Pipe truss structure design and selection method based on structured generative adversarial network

Abstract

The present application relates to the technical field of structural engineering optimization design and intelligent selection technology, in particular to a tubular truss structure design and selection method based on structured generative adversarial network, comprising: obtaining tubular truss load data and corresponding structure design schemes under different load conditions; encoding the load data and the design schemes respectively to obtain multi-dimensional dynamic load feature vectors and graph structure information; inputting the two into a generator to generate design schemes and selection schemes under the constraint of physical information and with the target of minimizing the use of steel; discriminating the generated schemes and the real schemes by a discriminator, calculating the adversarial loss, and alternately training the generator and the discriminator until convergence; finally inputting the preset load and the structure design scheme into the converged generator to output the optimal selection scheme. The method realizes end-to-end mapping from dynamic load to optimal design and selection scheme, significantly improves the design efficiency, and realizes structural lightweighting while ensuring engineering reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of structural engineering optimization design and intelligent selection, in particular to a truss tube structure design and selection method based on structured generative adversarial network. BACKGROUND

[0002] With the global wind power technology developing towards large-scale and super-high hub height, the design of wind turbine support structure faces unprecedented engineering challenges. In the context of pursuing higher power generation efficiency and wider utilization of wind energy resources, the support structure must withstand increasingly complex multi-dimensional dynamic loads, which puts extremely high requirements on the strength, stability, fatigue life and dynamic characteristics of the structure. The traditional tubular tower structure has been difficult to meet the economic and reliability requirements of the next generation of super-large wind power projects when the hub height exceeds 140m due to its inherent low material utilization, high manufacturing and transportation costs and other limitations.

[0003] Truss support structure is considered an ideal solution to overcome the bottleneck of traditional tubular tower due to its excellent lightweight, high stiffness and high material utilization characteristics, as shown in the truss support structure shown in the truss support structure. Figure 6 However, the existing truss structure design and optimization method still has significant technical defects. The truss structure has a highly complex topological configuration and numerous design parameters. Currently, its optimization design mainly relies on the following two technical paths:

[0004] Path one: iterative optimization based on numerical simulation; this path usually adopts the combination of finite element analysis software and traditional optimization algorithm, and the designer needs to establish a large number of numerical simulation models with different design parameters and perform iterative optimization under the condition of meeting multiple constraints such as strength, stability, fatigue and frequency; however, this method has the defects of low calculation efficiency, long cycle and high calculation cost, which is difficult to meet the demand of rapid response and multiple iterations of engineering design;

[0005] Path two: regression prediction based on traditional artificial intelligence; to solve the efficiency bottleneck of numerical simulation, some studies use neural network technology to build a proxy model to predict the structure performance or inversely predict the design parameters; however, this method has new limitations: the traditional AI model is essentially a "black box" regression model, which highly depends on the distribution of training data, and its prediction results cannot guarantee strict compliance with the physical laws and hard constraints in engineering specifications; and the existing model is difficult to directly process the input of any dynamic load time series, and lacks the end-to-end intelligent mapping capability from any load condition to the optimal structure parameters.

[0006] In summary, there is an urgent need for a pipe truss structure design and selection scheme that can break through the bottleneck of traditional numerical simulation calculation and embed the principles of engineering mechanics and design specifications within the model, achieving lightweight and engineering reliability collaborative intelligent optimization of pipe truss structures. SUMMARY

[0007] The purpose of the present application is to provide a pipe truss structure design and selection method based on structured generative adversarial networks to solve the technical problems identified in the background art.

[0008] The present application is implemented by the following technical scheme: a pipe truss structure design and selection method based on structured generative adversarial networks, comprising the following steps:

[0009] Obtain pipe truss load data under different load conditions and corresponding pipe truss structure design schemes, which include pipe truss structure topology, geometry and material properties;

[0010] Respectively encode the pipe truss load data and its corresponding pipe truss structure design scheme, including encoding the pipe truss load data using a multi-modal encoder to obtain a multi-dimensional dynamic load feature vector, and graph structure encoding the pipe truss structure design scheme to obtain graph structure information;

[0011] Take the multi-dimensional dynamic load feature vector and graph structure information as the input of the generator, consider the physical information constraint, and minimize the steel consumption as the optimization objective to obtain the generated pipe truss structure design scheme and the corresponding selection scheme, which includes steel consumption and cross-sectional parameters of each member;

[0012] Use the discriminator to process the generated pipe truss structure design scheme to obtain the generated pipe truss structure design scheme discrimination probability, and process the real pipe truss structure design scheme to obtain the real pipe truss structure design scheme discrimination probability;

[0013] Determine the adversarial loss according to the generated pipe truss structure design scheme discrimination probability and the real pipe truss structure design scheme discrimination probability, and alternately train the generator and the discriminator until the output of the generator meets the convergence condition;

[0014] For the preset pipe truss load data and the preset pipe truss structure design scheme, after encoding processing, input the converged generator to obtain the optimal selection scheme.

