Stiffened wallboard laying optimization method and device

By combining neural networks and genetic algorithms, the mechanical agent model is used to optimize the laying order of composite materials, the problem of many laying design variables and low computational efficiency in aircraft structure design is solved, and efficient optimization of reinforced wall panels is achieved.

CN119089806AInactive Publication Date: 2024-12-06CHINA AIRPLANT STRENGTH RES INST

Patent Information

Application Number
CN202411593979.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-08
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In aircraft structural design, the laminated layer design variables of composite material laminates are numerous and discrete, resulting in large calculation amounts and low calculation efficiency, making it difficult to achieve efficient optimization of reinforced wall panels.

Method used

Combining artificial neural networks and genetic algorithms, structural strain is calculated through pre-trained mechanical agent models, and cross-mutation is used to optimize the laying order to improve design efficiency.

Benefits of technology

It improves the efficiency of optimization of laying sequence, reduces calculation costs, and realizes the best laying solution that meets multiple design indicators, which is suitable for aircraft structural design.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089806A_ABST
    Figure CN119089806A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of aircraft structure design, and particularly relates to a stiffened wallboard laying optimization method and device. The method comprises the following steps: S1, obtaining a first specified number of layering schemes; s2, encoding each layering scheme according to a corresponding relationship between a preset layering angle and a specified number to obtain an initial population with a first specified number of codes; s3, calculating the structural strain corresponding to each individual in the population based on a pre-trained mechanical agent model; and S4, selecting a second specified number of individuals which are ranked in the top after the structural strain is ranked from large to small, performing crossover variation of the genetic algorithm, forming a new population, returning to the previous step, and performing loop execution until iterative convergence to obtain an optimal layering scheme. According to the method, the mechanical agent model with relatively high training precision is embedded into the generalized fitness function of the genetic algorithm to guide the genetic algorithm to optimize the layering sequence, so that the optimization efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the technical field of aircraft structure design, and in particular relates to a method and device for optimizing the plywork of a reinforced wall panel. Background Art

[0002] Composite materials have become the most direct means to achieve lightweight aircraft structures due to their excellent mechanical properties. In order to obtain the minimum weight under the constraints of strength and stiffness, the design variables such as ply thickness, ply ratio and ply sequence must be optimized when designing the ply of composite laminates. The ply sequence of composite materials directly affects the strength, stiffness and stability of the structure. Unreasonable ply will weaken the structural bearing capacity, anti-deformation ability and hinder production and manufacturing. Therefore, it is necessary to obtain the best ply sequence that meets multiple needs through optimization design. There are many variables in the optimization of ply sequence and the variables are discrete. In addition, it is necessary to meet multiple design indicators at the same time in the design. Therefore, it is very necessary to improve the design efficiency under multi-task conditions. In recent years, the optimization of discrete variables has been realized through intelligent algorithms such as genetic algorithms, particle swarm algorithms, and ant colony algorithms, which can simplify the "combinatorial explosion" problem caused by the optimization of discrete variables. However, this type of method has large computational complexity, time-consuming calculation, and low computational efficiency. It is not suitable for a large number of stiffened wall panel optimization design work in aircraft structure design. Summary of the invention

[0003] In order to solve at least one of the above technical problems, the present application provides a reinforced wall panel ply optimization method and device, which combines artificial neural networks and genetic algorithms to solve the problems of too many design variables and low computational efficiency.

[0004] The first aspect of the present application provides a method for optimizing the ply of a reinforced wall panel, mainly comprising: Step S1, obtaining a first specified number of ply laying schemes; Step S2, encoding each plying scheme according to the correspondence between the preset plying angle and the specified number, to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; Step S3, calculating the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; Step S4, select the second specified number of individuals ranked top after the structural strain is sorted from large to small, perform crossover mutation of the genetic algorithm, form a new population, return to the previous step and execute repeatedly until the iteration converges, and obtain the optimal ply plan.

[0005] Preferably, before step S3, the method further comprises: Step S31: on the premise that the ply thickness and ply angle are determined, multiple ply schemes are constructed based on different ply orders as training input parameters of the mechanical proxy model; Step S32, determining the structural strain of each ply scheme through finite element simulation as an expected output parameter; Step S33, input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

[0006] Preferably, in step S3, the mechanical proxy model is a neural network model built based on the Pytorch framework, the neural network model includes two hidden layers, each hidden layer contains 50 neurons, the activation function of the neural network model uses the rectified linear unit function ReLU, and the difference between the model output and the expected output parameters is represented by the mean square error function.

[0007] Preferably, in step S4, when performing crossover mutation of the genetic algorithm, the crossover rate is set to 70% and the mutation rate is set to 0.5%.

