A composite material model correction method under dynamic environment

Abstract

The application provides a composite material model correction method in a dynamic environment, which corrects physical property parameters of a structure by establishing a transmission relationship between the physical property parameters of the composite material and dynamic modal parameters, thereby reducing errors between a finite element model and a test model and significantly improving the precision of the finite element model; an agent model constructed by using a BP neural network does not need to depend on a mass matrix and a stiffness matrix of a system and can realize reverse identification of parameters based on a data-driven transmission relationship only. This method greatly shortens the time required for model correction and significantly improves the correction efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of composite material model correction, and particularly relates to a composite material model correction method in a dynamic environment. BACKGROUND

[0002] The composite material model correction in a dynamic environment is an important scientific problem in aerospace engineering, is a key technology for guiding lightweight design of a structure and ground test, and is an important basis for carrying out structure coupling field response analysis and predicting service performance and safety performance.

[0003] At present, a finite element analysis method of a typical composite material has been widely applied in the field of aerospace. However, it is almost impossible to establish a finite element model that can accurately describe an actual object only by relying on engineering experience. The error sources in finite element modeling are various, scholars such as Mottershead summarize model errors into three main aspects and make a detailed discussion: structural error, parameter error and order error. Among them, the structural error refers to the rationality of the selected mathematical model, and the order error is related to the discretization degree. These two types of errors can usually be avoided by reasonable modeling. In comparison, the parameter error caused by the difference of material parameters, boundary conditions and the like often becomes the focus of finite element model correction.

[0004] The design parameter type model correction method constructs an objective function, repeatedly iterates to calculate the correction amount, so as to effectively reduce the error. This method is suitable for correcting the physical parameters, geometric parameters and boundary conditions of the model, and is not limited by the correction object. The corrected parameters have clear physical meaning and can be directly compared with the actual situation, so it has been widely applied in engineering practice. However, the finite element model correction method based on sensitivity needs to repeatedly perform iterative calculation of the finite element model, and this process often consumes a lot of time.

[0005] In view of the problem that the existing design parameter type model correction method consumes a long time in the iterative process of the finite element model, it is urgent to explore an effective way to shorten the correction time, so as to improve the efficiency of the correction process. SUMMARY

[0006] The purpose of the application is to overcome the deficiencies in the prior art, and provide a composite material model correction method in a dynamic environment. The application scheme can solve the problems existing in the prior art.

[0007] The technical solution of the application is as follows:

[0008] A composite material model correction method in a dynamic environment, comprising the following steps:

[0009] Step one, carry out structural modal test on the typical structure of the composite material, and perform random vibration test under normal temperature or high temperature conditions to obtain dynamic modal parameters of the typical structure;

[0010] Step two, establish a fine finite element model, estimate the physical property parameter range of the composite material according to engineering experience, sample by Latin hypercube, and input the sample parameters into the fine finite element model of the typical structure of the composite material to obtain dynamic modal parameters of the skin-stiffened structure under different physical property parameters, and establish a sample space of the physical property parameters and corresponding modal parameters;

[0011] Step three, train a BP neural network proxy model through the sample space, map the physical property parameters to the modal parameters, the BP neural network proxy model includes two directions of forward propagation and backward propagation, the forward propagation refers to the information transmission from the input layer to the output layer to generate an output result, and the backward propagation refers to the adjustment and modification of the weights and thresholds through the backward error transmission of the neural network;

[0012] Step four, according to the BP neural network proxy model obtained after training, taking the minimum norm of the modal parameters output by different physical property parameters and the modal parameters obtained by test as an optimization objective, and optimizing by an optimization algorithm to obtain the optimal solution of the modal parameters of the typical structure of the composite material, thereby completing the correction of the fine finite element model, and obtaining the mechanical model of the typical structure of the composite material;

[0013] Step five, obtain the physical property parameters of the material at different temperatures by establishing the BP neural network proxy model under different conditions, fit the obtained physical property parameters of the material at different temperatures by the least square method, predict the material parameters under the temperature conditions without test, and obtain the mechanical model of the typical structure of the composite material under dynamic environment.

[0014] Further, the method for obtaining the optimal solution of the dynamic modal parameters of the typical structure of the composite material comprises the following steps:

[0015] The residual value between the real dynamic modal information parameters obtained by the structural modal test and the dynamic modal parameters calculated by the proxy model is minimized as the target of model correction, the model correction problem is converted into an optimization problem of minimizing the dynamic modal frequency residual, and the minimum value problem is solved by an optimization method.

