Intelligent shape finding design method for large-span spatial curved surface special-shaped structure

Abstract

The invention discloses an intelligent shape finding design method for a large-span space curved surface special-shaped structure. The method comprises the steps that an initial three-dimensional parameterized model of the curved surface special-shaped structure is constructed; establishing a mixed loss function taking the structural mechanical property index as a core; an intelligent shape finding model is constructed based on a conditional generative adversarial network, a generator is responsible for performing intelligent adjustment and variation of a curved surface form on an initial model, and a discriminator synchronously judges the mechanical rationality, geometric feasibility and building expressive force of a generated structure based on a mixed loss function; and through antagonistic iterative training of the generator and the discriminator, the structural form is driven to be comprehensively optimized based on targets such as mechanical properties and architectural forms, and finally an ideal structural form is output. According to the method, intelligent confrontation generation and reinforcement learning are deeply fused into structure form finding, and automatic, efficient and intelligent generation and performance-driven self-evolution of the structure form under complex building constraints are achieved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of architectural structural design and computer-aided design, and in particular to an intelligent form-finding method for large-span spatial curved irregular structures that integrates artificial intelligence, generative adversarial networks and reinforcement learning technologies. Background Technology

[0002] Large-span, curved, irregularly shaped spatial structures, such as airport terminals, stadiums, and cultural centers, are typical forms of modern landmark architecture. In pursuit of unique architectural expression, these structures often exhibit complex spatial curved surface forms, such as continuous flow, hyperbolic curves, and twists. For example, a certain international airport terminal uses a double-layered grid shell structure with complex curved surfaces supported by a giant arch shell. The core design principle of this type of "curved irregularly shaped structure" is "form finding," that is, finding a three-dimensional form that precisely matches the architectural concept while possessing efficient mechanical performance and economic efficiency.

[0003] Traditional form-finding methods, such as the force density method and the dynamic relaxation method, have obvious limitations: 1) The process is highly dependent on the engineer's experience, requiring a lot of manual iterations, which is inefficient and has a long design cycle; 2) They are essentially "analysis-oriented" rather than "generative-oriented", making it difficult to find the global optimal solution in a large space of form solutions and easily getting trapped in local optima; 3) For conditions with fixed support points (such as giant arch shells) and complex surface constraints, traditional methods are difficult to systematically explore the complex mapping relationship between form and performance, often sacrificing structural efficiency to meet architectural form requirements.

[0004] In recent years, generative artificial intelligence has made groundbreaking progress in the fields of image and video generation. Generative adversarial networks (GANs), through adversarial games between generators and discriminators, can learn distributions from data and create highly realistic and diverse new samples. This provides a new paradigm for structural form finding, viewing the generation of structural forms as a "creative" process that conforms to the laws of mechanics. Through adversarial learning by AI, it automatically explores forms that combine high mechanical performance and architectural aesthetics.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the general background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an efficient, intelligent, and performance-driven method for finding the shape of large-span spatial curved irregular structures. This method can deeply integrate high-dimensional mechanical performance requirements with intelligent adversarial generation technology to automatically generate structural forms with near-optimal mechanical performance under complex architecture and geometric constraints.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides an intelligent form-finding design method for large-span spatial curved irregular structures, comprising the following steps: S1: Parametric model construction: Based on the architectural concept scheme, extract the key control parameters of the structure and construct a parametric initial three-dimensional structural model. The initial three-dimensional structural model includes a mathematical control network describing the complex curved surface. S2: Establishing a hybrid loss function, creating a comprehensive hybrid loss function L. total The function must contain at least the mechanical performance loss term L. mech Geometric constraint loss term L geo And architectural effect loss item L aes L total =α·L mech +β·L geo + γ·L aes , where α, β, γ are weighting coefficients; S3: Intelligent Shape-Finding Model Construction. This involves constructing an intelligent shape-finding model based on a conditional generative adversarial network (GAN). The model includes: Generator G: Takes a random noise vector and the parameters of the initial 3D structural model as input, and generates a new structural model with variations in surface morphology through a deep learning network; Discriminator D: Taking the structural model output by the generator or the real optimal structural case as input, and based on the hybrid loss function, outputs a comprehensive score to evaluate the merits of the structure in terms of mechanical performance, geometric feasibility and architectural effect; S4: Adversarial Iterative Training and Performance-Driven Evolution. The initial 3D structural model is input into the intelligent shape-finding model. Through multiple adversarial trainings between the generator and the discriminator, and by introducing a reinforcement learning mechanism based on policy gradients, the discriminator's score is used as a reward signal to drive the generator to explore a morphological space with better mechanical performance, while simultaneously optimizing the discriminator's discrimination ability. S5: Optimal form output. After the training process converges, the optimized generator is applied to the initial model to output the final structural form that is optimal in terms of both mechanical performance and architectural aesthetics under given constraints.