[0015] According to a preferred embodiment, the pipe truss structure topology includes the shape of the web member arrangement, and the geometry includes the bottom outer diameter of the member, the top-to-bottom diameter ratio, the number of edges, and the number of segments.

[0016] According to a preferred embodiment, the multimodal encoder is based on a causal temporal convolutional network, and the multidimensional dynamic load feature vector output by the encoder is used to approximately characterize the structural engineering requirement parameters under arbitrary dynamic loads.

[0017] According to a preferred embodiment, the physical information constraint includes a strength constraint loss term, a stability constraint loss term, a fatigue damage loss term, and a frequency constraint loss term, expressed as follows:

[0018]

[0019] In the above formula, This represents the total loss due to physical information constraints. , , and The learnable weight coefficients for each loss term are constrained by physical information. This represents the strength constraint loss term. Represents the stability constraint loss term. This represents the fatigue damage loss term. This represents the frequency constraint loss term.

[0020] According to a preferred embodiment, the strength constraint loss term must satisfy the following condition: the ratio of the calculated stress to the allowable stress of all members is not greater than 1, expressed as:

[0021]

[0022] In the above formula, Represents a set of rods, For rods Calculated stress, This indicates the allowable stress of the member.

[0023] According to a preferred embodiment, the stability constraint loss term must satisfy the following: the ratio of the actual slenderness ratio to the allowable slenderness ratio of all members, and the ratio of the actual diameter-to-thickness ratio to the allowable diameter-to-thickness ratio, are both not greater than 1, expressed as:

[0024]

[0025] In the above formula, For rods The slenderness ratio, For the allowable slenderness ratio of the rod, For rods Aspect ratio, This refers to the allowable diameter-to-thickness ratio of the rod.

[0026] According to a preferred embodiment, the fatigue damage loss term needs to satisfy: the ratio of the accumulated fatigue damage of all bars to the allowable accumulated fatigue damage value is not greater than 1, expressed as:

[0027]

[0028] In the above formula, represents the accumulated fatigue damage of the bar , represents the allowable accumulated fatigue damage value.

[0029] According to a preferred embodiment, the frequency constraint loss term needs to satisfy: the absolute value of the difference between the first-order characteristic frequency of the tube truss structure and the excitation frequency is not less than 15% of the excitation frequency, expressed as:

[0030]

[0031] In the above formula, represents the excitation frequency of the wind turbine 1P or 3P, represents the first-order characteristic frequency of the tube truss structure.

[0032] According to a preferred embodiment, the total optimization target expression of the generator is as follows:

[0033]

[0034] In the above formula, represents the total optimization target of the generator, , and is the weight coefficient of each sub-target of the total optimization target, represents the steel material usage, represents the adversarial loss.

[0035] According to a preferred embodiment, the alternating training of the generator and the discriminator adopts a staged training strategy, specifically:

[0036] The first stage is set to zero, aiming to minimize the physical information constraint , so that the generator quickly learns and outputs a feasible tube truss structure design scheme and the corresponding selection scheme;

[0037] The second stage introduces a non-zero , which drives the generator to converge to the optimal solution of the adversarial loss constraint under the supervision of the discriminator, aiming to minimize .

[0038] The technical scheme of the pipe truss structure design and selection method based on the structured generative adversarial network has at least the following advantages and beneficial effects: (1) the method can realize end-to-end mapping from multi-dimensional dynamic load input to optimal selection scheme output, compared with the traditional numerical simulation method, the design efficiency and response speed are obviously improved while the time-consuming of iterative optimization calculation is effectively reduced; (2) by considering the physical information constraint and taking the minimum steel consumption as the optimization objective, the generator can accurately search and converge to the globally optimal and lightest scheme under the premise of ensuring engineering reliability; (3) by combining the multi-modal encoder and the generator, the model has the ability to process any dynamic load condition and complex structure topology, greatly expanding its application range in the next generation of wind power structure design. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The overall flowchart of the pipe truss structure design and selection method based on the structured generative adversarial network provided for the embodiment 1 of the present application is shown in the figure.