[0016] Further, the BP neural network agent model comprises an input layer, a hidden layer and an output layer, the forward propagation of information is from the input layer to the hidden layer and then to the output layer, the backward propagation of error is from the output layer to the hidden layer and then to the input layer, the forward propagation of information and the backward propagation of error are repeated alternately and circularly until the error meets the requirement set in advance or the number of neural network training reaches the maximum value, the optimization process ends and the optimal solution is output.

[0017] Further, in step three, the input physical property parameters are normalized, that is, the data in each column are divided by the maximum value in the column, and the normalized values are used as the input physical property parameters.

[0018] Further, in step four, the optimal solution is subjected to the inverse process of the normalization process, and the optimal solution is restored to the original numerical value to obtain the optimized material physical property parameters.

[0019] Further, the BP neural network agent model utilizes the nonlinear coding characteristics of the neural network to establish a nonlinear mapping from the physical property parameters to the modal parameters.

[0020] Compared with the prior art, the present application has the following beneficial effects:

[0021] 1) The present application performs reasonable normalization on data, so that the agent model established does not limit the numerical value of the elastic modulus, shear modulus, Poisson's ratio and thermal expansion coefficient and other parameters in the finite element model, significantly improves the applicability and flexibility of the correction method, and realizes unconstrained selection of the correction parameters.

[0022] 2) The present application utilizes the nonlinear distributed coding characteristics of the BP neural network for input and output parameters to construct a nonlinear mapping relationship from input to output. When the network size is large enough, information is embedded in the network weight value to form a high-redundancy information storage mode, thereby ensuring that all influencing factors are simultaneously coded. The BP neural network has strong fault tolerance due to its integrity, dynamics and correlation, thereby significantly improving the accuracy of model correction.

[0023] 3) The agent model constructed by the present application using the BP neural network does not need to rely on the mass matrix and stiffness matrix of the system, and can realize parameter reverse identification based on the data-driven transfer relationship. This method significantly shortens the time required for model correction and significantly improves the correction efficiency.

[0024] 4) The present application corrects the physical property parameters of the structure by establishing the transfer relationship between the physical property parameters of the composite material and the dynamic modal parameters, thereby reducing the error between the finite element model and the test model and significantly improving the accuracy of the finite element model. BRIEF DESCRIPTION OF DRAWINGS

[0025] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. It is to be expressly understood that the drawings are for purposes of illustration only and are not intended as a definition of the limits of the application. In the drawings, the same reference numbers are used to designate analogous elements throughout several views.

[0026] Figure 1 A technical route of a composite material model correction method in a dynamic environment is shown according to an embodiment of the application;

[0027] Figure 2 A typical skin-stiffened structure diagram is shown according to an embodiment of the application;

[0028] Figure 3 A BP neural network proxy model network diagram is shown according to an embodiment of the application. DETAILED DESCRIPTION

[0029] It should be noted that the embodiments and the features in the embodiments in the present application can be combined with each other without conflict. The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The description of the at least one exemplary embodiment is actually only illustrative, but not intended to limit the present application and its application or use in any way. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0030] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0031] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0032] like Figure 1 As shown, a method for correcting composite material models under dynamic environments includes the following steps:

[0033] Step 1: Conduct structural modal tests on typical composite material structures, and perform random vibration tests under normal or high temperature conditions to obtain the dynamic modal parameters of typical composite material structures.

[0034] Step 2: Establish a refined finite element model. Based on engineering experience, estimate the range of physical property parameters of the composite material. Through Latin hypercube sampling, input the sample parameters into the refined finite element model of a typical composite material structure to obtain the dynamic modal parameters of the typical composite material structure under different physical property parameters, and establish a sample space of physical property parameters and corresponding modal parameters.

[0035] Step 3: Train a BP neural network surrogate model through the sample space. Use physical property parameters as input and modal parameters as output to establish a mapping relationship between the two. The BP neural network surrogate model fits the data in two directions: forward propagation and backward propagation. Forward propagation refers to the information being passed from the input layer to the output layer in sequence to produce the output result. Backward propagation refers to the neural network adjusting and modifying the weights and thresholds layer by layer by propagating the error in the reverse direction.

[0036] Step four: Based on the BP neural network surrogate model obtained after training, the minimum norm of the modal parameters output by different physical property parameters and the modal parameters obtained by experiments is used as the optimization objective. The optimization algorithm is used to obtain the optimal solution of the modal parameters of typical composite material structures, thereby completing the correction of the refined finite element model and obtaining the mechanical model of typical composite material structures under dynamic environment.

