A neural network-based wing tool flag thickness prediction method

Abstract

The application discloses a wing tool flag thickness prediction method based on a neural network, which comprises the following steps: step one, a three-dimensional geometric model of a wing tool flag is established, and a plurality of simplified models with different thicknesses and corresponding neutral format models are obtained through simplification processing and parameter setting of the three-dimensional geometric model; step two, finite element calculation is performed on the neutral format models with different thicknesses, and statics solving is performed by applying concentrated loads to the neutral format models with different thicknesses, so that displacement results and stress results of the wing tool flag under the concentrated loads are calculated under different thicknesses; step three, a neural network learning rate optimization model is established according to the calculated displacement results and stress results in step two, and the output wing tool flag thickness is obtained through model training; and step four, the tool flag thickness prediction value under a specified displacement is obtained through the neural network learning rate optimization model obtained after training. In the technical scheme, the wing tool flag thickness is quickly predicted by considering factors such as material properties of the wing tool flag, rib height of the wing tool flag, rib width of the wing tool flag, displacement, load and weight, so that the tool flag design cycle is shortened, the tool flag design reliability is ensured, and the engineering design cost is reduced, and the wing tool flag thickness prediction method is fast and effective.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tooling structure design, and particularly relates to a wing tooling flag thickness prediction method based on a neural network.

[0002] Specifically, the present application relates to a method for quickly predicting the thickness of a wing tooling flag through an artificial neural network to perform optimization design. BACKGROUND

[0003] During wing assembly, insufficient tooling flag thickness on the assembly tooling may cause wing deformation out of tolerance and reduce assembly accuracy. In order to improve assembly quality, the thickness of the tooling flag needs to be optimized and designed.

[0004] Traditional tooling flag thickness design is determined through artificial experience, lacks analysis of the relationship between tooling flag thickness and wing deformation, and has the problem of inaccurate design value. Through finite element calculation of the thickness of the wing tooling flag, a long time cost is required, and the operation is difficult, and a large amount of iterative calculation is required for different design schemes. SUMMARY

[0005] The present application aims to solve the above technical problems, and provides a wing tooling flag thickness prediction method based on a neural network to solve the problem of inaccurate design value obtained through artificial experience design in the existing design of wing tooling flag thickness, and the problems of long time cost, difficult operation, and a large amount of iterative calculation in the finite element calculation method.

[0006] The technical scheme of the present application is as follows: in a first aspect, the present application provides a wing tooling flag thickness prediction method based on a neural network, comprising:

[0007] Step 1: a three-dimensional geometric model of the wing tooling flag is established, and through simplification processing and parameter setting of the three-dimensional geometric model, a plurality of simplified models with different thicknesses and corresponding neutral format models are obtained;

[0008] Step 2: finite element calculation is performed on the neutral format models with different thicknesses, statics solving is performed through the way of applying concentrated load to the neutral format models with different thicknesses, so as to calculate the displacement result and stress result of the wing tooling flag under the concentrated load under different thicknesses;

[0009] Step 3: a neural network learning rate optimization model is established according to the calculated displacement result and stress result in step 2, and the output wing tooling flag thickness is obtained through model training;

[0010] Step 4: the tooling flag thickness prediction value under the specified displacement is obtained through the neural network learning rate optimization model obtained after training.

[0011] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the step one comprises:

[0012] Step 1-1, a three-dimensional geometric model of the wing tool flag is established by using CAD software, and a simplified model of the wing tool flag is obtained by deleting all parts in the wing tool flag model except the bolts and the gaskets in the three-dimensional geometric model;

[0013] Step 1-2, the thickness of the tool flag end plate in the simplified model of the wing tool flag is taken as a variable parameter to establish a plurality of simplified models with different thicknesses of the tool flag end plate;

[0014] Step 1-3, the plurality of simplified models with different thicknesses of the tool flag end plate are saved as corresponding neutral format models.

[0015] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the simplified model of the wing tool flag obtained in the step 1-1 has the following requirements:

[0016] The simplified model has the same length, width and height as the actual wing tool flag, and the positions of the tool flag end plate, the tool flag rib, the tool flag front end face and the bolt hole in the simplified model are consistent with those of the actual wing tool flag.