[0008] Furthermore, the mechanical property loss term L mech The model rapidly calculates the predicted values ​​of key mechanical performance indicators by taking the coordinates of the control points of the structural morphology and the cross-sections of the members as inputs. These key mechanical performance indicators include the maximum equivalent stress, maximum displacement, critical buckling load, and strain energy of the structure.

[0009] Furthermore, the generator in step S3 adopts a convolutional neural network or graph neural network with a U-Net architecture. The input of the generator is a voxelized or point cloud-based structural model, and the output is new structural model data after morphological adjustment.

[0010] Furthermore, a reinforcement learning mechanism is introduced in step S4, specifically: the generator is regarded as an intelligent agent that executes the policy, the discriminator's comprehensive score of the generated structure is used as the reward given by the environment, and the generator's parameters are updated through the policy gradient method to maximize the accumulated expected reward, thereby realizing the directional exploration towards the high-performance morphological space.

[0011] Furthermore, in step S1, the key control parameters include the coordinates and weights of the control points of the NURBS surface, the topological connection relationship of the bifurcated columns, the library of rod cross-sectional dimensions, and the boundary conditions of the structure.

[0012] By adopting the above technical solution, the present invention has the following beneficial effects: Leading performance-driven design uses mechanical performance indicators directly as the optimization target of AI through surrogate models and loss functions, ensuring the structural efficiency of the generated results and achieving a leap from "constructable" to "optimal construction".

[0013] Intelligent morphological exploration, combining the adversarial mechanism of GAN with the targeted exploration of reinforcement learning, can efficiently search the morphological solution space far beyond the range of human experience, making it easy to discover innovative and high-performance "supernormal morphologies".

[0014] Multi-objective collaboration, through a hybrid loss function, naturally unifies multiple objectives such as structural performance, material usage, construction feasibility, and architectural effect within a single framework for optimization, automatically finding the optimal balance point.

[0015] High-efficiency design frees designers from the tedious cycle of "modeling-analysis-adjustment," enabling algorithm-driven design and greatly shortening the engineering design cycle.

[0016] Highly targeted: It is particularly suitable for solving the form-finding problems of modern spatial structures with complex curved surfaces, irregular shapes, and large spans, such as a certain coastal international airport. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the intelligent form-finding method based on mechanical properties for large-span spatial curved irregular structures provided in this application. Figure 2 This is a flowchart illustrating step S2 of the intelligent form-finding method for large-span spatial curved irregular structures based on mechanical properties provided in this application. Figure 3 This is a flowchart illustrating step S3 of the intelligent form-finding method for large-span spatial curved irregular structures based on mechanical properties provided in this application. Figure 4 This is a flowchart illustrating step S4 of the intelligent form-finding method for large-span spatial curved irregular structures based on mechanical properties provided in this application. Figure 5 This is a flowchart illustrating step S5 of the intelligent form-finding method for large-span spatial curved irregular structures based on mechanical properties provided in this application. Figure 6 This is a demonstration example of the form finding method based on mechanical properties for large-span spatial curved irregular structures provided in this application in the structural system of Terminal 3 at Tianjin Airport. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] This application provides an intelligent form-finding method for large-span spatial curved irregular structures based on mechanical performance. The method includes: constructing an initial three-dimensional parametric model containing the curved irregular structure; establishing a hybrid loss function with structural mechanical performance indicators as the core; constructing an intelligent form-finding model based on a conditional generative adversarial network (GAN), where the generator is responsible for intelligently mutating the surface morphology of the initial model, and the discriminator simultaneously evaluates the mechanical rationality, geometric feasibility, and architectural expressiveness of the generated structure based on the hybrid loss function; through adversarial iterative training between the generator and the discriminator, the structural morphology is driven to evolve towards the Pareto frontier, which achieves optimal mechanical performance and aesthetically pleasing architectural form, ultimately outputting the optimal structural morphology. This invention deeply integrates intelligent adversarial generation and reinforcement learning into structural form-finding, realizing automatic, efficient, and intelligent generation and performance-driven self-evolution of structural morphology under complex architectural constraints. It is particularly suitable for the design of large-span spatial curved irregular structures such as airport terminals and stadiums.

[0022] The core of the intelligent form-finding design method for large-span spatial curved irregular structures in this application lies in constructing an intelligent generative adversarial network driven by mechanical properties. The process is as follows: Figure 1 As shown, it includes the following steps: S1: Parametric model construction: Using parametric design platforms (such as Rhino / Grasshopper), key control parameters of the structure are defined based on the architectural design. For a bifurcated column-supported double-layer curved reticulated shell structure, the parameters include: the coordinates (Px, Py, Pz) and weights of the NURBS surface control points, the topological connection matrix of the bifurcated columns, the thickness and spacing of the upper and lower reticulated shell layers, the library of member cross-sectional dimensions, and the boundary support conditions of the structure. By adjusting these parameters, a series of 3D models with varying shapes but conforming to the basic architectural concept can be generated, serving as the foundational data for intelligent form-finding.

