a geometric feature diagram of the joist provided by the embodiment of the present application; wherein, (a) geometric features of the top plate and the bottom plate; (b) geometric features of the rib plate; (c) initial model of the joist;

[0060] Figure 11 is a loss function diagram of the acquisition process BEGAN of the intelligent generation scheme provided by the embodiment of the present application; wherein, (a) loss function of the generator and the discriminator; (b) global convergence measure;

[0061] Figure 12 is a flow chart of the acquisition process of the intelligent generation scheme provided by the embodiment of the present application;

[0062] Figure 13 is an example diagram of the intelligent generation of the rib plate layout provided by the embodiment of the present application; wherein, (a) rib plate layout 1; (b) rib plate layout 2; (c) rib plate layout 3; (d) rib plate layout 4; (e) rib plate layout 5; (f) rib plate layout 6; (g) rib plate layout 7; (h) rib plate layout 8;

[0063] Figure 14 is an example diagram of the intelligent generation scheme provided by the embodiment of the present application; wherein, (a) scheme 1; (b) scheme 2; (c) scheme 3; (d) scheme 4; (e) scheme 5; (f) scheme 6; (g) scheme 7; (h) scheme 8;

[0064] Figure 15 is a static analysis result diagram of the optimal scheme provided by the embodiment of the present application; wherein, (a) displacement nephogram under the construction condition; (b) equivalent stress nephogram under the construction condition; (c) displacement nephogram under the jacking condition; (d) equivalent stress nephogram under the jacking condition; (e) displacement nephogram under the stoppage condition; (f) equivalent stress nephogram under the stoppage condition. DETAILED DESCRIPTION

[0065] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be given below in combination with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0066] In view of the problems in the prior art, the present application provides a method and system for intelligent design and topology optimization of key nodes of an aerial building construction machine. The present application will be described in detail below in combination with the accompanying drawings.

[0067] Two specific application embodiments of the present application are:

[0068] Embodiment 1: Key node design of aerial building construction machine for large commercial complex

[0069] Application of topology optimization: For the specific needs of large commercial complexes, the topology optimization technology is used to design the joist nodes. Considering the complexity and multifunctionality of the building, the node design ensures the stability and aesthetics of the structure.

[0070] Application of generative adversarial network: The generative adversarial network is applied to learn the existing design schemes of commercial complexes and explore innovative node design schemes. This step provides a large number of innovative and practical design options.

[0071] Post-processing and reconstruction: The generated node design is subjected to edge regularization processing using the OpenCV2 library, and then three-dimensional reconstruction is performed using macro tools. This makes the design scheme more accurate and visualized, facilitating the evaluation of engineers and architects.

[0072] Embodiment 2: Key node optimization of aerial building construction machine for cross-sea bridge

[0073] Topology optimization for specific scenarios: For the special environment and load-bearing requirements of cross-sea bridges, the topology optimization design of nodes is performed. Natural factors such as wind load and sea current are considered in the design to ensure the stability and safety of the structure.

[0074] Exploration of innovative design: The generative adversarial network is used to learn and analyze the node design of existing cross-sea bridges to generate node design schemes with innovative features. Such exploration helps to propose designs that are more suitable for extreme environments.

[0075] Post-processing and visualized reconstruction: The generated design scheme is subjected to post-processing, including edge regularization, followed by three-dimensional reconstruction through automated tools, to facilitate detailed evaluation and modification of the design scheme by engineers.

[0076] The present application mainly improves the following problems and defects in the prior art, achieving significant technical progress:

[0077] Low design efficiency: In the background art, the design of key nodes of air building machines usually relies on traditional manual methods or simple computer-aided design (CAD), which are inefficient and have long design iteration cycles.

[0078] Limited optimization capability: Traditional design methods often fail to fully realize structural optimization, which may result in suboptimal designs in terms of material usage, load-bearing capacity, and stability.

[0079] Lack of design innovation: Due to the limitations of designers' experience and imagination, traditional design methods are difficult to explore innovative and efficient structural layout solutions.

[0080] To address the problems in the prior art, the technical solution adopted by the present application is as follows: a learning library established using topology optimization technology can effectively guide node design and improve the initial quality and efficiency of design. Deep learning and generative adversarial networks (GAN) are introduced to realize global optimization design exploration of complex structures and improve the innovation and diversity of design. Edge regularization processing is performed using the OpenCV2 library to improve the geometric accuracy and applicability of generated solutions. Macro tools and secondary development are used to realize batch automatic three-dimensional reconstruction, significantly improving the visualization and evaluation efficiency of design solutions.