[0040] Figure 2 The principle diagram of the edge condition message passing of the graph neural network provided for the embodiment 1 of the present application is shown in the figure.

[0041] Figure 3 The network connection diagram of the generator and the discriminator provided for the embodiment 1 of the present application is shown in the figure.

[0042] Figure 4 The principle diagram of the structure reliability constraint embedding of the physical information constraint loss provided for the embodiment 1 of the present application is shown in the figure.

[0043] Figure 5 The engineering constraint diagram of the physical information constraint loss function provided for the embodiment 1 of the present application is shown in the figure.

[0044] Figure 6 The structural diagram of the truss support structure referred to in the background art is shown in the figure. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0046] Embodiment 1

[0047] The embodiments of the present application provide a pipe truss structure design and selection method based on a structured generative adversarial network, Figure 1The overall flowchart of the pipe truss structure design and selection method is shown in FIG. 1. Figure 1 The pipe truss structure design and selection method includes the following steps:

[0048] Step S1, data preparation and coding;

[0049] Firstly, a data set for model training is obtained, which includes multiple groups of pipe truss load time series under different load conditions, and a verified pipe truss structure design scheme corresponding to each group of load time series. The load time series can be obtained from a wind turbine overall simulation software, such as GH-Bladed software. The structure design scheme includes topology (such as web member arrangement shape), geometry (bar member bottom outer diameter, top-to-bottom diameter ratio, number of edges, number of segments), and material properties.

[0050] Subsequently, the two types of data are coded and processed to convert them into feature representations suitable for neural network processing, as follows:

[0051] Load coding: a multi-modal encoder based on a causal time series convolution network is used to process the load time series, extract and compress its dynamic features, and form a fixed-dimensional multi-dimensional dynamic load feature vector. This vector aims to represent the comprehensive effect of the load on the structure;

[0052] Structure coding: the structure design scheme is abstracted into a graph structure, with nodes representing connection points and edges representing bar members. A graph neural network is used to encode the graph structure to obtain graph structure information containing topological and geometric relationships.

[0053] Step S2, generator design;

[0054] In this embodiment, the step aims to build a generator model using a graph neural network GNN with an edge-conditioned message passing mechanism; see FIG. 2. Figure 2 The generator model takes the multi-dimensional dynamic load feature vector and graph structure information obtained in step S1 as its conditional input, and under the premise of considering physical information constraints, minimizes the steel usage as the optimization direction, and outputs a generated pipe truss structure design scheme and a corresponding selection scheme, which includes steel usage and bar member cross-section parameters.

[0055] In combination with FIG. 3, Figure 2 The edge-conditioned message passing mechanism of the graph neural network GNN used by the generator is implemented as follows: the multi-dimensional dynamic load feature vector and the graph structure information are taken as initial inputs. For nodes and nodes in the graph structure, edge corresponds to a bar member of the pipe truss, and the edge-conditioned information it carries includes features related to the geometric parameters and material properties of the bar member. It will receive from the node and nodes The message, and generate a message accordingly. Subsequently, regarding the message With nodes The node's own state is aggregated, and the node state is iteratively optimized using an update function. After the GNN iterates through L layers, the output layer is decoded by an MLP to finally output the design scheme and selection scheme for the truss structure.

[0056] The physical information constraints are embedded in the form of a loss function, including:

[0057] The strength constraint loss term must satisfy the following condition: the ratio of the calculated stress to the allowable stress of all members is not greater than 1, expressed as:

[0058]

[0059] In the above formula, Represents a set of rods, For rods The calculated stress, This indicates the allowable stress of the member.

[0060] The stability constraint loss term must satisfy the following condition: the ratio of the actual slenderness ratio to the allowable slenderness ratio of all members, and the ratio of the actual diameter-to-thickness ratio to the allowable diameter-to-thickness ratio, must not exceed 1, expressed as:

[0061]

[0062] In the above formula, For rods The slenderness ratio, For the allowable slenderness ratio of the rod, For rods Aspect ratio, This refers to the allowable diameter-to-thickness ratio of the rod.

