Modular neural network driven co-design method of sandwich beam with bamboo-like core

Abstract

The application discloses a modular neural network driven three-element collaborative design method for a bamboo-imitating sandwich wing spar, relates to the technical field of aircraft design, and comprises the following steps: constructing a linear bearing performance sub neural network; constructing a nonlinear effect sub neural network; constructing a defect reduction effect sub neural network; fusing the linear bearing performance sub neural network, the nonlinear effect sub neural network and the defect reduction effect sub neural network to obtain a modular neural network; inputting input parameters into the modular neural network to obtain output parameters through the modular neural network; and iteratively designing a spar parameter through a spar parameter iterative design module to complete the optimization design of the spar. The modular neural network driven three-element collaborative design method for the bamboo-imitating sandwich wing spar is adopted, the bamboo-imitating sandwich wing spar designed by the method is generally applicable to a large-aspect-ratio low-wing load wing, the prediction accuracy of a structural damage boundary is high, and the method has the advantages of high accuracy, high efficiency and good robustness.

Description

Technical Field

[0001] This invention relates to the field of aircraft design technology, and in particular to a modular neural network-driven collaborative design method for the three elements of a bamboo-inspired sandwich wing spars. Background Technology

[0002] Near-space solar-powered unmanned aerial vehicles (UAVs) are characterized by long flight time and zero pollution, and theoretically can achieve uninterrupted flight, playing an increasingly important role in space offense and defense and information warfare. The performance indicators of solar-powered UAVs are highly sensitive to structural weight, and the wing spars are the heaviest components in the entire structure of a solar-powered UAV; therefore, wing spar design is a key focus in the design of this type of aircraft.

[0003] However, compared to traditional UAVs, solar-powered UAVs differ significantly in their structural form and design methods. Firstly, the wing loading of solar-powered UAVs is one to two orders of magnitude lower than that of traditional UAVs; secondly, the aspect ratio of solar-powered UAVs is more than 50% higher than that of traditional UAVs. These two unique characteristics present the following challenges to the design of solar-powered UAV wing spars:

[0004] First, solar-powered drones have wings with a large aspect ratio and low wing loading. If the wing spars are designed according to strength and stiffness requirements, their wall thickness is usually thin, which will make buckling failure the main failure mode, causing instability to occur before strength failure.

[0005] Second, this type of wing uses thin-walled beams with a large aspect ratio. When these beams are subjected to bending, the beam cross-section will undergo flattening deformation, resulting in significant nonlinear effects. This causes actual buckling failure to occur prematurely, thereby further reducing the buckling boundary.

[0006] Third, thin-walled structures also face the problem of defect sensitivity, that is, the actual structure is affected by geometric and material deviations introduced by processing, transportation and other processes, and the actual load-bearing capacity may be further reduced.

[0007] To address the aforementioned technical challenges, a common solution currently is to improve buckling performance through sandwich structures. This is because sandwich structures can achieve a significant increase in structural thickness at a relatively low weight cost, effectively improving the buckling load of thin-walled wing beams with high slenderness ratios. The improvement effect is significantly better than that achieved by thickening carbon fiber and adding stiffeners. Furthermore, sandwich structures also contribute to reducing structural defect sensitivity. However, sandwich wing beams with high slenderness ratios still exhibit significant geometric nonlinear effects, meaning the cross-section is prone to flattening deformation, which can cause the actual buckling load of the wing beam to be lower than the linearly predicted value. Therefore, some researchers have proposed a design approach that involves adding a bamboo-like structure to the thin-walled wing beam cross-section to suppress flattening deformation, thereby improving the buckling boundary.