[0081] The present application solves the technical problems brought about by the prior art and achieves significant technical progress:

[0082] As shown in Figure 1 , the air building machine key node intelligent design and topology optimization method provided by the embodiments of the present application,

[0083] S1, topology optimization of air building machine key nodes: a high-quality big data learning library of air building machine key nodes is established using topology optimization technology;

[0084] S2, boundary balance generative adversarial network: a generative design method based on deep learning is introduced to fully learn the topological features of existing design solutions and explore the unknown latent space globally based on these features, thereby deriving a large number of globally optimized design solutions;

[0085] S3, post-processing of intelligent generated solutions: edge regularization processing is performed on the generated ribbed plate layout solutions using the Python third-party OpenCV2 library;

[0086] S4, reconstruction and evaluation of intelligent generated solutions: to realize batch automatic three-dimensional reconstruction of generated ribbed plate layout solutions, macro tools are used and secondary development is performed.

[0087] As shown in Figure 2 , the air building machine key node intelligent design and topology optimization system provided by the embodiments of the present application comprises:

[0088] Topology optimization module: for establishing high-quality big data learning library of key nodes of air construction building by topology optimization technology;

[0089] Adversarial network generation module: for introducing deep learning-based generative design method, fully learning the topology characteristics of existing design schemes, and on the basis of inheriting these characteristics, globally exploring the unknown potential space, thereby deriving a large number of globally optimized design schemes;

[0090] Intelligent generated scheme post-processing module: for performing edge regularization processing on the generated rib plate layout scheme through the Python third-party OpenCV2 library;

[0091] Intelligent generated scheme reconstruction and evaluation module: for realizing batch automatic three-dimensional reconstruction of the generated rib plate layout scheme, using macro tools and developing them again.

[0092] 1. Topology optimization of key nodes of air construction building

[0093] The application establishes a high-quality big data learning library of key nodes of air construction building by topology optimization technology, and the implementation process is as shown in Figure 3 The implementation process is as shown in Figure 3 First, the design area of the initial topology optimization node is determined according to the layout and structural characteristics of the joist node, and the design area is discretized, then the node is optimized and designed by applying corresponding loads and constraints, and finally the two-dimensional rib plate layout characteristics of the obtained topology optimization node are extracted. Common topology optimization methods include variable density method (SIMP), homogenization method, level set method, etc. Because SIMP method has high calculation efficiency and stability, SIMP method is selected for topology optimization. The stiffness of the key node of the air construction building is an important factor affecting its bearing capacity, so the maximum stiffness (minimum flexibility) is selected as the preferred objective function; in order to realize the lightweight design of the structure, the limit value of the volume fraction of the structure is used as the constraint condition; the relative density of the element is used as the design variable, and the mathematical model is as follows.

[0094] (1)

[0095] (2)

[0096] (3)

[0097] (4)

[0098] Wherein Minimizes a certain physical quantity; Subject to certain constraint conditions; is the relative density of the unit, ; is the flexibility function of the structure; is the external load matrix of the structure; is the global stiffness matrix of the structure; is the displacement matrix of the structure; is the design variable is the optimized volume of the corresponding unit; is the sum of the optimized volumes of all units; is the initial volume of the entire structure; is the volume fraction limit value; is the total number of units after finite element discretization of the structure.

[0099] After topology optimization, it is necessary to unify the size of the topology optimization image and perform data augmentation. The Image class in the third-party library PIL of Python is used to scale the topology optimization result to 128 voxels x 128 voxels, while setting the nearest neighbor difference (NEAREST) resampling filter to ensure that the scaled image forms clearer and more regular hard edges. Secondly, the data set data structure is defined as a training set picture-training set label feature pair, and the training set label is the corresponding working condition type during topology optimization. Then the data set is expanded through data augmentation method to reduce the overfitting phenomenon of the network. The data augmentation method mainly includes geometric transformation and pixel transformation. Geometric transformation includes rotation, flipping, scaling, translation and the like; pixel transformation includes noise disturbance, Gaussian blur, color change, brightness adjustment and the like. Considering the symmetry requirement of the nodes, the present application mainly adopts rotation and flipping for data augmentation, and the enhanced data set is used as the final training set. Finally, the gray value of the training set image is converted to the range of [0, 1] to speed up the convergence process of the BEGAN algorithm.