[0017] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the step two comprises:

[0018] Step 2-1, each neutral format model is imported into a finite element calculation software, and the Young's modulus, the Poisson's ratio and the material density of the wing tool flag are set according to the material type of the wing tool flag;

[0019] Step 2-2, the imported neutral format model is divided into grid elements, and the maximum size of the length, width and height of the grid elements in the neutral format model does not exceed 10% of the length, width and height of the actual wing tool flag;

[0020] Step 2-3, a concentrated load is applied to the front end face of the tool flag in the neutral format model, and the displacement result and the stress result of the wing tool flag represented by the neutral format model under the concentrated load are calculated by solving the statics.

[0021] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the step three comprises:

[0022] Step 3-1, each grid element in the neutral format model is numbered, and the displacement components corresponding to the number are stored in the following displacement matrix [U],

[0023]

[0024] Wherein, the first column is the number of 1 to n grid cells in the neutral format model; the second to fourth columns are the displacement components corresponding to the grid cell number; wherein, u i,1 , u i,2 , u i,3 respectively represent the displacement components of the i-th grid cell in the heading direction, the spanwise direction and the gravity direction;

[0025] Step 3-2, construct a neural network learning rate optimization model, and set the neural network learning rate of the model as:

[0026]

[0027] Wherein, δ is the learning rate in the neural network, δ max , δ min are the maximum learning rate and the minimum learning rate respectively; n is the iteration number, ξ is the coefficient, used to adjust the influence of the iteration number on the learning rate; the neural network learning rate optimization model includes three layers: input layer, hidden layer and output layer;

[0028] Step 3-3, determine the number of neuron nodes N n of the hidden layer, the input layer is the displacement matrix [U] obtained in step 3-1, and the output layer is the wing tool flag thickness, the neural network is trained to obtain the output wing tool flag thickness.

[0029] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the number of neuron nodes N n of the hidden layer in step 3-1 is:

[0030]

[0031] Wherein, N in is the number of input layer nodes in the neural network; N out is the number of output layer nodes in the neural network, and C is the coefficient.

[0032] Optionally, in the neural network-based wing tool flag thickness prediction method described above, step four includes:

[0033] In the neural network learning rate optimization model obtained after training, the load size, tool flag rib height, tool flag rib width, material properties, and wing tool flag displacement, weight of the wing tool flag to be designed are input as input parameters, and the output wing tool flag thickness is obtained, that is, the thickness prediction value.

[0034] Optionally, in the neural network-based wing tool flag thickness prediction method described above, it further includes:

[0035] Step five, using the predicted tool flag thickness prediction value, designing the wing tool flag, and checking the strength of the designed wing tool flag.

[0036] Optionally, in the neural network-based wing tool flag thickness prediction method described above, the step five comprises:

[0037] Using the predicted tool flag thickness prediction value, carrying out tool flag design, and importing the designed wing tool flag model into the finite element software for strength checking to obtain an equivalent stress value through calculation;

[0038] If the equivalent stress value does not exceed the material strength design value, the final tool flag thickness result is obtained; if the equivalent stress value exceeds the strength design value, the tool flag thickness is increased by 1%; and steps four and five are repeated until the calculated equivalent stress value meets the strength design requirement.

[0039] In a second aspect, the embodiments of the present application also provide a computer-readable storage medium, comprising: a memory and a processor;

[0040] The memory is used to store a computer-readable program.

[0041] The processor is used to realize the neural network-based wing tool flag thickness prediction method according to any one of the above when executing the computer-readable program.

[0042] The embodiments of the present application provide a neural network-based wing tool flag thickness prediction method, which considers factors such as wing tool flag material properties, tool flag rib height, tool flag rib width, displacement, load, weight, etc., establishes a neural network model according to the wing tool flag finite element calculation result, determines the tool flag thickness under the specified displacement through the established neural network model, thereby realizing the prediction of the tool flag thickness before formal design, selecting the best thickness scheme in the design, avoiding the repeated iteration process of design and stiffness calculation, shortening the tool flag design cycle, ensuring the reliability of the tool flag design, and reducing the engineering design cost. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings are used to provide a further understanding of the technical solutions of the present application, and constitute a part of the specification, and are used to explain the technical solutions of the present application together with the embodiments of the present application, and do not constitute a limitation to the technical solutions of the present application.

