Neural Network-Based Multiscale Modeling Method, Device, and Medium for Composite Materials

Abstract

The present invention relates to the technical field of composite material analysis, and particularly relates to a multi-scale modeling method, device and medium for composite materials based on neural networks. The method includes: obtaining the structural data of the microscopic unit cell of the composite material; constructing a microscopic finite element damage evolution model; sampling within the loading strain range of the microscopic unit cell to obtain multiple groups of microscopic strain input data; using the multiple groups of microscopic strain input data as the input of the microscopic finite element damage evolution model, and through finite element simulation calculation, obtaining the corresponding stress output data; training the neural network based on the multiple groups of microscopic strain input data and the corresponding stress output data to obtain a microscopic stress transfer network; obtaining the structural data of the mesoscopic unit cell of the composite material; constructing a mesoscopic finite element damage evolution model; and the mesoscopic finite element damage evolution model updates the stress at the integration points based on the microscopic stress transfer network. The present invention can more accurately realize the multi-scale modeling of composite materials.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of composite material analysis, and particularly to a multi-scale modeling method for composite materials based on neural networks, an electronic device, and a storage medium. Background Art

[0002] The purpose of multi-scale modeling of composite materials is to establish a quantitative relationship between the macroscopic properties of composite materials and the properties of their constituent materials and structures at various scales, and to reveal the internal mechanism that different organizational forms at different scales lead to different macroscopic properties.

[0003] The key to multi-scale modeling lies in the information transfer between different scales. Taking carbon fiber reinforced composite materials as an example, the microscopic scale usually refers to the scale of fiber filaments and resin matrix, the mesoscopic scale refers to the scale of fiber bundles and resin matrix, and the macroscopic scale usually refers to the scale of structural components or general test pieces. At present, in the process of multi-scale modeling, when transferring information from a low scale to a high scale (i.e., from the microscopic scale to the mesoscopic scale, from the mesoscopic scale to the macroscopic scale), some simplifications or approximations are generally made. For example, taking a single cell at the microscopic scale as the research object, homogenizing two materials at the low scale into one material at the high scale, and only paying attention to a few characteristics of the material. The coupling degree between the low scale and the high scale is low, and when constructing the material model, the introduced damage criterion often has poor universality for different materials, which leads to low accuracy of the multi-scale modeling of composite materials and cannot accurately reflect the characteristics of composite materials. Summary of the Invention

[0004] Based on the problem of low accuracy of multi-scale modeling of composite materials, the present invention provides a multi-scale modeling method for composite materials based on neural networks, an electronic device, and a storage medium, which can realize information transfer between different scales based on neural networks, improve the coupling degree between different scale models, thereby improving the calculation accuracy, and can save computing resources.

[0005] In the first aspect, the embodiments of the present invention provide a multi-scale modeling method for composite materials based on neural networks, including:

[0006] Obtain the structural data of the microscopic unit cell of the composite material;

[0007] Based on the structural data of the microscopic unit cell, construct a microscopic finite element damage evolution model and set periodic boundary conditions;

[0008] Determine the loading strain range in six directions of the microscopic unit cell;

[0009] Sample within the loading strain range of the microscopic unit cell to obtain multiple groups of microscopic strain input data; each group of microscopic strain input data includes strain data in six directions of the microscopic unit cell;

[0010] Using multiple sets of microscopic strain input data as the input of the microscopic finite element damage evolution model, through finite element simulation calculation, the corresponding stress output data is obtained;

[0011] Based on multiple sets of microscopic strain input data and the corresponding stress output data, the neural network is trained to obtain a microscopic stress transfer network;

[0012] Obtain the structural data of the mesoscopic unit cell of the composite material;

[0013] Based on the structural data of the mesoscopic unit cell, a mesoscopic finite element damage evolution model is constructed and periodic boundary conditions are set; the mesoscopic finite element damage evolution model updates the stress at the integration points based on the microscopic stress transfer network.