[0100] 2. Boundary equilibrium generative adversarial network

[0101] A high-quality data library is successfully constructed using topology optimization techniques. However, these results are relatively few and limited, and are obtained under a specific layout, so the optimization effect has limitations. To overcome these limitations, the present invention introduces a generative design method based on deep learning. This method not only learns the topological features of existing design schemes, but also explores the unknown potential space globally on the basis of inheriting these features, thereby deriving a large number of globally optimized design schemes. Specifically, the present invention will select a suitable generative adversarial network architecture and train it using the data set we constructed to better exploit its potential for exploring and generating design schemes. After establishing the large data learning library, a suitable generative adversarial network architecture needs to be selected and trained using the data set. BEGAN is a simple and robust GAN architecture that can achieve fast and stable convergence through standard training steps, and can control the balance between visual quality and image diversity, so the deep learning model is used to intelligently generate new rib plate layout schemes. The discriminator of BEGAN uses a similar autoencoder as EBGAN, and the generator only uses a decoder. It adopts a loss distribution matching autoencoder derived from the Wasserstein distance, and measures the loss difference between the sample and the autoencoder output, and then obtains the lower limit of the Wasserstein distance between the autoencoder loss distribution of the real sample and the generated sample, and takes this lower limit as the optimization target. The objective function of BEGAN is as follows:

[0102] (5)

[0103] for (6)

[0104] for (7)

[0105] (8)

[0106] Wherein represents the loss of training pixel-level autoencoder; represents the autoencoder function; represents real / fake samples; represents the target criterion; represents the discriminator loss function; represents the generator loss function; represents real images; represents random samples sampled from a uniform distribution [-1, 1]; and represent samples from ; and denote the parameters of the discriminator and the generator; variables for controlling the degree of emphasis during gradient descent, by constant adjustment to maintain balance , The initial value of is 0, denote the number of iteration steps; is a proportional gain (i.e. learning rate); hyperparameter diversity ratio, used to balance the performance between the generator and the discriminator and control the quality and diversity of generated samples. Then the global convergence measure is:

[0107] (9)

[0108] The BEGAN network structure based on the deep learning framework TensorFlow 2.6.0 is shown in Figure 4 , in which the discriminator is an autoencoder composed of an encoder and a decoder, and the generator uses the same network structure as the decoder of the discriminator. The input of the encoder is composed of two parts, one part is the real image from the dataset, that is, a batch of 128×128×1 joist node topology optimization images; the other part is the false image generated by the generator. Then the feature extraction is performed through the stacked network composed of 3 convolutional layers and max pooling layers in succession, the activation function of each convolutional layer is set to exponential linear unit (ELU), which adds a nonlinear factor to the deep learning model, and a Batch Normalization batch normalization layer is connected after it, which facilitates the formation of more stable gradient flow in the deep learning process. Finally, a flat layer is used to compress the high-dimensional input image into a feature vector. The input of the decoder is the feature vector, which is mapped through a fully connected layer and reshaped to output a feature map of 16×16×32. The next three network structures are stacked by convolutional layers and up-sampling layers, the activation function of each convolutional layer is set to exponential linear unit (ELU), and a Batch Normalization batch normalization layer is connected after it, and finally a convolutional layer with an activation function of tanh is used to output the joist node new configuration image, so as to expand the feature effect in the iteration process with significant feature difference.

[0109] 3. Post-processing of the intelligent generation scheme

[0110] ​The present application performs edge regularization processing on the generated rib plate layout scheme through the Python third-party OpenCV2 library. First, the threshold segmentation function in OpenCV2 is used to perform binaryzation processing on the rib plate layout image. For the generated rib plate layout scheme, black represents the rib plate design area, and white represents the background area. By converting the image to a binary form, the outline of the rib plate design area can be highlighted, making it easier to be processed and analyzed by computer vision algorithms. Then, the small connected domains in the image are filtered to remove impurities by using the measurement function of scikit-image (skimage). The specific details are as follows: each independent black area in the image is numbered, and the area of each numbered black area is calculated, then the black area with an area greater than a set threshold is retained by using a judgment statement, and the corresponding number of the retained black area is used for region integration. Finally, the binary open operation of OpenCV2 is used to remove noise again to improve the image quality, and the rib plate layout image after the binary open operation processing becomes more clear and regular. The above three operations can be integrated into a code file to realize batch regularization processing of the generated rib plate layout scheme.