[0008] Preferably, in step S4, the second specified quantity is 15%-20% of the first specified quantity.

[0009] The second aspect of the present application provides a reinforced wall panel ply optimization device, which mainly includes: A ply plan building module is used to obtain a first specified number of ply plans; A genetic algorithm population generation module is used to encode each plying scheme according to the correspondence between the preset plying angle and the specified number to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; The structural strain prediction module is used to calculate the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; The genetic algorithm loop control module is used to select the second specified number of individuals ranked at the top after the structural strain is sorted from large to small to perform crossover mutation of the genetic algorithm to form a new population, return to the previous step and execute cyclically until the iteration converges to obtain the optimal ply plan.

[0010] Preferably, the genetic algorithm population generation module includes a training module for training the mechanical proxy model, and the training module includes: An input parameter determination unit is used to construct multiple plying schemes based on different plying orders under the premise that the ply thickness and plying angle are determined, and serve as training input parameters for the mechanical proxy model; An expected output parameter simulation generation unit is used to determine the structural strain of each ply solution as the expected output parameter through finite element simulation; The mechanical proxy model parameter adjustment unit is used to input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

[0011] Preferably, the mechanical agent model is a neural network model built based on the Pytorch framework, which includes two hidden layers, each hidden layer contains 50 neurons, and the activation function of the neural network model uses the rectified linear unit function ReLU, and the difference between the model output and the expected output parameters is represented by the mean square error function.

[0012] Preferably, in the genetic algorithm loop control module, the crossover rate of the genetic algorithm is set to 70%, and the mutation rate is set to 0.5%.

[0013] Preferably, the second specified amount is 15%-20% of the first specified amount.

[0014] The present application embeds a mechanical proxy model with relatively high training accuracy into the generalized fitness function of the genetic algorithm to guide the genetic algorithm to optimize the ply sequence, thereby improving the optimization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a preferred embodiment of the reinforced wall panel ply optimization method of the present application. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the implementation of this application clearer, the technical scheme in the implementation of this application will be described in more detail in combination with the drawings in the implementation of this application. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described implementation is a part of the implementation of this application, not all of the implementations. The implementation described below with reference to the drawings is exemplary and is intended to be used to explain this application, and cannot be understood as a limitation on this application. Based on the implementation in this application, all other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The implementation of this application is described in detail below in combination with the drawings.

[0017] The first aspect of the present application provides a method for optimizing the ply of a reinforced wall panel, the method mainly comprising: Step S1, obtaining a first specified number of ply laying schemes; Step S2, encoding each plying scheme according to the correspondence between the preset plying angle and the specified number, to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; Step S3, calculating the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; Step S4, select the second specified number of individuals ranked top after the structural strain is sorted from large to small, perform crossover mutation of the genetic algorithm, form a new population, return to the previous step and execute repeatedly until the iteration converges, and obtain the optimal ply plan.

[0018] The main body of the above steps of this application is genetic algorithm, such as Figure 1 As shown, each original ply given in step S1 is taken as an individual, which is encoded in step S2 to form an initial population, and then the structural strain of each individual in the initial population is calculated through the mechanical proxy model in step S3, and some individuals with the highest ranking after the structural strain is sorted from large to small are selected in step S4, and a new population is re-formed through crossover mutation, and the process returns to step S3, and steps S3 and S4 are executed cyclically to complete the optimization of the ply scheme.

[0019] According to the above description, it can be seen that the optimization variable in the entire optimization process is the ply sequence, and the optimization target is the maximum strain value, that is, the fitness function in the genetic algorithm is to calculate the strain value. The ultimate goal is to find the ply sequence corresponding to the maximum strain value. During the iteration process, multiple individuals with larger strain values ​​are found to participate in cross-mutation. The constraint condition is that the principal strain value of the structure is less than the allowable strain value.

[0020] In step S2, common laying angles in engineering include 0°, 45°, -45° and 90°. The corresponding relationship between the laying angle and the specified number can be, for example, 0° corresponds to 0, 45° corresponds to 1, -45° corresponds to 2, and 90° corresponds to 3. If a laying scheme is [45 / -45 / 0 / -45 / 0 / 45 / 90 / 0] (unit: degree), the corresponding code is [1 / 2 / 0 / 2 / 0 / 1 / 3 / 0].

[0021] In step S3, a mechanical proxy model is used to quickly predict the structural strain according to the ply sequence. This process replaces the finite element analysis process in the prior art and is the key to achieving efficient optimization of the ply sequence of the reinforced wall panels. It is understandable that before executing step S3, it is very important to train an accurate mechanical proxy model for strain prediction. An accurate mechanical proxy model mainly considers two points: one is the need for sufficient and reliable training samples, and the other is the design of excellent model hyperparameters, which are explained below.