[0044] Figure 1 A flowchart of a neural network-based wing tool flag thickness prediction method is provided for the present application.

[0045] Figure 2 A schematic diagram of a geometric model of a wing tool flag in the wing tool flag thickness prediction method based on a neural network provided by an embodiment of the present application;

[0046] Figure 3 A schematic diagram of a finite element mesh element model in the wing tool flag thickness prediction method based on a neural network provided by an embodiment of the present application;

[0047] Figure 4 A flowchart of neural network model calculation in the wing tool flag thickness prediction method based on a neural network provided by an embodiment of the present application;

[0048] Figure 5 A schematic diagram of the prediction accuracy of a neural network model in the wing tool flag thickness prediction method based on a neural network provided by an embodiment of the present application.

[0049] BRIEF DESCRIPTION OF DRAWINGS

[0050] 1, wing tool flag; 2, tool flag end plate; 3, tool flag rib; 4, tool flag front end face; 5, bolt hole; 6, mesh element. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the embodiments of the present application will be described in detail below with reference to the drawings. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0052] As explained in the above background, the important role of the wing tool flag in the wing assembly and the necessity of designing the wing tool flag thickness. However, the existing design method for the wing tool flag thickness, the inaccurate design value obtained by the manual experience design method, and the time cost, the difficulty of operation, and the need for a large number of iterative calculations of the finite element calculation method.

[0053] Based on the above problems, it is urgent to provide a method for quickly obtaining the wing tool flag thickness prediction. With the development of artificial intelligence, neural network technology has become a widely used model prediction technology. In the present application, neural network is applied to tool flag thickness prediction, a neural network model considering factors such as wing tool flag material properties, tool flag rib height, tool flag rib width, displacement, load, weight, etc. is established to quickly predict the tool flag thickness. The tool flag thickness is predicted before formal design, the best thickness scheme is selected in the design, the process of repeated iteration of design and stiffness calculation is avoided, the tool flag design cycle is shortened, the tool flag design reliability is ensured, and the engineering design cost is reduced. It is a fast and effective wing tool flag thickness prediction method.

[0054] The following specific embodiments of the present application can be combined with each other, and some embodiments may not be described again for the same or similar concepts or processes.

[0055] Figure 1 A flowchart of a wing tool flag thickness prediction method based on a neural network is provided for the present application. The wing tool flag thickness prediction method based on a neural network provided by the embodiment of the present application comprises the following steps:

[0056] Step one, a three-dimensional geometric model of the wing tool flag is established, and through the simplification processing and parameter setting of the three-dimensional geometric model, a plurality of simplified models with different thicknesses and corresponding neutral format models are obtained;

[0057] Step two, finite element calculation is performed on the neutral format models with different thicknesses, and statics solving is performed by applying concentrated loads in the neutral format models with different thicknesses to calculate the displacement results and stress results of the wing tool flag under concentrated loads at different thicknesses;

[0058] Step three, a neural network learning rate optimization model is established according to the displacement results and stress results calculated in step two, and the output wing tool flag thickness is obtained through model training;

[0059] Step four, the tool flag thickness prediction value under the specified displacement is obtained through the neural network learning rate optimization model obtained after training.

[0060] In one implementation manner of the embodiment of the present application, the implementation process of the above step one can include:

[0061] Step 1-1, a three-dimensional geometric model of the wing tool flag is established by using CAD software, and a simplified model of the wing tool flag is obtained by deleting all parts in the wing tool flag model except the bolts and gaskets in the three-dimensional geometric model;

[0062] Step 1-2, the tool flag end plate thickness in the simplified model of the wing tool flag is taken as a variable parameter to establish a plurality of simplified models with different tool flag end plate thicknesses;

[0063] Step 1-3, the plurality of simplified models with different tool flag end plate thicknesses are saved as corresponding neutral format models.

[0064] In the specific implementation of this implementation manner, the simplified model of the wing tool flag obtained in the above step 1-1 has the following requirements:

[0065] The simplified model has the same length, width and height dimensions as the actual wing tool flag, and the tool flag end plate, tool flag rib, tool flag front end face and bolt hole position in the simplified model are consistent with the actual wing tool flag.