[0014] Optionally, the multi-scale modeling method of composite materials based on neural network further includes:

[0015] Determine the loaded strain range in six directions of the mesoscopic unit cell;

[0016] Sampling within the loaded strain range of the mesoscopic unit cell to obtain multiple sets of mesoscopic strain input data; each set of mesoscopic strain input data includes strain data in six directions of the mesoscopic unit cell;

[0017] Using multiple sets of mesoscopic strain input data as the input of the mesoscopic finite element damage evolution model, through finite element simulation calculation, the corresponding stress output data is obtained;

[0018] Based on multiple sets of mesoscopic strain input data and the corresponding stress output data, the constructed neural network is trained to obtain a mesoscopic stress transfer network;

[0019] Obtain the structural data of the macroscopic structural component of the composite material;

[0020] Based on the structural data of the macroscopic structural component, a macroscopic finite element damage evolution model is constructed and boundary conditions are set; the macroscopic finite element damage evolution model updates the stress at the integration points based on the mesoscopic stress transfer network.

[0021] Optionally, constructing a microscopic finite element damage evolution model based on the structural data of the microscopic unit cell includes:

[0022] Respectively obtain the experimental data of each component material in the composite material;

[0023] Based on the experimental data of each component material, respectively determine the mechanical parameters of each component material;

[0024] Based on the mechanical parameters of each component material and the structural data of the microscopic unit cell, a microscopic finite element damage evolution model is constructed.

[0025] Optionally, after constructing the microscopic finite element damage evolution model based on the structural data of the microscopic unit cell and setting the periodic boundary conditions, before using multiple groups of microscopic strain input data as the input of the microscopic finite element damage evolution model and obtaining the corresponding stress output data through finite element simulation calculation, it further includes:

[0026] Obtain the experimental data of the unidirectional tape made of composite materials; the structure of the unidirectional tape corresponds to the constructed microscopic finite element damage evolution model;

[0027] Verify the constructed microscopic finite element damage evolution model based on the experimental data of the unidirectional tape.

[0028] Optionally, after constructing the mesoscopic finite element damage evolution model based on the structural data of the mesoscopic unit cell and setting the periodic boundary conditions, before using multiple groups of mesoscopic strain input data as the input of the mesoscopic finite element damage evolution model and obtaining the corresponding stress output data through finite element simulation calculation, it further includes:

[0029] Obtain the experimental data of the laminated plate made of composite materials; the structure of the laminated plate corresponds to the constructed mesoscopic finite element damage evolution model;

[0030] Verify the constructed mesoscopic finite element damage evolution model based on the experimental data of the laminated plate.

[0031] Optionally, the composite material multi-scale modeling method based on neural network further includes:

[0032] Obtain the experimental data of the macroscopic structural component made of composite materials;

[0033] Verify the constructed finite element damage evolution model based on the experimental data of the macroscopic structural component.

[0034] Optionally, if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcement phase and a matrix phase, and the reinforcement phase is a fiber;

[0035] The microscopic finite element damage evolution model uses a cuboid unit cell, and the center and four corners of the cross-section of the cuboid unit cell are all center points of circles. The center includes a complete circular cross-section of the fiber, and each of the four corners has a quarter circular cross-section of the fiber.

[0036] Optionally, if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcement phase and a matrix phase, and the reinforcement phase is a fiber. The fiber and the matrix phase form a fiber bundle, and the matrix phase is filled between the fiber bundles;

[0037] The mesoscopic finite element damage evolution model uses a plate-shaped unit cell, and the plate-shaped unit cell is woven by multiple segments of fiber bundles, and the void area is filled with the matrix phase.

[0038] In a second aspect, an embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0039] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed on a computer, the computer is made to execute the method described in any embodiment of this specification.

[0040] An embodiment of the present invention provides a multi-scale modeling method for composite materials based on a neural network, an electronic device, and a storage medium. The present invention uses a neural network to learn the correspondence between strain and stress at the microscale. During the process of modeling at the mesoscale and calculating damage evolution through finite element simulation, stress update is realized based on the trained neural network, so as to achieve the effect of embedding a microcell into a mesocell, improving the accuracy of multi-scale modeling. The present invention realizes information transfer between different scales based on a neural network, improves the coupling degree between different scale models, thereby improving the calculation accuracy, saving computing resources, and being able to provide technical support for the analysis and research of composite materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 is a flowchart of a multi-scale modeling method for composite materials based on a neural network provided by an embodiment of the present invention;

[0043] Figure 2 is a flowchart of another multi-scale modeling method for composite materials based on a neural network provided by an embodiment of the present invention;