[0023] S2: Establishment of the hybrid loss function: Establish a hybrid loss function L total ,like Figure 2 As shown. It is the navigation system that drives the entire AI model to evolve towards the optimal solution space. The function is a weighted sum of multiple physically meaningful losses: Mechanical property loss L mech This is the core of the invention. By training a lightweight neural network (i.e., a surrogate model) offline, the computationally intensive finite element analysis is replaced, enabling rapid prediction of the structure's maximum equivalent stress σ_max, maximum displacement d_max, and critical buckling load factor λ. mech It can be designed as: L mech =w1 * max(0, (σ_max / σ_allow - 1))+ w2 * max(0, (d_max / d_allow - 1)) - w3 * λ, where σ_allow and d_allow are the allowable stress of the material and the allowable displacement of the specification, and w1, w2, w3 are weights. This design encourages the structure to pursue higher stability while meeting the strength and stiffness requirements.

[0024] Geometric constraint loss L geo Ensure that the generated structure meets geometric and construction requirements, such as minimum clearance between members, constructability of nodal domains, fixed position constraints of arch support points, and surface smoothness (measured by the rate of curvature change).

[0025] Loss of architectural effect L aesThe similarity (e.g., Chamfer Distance) between the generated structure and the target reference surface (such as the initial conceptual model) provided by the architect in terms of control point coordinates or point cloud level is calculated and defined in combination with the Gaussian curvature distribution characteristics of the surface to ensure that the result does not violate the architectural intent.

[0026] S3: Intelligent Shape-Finding Model Construction (based on cGAN): Construct a conditional generative adversarial network model, the structure of which is as follows: Figure 3 As shown.

[0027] Generator G: A convolutional neural network or graph neural network with a U-Net structure. It takes voxelized data or point cloud data of the initial structural model and a random noise vector z as input. The noise z provides randomness and diversity in morphological variation. The generator outputs a new structural model G(z|initial_model) with intelligent surface morphological adjustment through a series of encoding and decoding operations.

[0028] Discriminator D: Employs a convolutional neural network or a graph neural network. It receives a structural model (from a generator or a pre-set library of excellent structural examples) and outputs a scalar score D(X). This score not only represents the probability that the structure "looks like a truly excellent structure," but more importantly, one of its training objectives is to accurately evaluate and reflect the L of the input structure. total Value, that is, it is an "expert discriminator" that understands mechanics and standards.

[0029] S4: Adversarial Iterative Training and Performance-Driven Evolution This process is one of AI's "self-play" and competitive evolution, such as... Figure 4 As shown.

[0030] Generation: The generator intelligently mutates the initial model according to the current strategy to generate a batch of new structural forms.

[0031] Evaluation: The discriminator evaluates these new structures and pre-set excellent structural examples, and gives a comprehensive score. Combat and Evolution Update: (1) Update the discriminator D: The goal is to maximize its ability to distinguish between "good cases" and "generated structures," while simultaneously ensuring that its score for good cases is lower than that of L. total Values ​​are related.

[0032] (2) Updating the Generator G: This invention introduces the policy gradient method from reinforcement learning. The generator is viewed as an agent, its behavior of generating morphologies is the policy, and the discriminator's overall score is the reward for that behavior. The goal of updating the generator is to adjust its parameters to maximize the probability of generating high-reward (i.e., high-performance) structural morphologies. This approach effectively guides the generator to conduct directional exploration in a broad morphology space.

[0033] Repeat steps 1-3 until the generated structure is stable and its mechanical and aesthetic properties meet or exceed the preset convergence criteria.

[0034] S5: Optimal output form: After training converges, the optimal generator model is applied to the initial concept model to output the final structural form, such as... Figure 5 As shown, this form is a structural form that is mechanically efficient and aesthetically pleasing, obtained through massive intelligent exploration and performance-driven evolution under multiple constraints.

[0035] The present invention will now be described in detail using the example of a double-layer curved reticulated shell structure supported by a giant arch shell of a certain airport's T3 terminal building.

[0036] Preparation: Build a parametric model in the Rhino / Grasshopper environment. Define the roof shape using NURBS surfaces, and control the surface shape by adjusting control points and their weights. Define the base point positions, bifurcation logic, and angles of the bifurcated columns. Output the model as point cloud data or convert it into a voxel mesh as input data for the AI ​​model.

[0037] Model training: In the Python environment, the cGAN model is built using the PyTorch deep learning framework.

[0038] Generator G: Uses a 3D U-Net, with inputs of 100-dimensional noise z and 128x128x128 voxel data of the initial model, and outputs voxel data of the same size representing the new morphology.