[0008] In recent years, research teams have combined the aforementioned bamboo-like structure and sandwich structure to propose a bamboo-like sandwich structure suitable for wings with high aspect ratios and low wing loading. Their load-bearing capacity has been verified through numerical simulation and mechanical experiments. However, existing design methods for this type of beam are still based on traditional wing spars, which have the following problems: the influence of the bamboo-like structure and sandwich structure on the nonlinear effects of the wing spars is not quantitatively considered; the mechanism by which the bamboo-like structure and sandwich structure affect the wing spars' sensitivity to defects is not reflected in the design; the upper boundary of structural failure can only be roughly estimated using empirical safety factors, leading to an ambiguous failure boundary; the designed wing spars are conservative and require continuous iterative optimization through practical applications; and the structural weight advantage is weakened, easily causing structural overweight problems. Summary of the Invention

[0009] The purpose of this invention is to provide a modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam, thereby solving the problems mentioned in the background art.

[0010] To achieve the above objectives, this invention provides a modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam, comprising the following steps:

[0011] Step S1: Construct a linear load-bearing capacity sub-neural network;

[0012] Step S2: Construct a nonlinear effect sub-neural network;

[0013] Step S3: Construct a defect reduction effect sub-neural network;

[0014] Step S4: Fuse the linear load-bearing capacity sub-neural network, the nonlinear effect sub-neural network, and the defect reduction effect sub-neural network to obtain a modular neural network;

[0015] Step S5: Input the input parameters into the modular neural network and obtain the output parameters through the modular neural network;

[0016] Step S6: Iterate through step S5 to complete the wing sparsity optimization design.

[0017] Preferably, step S1 includes:

[0018] Step S11: Construct a sample library containing various panel ply parameters, and obtain the linear stress matrix of the material and the linear buckling load of the structure through finite element analysis to form a dataset;

[0019] Step S12: Build a model based on the network architecture. The input layer receives panel ply parameters, the hidden layer processes the relationship between panel ply parameters through multiple neurons, and the output layer outputs the linear strain of the material and the linear buckling load of the structure.

[0020] Step S13: Through training and optimization of the dataset, the linear bearing capacity sub-neural network outputs the corresponding linear strain of the material and the linear buckling load of the structure according to the input panel layup parameters.

[0021] Preferably, step S2 includes:

[0022] Step S21: Fix the panel layup parameters and design a sample library containing different sandwich structure parameters and bamboo-like structure parameters;

[0023] Step S22: Using the finite element analysis method, obtain the nonlinear stability factor and form a dataset;

[0024] Step S23: Build a network according to the network architecture. The input layer receives the parameters of the sandwich structure and the parameters of the bamboo-like structure. The hidden layer processes the correlation between the parameters through multiple neurons. The output layer outputs the nonlinear stability factor of material stress and the nonlinear stability factor of buckling load.

[0025] Step S24: Perform training optimization so that the nonlinear effect sub-neural network outputs the corresponding nonlinear stability factor based on the input sandwich structure parameters and bamboo-like structure parameters.

[0026] Preferably, step S3 includes:

[0027] Step S31: Generate a large-scale sample of a wing beam structure with geometric defects, fix the panel ply parameters, adjust the sandwich structure parameters and the bamboo-like structure parameters, and use the finite element analysis method to obtain the defect reduction factor.

[0028] Step S32: Build a network according to the network architecture. The input layer receives the parameters of the sandwich structure and the parameters of the bamboo-like structure. The hidden layer processes the correlation between the parameters and the defect reduction effect through multiple layers of neurons. The output layer outputs the defect reduction factor accordingly.

[0029] Step S33: Perform training and optimization so that the defect reduction effect sub-neural network outputs the corresponding defect reduction factor based on the input sandwich structure parameters and bamboo-like structure parameters.

[0030] Preferably, in step S31, generating a large-scale sample of a spar structure containing geometric defects using a defect random sample generation module includes:

[0031] Step S311: Use three-dimensional laser scanning technology to obtain the geometric shape of the sandwich and panel, and compare it with the ideal shape to obtain the distribution of geometric defects in the thin-walled tube beam;

[0032] Step S312: Perform a Fourier transform on the geometric defect distribution to obtain the frequency domain characteristics and power spectrum of the geometric defects;

[0033] Step S313: Generate large-scale geometric defect samples by performing inverse transformation on the power spectrum, and randomly generate sandwich beam samples with the same defect pattern by randomly assigning phase.