[0044] Figure 3 is a schematic diagram of a microcell structure;

[0045] Figure 4 is a schematic diagram of a mesocell structure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] As described above, currently in the multi-scale modeling process, when transferring information from a low scale to a high scale (i.e., from the microscopic scale to the mesoscopic scale, and from the mesoscopic scale to the macroscopic scale), some reduction or approximation processing is generally performed. For example, taking a microscopic-scale unit cell as the research object, two materials at the low scale are homogenized into one material at the high scale, and only a few properties of the material are concerned. The coupling degree between the low scale and the high scale is relatively low, and when constructing a material model, the introduced damage criterion often has poor universality for different materials, which leads to low accuracy in the multi-scale modeling of composite materials and cannot accurately reflect the characteristics of composite materials. In view of this, the present invention provides a multi-scale modeling method that uses a neural network as a surrogate model for the low-scale model to transfer information to the high-scale model, thereby improving the coupling degree between different-scale models.

[0048] The following describes the specific implementation of the above concept.

[0049] Please refer to Figure 1 , the embodiments of the present invention provide a multi-scale modeling method for composite materials based on a neural network, and the method includes:

[0050] Step 100, obtaining the structural data of the microscopic unit cell of the composite material;

[0051] The structural data of the microscopic unit cell includes data such as the geometric dimensions of the microscopic unit cell, the mechanical properties of the component materials, and the fiber volume fraction.

[0052] Step 102, based on the structural data of the microscopic unit cell, constructing a microscopic finite element damage evolution model and setting periodic boundary conditions;

[0053] The microscopic finite element damage evolution model can be constructed using the existing commercial finite element software Abaqus. The constructed microscopic finite element damage evolution model can be used to simulate the changes in the test piece at the microscopic scale during the failure and loss evolution process, so as to generate the data required for neural network training.

[0054] Step 104, determining the loaded strain range in six directions of the microscopic unit cell;

[0055] The loaded strain range determined in this step 104 can be obtained from experiments or can be determined by combining existing literature.

[0056] Step 106: Sample within the loading strain range of the microscopic unit cell to obtain multiple groups of microscopic strain input data; each group of microscopic strain input data includes strain data in six directions of the microscopic unit cell.

[0057] The specific number of groups of microscopic strain input data obtained in this step 106 can be set according to actual needs and will not be further limited here.

[0058] Step 108: Use multiple groups of microscopic strain input data as the input of the microscopic finite element damage evolution model, and through finite element simulation calculation, obtain the corresponding stress output data; one group of microscopic strain input data corresponds to one group of stress output data, and one group of stress output data includes stress data in six directions.

[0059] In this step 108, a sufficient amount of strain data is used as the input to perform batch calculations on the microscopic unit cell in the constructed microscopic finite element damage evolution model, and the stress output of the microscopic unit cell under different strains is obtained, that is, the sample data for training the neural network is obtained through finite element simulation calculation; one group of sample data consists of one group of microscopic strain input data and the corresponding one group of stress output data.

[0060] Step 110: Based on multiple groups of microscopic strain input data and the corresponding stress output data, train the neural network to obtain the microscopic stress transfer network.

[0061] In this step 110, an appropriate neural network structure can be built using the Python language in the prior art, and then the constructed neural network is trained using the sample data set composed of the input strain and the output stress; the training of the neural network can refer to the prior art and will not be further elaborated here.

[0062] Step 112: Obtain the structural data of the mesoscopic unit cell of the composite material.

[0063] The structural data of the mesoscopic unit cell includes data such as the geometric dimensions of the mesoscopic unit cell, the mechanical properties of the component materials, and the fiber volume fraction.

[0064] Step 114: Based on the structural data of the mesoscopic unit cell, construct a mesoscopic finite element damage evolution model and set periodic boundary conditions; the stress at the integration points of the mesoscopic finite element damage evolution model is updated based on the microscopic stress transfer network.

[0065] The construction of the mesoscopic finite element damage evolution model can utilize the existing commercial finite element software Abaqus. The constructed mesoscopic finite element damage evolution model can be used to simulate the changes in the test piece at the mesoscopic scale during the failure and loss evolution process. The neural network (i.e., the microscopic stress transfer network) built using the Python language can be embedded into the Abaqus user material subroutine to achieve stress update.