[0063] For fatigue damage loss terms, see [link / reference]. Figure 5 As shown, according to DNV-RP-C203 and IIW criteria, fatigue assessment is performed using the nominal stress method and SN curves. The following condition must be met: the ratio of the cumulative fatigue damage to the allowable cumulative fatigue damage value for all members must not exceed 1, expressed as:

[0064]

[0065] In the above formula, Indicating rods Cumulative fatigue damage, This indicates the allowable cumulative fatigue damage value. (From...) Figure 5As shown in (b) in FIG. 1, the fatigue grade FAT of the tower column component is 71, and the FAT grades of the diagonal bars DB and the cross bars CB are 45.

[0066] a frequency constraint loss term, see Figure 5 As shown in FIG. 2, to avoid resonance, the absolute value of the difference between the first-order characteristic frequency of the tube truss structure and the excitation frequency should be greater than 15% of the excitation frequency, expressed as:

[0067]

[0068] In the above formula, represents the excitation frequency of the wind turbine 1P or 3P, represents the first-order characteristic frequency of the tube truss structure. It is calculated by Figure 5 As shown in (a) in FIG. 3, taking a 5.5 MW wind turbine as an example, the rated speed of the wind turbine is 9.57 rpm, and the corresponding excitation frequencies are 0.1595 Hz (1P) and 0.4785 Hz (3P), respectively.

[0069] see Figure 4 As shown in FIG. 4, in this embodiment, the above loss terms are combined as a physical information constraint total loss to guide the generator model to follow the engineering specifications, and the expression is as follows:

[0070]

[0071] In the above formula, , , and are the learnable weight coefficients of the physical information constraint loss terms.

[0072] It should be noted that, by combining the multi-modal encoder and the generator, the model has the ability to process any dynamic load case and complex structure topology, greatly expanding its application range in the next generation of wind power structure design.

[0073] Step S3, discriminator design;

[0074] In this embodiment, this step aims to build a discriminator model for evaluating the rationality of the generated tube truss structure design scheme output by the generator model; specifically, see Figure 3 As shown in FIG. 5, it includes:

[0075] Step S301, input the generated tube truss structure design scheme generated in step S2 into the discriminator model to obtain the generated tube truss structure design scheme discrimination probability;

[0076] Step S302, input the real pipe truss structure design scheme into the same discriminator model to obtain a real pipe truss structure design scheme discrimination probability;

[0077] Step S303, calculate the adversarial loss according to the generated pipe truss structure design scheme discrimination probability and the real pipe truss structure design scheme discrimination probability, so as to measure the difference between the generated scheme and the real scheme distribution.

[0078] In combination Figure 3 As shown in the figure, after the pipe truss load data and the corresponding structure design scheme provided by the data set are encoded through step S1, the multi-dimensional dynamic load feature vector and the graph structure information are output, and are sent to the generator; the generator obtains the generated pipe truss structure design scheme and the corresponding selection scheme according to the total optimization target, and inputs the generated pipe truss structure design scheme and the corresponding selection scheme into the discriminator together with the real pipe truss structure design scheme in the data set; the discriminator outputs the respective discrimination probability by discriminating the two types of schemes; the adversarial loss and the physical information constraint loss calculated by the two types of discrimination probability are fed back to the generator, and the generator is optimized.

[0079] Step S4, alternating training;

[0080] In this embodiment, based on steps S2 and S3, the generator model and the discriminator model are alternately iteratively trained;

[0081] The total optimization target expression of the generator model is as follows:

[0082]

[0083] In the above formula, denotes the total optimization target of the generator model, , and is the weight coefficient of each sub-target of the total optimization target, denotes the steel consumption, denotes the adversarial loss.

[0084] It should be noted that, by taking into account the physical information constraint and minimizing the steel consumption as the optimization target, the generator can accurately search and converge to the globally optimal and lightest scheme under the premise of ensuring engineering reliability.

[0085] In order to improve the training efficiency and stability, this embodiment adopts a phased training strategy, which is as follows:

[0086] First phase: set to zero, so as to minimize the physical information constraint as the target, so that the generator model quickly learns and outputs a feasible pipe truss structure design scheme and the corresponding selection scheme;

[0087] Second stage: introduce non-zero , drive the generator model under the supervision of the discriminator model, aiming to minimize , converging to the optimal solution of the adversarial loss constraint.

[0088] Step S5, optimal scheme generation;

[0089] In this embodiment, for the preset pipe truss load data and the preset pipe truss structure design scheme, after the same coding processing as step S1, the optimal scheme can be obtained by inputting the converged generator model.