[0111] BEGAN has strong feature extraction and model generation capabilities, so it can not only create a large number of novel rib plate layout schemes, but also may produce some similar designs. In order to obtain design schemes with creativity and diversity, the intersection over union of feature point data is used as a similarity evaluation index to filter the similarity of a large number of generated schemes. First, the generated schemes are compared pixel by pixel, then the similarity is calculated by intersection over union, and finally the design schemes with high similarity are filtered by a pre-set similarity threshold. In addition, since symmetry can improve the stability, balance and beauty of the structure, the design schemes filtered by similarity are further filtered by symmetry to improve the quality, aesthetics and applicability of the design, so that they better meet the functional, aesthetic and manufacturing requirements. The method of symmetry filtering refers to the similarity filtering, and the comparison of the generated schemes is adjusted to the comparison of the images on both sides of the two central axes of a single generated scheme pixel by pixel. When the intersection over union results of the two meet the symmetry threshold requirements at the same time, the design scheme is retained.

[0112] 4. Reconstruction and evaluation of intelligently generated schemes

[0113] To realize the batch automatic 3D reconstruction of the ribbed plate layout scheme, the macro tool is used and secondary development is carried out. The process is roughly divided into three parts: first, the filtered ribbed plate layout scheme is scaled and the feature profile is extracted in batches. Then, the following process is recorded as a macro command, that is, the feature profile of a single scheme is stretched through SolidWorks (SW), and the ribbed plate is exported together with the top plate and the bottom plate as a.STEP entity file. Finally, the SW macro command is secondarily developed to realize the batch automatic 3D reconstruction of the ribbed plate layout scheme. The 3D reconstruction process of the ribbed plate layout scheme is as shown in Figure 5 .

[0114] The key node of the air building machine is often in a multi-working condition environment, so it is necessary to comprehensively evaluate multiple attributes of the node under the multi-working condition, and then determine an optimal scheme to help the designer make a decision. The core goal of the TOPSIS method is to find an optimal scheme, that is, under the comprehensive consideration of multiple evaluation indexes, it is close to the ideal scheme and far away from the inferior scheme. This comprehensive measurement is usually in the form of a comprehensive score, and the highest score of the selected scheme is considered as the best choice. The present application uses the method to comprehensively consider the quality, maximum displacement and maximum equivalent stress three key evaluation indexes from the multi-working condition angle, and then selects the optimal scheme from the intelligently generated node design scheme. The basic steps of the TOPSIS method are as shown in Figure 6 .

[0115] Referring to Figure 7 , the present embodiment case provides an intelligent design and topology optimization technology for a key node of an air building machine, and the case is applied to a lightweight air building machine for a 358m high-rise building construction. The case mainly includes the following steps:

[0116] Step 1, data set establishment of the key node of the joist. In order to facilitate the bolt connection between the joist top plate and the steel platform and the support column in the later stage, the area near the bolt hole of the top plate and the bottom plate is set as a non-design area, and the specific range is a circular area with the long side of the bottom plate hole as the diameter. Then, the hole expansion area is subjected to the regularization treatment of the circumscribed rectangle. The determination process of the planar design area is as shown in Figure 8 , Figure 8 The rectangular area in (d) is a non-design area, and the other areas are design areas. Finally, referring to the initial ribbed plate arrangement scheme of the joist, the layout form of the ribbed plate is roughly divided into ten types as shown in Figure 9 , and the thickness of each ribbed plate is 10mm.

[0117] The topology optimization process of the joist is introduced by taking type 1 as an example. First, the initial model of the joist under the layout is established through SolidWorks, as shown in Figure 10 (c), the whole model is divided into an optimization area (ribbed plate) and a non-optimization area (top plate and bottom plate), and the geometric characteristics of the joist are asFigure 10 (a) and (b) 10, Table 1 summarizes the detailed joist geometry parameters. Secondly, the joist initial model is imported into Ansys Workbench, and the material of the model is structural steel, and the related properties of the material are shown in Table 2. Then meshing is carried out, in order to obtain a more regular rib plate optimization layout, set each rib plate as the minimum optimization mesh element, and use 20-node second-order hexahedral solid element to obtain higher calculation accuracy. Finally, the topology optimization under each working condition is carried out, according to the actual working condition and boundary condition of the joist, the corresponding load of the working condition is applied on the top plate of the joist, and the bottom plate is set as the fixed end, with a certain volume fraction as the constraint condition, and the stiffness (minimum flexibility) is maximized as the objective function to carry out topology optimization.