[0066] In an implementation form of the embodiment of the application, the implementation process of step two can include:

[0067] Step 2-1, import each neutral format model into the finite element calculation software respectively, set the Young's modulus, Poisson's ratio and material density of the wing tool flag according to the material type of the wing tool flag;

[0068] Step 2-2, divide the grid elements of the imported neutral format model, the maximum size of the length, width and height of the grid elements in the neutral format model does not exceed 10% of the actual length, width and height of the wing tool flag;

[0069] Step 2-3, apply a concentrated load on the front end surface of the tool flag of the neutral format model, and solve it by statics solving method to calculate the displacement result and stress result of the wing tool flag represented by the neutral format model under the concentrated load.

[0070] In an implementation form of the embodiment of the application, the implementation process of step three can include:

[0071] Step 3-1, number each grid element in the neutral format model, and store the displacement components corresponding to the number in the displacement matrix [U] as follows,

[0072]

[0073] Wherein, the first column is the number of 1 to n grid elements in the neutral format model; the second to fourth columns are the displacement components corresponding to the grid element number; wherein, u i,1 , u i,2 , u i,3 respectively represent the heading, spanwise and gravity direction displacement components of the i-th grid element;

[0074] Step 3-2, construct a neural network learning rate optimization model, and set the neural network learning rate of the model as:

[0075]

[0076] Wherein, δ is the learning rate in the neural network, δ max , δ min are the maximum learning rate and the minimum learning rate respectively; n is the iteration number, ξ is the coefficient for adjusting the influence of iteration number on learning rate; the neural network learning rate optimization model includes three layers: input layer, hidden layer and output layer;

[0077] Step 3-3, determine the number of neuron nodes N nThe input layer is the displacement matrix [U] obtained in step 3-1, and the output layer is the wing tool flag thickness, and the neural network is trained to obtain the output wing tool flag thickness.

[0078] In the implementation, the number of neuron nodes N n of the hidden layer in step 3-1 is

[0079]

[0080] wherein N in is the number of input layer nodes in the neural network; N out is the number of output layer nodes in the neural network, and C is a coefficient.

[0081] In an implementation of the embodiment of the present application, the implementation of step four is:

[0082] In the neural network learning rate optimization model obtained after training, the load size of the wing tool flag to be designed, the tool flag rib height, the tool flag rib width, the material properties, and the displacement and weight of the wing tool flag are input parameters, and the output wing tool flag thickness is a thickness prediction value.

[0083] Further, the neural network-based wing tool flag thickness prediction method provided in the embodiment of the present application can further include:

[0084] Step five, using the predicted tool flag thickness prediction value to design the wing tool flag, and checking the strength of the designed wing tool flag.

[0085] The specific implementation of step five includes:

[0086] Using the predicted tool flag thickness prediction value to carry out tool flag design, and importing the designed wing tool flag model into the finite element software to check the strength, and calculating to obtain an equivalent stress value;

[0087] If the equivalent stress value does not exceed the material strength design value, the final tool flag thickness result is obtained; if the equivalent stress value exceeds the strength design value, the tool flag thickness is increased by 1%; and steps four and five are repeated until the calculated equivalent stress value meets the strength design requirement

[0088] The wing tool flag thickness prediction method based on the neural network provided by the embodiment of the present application considers factors such as material properties of the wing tool flag, rib height of the tool flag, rib width of the tool flag, displacement, load, weight and the like, establishes a neural network model according to the finite element calculation result of the wing tool flag, determines the tool flag thickness under the specified displacement through the established neural network model, thereby achieving the prediction of the tool flag thickness before formal design, selecting the optimal thickness scheme in the design, avoiding the repeated iteration process of design and stiffness calculation, shortening the tool flag design cycle, ensuring the reliability of the tool flag design, and reducing the engineering design cost; that is, the wing tool flag thickness prediction method provided by the present application can provide a theoretical basis for optimization design, and is a fast and effective wing tool flag thickness prediction method.

[0089] Based on the wing tool flag thickness prediction method based on the neural network provided in the above embodiment of the present application, the embodiment of the present application further provides a computer readable storage medium, characterized by comprising a memory and a processor.

[0090] The memory is configured to store a computer readable program.

[0091] The processor is configured to implement the wing tool flag thickness prediction method based on the neural network provided in any of the above embodiments when executing the computer readable program.