[0037] Furthermore, the method for obtaining the optimal solution of the dynamic modal parameters of a typical composite material structure is as follows:

[0038] The residual value between the real dynamic modal information parameters obtained by the structural modal test and the dynamic modal parameters calculated by the proxy model is minimized as the target of the model correction, the model correction problem is converted into an optimization problem of minimizing the dynamic modal frequency residual, and then the minimum value problem is solved by an optimization method.

[0039] Further, in an embodiment, the BP neural network proxy model comprises an input layer, a hidden layer and an output layer, the forward propagation of information is from the input layer to the hidden layer and then to the output layer, the backward propagation of error is from the output layer to the hidden layer and then to the input layer, the forward propagation of information and the backward propagation of error are repeatedly alternated and circulated until the error meets the requirements set in advance or the number of neural network training reaches the maximum value, the optimization process ends, and the optimal solution is output.

[0040] Further, in an embodiment, in step three, the input physical property parameters are normalized, that is, the data in each column are divided by the maximum value in the column, and the normalized values are used as the input physical property parameters. By reasonably normalizing the data, the established proxy model does not limit the numerical size of the elastic modulus, shear modulus, Poisson's ratio and thermal expansion coefficient and other parameters in the finite element model, significantly improving the applicability and flexibility of the correction method, and realizing the unconstrained selection of the correction parameters.

[0041] Further, in an embodiment, in step four, the optimal solution is subjected to the inverse process of the normalization process, and the optimal solution is restored to the original numerical size to obtain the optimized material physical property parameters.

[0042] Further, in an embodiment, the BP neural network proxy model uses the characteristics of nonlinear coding of neural networks to establish a nonlinear mapping of physical property parameters to modal parameters. By using the characteristics of nonlinear distributed coding of BP neural network for input and output parameters, a nonlinear mapping relationship from input to output is constructed. When the network size is large enough, information is embedded in the network weights, forming a high-redundancy information storage mode, thereby ensuring that all influencing factors are simultaneously coded. The BP neural network has strong fault tolerance due to its integrity, dynamics and correlation, thereby significantly improving the accuracy of model correction.

[0043] In order to have a further understanding of the method for modeling the typical structure of composite materials in a dynamic environment, specific examples and drawings are used in the following detailed description.

[0044] Taking a typical composite skin-stiffened structure as an example:

[0045] In this embodiment, the typical skin-stiffened structure is as shown in Figure 2As shown, the typical skin-stiffened structure includes a skin and a stiffener, both made of C / SiC composite material, and the material physical property parameters at room temperature are: elastic modulus E1=E2=55 GPa, shear modulus G 12 =100 GPa, G 13 =G 23 =20 GPa, isotropic Poisson's ratio μ=0.05, and isotropic thermal expansion coefficient α=1.3E-6.

[0046] The composite material model correction method under dynamic environment first determines E1, E2, G 12 , and α as the material physical property parameters to be optimized according to engineering experience, and estimates the value range of the to-be-optimized parameters. In this embodiment, the value range of the elastic modulus E1=E2 is 40-80 GPa, the value range of the shear modulus G 12 is 80-150 GPa, and the value range of the thermal expansion coefficient α is 1E-6-2E-6.

[0047] The composite material model correction method under dynamic environment, after determining the value range and the to-be-optimized parameters, establishes a sample space in the value range of the parameters through Latin hypercube sampling. In this embodiment, a sample space containing 600 groups of data is established.

[0048] The composite material model correction method under dynamic environment, after sampling is completed, establishes a finite element model through the existing technology, calculates the modal parameters of each sample point in the sample space, and establishes a training set for training the BP neural network proxy model. In one embodiment, the skin-stiffened structure is established as a finite element model through a shell element to improve the calculation efficiency.

[0049] The composite material model correction method under dynamic environment, the selection of the number of nodes in the hidden layer is the most uncertain part in the entire proxy model. In this embodiment, the number of nodes in the hidden layer is determined to be 10 through an empirical formula, and the constructed neural network is as shown in Figure 3 .

[0050] The composite material model correction method under dynamic environment, in order to achieve better training and fitting effect, the training data is normalized before training, and the data in each column is divided by the maximum value in all the values in the column, so as to solve the problem that the sensitivity of different parameters is different in the fitting process due to the large difference in the magnitude of each parameter.