[0039] Discriminator D: Employs a 3D CNN, uses spectral normalization to stabilize the training process, and ultimately outputs a comprehensive score.

[0040] Proxy Model: A standalone 3D CNN regression model is pre-trained using 50,000 structural samples generated through parametric methods and subjected to finite element analysis, along with their mechanical indices (σ_max, d_max, λ). This model is frozen during cGAN training for rapid computation of L. mech .

[0041] Training process: Set the batch size to 8 and use Wasserstein GAN with Gradient Penalty (WGAN-GP) loss to enhance training stability. Train the discriminator and generator alternately. Use the Adam optimizer. Train for 20,000 epochs. For generator updates, update the generator once using the policy gradient method every 5 discriminator training iterations.

[0042] Form-finding process: The prepared initial model of the airport terminal is input into the trained system. The system runs processes S1 to S4 in the background. Engineers can monitor the evolution of the generated form and the changes in key mechanical indicators (such as maximum stress) in real time through a visual interface.

[0043] Results Output and Validation: After training, the final output form is shown below. Figure 6 As shown. The morphological results were imported into the commercial finite element software MIDAS for high-precision verification analysis. Practice shows that the structural morphology generated by the method of this invention, under the premise of satisfying all constraints, generally has better mechanical properties than the morphology obtained by traditional methods, and the material usage has an optimization potential of 5%-15%.

[0044] In summary, this invention provides a novel, efficient, and intelligent solution for the design of large-span spatial curved irregular structures by deeply coupling AI adversarial generation, reinforcement learning, and structural mechanical properties. It has broad engineering application prospects and significant industrial value.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent form-finding design method for large-span spatial curved irregular structures, characterized in that, Includes the following steps: S1: Parametric model construction: Based on the architectural concept scheme, extract the key control parameters of the structure and construct a parametric initial three-dimensional structural model. The initial three-dimensional structural model includes a mathematical control network describing the complex curved surface. S2: Establishing a hybrid loss function, creating a comprehensive hybrid loss function L. total The function must contain at least the mechanical performance loss term L. mech Geometric constraint loss term L geo And architectural effect loss item L aes L total =α·L mech +β·L geo +γ·L aes , where α, β, γ are weighting coefficients; S3: Intelligent Shape-Finding Model Construction. This involves constructing an intelligent shape-finding model based on a conditional generative adversarial network (GAN). The model includes: Generator G: Takes a random noise vector and the parameters of the initial 3D structural model as input, and generates a new structural model with variations in surface morphology through a deep learning network; Discriminator D: Taking the structural model output by the generator or the real optimal structural case as input, and based on the hybrid loss function, outputs a comprehensive score to evaluate the merits of the structure in terms of mechanical performance, geometric feasibility and architectural effect; S4: Adversarial Iterative Training and Performance-Driven Evolution. The initial 3D structural model is input into the intelligent shape-finding model. Through multiple adversarial trainings between the generator and the discriminator, and by introducing a reinforcement learning mechanism based on policy gradients, the discriminator's score is used as a reward signal to drive the generator to explore a morphological space with better mechanical performance, while simultaneously optimizing the discriminator's discrimination ability. S5: Optimal form output. After the training process converges, the optimized generator is applied to the initial model to output the final structural form that is optimal in terms of both mechanical performance and architectural aesthetics under given constraints.

2. The intelligent form-finding design method for large-span spatial curved irregular structures according to claim 1, characterized in that, The mechanical property loss term L mech The model rapidly calculates the predicted values ​​of key mechanical performance indicators by taking the coordinates of the control points of the structural morphology and the cross-sections of the members as inputs. These key mechanical performance indicators include the maximum equivalent stress, maximum displacement, critical buckling load, and strain energy of the structure.

3. The intelligent form-finding design method for large-span spatial curved irregular structures according to claim 1, characterized in that, The generator in step S3 uses a convolutional neural network or graph neural network with a U-Net architecture. The input of the generator is a voxelized or point cloud-based structural model, and the output is new structural model data after morphological adjustment.

4. The intelligent form-finding design method for large-span spatial curved irregular structures according to claim 1, characterized in that, In step S4, a reinforcement learning mechanism is introduced, specifically: the generator is regarded as an intelligent agent that executes the policy, the discriminator's comprehensive score of the generated structure is used as the reward given by the environment, and the generator's parameters are updated through the policy gradient method to maximize the accumulated expected reward, thereby achieving targeted exploration towards the high-performance morphological space.

5. The intelligent form-finding design method for large-span spatial curved irregular structures according to claim 1, characterized in that, In step S1, the key control parameters include the coordinates and weights of the control points of the NURBS surface, the topological connection relationship of the bifurcated columns, the library of rod cross-sectional dimensions, and the boundary conditions of the structure.

Images