[0066] In a specific embodiment, a neural network can be written and trained using the Python language, the neural network can be converted and compiled into a C language dynamic link library, and then the C language dynamic link library containing the neural network can be imported into the Abaqus user material subroutine for use, so as to realize the use of the neural network in the Abaqus user subroutine.

[0067] The neural network includes an input layer, a hidden layer, and an output layer. The parameters in the neural network are continuously iteratively updated by the backpropagation method through the data set, and finally the network structure can better reflect the implicit law of the data. In the present invention, the low-scale finite element model of the multi-scale model is regarded as a non-linear system, and the neural network has a high accuracy in simulating non-linear relationships. Therefore, in the embodiment of the present invention, the neural network is used as the surrogate model of the low-scale model to transmit information to the high-scale model. The obtained mesoscopic finite element damage evolution model embeds a microscopic stress transfer network to realize stress update. That is to say, the calculation result of the neural network (i.e., stress output) is used to replace the calculation result of the original general mechanical transfer model, which has the effect of embedding the stress-strain relationship of the microscopic unit cell into the integration point of the mesoscopic unit cell, and saves computing resources, can improve the coupling degree between different scale models, and further improve the calculation accuracy.

[0068] Optionally, as Figure 2 shown, the multi-scale modeling method for composite materials provided by the present invention further includes:

[0069] Step 116, determining the loading strain ranges in six directions of the mesoscopic unit cell;

[0070] Step 118, sampling within the loading strain range of the mesoscopic unit cell to obtain multiple groups of mesoscopic strain input data; each group of mesoscopic strain input data includes strain data in six directions of the mesoscopic unit cell;

[0071] Step 120, using multiple groups of mesoscopic strain input data as the input of the mesoscopic finite element damage evolution model, and through finite element simulation calculation, obtaining the corresponding stress output data;

[0072] The mesoscopic finite element damage evolution model performs finite element simulation calculation, and updates the stress based on the microscopic stress transfer network;

[0073] This step 120 takes a sufficient amount of strain data as the input, performs batch calculations on the mesoscopic unit cells in the constructed mesoscopic finite element damage evolution model, and obtains the stress output of the mesoscopic unit cells under different strains, that is, obtains the sample data for training the newly built neural network through finite element simulation calculation;

[0074] Step 122: Based on multiple groups of mesoscopic strain input data and corresponding stress output data, train the established neural network to obtain a mesoscopic stress transfer network;

[0075] This step 122 uses the sample data obtained in step 120 to train the newly established neural network. Different from the microscopic stress transfer network, the neural network learns the non-linear relationship between the input strain and output stress of the mesoscopic unit cell;

[0076] Step 124: Obtain the structural data of the macroscopic structural component of the composite material;

[0077] The structural data of the macroscopic structural component, such as size, geometric shape, etc., can be set according to actual simulation needs;

[0078] Step 126: Based on the structural data of the macroscopic structural component, construct a macroscopic finite element damage evolution model and set boundary conditions; the macroscopic finite element damage evolution model updates the stress at the integration points based on the mesoscopic stress transfer network.

[0079] In the above embodiments, information is transferred from the microscopic scale to the mesoscopic scale through one neural network (i.e., the microscopic stress transfer network), and information is transferred from the mesoscopic scale to the macroscopic scale through another neural network (i.e., the mesoscopic stress transfer network). When the macroscopic finite element damage evolution model performs finite element simulation calculations, stress updates are performed based on the mesoscopic stress transfer network, thus realizing the coupling between different scale models, saving computing resources while improving computing accuracy.

[0080] Optionally, step 102 includes:

[0081] Respectively obtain the experimental data of each component material in the composite material, such as data obtained from static experiments, fatigue experiments, and / or impact experiments, etc.;

[0082] Based on the experimental data of each component material, respectively determine the mechanical parameters of each component material;

[0083] Based on the mechanical parameters of each component material and the structural data of the microscopic unit cell, construct a microscopic finite element damage evolution model.