[0092] The implementation of the wing tool flag thickness prediction method based on the neural network provided by the embodiment of the present application is schematically described below through an implementation example.

[0093] Implementation Example

[0094] The wing tool flag thickness prediction method based on the neural network provided by the embodiment of the present application is described with reference to the flowchart shown in Figure 1 The method comprises the following steps:

[0095] Step 1: a three-dimensional geometric model of the wing tool flag is established, and through the simplification processing and parameter setting of the three-dimensional geometric model, a plurality of simplified models with different thicknesses and corresponding neutral format models are obtained;

[0096] Step 2: finite element calculation is performed on the neutral format models with different thicknesses, and statics solving is performed by applying concentrated load to the neutral format models with different thicknesses, so as to calculate the displacement result and stress result of the wing tool flag under the concentrated load under different thicknesses;

[0097] Step 3: a neural network learning rate optimization model is established according to the displacement result and stress result calculated in step 2, and the output wing tool flag thickness is obtained through model training;

[0098] Step four, through the neural network learning rate optimization model obtained after training, the jig flag thickness prediction value under the specified displacement is obtained.

[0099] In this embodiment example, the specific process of establishing the plurality of simplified models with different thicknesses and the corresponding neutral format models in step one is as follows:

[0100] Step 1-1, a three-dimensional geometric model of the wing jig flag is established by using CATIA V5 2018 software, wherein all parts except the bolts and the gaskets in the wing jig flag model are deleted as a simplified model.

[0101] It should be noted that in this step, it is required to ensure that the simplified model has the same length, width and height as the original wing jig flag, and the jig flag end plate 2, the jig flag rib 3, the jig flag front end face 4 and the bolt hole position 5 in the simplified model are consistent with the wing jig flag. Figure 2 As shown in the schematic diagram of the geometric model of the wing jig flag in the neural network-based wing jig flag thickness prediction method provided by the embodiment of the present application.

[0102] Step 1-2, the thickness of the jig flag end plate is taken as a variable parameter, and the input parameter is 30mm to 50mm with an interval of 5mm to establish a plurality of simplified models with different jig flag end plate thicknesses.

[0103] Step 1-3, the plurality of simplified models with different jig flag end plate thicknesses are saved as corresponding neutral format models, and in this embodiment example, the models are all selected as.STP format files.

[0104] In this embodiment example, the implementation of the finite element calculation on the neutral format models with different thicknesses in step two includes the following steps:

[0105] Step 2-1, each neutral format model is respectively imported into the finite element calculation software, and the Young's modulus and Poisson's ratio of the wing jig flag are set as 210GPa and 0.30 respectively, and the material density is taken as 7.8x10 -9 t / mm 3 .

[0106] Step 2-2, the imported neutral format model is meshed, and the maximum size of the grid element in the neutral format model does not exceed 10% of the length, width and height of the actual wing jig flag, as Figure 3 As shown in the schematic diagram of the grid element model of the finite element in the neural network-based wing jig flag thickness prediction method provided by the embodiment of the present application.

[0107] Step 2-3, a concentrated load is applied on the front end face of the tooling flag of the neutral format model, the load size is 150N, and the displacement result and the stress result of the wing tooling flag under the concentrated load are solved by a statics solving method.

[0108] In the implementation example, the implementation manner of the neural network learning rate optimization model according to the calculated displacement result and stress result in step three is as follows:

[0109] Step 3-1, each grid element in the neutral format model is numbered, and the displacement components corresponding to the numbers are stored in the following displacement matrix [U],

[0110]

[0111] wherein the first column is the number of 1 to n grid elements in the neutral format model; the second to fourth columns are the displacement components corresponding to the grid element numbers; wherein, u i,1 , u i,2 , and u i,3 respectively represent the heading, spanwise, and gravity direction displacement components of the i-th grid element.