[0022] In some optional implementations, the process further includes before step S3: Step S31: on the premise that the ply thickness and ply angle are determined, multiple ply schemes are constructed based on different ply orders as training input parameters of the mechanical proxy model; Step S32, determining the structural strain of each ply scheme through finite element simulation as an expected output parameter; Step S33, input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

[0023] The above embodiment provides a process of how to form sufficient and reliable training samples of a mechanical proxy model and perform training. After the layup scheme is formed in step S31, it is necessary to calculate reliable structural strains for each layup scheme in step S32. This process is based on finite element simulation. The layup scheme and the structural strain form training samples. The training samples are constructed into a training set, a test set, and a validation set in a ratio of 7:2:1.

[0024] In some optional implementations, the multiple ply laying schemes formed in step S31 may be further screened to remove ply laying schemes that do not meet the restrictions of balance, symmetry, surface 45 degrees, and number of continuous plies.

[0025] In step S32, it is necessary to construct a finite element model of the structure, using two-dimensional shell elements and applying necessary boundary conditions and loads, such as a four-sided simply supported reinforced wall panel structure. Then, the ply property scheme of each design point is assigned to the above finite element model in turn, and static analysis is carried out using finite element analysis software to obtain the structural strain. It can be understood that the structural strain output by the finite element model is the expected output parameter.

[0026] Finally, in step S33, the structural strain output by the mechanical proxy model is compared with the structural strain output by the finite element model. The comparison result is subjected to partial derivative calculation to continuously correct the parameters of the mechanical proxy model until an accurate mechanical proxy model is trained.

[0027] In some optional embodiments, in step S3, the mechanical agent model is a neural network model built based on the Pytorch framework, and the neural network model includes two hidden layers, each hidden layer contains 50 neurons, and the activation function of the neural network model uses the rectified linear unit function ReLU, and the difference between the model output and the expected output parameters is represented by the mean square error function.

[0028] In some optional implementations, in step S4, when performing crossover mutation of the genetic algorithm, the crossover rate is set to 70% and the mutation rate is set to 0.5%. In addition, if the ply scheme satisfies symmetry, half of the number of plies can be passed in to simplify the amount of calculation. In this embodiment, crossover is to generate a new ply scheme by exchanging the ply angles of corresponding positions in the two ply designs; mutation is to change the ply angle of a single layer in the ply design to generate a new ply scheme. After the population undergoes crossover mutation, the number of individuals in it is restored to the initial population number, which is the first specified number in step S1.

[0029] In some optional implementations, in step S4, the second specified quantity is 15%-20% of the first specified quantity.

[0030] For example, assuming that the initial population contains 50 individuals, that is, the first specified number is 50, in step S4, the individuals with the top 20% fitness values ​​are retained, and the second specified number is 50*20%=10. That is, 10 individuals with larger structural strain are retained to participate in the subsequent crossover mutation operation.

[0031] It should also be noted that the convergence condition of step S4 may be that the difference between the maximum fitness value of the last generation and the maximum fitness value of the previous generation is less than 1%, or the number of algorithm iterations is greater than 2000 times.

[0032] The present application embeds a mechanical proxy model with relatively high training accuracy into the generalized fitness function of the genetic algorithm to guide the genetic algorithm to optimize the ply sequence, thereby improving the optimization efficiency.

[0033] The second aspect of the present application provides a reinforced wall panel ply optimization device corresponding to the above method, mainly comprising: A ply plan building module is used to obtain a first specified number of ply plans; A genetic algorithm population generation module is used to encode each plying scheme according to the correspondence between the preset plying angle and the specified number to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; The structural strain prediction module is used to calculate the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; The genetic algorithm loop control module is used to select the second specified number of individuals ranked at the top after the structural strain is sorted from large to small to perform crossover mutation of the genetic algorithm to form a new population, return to the previous step and execute cyclically until the iteration converges to obtain the optimal ply plan.

[0034] In some optional implementations, the genetic algorithm population generation module includes a training module for training the mechanical proxy model, and the training module includes: An input parameter determination unit is used to construct multiple plying schemes based on different plying orders under the premise that the ply thickness and plying angle are determined, and serve as training input parameters for the mechanical proxy model; An expected output parameter simulation generation unit is used to determine the structural strain of each ply solution as the expected output parameter through finite element simulation; The mechanical proxy model parameter adjustment unit is used to input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

[0035] In some optional embodiments, the mechanical agent model is a neural network model built based on the Pytorch framework, which includes two hidden layers, each hidden layer contains 50 neurons, and the activation function of the neural network model uses the rectified linear unit function ReLU, and the difference between the model output and the expected output parameters is represented by the mean square error function.