[0084] If an accurate microscopic finite element damage evolution model of the composite material is to be constructed, not only the structural data such as the proportion of each component material in the composite material needs to be determined, but also the mechanical parameters of each component material in the composite material, such as elastic modulus, strength, etc., need to be determined. Based on the experimental data of each component material, it is possible to better determine the numerical values of the required mechanical parameters, and the obtained model is more targeted and closer to the actual situation of the composite material. The experiments preferably refer to ASTM standards. Of course, in other embodiments, the mechanical parameters of each component material can also be determined in combination with existing literature.

[0085] Optionally, step 106 performs sampling within the load-bearing strain range of the microscopic unit cell, including:

[0086] Performing sampling within the load-bearing strain range of the microscopic unit cell by the Latin hypercube sampling method.

[0087] Correspondingly, step 118 performs sampling within the load-bearing strain range of the mesoscopic unit cell, also including:

[0088] Performing sampling within the load-bearing strain range of the mesoscopic unit cell by the Latin hypercube sampling method.

[0089] Using the Latin hypercube sampling method, a sufficient amount of strain input data can be obtained quickly.

[0090] Optionally, after step 102 and before step 108, it further includes:

[0091] Obtaining experimental data of the unidirectional tape made of composite materials; the structure of the unidirectional tape corresponds to the constructed microscopic finite element damage evolution model;

[0092] Based on the experimental data of the unidirectional tape, verifying the constructed microscopic finite element damage evolution model.

[0093] In the above embodiment, the unidirectional tape is a solid test piece, and the experimental data of the unidirectional tape is used to test whether the constructed microscopic finite element damage evolution model can accurately simulate the transfer relationship between strain and stress. The geometric structure of the microscopic unit cell corresponds to the geometric structure of the unidirectional tape. The cross-section of the unidirectional tape can be regarded as composed of the cross-sections of multiple microscopic unit cells. The microscopic unit cell plus periodic boundary conditions can simulate the unidirectional tape. Further, if the verification result shows that the microscopic finite element damage evolution model cannot accurately simulate the failure and damage evolution process of the unidirectional tape, it is necessary to return to step 102 to adjust the constructed microscopic finite element damage evolution model. By verifying the calculation results of the microscopic finite element damage evolution model with the actually obtained experimental data, not only can the simulation accuracy of the microscopic finite element damage evolution model be ensured, but also it can be ensured that the trained microscopic stress transfer network can accurately reflect the strain-stress relationship of the microscopic unit cell.

[0094] Optionally, after step 114 and before step 120, it further includes:

[0095] Obtaining experimental data of the laminated plate made of composite materials; the structure of the laminated plate corresponds to the constructed mesoscopic finite element damage evolution model;

[0096] Based on the experimental data of the laminated plate, verifying the constructed mesoscopic finite element damage evolution model.

[0097] In the above embodiments, the laminate is a solid test piece, and the experimental data of the laminate is used to verify whether the constructed mesoscopic finite element damage evolution model can accurately simulate the transfer relationship between strain and stress. The geometric structure of the mesoscopic unit cell corresponds to that of the unidirectional tape. The cross-section of the laminate can be regarded as composed of the cross-sections of multiple mesoscopic unit cells. By adding periodic boundary conditions to the mesoscopic unit cell, the laminate can be simulated. Further, if the verification result shows that the mesoscopic finite element damage evolution model cannot accurately simulate the failure and damage evolution process of the laminate, it is necessary to return to step 114 to adjust the constructed mesoscopic finite element damage evolution model.

[0098] Optionally, the method further includes:

[0099] Obtaining experimental data of a macroscopic structural member made of a composite material;

[0100] Based on the experimental data of the macroscopic structural member, verifying the constructed finite element damage evolution model.

[0101] The experimental data can determine the mechanical properties of the solid macroscopic structural member, such as various modulus and strength properties, etc. Combining with observation techniques such as DIC, the strain field distribution on the surface of the macroscopic structural member can be obtained. In the above embodiments, using the experimental data of the macroscopic structural member to verify whether the finite element damage evolution model is accurate can ensure that the obtained multi-scale model can effectively perform simulation calculations and provide technical support for composite material analysis.