[0112] Step 3-1, a neural network learning rate optimization model is constructed, and the neural network learning rate of the model is set as:

[0113]

[0114] wherein, δ is the learning rate in the neural network, δ max , and δ min are the maximum learning rate and the minimum learning rate, respectively, and are 0.1 and 0.005, respectively; n is the iteration number, and is 10000; ξ is a coefficient, and is used to adjust the influence of the iteration number on the learning rate, and is 0.2 to 0.5. The learning rate optimization model includes three layers: an input layer, a hidden layer, and an output layer, as shown in Figure 4 Fig. 1 is a flowchart of neural network model calculation in a wing tooling flag thickness prediction method based on a neural network provided by the embodiment of the application; the input layer is the matrix [U] obtained in step 3-1, and the output layer is the tooling flag thickness.

[0115] Step 3-3, the number N n of neuron nodes of the hidden layer is determined. n The input layer is the displacement matrix [U] obtained in step 3-1, the output layer is the wing tooling flag thickness, the neural network is trained, the output wing tooling flag thickness is obtained, the neural network training is started, and the number N n of neuron nodes of the hidden layer is:

[0116]

[0117] wherein, Nin N is the number of input layer nodes in the neural network, and N in = 6; N out is the number of output layer nodes in the neural network, N out = 1, C is an adjustment coefficient, and C = 9. Figure 5 The training result is shown in FIG. 2, which is a schematic diagram of the prediction accuracy of the neural network model in the wing tool flag thickness prediction method based on the neural network provided by the embodiment of the present application. It can be seen from FIG. 2 that the absolute error of the training is within ±0.01 mm, and the relative error is controlled within ±10%, that is, the model can effectively predict the tool flag thickness. Figure 5

[0118] In the implementation example, the implementation manner of obtaining the tool flag thickness prediction value under the specified displacement by the neural network learning rate optimization model obtained after the training of step four is as follows:

[0119] Step 4-1, input the structure parameters required to be designed for the tool flag, and perform thickness prediction as the input layer. For the steel tool flag, the Young's modulus and Poisson's ratio are 210 GPa and 0.3 respectively, the material density is 7.8×10 -9 t / mm 3 , the rib width and height are 10 mm and 16 mm respectively, the load size is 150 N, the maximum deformation amount is required to be 0.2 mm, and the weight is required to be less than 100 kg. Therefore, the input condition is [0.065 10 16 150 210 0.3 7.8×10 -9 ], and the output thickness is 35.2 mm. It is indicated that the thickness greater than 35.2 mm can meet the requirements, and therefore the thickness is taken as 40 mm.

[0120] Further, the method provided by the implementation example further includes the following steps:

[0121] Step 5, strength checking is performed on the predicted tool flag thickness, the maximum stress obtained by strength calculation is 115 MPa, which does not exceed the design allowable value, and therefore the final tool flag thickness is obtained.

[0122] The technical solution provided by the present application considers the factors such as the material properties of the wing tool flag, the rib height of the tool flag, the rib width of the tool flag, the displacement, the load, and the weight, and quickly predicts the tool flag thickness, thereby shortening the tool flag design cycle, ensuring the reliability of the tool flag design, and reducing the engineering design cost. It is a fast and effective wing tool flag thickness prediction method.

[0123] ​Although the present application has been described with reference to the above embodiments, the contents are only the embodiments for facilitating the understanding of the present application, and are not intended to limit the present application. Any modification and change in the form and details of the embodiments can be made by any person skilled in the art without departing from the spirit and scope of the present application. The patent protection scope of the present application shall be subject to the scope defined by the appended claims.

Claims

1. A neural network-based wing tooling flag thickness prediction method, characterized by, The application relates to a method for predicting the thickness of a wing tooling flag. The method comprises the following steps: Step one, a three-dimensional geometric model of a wing tooling flag is established, and a plurality of simplified models with different thicknesses and corresponding neutral format models are obtained through simplification processing and parameter setting of the three-dimensional geometric model; Step two, finite element calculation is carried out on the neutral format models with different thicknesses, and the displacement results and stress results of the wing tooling flag under concentrated loads at different thicknesses are calculated through statics solving by applying concentrated loads to the neutral format models with different thicknesses; Step three, a neural network learning rate optimization model is established according to the displacement results and stress results calculated in step two, and the output wing tooling flag thickness is obtained through model training; Step four, the wing tooling flag thickness prediction value under a specified displacement is obtained through the neural network learning rate optimization model obtained after training. Step 3-1, number each grid cell in the neutral format model, and store the displacement component corresponding to the number in the displacement matrix as follows In this case, ； Wherein, the first column is the number of 1 to n grid cells in the neutral format model; the second to fourth columns are displacement components corresponding to the grid cell number; wherein, u i,1 、u i,2 、u i,3 respectively represent the displacement components of the first i grid cell in the heading, spanwise, and gravity directions. The step three comprises: ； wherein, is a learning rate in a neural network, , is a maximum learning rate and a minimum learning rate, respectively; n is a number of iterations, is a coefficient for adjusting an influence of the number of iterations on the learning rate; the neural network learning rate optimization model comprises three layers: an input layer, a hidden layer, and an output layer; Step 3-3, determine the number of neuron nodes N of the hidden layer n The input layer is the displacement matrix obtained in step 3-1 The output layer is the wing tooling flag thickness, and the neural network training is performed to obtain the output wing tooling flag thickness.