[0036] In some optional implementations, in the genetic algorithm loop control module, the crossover rate of the genetic algorithm is set to 70%, and the mutation rate is set to 0.5%.

[0037] In some optional embodiments, the second specified amount is 15%-20% of the first specified amount.

[0038] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A reinforced wall panel ply optimization method, characterized in that: include: Step S1, obtaining a first specified number of ply laying schemes; Step S2, encoding each plying scheme according to the correspondence between the preset plying angle and the specified number, to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; Step S3, calculating the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; Step S4, select the second specified number of individuals ranked top after the structural strain is sorted from large to small, perform crossover mutation of the genetic algorithm, form a new population, return to the previous step and execute repeatedly until the iteration converges, and obtain the optimal ply plan.

2. The method for optimizing the ply of a reinforced wall panel according to claim 1, characterized in that: Before step S3, the method further comprises: Step S31: on the premise that the ply thickness and ply angle are determined, multiple ply schemes are constructed based on different ply orders as training input parameters of the mechanical proxy model; Step S32, determining the structural strain of each ply scheme through finite element simulation as an expected output parameter; Step S33, input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

3. The method for optimizing the ply of reinforced wall panels according to claim 2, characterized in that: In step S3, the mechanical agent model is a neural network model built based on the Pytorch framework. The neural network model includes two hidden layers, each hidden layer contains 50 neurons, and the activation function of the neural network model uses the rectified linear unit function ReLU. The difference between the model output and the expected output parameters is represented by the mean square error function.

4. The method for optimizing the plywork of reinforced wall panels according to claim 1, characterized in that: In step S4, when performing crossover mutation of the genetic algorithm, the crossover rate is set to 70% and the mutation rate is set to 0.5%.

5. The method for optimizing the plywork of reinforced wall panels according to claim 1, characterized in that: In step S4, the second specified quantity is 15%-20% of the first specified quantity.

6. A reinforced wall panel ply optimization device, characterized in that: include: A ply plan building module is used to obtain a first specified number of ply plans; A genetic algorithm population generation module is used to encode each plying scheme according to the correspondence between the preset plying angle and the specified number to obtain an initial population with a first specified number of codes, wherein each code corresponds to an individual in the population; The structural strain prediction module is used to calculate the structural strain corresponding to each individual in the population based on the pre-trained mechanical agent model; The genetic algorithm loop control module is used to select the second specified number of individuals ranked at the top after the structural strain is sorted from large to small to perform crossover mutation of the genetic algorithm to form a new population, return to the previous step and execute cyclically until the iteration converges to obtain the optimal ply plan.

7. The device for optimizing the ply of a reinforced wall panel according to claim 6, characterized in that: The genetic algorithm population generation module includes a training module for training the mechanical proxy model, and the training module includes: An input parameter determination unit is used to construct multiple plying schemes based on different plying orders under the premise that the ply thickness and plying angle are determined, and serve as training input parameters for the mechanical proxy model; An expected output parameter simulation generation unit is used to determine the structural strain of each ply solution through finite element simulation as the expected output parameter; The mechanical proxy model parameter adjustment unit is used to input the training input parameters into the mechanical proxy model to obtain the model output, adjust the parameters of the mechanical proxy model based on the difference between the model output and the expected output parameters, and obtain the mechanical proxy model after multiple trainings.

8. The device for optimizing the ply of a reinforced wall panel according to claim 7, characterized in that: The mechanical agent model is a neural network model built based on the Pytorch framework. The neural network model includes two hidden layers, each hidden layer contains 50 neurons, and the activation function of the neural network model uses the rectified linear unit function ReLU. The difference between the model output and the expected output parameters is represented by the mean square error function.

9. The device for optimizing the ply of a reinforced wall panel according to claim 6, characterized in that: In the genetic algorithm loop control module, the crossover rate of the genetic algorithm is set to 70%, and the mutation rate is set to 0.5%.

10. The device for optimizing the ply of a reinforced wall panel according to claim 6, characterized in that: The second specified amount is 15%-20% of the first specified amount.

Citation Information

Patent Citations

  • Performance optimization design method for harmonic reducer flexible wheel with composite layer structure

    CN106709106A

  • Method for optimally designing angles of laying layers of single-layer boards

    CN107844664A

  • Laying optimization design method of composite pressure-resistant shell

    CN110321613A

  • Genetic algorithm-based wind driven generator fairing laying optimization design method

    CN110704957A

  • Unmanned aerial vehicle wing ply optimization method based on genetic algorithm

    CN111199074A

Cited By

  • Layer design criterion mining method based on generative and interpretable artificial intelligence

    CN120260746A