[0102] In some alternative embodiments, as Figure 3 shown, if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcing phase and a matrix phase. The reinforcing phase is a fiber. For example, the reinforcing phase can be a carbon fiber, and the matrix phase can be a resin material, forming a carbon fiber-reinforced composite material;

[0103] The microscopic finite element damage evolution model can adopt a cuboid unit cell. The center and four corners of the cross-section of the cuboid unit cell are all center points. The center includes a complete fiber circular cross-section, and each of the four corners has a quarter fiber circular cross-section. That is to say, the center of the cuboid unit cell is a complete fiber, and each of the four corners is a quarter of a fiber. The microscopic finite element damage evolution model can also adopt a unit cell with randomly distributed fiber filaments in the finite element model.

[0104] Fiber-reinforced composite materials are a widely used type of composite material. The commonly used model is a cuboid unit cell with a complete fiber circular cross-section at the center and four symmetrically distributed quarter fiber circular cross-sections at the corners, which can be used to simulate fiber-reinforced composite materials with different materials and different ratios. The only difference lies in the specific materials and geometric dimensions of the fiber and matrix parts in the cuboid unit cell.

[0105] In some alternative embodiments, such as Figure 4 shown, if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcing phase and a matrix phase. The reinforcing phase can be selected as carbon fiber, and the matrix phase can be selected as a resin material. Fiber bundles are composed of fibers and the matrix phase, and the matrix phase is also filled between the fiber bundles (for the convenience of display, Figure 4 only the fiber bundles are shown in

[0106]

[0106] , and the matrix phase filled between the fiber bundles is not shown);

[0107] The mesoscopic finite element damage evolution model adopts a plate-shaped unit cell, which is composed of multiple fiber bundles woven together, and the void area is filled with the matrix phase.

[0108] The fiber bundle fabric structure is a widely used mesoscopic unit cell model. By changing the size, distribution of the fiber bundles, and the matrix material, different fiber-reinforced composite materials can be simulated.

[0109] In other embodiments, other forms of microscopic finite element damage evolution models and mesoscopic finite element damage evolution models can also be used to achieve the simulation of different composite materials.

[0109] An embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, a composite material multi-scale modeling method according to any embodiment of the present invention is implemented.

[0110] An embodiment of the present invention further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by the processor, the processor is enabled to execute a composite material multi-scale modeling method according to any embodiment of the present invention.

[0111] Specifically, a system or device equipped with a storage medium can be provided. A software program code for implementing the functions of any one of the above embodiments is stored on the storage medium, and the computer (or CPU or MPU) of the system or device is enabled to read and execute the program code stored in the storage medium.

[0112] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments. Therefore, the program code and the storage medium storing the program code constitute a part of the present invention.

[0113] Embodiments of the storage medium for providing the program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the program code can be downloaded from a server computer via a communication network.

[0114] In addition, it should be clear that not only can part or all of the actual operations be completed by executing the program code read by a computer, but also by an operating system operating on the computer based on the instructions of the program code, thereby implementing the functions of any one of the above embodiments.

[0115] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion module connected to the computer, and then based on the instructions of the program code, the CPU etc. installed on the expansion board or the expansion module execute part or all of the actual operations, thereby implementing the functions of any one of the above embodiments.

[0116] In summary, the present invention provides a multi-scale modeling method, device and medium for composite materials based on neural networks, which can realize the multi-scale simulation of composite materials, establish the quantitative relationship between the macroscopic properties of composite materials and the properties of their component materials and the structures at each scale, and reveal the internal mechanism that different organizational forms at each scale of the material lead to different macroscopic properties. The present invention regards the low-scale finite element model of the multi-scale model as a non-linear system, and uses a neural network to simulate the non-linear relationship of this system, that is, uses a neural network as the surrogate model of the low-scale model to transmit information to the high-scale model, which can achieve the effect of improving the coupling degree between different-scale models and improving the calculation accuracy while saving computing resources.