[0034] Step S314: In the evolutionary power spectrum, the spatial distribution of the power spectrum amplitude is further defined, and a spatially uneven geometric initial defect is generated based on the evolutionary power spectrum. The influence of the spatial distribution characteristics of the geometric defect on the post-buckling behavior is studied.

[0035] Step S315: Detect the initial bonding defect characteristics and their distribution using ultrasonic flaw detection technology, and quantitatively describe the geometric and distribution characteristics of the bonding defects using statistical methods, providing a basis for the definition and generation of bonding defects.

[0036] Preferably, the finite element analysis method includes:

[0037] A finite element model of a bamboo-joint sandwich beam was created in Abaqus. The panel and the sandwich core were modeled using S4R continuous shell elements, and the bamboo joints were modeled using Timoshenko beam elements.

[0038] A fixed boundary condition is applied to one end of the beam, and a multi-point constraint method is used to establish beam constraint between the boundary and a reference point located at the center at the other end, and a bending load is applied to that reference point.

[0039] The analysis step calculates linear stress, and the analysis step calculates linear buckling load;

[0040] In the analysis step, the velocity boundary condition is replaced with the load at the reference point. The load-displacement curve under bending condition is obtained by monitoring the support reaction force at the reference point. The load corresponding to the point where the load decreases sharply is taken as the nonlinear buckling load, and the stress output in the intermediate step is taken as the nonlinear stress.

[0041] Preferably, in step S4, the linear bearing capacity sub-neural network, the nonlinear effect sub-neural network, and the defect reduction effect sub-neural network are fused using the physical rule fusion module.

[0042] Preferably, step S5 includes:

[0043] Input panel layup into the linear load-bearing capacity sub-neural network, and output linear strain of material and linear buckling load of structure;

[0044] Input bamboo-like structure parameters and sandwich layer structure parameters into the nonlinear effect sub-neural network, and output a nonlinear stability factor;

[0045] Input bamboo-like structure parameters and sandwich layer structure parameters into the defect reduction effect sub-neural network, and output the defect reduction factor;

[0046] Among them, the panel ply parameters include the number of ply layers and the ply direction, the bamboo-like structure parameters include the spacing and thickness, and the sandwich layer structure parameters include the spacing and thickness.

[0047] Preferably, the strength failure boundary and buckling failure boundary are determined by the linear strain of the material, the linear buckling load of the structure, the nonlinear stability factor, and the defect reduction factor.

[0048] Preferably, in step S6, the wing spars parameter iterative design module is used to iterate and complete the wing spars optimization design.

[0049] Therefore, the present invention employs the above-mentioned modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam, which has the following beneficial effects:

[0050] (1) The bamboo-like sandwich wing spars designed by this method are generally applicable to wings with large aspect ratio and low wing load. Compared with existing design methods, the structural failure boundary prediction accuracy is higher, and it has the advantages of high accuracy, high efficiency and good robustness.

[0051] (2) This method establishes a sub-neural network model that characterizes the intrinsic relationship between the parameters of the bamboo-like structure, the parameters of the sandwich structure, and the nonlinear material strain and nonlinear buckling moment through batch analysis of samples, thereby quantifying the influence of the wing beam design variables on the nonlinear effect.

[0052] (3) This method obtains the frequency domain characteristics and power spectrum of geometric defects by performing Fourier transform on the geometric defect distribution of a limited sample, and obtains a large-scale geometric defect sample by inverse transforming the power spectrum, which effectively reduces the difficulty and cost of generating defect samples.

[0053] (4) This method establishes a sub-neural network model that characterizes the intrinsic relationship between the parameters of the bamboo-like structure, the parameters of the sandwich structure, the ultimate strain of the defective structure, and the buckling boundary of the defective structure through batch analysis of random samples, thereby quantifying the influence of the wing beam design variables on the defect reduction effect.