[0090] In summary, the method can realize end-to-end mapping from multi-dimensional dynamic load input to optimal scheme output. Compared with the traditional numerical simulation method, the design efficiency and response speed are obviously improved while effectively reducing the time-consuming of iterative optimization calculation.

[0091] In addition, the preset pipe truss structure design scheme and the corresponding optimal scheme can be further subjected to regional seismic damage assessment to determine the seismic economic loss under the action of the earthquake. Here, no more details are given.

[0092] The above is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for designing and selecting tubular truss structures based on structured generative adversarial networks, characterized in that, Includes the following steps: Obtain the load data of the tubular truss under different load conditions and the corresponding tubular truss structural design scheme, wherein the tubular truss structural design scheme includes the topology, geometry and material properties of the tubular truss structure; The truss load data and its corresponding truss structural design scheme are encoded, including encoding the truss load data using a multimodal encoder to obtain a multidimensional dynamic load feature vector, and encoding the truss structural design scheme into a graph structure to obtain graph structure information. Using the multidimensional dynamic load feature vector and graph structure information as input to the generator, taking into account physical information constraints, and with minimizing steel usage as the optimization objective, a design scheme for a generated tubular truss structure and a corresponding selection scheme are obtained. The selection scheme includes the amount of steel used and the cross-sectional parameters of each member. The generated truss structure design scheme is processed by a discriminator to obtain the discrimination probability of the generated truss structure design scheme, and the real truss structure design scheme is processed to obtain the discrimination probability of the real truss structure design scheme. The adversarial loss is determined based on the discrimination probability of the generated tubular truss structure design scheme and the discrimination probability of the actual tubular truss structure design scheme. The generator and discriminator are trained alternately until the output of the generator satisfies the convergence condition. The preset truss load data and preset truss structural design scheme are encoded and then input into a convergent generator to obtain the optimal solution.

2. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 1, characterized in that, The topology of the tubular truss structure includes the shape of the web members, and the geometry includes the bottom outer diameter of the members, the ratio of the top to the bottom diameter, the number of sides, and the number of segments.

3. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 1, characterized in that, The multimodal encoder is based on a causal temporal convolutional network, and its output multidimensional dynamic load feature vector is used to approximately characterize the structural engineering requirement parameters under arbitrary dynamic loads.

4. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 1, characterized in that, The physical information constraints include strength constraint loss terms, stability constraint loss terms, fatigue damage loss terms, and frequency constraint loss terms, expressed as follows: In the above formula, This represents the total loss due to physical information constraints. , , and The learnable weight coefficients for each loss term are constrained by physical information. This represents the strength constraint loss term. Represents the stability constraint loss term. This represents the fatigue damage loss term. This represents the frequency constraint loss term.

5. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 4, characterized in that, The expression for the strength constraint loss term is as follows: In the above formula, Represents a set of rods, For rods The calculated stress, This indicates the allowable stress of the member.

6. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 4, characterized in that, The expression for the stability constraint loss term is as follows: In the above formula, Represents a set of rods, For rods The slenderness ratio, For the allowable slenderness ratio of the rod, For rods Aspect ratio, This refers to the allowable diameter-to-thickness ratio of the rod.

7. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 4, characterized in that, The expression for the fatigue damage loss term is as follows: In the above formula, Represents a set of rods, Indicating rods Cumulative fatigue damage, This indicates the allowable cumulative fatigue damage value.

8. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 4, characterized in that, The expression for the frequency constraint loss term is as follows: In the above formula, This indicates the excitation frequency of a 1P or 3P wind turbine generator. This represents the first-order characteristic frequency of the tubular truss structure.

9. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 4, characterized in that, The overall optimization objective expression for the generator is as follows: In the above formula, This represents the overall optimization objective of the generator. , and The weight coefficients of each sub-objective of the overall optimization objective are given. Indicates the amount of steel used. It signifies resistance to loss.

10. The method for designing and selecting tubular truss structures based on structured generative adversarial networks as described in claim 9, characterized in that, The alternating training of the generator and discriminator adopts a phased training strategy, specifically: Phase 1 Setup Set to zero to minimize physical information constraints. The goal is to enable the generator to quickly learn and output feasible tubular truss structure design schemes and corresponding selection schemes; The second stage introduces non-zero values. This drives the generator to minimize, under the supervision of the discriminator. With the objective of convergence to the optimal solution that overcomes the loss constraint.

Images