[0118] Table 1. Geometric parameters of joist

[0119]

[0120] Table 2. Related properties of materials

[0121]

[0122] The data set of the joist topology optimization model under three working conditions generated by the SIMP method is made, and the manufacturing process constraints such as penalty factor, checkerboard control and minimum member size are introduced into the optimization problem of each working condition. Collect the topology model pictures of different working conditions, different density thresholds and different manufacturing process constraints, and perform expansion processing, in order to make the generated layout scheme located in the plane design area and meet the symmetry requirement, select the enhancement methods of 180 degree rotation, horizontal flip and vertical flip to carry out data enhancement, while keeping the sample label unchanged, not only can inherit the topology optimization characteristics of the joist model, but also can increase the training data and improve the model generalization ability. After data enhancement, the training set has a total of 9000 PNG format pictures, divided into 3 working conditions, considering X and Y wind directions for each working condition. In addition, there are ten types of layout, and each type of picture is not less than 100, and each size is 128 voxels x 128 voxels.

[0123] Step 2, BEGAN model training. The present application adopts the high encapsulation degree framework tf.keras of TensorFlow to quickly build BEGAN, and configures GPU computing to accelerate the convergence process of the deep learning model. In order to present more details and changes of the pixel contour of the two-dimensional rib plate layout, a high-resolution image of 128 voxels x 128 voxels is used as input and output, and the gray value of the pixel image in the training set is normalized to [0, 1] before training, in order to improve the convergence efficiency of the deep learning model. The generator and the discriminator both adopt the Adam optimizer, the learning rate is 0.0002, the first moment estimation exponential decay rate is 0.5, the diversity ratio γ=0.75, and the variable kt The initial value is 0, and the proportional gain λ of k is... k =0.001. All models were trained using mini-batch stochastic gradient descent (SGD) with 10 training batches and 150 training epochs. All model weights were initialized using the Glorot normal distribution initialization method. Specific experimental environment parameters are listed in Table 3.

[0124] Table 3. Experimental Environmental Parameters

[0125]

[0126] Ten 100-dimensional random noise values ​​were initialized using a uniform distribution with a minimum value of -1 and a maximum value of 1. These initial values, along with a batch-processed training set, were then input into BEGAN for training. During BEGAN training, the loss function consisted of a generator loss function, a discriminator loss function, and a global convergence metric. Their evolution is as follows: Figure 11 As shown. By Figure 11 As can be seen, the generator loss function, discriminator loss function, and global convergence metric all steadily decrease, indicating that the BEGAN deep learning model is gradually converging. After 1500 iterations, the loss function curves are basically convergent, indicating that the learning capabilities of the generator and discriminator are gradually increasing. After 2000 iterations, the loss function curves are completely convergent, indicating that the game between the generator and discriminator has reached a Nash equilibrium, capable of continuously generating reliable design solutions. Furthermore, the convergence values ​​of both loss functions are not zero, indicating that the intelligently generated solutions still differ from the input samples, meaning that the generated design solutions possess novelty and diversity.

[0127] Step 3: Intelligent Design and Evaluation of Key Nodes of the Sky Building Machine. The process for obtaining the intelligent generation solution is as follows: Figure 12 As shown. First, a uniform distribution with a minimum value of -1, a maximum value of 1, and a tensor shape of 25×100 is set as the random seed and input into the trained generator model, resulting in 25 rib layout images (128 pixels × 128 pixels) with different shapes. It's worth noting that the number and style of the generated rib layout images can be controlled by adjusting the shape of the tensor in the random seed (i.e., batch size and latent space). Then, these images undergo edge regularization and similarity and symmetry filtering, ultimately yielding 8 rib design schemes, such as... Figure 13 As shown. Finally, batch 3D reconstruction was performed on these 8 schemes to obtain 3D solid models that can be used for subsequent finite element analysis, as shown. Figure 14 As shown.