[0054] (5) This method integrates the sub-neural network model representing linear load-bearing performance, the sub-neural network model representing nonlinear effects, and the sub-neural network model representing defect reduction effects through modular neural network design, and realizes the collaborative design of the three elements.

[0055] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to the present invention.

[0057] Figure 2This invention provides a finite element analysis model of a bamboo-joint sandwich wing beam based on the modular neural network-driven collaborative design method for the three elements of a bamboo-joint sandwich wing beam.

[0058] Figure 3 This is a flowchart illustrating the generation of large-scale wing beam structure samples with geometric defects using the modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to the present invention.

[0059] Figure 4 This is a schematic diagram of the linear load-bearing performance sub-neural network of the modular neural network-driven collaborative design method for the three elements of the bamboo-like sandwich wing beam of the present invention.

[0060] Figure 5 This is a schematic diagram of the nonlinear effect sub-neural network of the modular neural network driven collaborative design method for the three elements of the bamboo-like sandwich wing beam of the present invention.

[0061] Figure 6 This is a schematic diagram of the defect reduction effect sub-neural network of the modular neural network driven collaborative design method for the three elements of the bamboo-like sandwich wing beam of the present invention. Detailed Implementation

[0062] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0063] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0064] Example

[0065] Please see Figures 1-6 This invention provides a modular neural network-driven collaborative design method for the three elements of a bamboo-inspired sandwich wing sparb. Its primary application is the novel bamboo-inspired sandwich wing sparb structure used in the high aspect ratio, low wing loading wings of solar-powered unmanned aerial vehicles (UAVs). The method includes the following steps:

[0066] Step S1: Construct a linear load-bearing capacity sub-neural network. This includes:

[0067] Step S11: Construct a sample library of wing beam structures containing various panel ply parameters, specifically the number and direction of ply layers. Calculate the linear stress matrix of the material for each sample using finite element analysis. S cr (like S 11 , S 12 (equal parameters) and linear buckling load of the structure M cr This forms a dataset.

[0068] Step S12, according to... Figure 4 The network architecture shown in the model receives panel ply parameters in the input layer, processes the correlation between panel ply parameters through multiple layers of neurons in the hidden layer, and outputs the linear strain of the material and the linear buckling load of the structure. These outputs can be used to determine the strength failure boundary.

[0069] Step S13: Through training and optimization of the dataset, the final linear load-bearing performance sub-neural network is obtained, which can output the corresponding linear strain of the material and the linear buckling load of the structure according to the input panel ply parameters, providing a basis for evaluating the linear load-bearing performance of the wing beam.

[0070] Step S2: Construct a nonlinear effect sub-neural network. This includes:

[0071] Step S21: Fix the panel layup parameters to eliminate their interference with nonlinear effects, and design a sample library of wing beam structures containing different sandwich structure parameters (equivalent modulus, thickness) and bamboo-like structure parameters (spacing, thickness).

[0072] Step S22: Calculate the linear stress using the finite element analysis method. S cr Linear buckling load M cr Nonlinear stress S k Nonlinear buckling load M k This yields the nonlinear stability factor, including the material stress nonlinear stability factor. ε cr (Calculated from nonlinear stress and linear stress) and nonlinear stability factor of buckling load η cr (Calculated from nonlinear buckling load and linear buckling load), forming a dataset.

[0073] Step S23, according to... Figure 5The network architecture shown constructs a network where the input layer receives parameters such as core modulus, core thickness, bamboo-like joint spacing, and bamboo-like joint thickness. The hidden layer processes the correlation between these parameters through multiple layers of neurons, and the output layer outputs the material stress nonlinear stability factor. ε cr and buckling load nonlinear stability factor η cr This study aims to quantify the influence of bamboo-like structures and sandwich structures on the nonlinear effects of the wing beam.

[0074] Step S24: Perform training optimization so that the nonlinear effect sub-neural network outputs the corresponding nonlinear stability factor based on the input sandwich structure parameters and bamboo-like structure parameters, providing support for the evaluation of nonlinear effects.