[0051] The composite material model correction method in a dynamic environment selects 400 groups of data as a training set and 200 groups of data as a test set after normalizing the data, trains and verifies the BP neural network proxy model by taking physical parameters as input modes and parameters as output, and establishes a nonlinear mapping from the physical parameters to the modal parameters by using the nonlinear coding characteristics of the neural network, so that the prediction accuracy of the proxy model is improved.

[0052] The composite material model correction method in a dynamic environment minimizes the residual value between the real dynamic modal information parameters obtained from the structural modal test and the dynamic modal parameters calculated by the proxy model as the target of model correction, converts the model correction problem into an optimization problem of minimizing the dynamic modal frequency residual, and solves the minimum value problem by using a particle swarm optimization algorithm. The optimal solution is obtained, and then the inverse process of the normalization process is performed to restore the optimal solution to the original numerical size, so that the optimized material physical parameters are obtained.

[0053] The composite material model correction method in a dynamic environment obtains the dynamic parameters of the structure under different temperature conditions by performing dynamic experiments under different temperature conditions, obtains the physical parameters of the material under different temperatures by establishing the BP neural network proxy model under different conditions, and uses the least square method to fit the obtained physical parameters of the material at different temperatures, so that the material parameters under the temperature conditions not subjected to the test can be predicted.

[0054] In summary, the composite material model correction method in a dynamic environment provided by the present application has at least the following advantages compared with the prior art:

[0055] The above is only a preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for correcting composite material models under dynamic environments, characterized in that, The method comprises the following steps: Step one, structural modal test is carried out on the composite skin-stiffened structure to obtain the dynamic modal parameters of the skin-stiffened structure under normal temperature or high temperature conditions; Step two, a fine finite element model is established, the material property parameter range of the composite material is estimated according to engineering experience, Latin hypercube sampling is carried out, and the sample parameters are input into the fine finite element model of the composite skin-stiffened structure to obtain the dynamic modal parameters of the skin-stiffened structure under different material property parameters, and a sample space of the material property parameters and the corresponding modal parameters is established; Step three, a BP neural network proxy model is trained through the sample space to establish the mapping relationship between the material property parameters and the modal parameters, the BP neural network proxy model is fitted to data in two directions of forward propagation and backward propagation, the forward propagation refers to the information being transmitted from the input layer to the output layer in sequence to generate an output result, and the backward propagation refers to the neural network adjusting and modifying the weights and thresholds layer by layer through the backward transmission of errors; Step four, the modal parameters output by the BP neural network proxy model under different material property parameters and the modal parameters obtained through the test are used as the optimization target, and an optimization algorithm is used for optimization to obtain the optimal solution of the modal parameters of the skin-stiffened structure, so that the fine finite element model is corrected, and a mechanical model of the skin-stiffened structure is obtained; Step five, the material property parameters of the skin-stiffened structure under different temperatures are obtained by establishing the BP neural network proxy model under different conditions, the material property parameters of the skin-stiffened structure under different temperatures are fitted by using the least square method, the material parameters under the temperature conditions that have not been tested are predicted, and a mechanical model of the skin-stiffened structure under a dynamic environment is obtained.

2. The method of claim 1, wherein, The method for obtaining the optimal solution of the dynamic modal parameters of the skin-stiffened structure comprises the following steps: The residual value between the real dynamic modal information parameters obtained through the structural modal test and the dynamic modal parameters calculated by the proxy model is minimized as the target of model correction, the model correction problem is converted into an optimization problem of minimizing the dynamic modal frequency residual, and the minimum value problem is solved by using an optimization method.

3. The method of claim 2, wherein, The BP neural network proxy model comprises an input layer, a hidden layer and an output layer, the forward propagation of information is from the input layer to the hidden layer and then to the output layer, the backward propagation of errors is from the output layer to the hidden layer and then to the input layer, the forward propagation of information and the backward propagation of errors are repeatedly and alternately performed in cycles until the error meets the requirements set in advance or the number of neural network training reaches the maximum value, the optimization process is ended, and the optimal solution is output.

4. The method of claim 1, wherein In step three, the input material property parameters are normalized, that is, the data in each column are divided by the maximum value in the column, and the normalized values are used as the input material property parameters.

5. The method of claim 4, wherein, In step four, the optimal solution is subjected to the inverse process of the normalization process, and the optimal solution is restored to the original value to obtain the optimized material property parameters.

6. The method of claim 3, wherein The BP neural network proxy model uses the nonlinear coding characteristics of the neural network to establish a nonlinear mapping from the material property parameters to the modal parameters.

Images