2. The neural network-based wing tooling flag thickness prediction method of claim 1, wherein, Step 3-2, a neural network learning rate optimization model is constructed, and the neural network learning rate of the model is set as: The step one comprises: Step 1-1, a three-dimensional geometric model of a wing tooling flag is established by using CAD software, and a simplified model of the wing tooling flag is obtained by deleting all parts except bolts and gaskets in the three-dimensional geometric model; Step 1-2, the thickness of the tooling flag end plate in the simplified model of the wing tooling flag is taken as a variable parameter to establish a plurality of simplified models with different tooling flag end plate thicknesses; 3. The neural network-based wing tooling flag thickness prediction method of claim 2, wherein, Step 1-3, the plurality of simplified models with different tooling flag end plate thicknesses are saved as corresponding neutral format models. The simplified model obtained in the step 1-1 has the following requirements:

4. The neural network-based wing tooling flag thickness prediction method of claim 2, wherein, The simplified model has the same length, width and height as the actual wing tooling flag, and the tooling flag end plate, tooling flag rib, tooling flag front end face and bolt hole position in the simplified model are consistent with the actual wing tooling flag. The step two comprises: Step 2-1, each neutral format model is imported into a finite element calculation software, and the Young's modulus, Poisson's ratio and material density of the wing tooling flag are set according to the material type of the wing tooling flag; Step 2-2, the imported neutral format model is subjected to mesh element division, and the maximum size of the mesh element length, width and height in the neutral format model does not exceed 10% of the length, width and height of the actual wing tooling flag; 5. The neural network-based wing tooling flag thickness prediction method of claim 1, wherein, The number of neuron nodes N of the implicit layer in step 3-1 n is: ； wherein, is the number of input layer nodes in the neural network; is the number of output layer nodes in the neural network, is a coefficient.

6. The neural network-based wing tooling flag thickness prediction method of claim 1, wherein, Step 2-3, a concentrated load is applied to the tooling flag front end face of the neutral format model, and the displacement results and stress results of the wing tooling flag represented by the neutral format model under the concentrated load are calculated through statics solving. The step four comprises:

7. The neural network-based wing tooling flag thickness prediction method of any one of claims 1-6, wherein, In the neural network learning rate optimization model obtained after training, the load size, tooling flag rib height, tooling flag rib width, material properties, displacement and weight of the wing tooling flag to be designed are input as input parameters, and the output wing tooling flag thickness is obtained, that is, the thickness prediction value. The application further comprises:

8. The neural network-based wing tooling flag thickness prediction method of claim 7, wherein, Step five, the designed wing tooling flag is designed by using the predicted thickness prediction value, and the strength of the designed wing tooling flag is checked. The step five comprises: The predicted tool flag thickness prediction value is used to carry out tool flag design, and the wing tool flag model obtained by the design is introduced into a finite element software to perform strength checking, and an equivalent stress value is calculated and obtained. If the equivalent stress value does not exceed the material strength design value, a final tool flag thickness result is obtained; if the equivalent stress value exceeds the strength design value, the tool flag thickness is increased by 1%; and steps four and five are repeated until the calculated equivalent stress value meets the strength design requirement.

9. A computer-readable storage medium, characterized in that, Comprise: A memory and a processor; Wherein the memory is configured to store a computer readable program; The processor is configured to implement the neural network-based wing tool flag thickness prediction method according to any one of claims 1-8 when executing the computer readable program.

Images