[0075] Step S3: Construct the defect reduction effect sub-neural network. This includes:

[0076] Step S31: Generate a large-scale sample of a wing beam structure with geometric defects through the defect random sample generation module, fix the panel ply parameters, adjust the sandwich structure parameters (equivalent modulus, thickness) and the bamboo-like structure parameters (spacing, thickness), and use the finite element analysis method to obtain the ultimate strain and buckling failure boundary of the defective structure (which can be characterized by parameters such as eigenvalue λ), and compare them with the corresponding parameters of the defect-free structure to calculate the defect reduction factor.

[0077] A large-scale sample of spar structure containing geometric defects is generated using a defect random sample generation module to reduce the difficulty and cost of sample generation. This includes the following steps:

[0078] Step S311, 3D laser scanning: Use 3D laser scanning technology to obtain the geometry of the interlayer and panel, and compare it with the ideal shape to obtain the distribution of geometric defects in the thin-walled tube beam;

[0079] Step S312, Original Surface Reconstruction: Perform Fourier transform on the geometric defect distribution to obtain the frequency domain characteristics and power spectrum of the geometric defects;

[0080] Step S313, Original Defect Power Spectrum: Generate large-scale geometric defect samples by performing inverse transformation on the power spectrum, and randomly generate sandwich beam samples with the same defect pattern by randomly assigning phase;

[0081] Step S314: Randomly generate defect samples: In the evolutionary power spectrum, the spatial distribution of the power spectrum amplitude is further defined, and geometric initial defects with uneven spatial distribution are generated based on the evolutionary power spectrum to study the influence of the spatial distribution characteristics of geometric defects on post-buckling behavior.

[0082] Step S315, Adjusting the Defect Location: The initial bonding defect characteristics and their distribution are detected by ultrasonic flaw detection technology. The geometric and distribution characteristics of the bonding defects are quantitatively described by statistical methods, providing a basis for the definition and generation of bonding defects.

[0083] Step S32, according to... Figure 6 The network architecture shown is used to build a network. The input layer receives parameters such as core modulus, core thickness, bamboo-like joint spacing, and bamboo-like joint thickness. The hidden layer processes the correlation between the parameters and the defect reduction effect through multiple layers of neurons. The output layer outputs the defect reduction factor, thereby quantifying the influence of bamboo-like joint structure and core structure on the defect sensitivity of the spar.

[0084] Step S33: Perform training and optimization so that the defect reduction effect sub-neural network outputs the corresponding defect reduction factor based on the input sandwich structure parameters and bamboo-like structure parameters, providing a basis for evaluating the defect reduction effect.

[0085] The finite element analysis methods for the beam ends in steps S11, S22, and S31 include:

[0086] A finite element model of a bamboo-joint sandwich beam was created in Abaqus. The faceplate and core were modeled using S4R continuous shell elements, while the bamboo joints were modeled using Timoshenko beam elements. Figure 2 As shown, the fixed constraint of the reference point includes: Axial displacement , Axial displacement , Axial displacement , around Axial angular displacement , around Axial angular displacement , around Axial angular displacement A fixed boundary condition is applied to one end of the beam, and a multi-point constraint method is used at the other end to establish a beam constraint (MPC-Beam) between the boundary and a reference point located at the center. Bending loads are then applied to this reference point. Linear stresses are calculated using the "Static, General" analysis step. S cr The linear buckling load is calculated using the "Buckling" analysis step. M cr Subsequently, the "Dynamics, Explicit" analysis step was used to replace the load at the reference point with the velocity boundary condition. By monitoring the support reaction force at the reference point, the load-displacement curve under bending conditions was obtained, and the load corresponding to the point of sharp load reduction was used as the starting point. M kAs a nonlinear buckling load, the stress output at the intermediate step is used as the nonlinear stress. S k .