[0117] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0118] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk or optical disk that can store program code.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multi-scale modeling method for composite materials based on neural networks, characterized in that, Including: Obtaining the structural data of the microscopic unit cell of the composite material; Based on the structural data of the microscopic unit cell, constructing a microscopic finite element damage evolution model and setting periodic boundary conditions; Determining the loading strain ranges in six directions of the microscopic unit cell; Sampling within the loading strain range of the microscopic unit cell to obtain multiple groups of microscopic strain input data; each group of microscopic strain input data includes strain data in six directions of the microscopic unit cell; Using the multiple groups of microscopic strain input data as the input of the microscopic finite element damage evolution model, and through finite element simulation calculation, obtaining the corresponding stress output data; Based on the multiple groups of microscopic strain input data and the corresponding stress output data, training a neural network to obtain a microscopic stress transfer network; Obtaining the structural data of the mesoscopic unit cell of the composite material; Based on the structural data of the mesoscopic unit cell, constructing a mesoscopic finite element damage evolution model and setting periodic boundary conditions; The stress at the integration points of the mesoscopic finite element damage evolution model is updated based on the microscopic stress transfer network.

2. The method according to claim 1, characterized in that Also including: Determining the loading strain ranges in six directions of the mesoscopic unit cell; Sampling within the loading strain range of the mesoscopic unit cell to obtain multiple groups of mesoscopic strain input data; Each group of mesoscopic strain input data includes strain data in six directions of the mesoscopic unit cell; Using the multiple groups of mesoscopic strain input data as the input of the mesoscopic finite element damage evolution model, and through finite element simulation calculation, obtaining the corresponding stress output data; Based on the multiple groups of mesoscopic strain input data and the corresponding stress output data, training the constructed neural network to obtain a mesoscopic stress transfer network; Obtaining the structural data of the macroscopic structural component of the composite material; Based on the structural data of the macroscopic structural component, constructing a macroscopic finite element damage evolution model and setting boundary conditions; the stress at the integration points of the macroscopic finite element damage evolution model is updated based on the mesoscopic stress transfer network.

3. The method according to claim 2, characterized in that, The constructing a microscopic finite element damage evolution model based on the structural data of the microscopic unit cell includes: Respectively obtaining the experimental data of each component material in the composite material; Based on the experimental data of each component material, respectively determining the mechanical parameters of each component material; Based on the mechanical parameters of each component material and the structural data of the microscopic unit cell, constructing a microscopic finite element damage evolution model.

4. The method according to claim 3, characterized in that, After constructing the microscopic finite element damage evolution model based on the structural data of the microscopic unit cell and setting periodic boundary conditions, and before using the multiple groups of microscopic strain input data as the input of the microscopic finite element damage evolution model and obtaining the corresponding stress output data through finite element simulation calculation, it further includes: Obtaining the experimental data of the unidirectional tape made of the composite material; the structure of the unidirectional tape corresponds to the constructed microscopic finite element damage evolution model; Based on the experimental data of the unidirectional tape, verifying the constructed microscopic finite element damage evolution model.

5. The method according to claim 4, characterized in that, After constructing the mesoscopic finite element damage evolution model based on the structural data of the mesoscopic unit cell and setting periodic boundary conditions, and before using the multiple groups of mesoscopic strain input data as the input of the mesoscopic finite element damage evolution model and obtaining the corresponding stress output data through finite element simulation calculation, it further includes: Obtain experimental data of a laminate made of a composite material; the structure of the laminate corresponds to the constructed mesoscopic finite element damage evolution model; Based on the experimental data of the laminate, verify the constructed mesoscopic finite element damage evolution model.

6. The method according to claim 5, characterized in that, It further includes: Obtain experimental data of a macroscopic structural member made of a composite material; Based on the experimental data of the macroscopic structural member, verify the constructed finite element damage evolution model.

7. The method according to claim 1, wherein if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcing phase and a matrix phase, and the reinforcing phase is a fiber; The microscopic finite element damage evolution model uses a cuboid unit cell, and the centers and four corners of the cross-section of the cuboid unit cell are all center points. The center includes a complete fiber circular cross-section, and each of the four corners has a quarter fiber circular cross-section.

8. The method according to claim 2, wherein if the composite material is a fiber-reinforced composite material, the component materials are divided into a reinforcing phase and a matrix phase, and the reinforcing phase is a fiber. Fiber bundles are composed of fibers and the matrix phase, and the matrix phase is filled between the fiber bundles; The mesoscopic finite element damage evolution model uses a plate-shaped unit cell, and the plate-shaped unit cell is woven by multiple fiber bundles, and the void area is filled with the matrix phase.

9. An electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that, When the processor executes the computer program, the method described in any one of claims 1-8 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed on a computer, the computer is made to execute the method described in any one of claims 1-8.

Images