[0128] The optimal scheme is selected by evaluating the comprehensive performance of each alternative scheme under different working conditions. First, 9 alternative schemes and 7 evaluation indexes are determined, wherein the 9 alternative schemes are an initial design scheme and intelligent generation schemes 1-8, and the 7 evaluation indexes are mass (m), maximum displacement under working condition 1 ( ), maximum equivalent stress under working condition 1 ( ), maximum displacement under working condition 2 ( ), maximum equivalent stress under working condition 2 ( Figure 15 ), maximum displacement under working condition 3 (

[0129] ), and maximum equivalent stress under working condition 3 ( ). Second, finite element static simulation is performed on the initial design scheme and the 8 intelligent generation schemes by using Ansys Workbench, and the static calculation results are listed in Table 4. Then, the TOPSIS method is applied to comprehensively score and sort the 9 schemes according to the 7 evaluation indexes, and the sorting results are listed in Table 5. Finally, the optimal design scheme is selected according to the sorting results, the static analysis results of the optimal design scheme under the three working conditions are shown in

[0130] , and the optimization degrees of the optimal design scheme under different working conditions compared with the initial design scheme are listed in Table 6, so as to verify the optimization level of the intelligent design method.

[0131] Table 4. Static analysis results of each alternative scheme under different working conditions

[0132]

[0133] Table 5. Comprehensive measurement and sorting results of each alternative scheme

[0134]

[0135] The application embodiment of the present application provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the intelligent design and topological optimization method of the key node of the air building machine.

[0136] The application embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the intelligent design and topological optimization method of the key node of the air building machine.

[0137] The application embodiment of the present application provides an information data processing terminal, which is used for realizing the intelligent design and topological optimization system of the key node of the air building machine.

[0138] It should be noted that embodiments of the present application can be realized by hardware, software, or a combination of software and hardware. The hardware portion can be realized by a special logic; the software portion can be stored in a memory and executed by a proper instruction execution system, such as a microprocessor or a specially designed hardware. A person of ordinary skill in the art can understand that the above-mentioned apparatus and method can be realized by computer executable instructions and / or included in processor control code, such as a carrier medium, such as a magnetic disk, CD or DVD-ROM, programmable memory, such as read-only memory (firmware), or data carrier, such as optical or electronic signal carrier. The apparatus of the present application and its modules can be realized by a hardware circuit, such as a very large scale integrated circuit or a gate array, a semiconductor, such as a logic chip, transistor, etc., or a programmable hardware device, such as a field programmable gate array, programmable logic device, etc., or by software executed by various types of processors, or by a combination of the above-mentioned hardware circuit and software, such as firmware.

[0139] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any modification, equivalent replacement, and improvement within the technical range disclosed by the present application, and within the spirit and principle of the present application, should be covered within the protection scope of the present application.

Claims

1. An intelligent design and topology optimization method for key nodes of an aerial building construction machine, characterized in that, Comprise: S1, topology optimization of key nodes of air construction machine: use topology optimization technology to establish a high-quality big data learning library of key nodes of air construction machine; S2, boundary balance generative adversarial network: introduce a deep learning-based generative design method, fully learn the topological characteristics of existing design schemes, and on the basis of inheriting these characteristics, globally explore the unknown potential space, thereby deriving a large number of globally optimized design schemes; S3, post-processing of intelligent generated scheme: edge regularization processing of the generated rib plate layout scheme is performed through the Python third-party OpenCV2 library; S4, reconstruction and multi-attribute decision evaluation of intelligent generated scheme: to realize batch automatic three-dimensional reconstruction of the generated rib plate layout scheme, macro tools are used and they are developed again.

2. The method of claim 1, wherein the method further comprises: S1 specifically comprises: first, determining the design area of the initial topology optimization node according to the layout and structural characteristics of the joist node, and then discretizing the design area, then optimizing the design of the node by applying appropriate loads and constraints, and finally extracting the two-dimensional rib plate layout characteristics of the obtained topology optimization node.

3. The method of claim 1, wherein the method further comprises: S1 also includes: the stiffness of the key node of the air construction machine is an important factor affecting its bearing capacity, so the maximum stiffness is taken as the preferred objective function; in order to realize lightweight design of the structure, the volume fraction limit value is taken as the constraint condition; the relative density of the unit is taken as the design variable, and its mathematical model is as follows: S.t.F=KU ∑V(ρ i ) / V0≤f 0 < p i ≤ 1, i = 1, 2, 3,..., n Min S.t. ρ = (ρ1, ρ2, …, ρn)T n ) T ; C is the flexibility function of the structure; F is the external load matrix of the structure; K is the global stiffness matrix of the structure; U is the displacement matrix of the structure; V(ρ i ) is the optimized volume of the element corresponding to the design variable ρ i ; ∑V(ρ i ) is the sum of the optimized volumes of all elements; V0 is the initial volume of the whole structure; f is the volume fraction limit value; n is the total number of elements after the finite element discretization of the structure.