[0087] Step S4: Using the physical rule fusion module, the linear bearing capacity sub-neural network, the nonlinear effect sub-neural network, and the defect reduction effect sub-neural network are fused according to the composite material structure design rules to obtain a modular neural network, which is used to predict the actual failure boundary of the bamboo-joint sandwich wing beam, thereby realizing the collaborative design of the three elements, including linear bearing characteristics, nonlinear effects, and defect reduction effects.

[0088] Step S5: Input the input parameters into the modular neural network, and obtain the output parameters through the modular neural network. For the novel bamboo-joint sandwich beam, the design parameters include three categories: panel ply parameters, bamboo-joint structural parameters, and core layer structural parameters. Specifically, they include:

[0089] The panel layup is input into the linear load-bearing capacity sub-neural network, and the linear strain of the material and the linear buckling load of the structure are output.

[0090] Input bamboo-like structural parameters and core layer structural parameters into the nonlinear effect sub-neural network, and output a nonlinear stability factor.

[0091] Input the bamboo-like structure parameters and the core layer structure parameters into the defect reduction effect sub-neural network, and output the defect reduction factor.

[0092] Among them, the panel ply parameters, including the number of ply layers and the ply direction, mainly determine the linear load-bearing capacity of the wing beam. The bamboo-like structural parameters include the spacing and thickness, and the sandwich layer structural parameters include the spacing and thickness. Neither the bamboo-like structural parameters nor the sandwich layer structural parameters provide much in-plane load-bearing capacity to the wing beam panel; therefore, they are considered to only affect the nonlinear effects and defect reduction effects of the wing beam.

[0093] The strength failure boundary and buckling failure boundary are determined by the linear strain of the material, the linear buckling load of the structure, the nonlinear stability factor, and the defect reduction factor.

[0094] Step S6: Iterate through step S5 using the wing sparb parameter iterative design module to complete the wing sparb optimization design.

[0095] Therefore, the present invention adopts the above-mentioned modular neural network-driven three-element collaborative design method for bamboo-like sandwich wing spars. The bamboo-like sandwich wing spars designed using this method are generally applicable to wings with high aspect ratio and low wing load. Compared with existing design methods, the structural failure boundary prediction accuracy is higher, and it has the advantages of high accuracy, high efficiency and good robustness.

[0096] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A modular neural network-driven collaborative design method for three elements of a bamboo-like sandwich wing beam, characterized in that, Includes the following steps: Step S1: Construct a linear load-bearing capacity sub-neural network; Step S2: Construct a nonlinear effect sub-neural network; Step S3: Construct a defect reduction effect sub-neural network; Step S4: Fuse the linear load-bearing capacity sub-neural network, the nonlinear effect sub-neural network, and the defect reduction effect sub-neural network to obtain a modular neural network; Step S5: Input the input parameters into the modular neural network and obtain the output parameters through the modular neural network; Step S6: Iterate through step S5 to complete the wing sparsity optimization design; Step S1 includes: Step S11: Construct a sample library containing various panel ply parameters, and obtain the linear stress matrix of the material and the linear buckling load of the structure through finite element analysis to form a dataset; Step S12: Build a model based on the network architecture. The input layer receives panel ply parameters, the hidden layer processes the relationship between panel ply parameters through multiple neurons, and the output layer outputs the linear strain of the material and the linear buckling load of the structure. Step S13: Through training and optimization of the dataset, the linear bearing capacity sub-neural network outputs the corresponding linear strain of the material and the linear buckling load of the structure according to the input panel ply parameters. Step S2 includes: Step S21: Fix the panel layup parameters and design a sample library containing different sandwich structure parameters and bamboo-like structure parameters; Step S22: Using the finite element analysis method, obtain the nonlinear stability factor and form a dataset; Step S23: Build a network according to the network architecture. The input layer receives the parameters of the sandwich structure and the parameters of the bamboo-like structure. The hidden layer processes the correlation between the parameters through multiple neurons. The output layer outputs the nonlinear stability factor of material stress and the nonlinear stability factor of buckling load. Step S24: Perform training optimization so that the nonlinear effect sub-neural network outputs the corresponding nonlinear stability factor based on the input sandwich structure parameters and bamboo-like structure parameters. Step S3 includes: Step S31: Generate a large-scale sample of a wing beam structure with geometric defects, fix the panel ply parameters, adjust the sandwich structure parameters and the bamboo-like structure parameters, and use the finite element analysis method to obtain the defect reduction factor. Step S32: Build a network according to the network architecture. The input layer receives the parameters of the sandwich structure and the parameters of the bamboo-like structure. The hidden layer processes the correlation between the parameters and the defect reduction effect through multiple layers of neurons. The output layer outputs the defect reduction factor accordingly. Step S33: Perform training and optimization so that the defect reduction effect sub-neural network outputs the corresponding defect reduction factor based on the input sandwich structure parameters and bamboo-like structure parameters.

2. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 1, characterized in that, Step S31, which generates a large-scale sample of spar structure containing geometric defects using a defect random sample generation module, includes: Step S311: Use three-dimensional laser scanning technology to obtain the geometric shape of the sandwich and panel, and compare it with the ideal shape to obtain the distribution of geometric defects in the thin-walled tube beam; Step S312: Perform a Fourier transform on the geometric defect distribution to obtain the frequency domain characteristics and power spectrum of the geometric defects; Step S313: Generate large-scale geometric defect samples by performing inverse transformation on the power spectrum, and randomly generate sandwich beam samples with the same defect pattern by randomly assigning phase. Step S314: In the evolutionary power spectrum, the spatial distribution of the power spectrum amplitude is further defined, and a spatially uneven geometric initial defect is generated based on the evolutionary power spectrum. The influence of the spatial distribution characteristics of the geometric defect on the post-buckling behavior is studied. Step S315: Detect the initial bonding defect characteristics and their distribution using ultrasonic flaw detection technology, and quantitatively describe the geometric and distribution characteristics of the bonding defects using statistical methods, providing a basis for the definition and generation of bonding defects.

3. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 2, characterized in that, The finite element analysis method includes: A finite element model of a bamboo-joint sandwich beam was created in Abaqus. The panel and the sandwich core were modeled using S4R continuous shell elements, and the bamboo joints were modeled using Timoshenko beam elements. A fixed boundary condition is applied to one end of the beam, and a multi-point constraint method is used to establish beam constraint between the boundary and a reference point located at the center at the other end, and a bending load is applied to that reference point. The analysis step calculates linear stress, and the analysis step calculates linear buckling load; In the analysis step, the velocity boundary condition is replaced with the load at the reference point. The load-displacement curve under bending condition is obtained by monitoring the support reaction force at the reference point. The load corresponding to the point where the load decreases sharply is taken as the nonlinear buckling load, and the stress output in the intermediate step is taken as the nonlinear stress.

4. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 3, characterized in that: In step S4, the linear bearing capacity sub-neural network, the nonlinear effect sub-neural network, and the defect reduction effect sub-neural network are fused using the physical rule fusion module.

5. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 4, characterized in that, Step S5 includes: Input panel layup into the linear load-bearing capacity sub-neural network, and output linear strain of material and linear buckling load of structure; Input bamboo-like structure parameters and sandwich layer structure parameters into the nonlinear effect sub-neural network, and output a nonlinear stability factor; Input bamboo-like structure parameters and sandwich layer structure parameters into the defect reduction effect sub-neural network, and output the defect reduction factor; Among them, the panel ply parameters include the number of ply layers and the ply direction, the bamboo-like structure parameters include the spacing and thickness, and the sandwich layer structure parameters include the spacing and thickness.

6. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 5, characterized in that: The strength failure boundary and buckling failure boundary are determined by the linear strain of the material, the linear buckling load of the structure, the nonlinear stability factor, and the defect reduction factor.

7. The modular neural network-driven collaborative design method for the three elements of a bamboo-like sandwich wing beam according to claim 6, characterized in that: In step S6, the wing spars parameter iterative design module is used to iterate and complete the wing spars optimization design.

Images