4. The method of claim 1, wherein the method further comprises: determining a number of the air-slab building key nodes; and determining a number of the air-slab building key node intelligent design and topology optimization methods. S2 specifically includes: after establishing the big data learning library, a suitable generative adversarial network architecture needs to be selected and trained using the high-quality big data learning library of the key node of the air construction machine, and a BEGAN deep learning model is used to intelligently generate a new rib plate layout scheme, and the objective function of BEGAN is as follows: L(v) = |v - D(v)| η L D = L(x) - k t L(G(z D )) for θ D L G = L(G(z G )) for θ G k t+1 = k t + λ k (γL(x) - L(G(z G ))) where L(v) denotes the loss of training pixel-level autoencoder; D(v) denotes the autoencoder function; v denotes real / fake samples; η denotes the target criterion; L D represents the discriminator loss function; L G represents the generator loss function; x denotes real images; z denotes random samples sampled from uniform distribution [-1, 1]; z D and z G denote samples from z; θ D and θ G denote parameters of the discriminator and the generator; variable k t ∈[0, 1] is used to control the degree of emphasis of L(g(z D )) during gradient descent, which is constantly adjusted by k t to maintain the balance E[L(G(z))] = γE[L(x)], the initial value of k0 is 0, t represents the number of iterations; λ k is the proportional gain of k; the hyperparameter γ∈[0, 1] represents the diversity ratio, which is used to balance the performance between the generator and the discriminator and control the quality and diversity of generated samples, and the global convergence measure is: M global = L(x) + |yL(x) - L(G(z G ))|.

5. The method of claim 1, wherein the method further comprises: determining a number of the air-slab building key nodes; and determining a number of the air-slab building key node intelligent design and topology optimization methods. S3 specifically includes: first, the threshold segmentation function in OpenCV2 is used to binarize the rib plate layout image; for the generated rib plate layout scheme, black represents the rib plate design area, and white represents the background area; by converting the image to a binary form, the shape outline of the rib plate design area can be highlighted, making it easier to be processed and analyzed by computer vision algorithms; then, the measurement function of scikit-image is used to filter small connected domains in the image to remove impurities; the specific details are as follows: each independent black area in the image is numbered, and the area of each numbered black area is calculated, then the black area with an area greater than a set threshold is retained by a judgment statement, and the corresponding number of the retained black area is used for region integration; finally, the binary open operation of OpenCV2 is used to remove noise again to improve the image quality, and the rib plate layout image after the binary open operation becomes clearer and more regular.

6. The method of claim 1, wherein, S4 is roughly divided into three parts: first, the filtered rib plate layout scheme is scaled and feature contour extraction is performed in batches; Then, the following process is recorded as a macro command, i.e. stretching the feature profile of a single scheme by SW, and exporting the rib plate together with the top plate and the bottom plate as a STEP entity file; finally, the SW macro command is developed to realize the batch automatic three-dimensional reconstruction of the rib plate layout scheme.

7. A system for intelligent design and topology optimization of key nodes of an aerial construction machine, using the method according to any one of claims 1 to 6, characterized in that, Comprise: Topology optimization module: for establishing a high-quality big data learning library of key nodes of the air building machine by using topology optimization technology; Adversarial network generation module: for introducing a generative design method based on deep learning, fully learning the topological features of existing design schemes, and on the basis of inheriting these features, globally exploring the unknown latent space, thereby deriving a large number of globally optimized design schemes; Intelligent generated scheme post-processing module: for performing edge regularization processing on the generated rib plate layout scheme by using the Python third-party OpenCV2 library; Intelligent generated scheme reconstruction and multi-attribute decision evaluation module: for realizing the batch automatic three-dimensional reconstruction of the generated rib plate layout scheme by using the macro tool and developing it.

8. A computer device, the computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the intelligent design and topology optimization method of the key nodes of the air building machine according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the intelligent design and topology optimization method of the key nodes of the air building machine according to any one of claims 1-6.

10. An information data processing terminal for implementing the intelligent design and topology optimization system of the key nodes of the air building machine according to claim 7.

Images