A machine learning-based design method for multi-layer composite structures

Abstract

The present invention discloses a design method for a multi-layer composite structure based on machine learning. A machine learning model for a multi-layer multi-component composite structure is established. The one-dimensional data matrix of the training set m is input into the machine learning model of the multi-layer multi-component composite structure to obtain the mechanical properties of the training set m. The mapping relationship between the one-dimensional data matrix and the mechanical properties of the training set m is optimized using the target mechanical properties of the training set m. Then, the generalization ability of the machine learning model of the multi-layer composite structure is verified using the test set n, completing the design of the multi-layer composite structure. Using the present invention, the material properties can be predicted quickly and accurately, and the material structure can be optimized. By obtaining a large amount of material data and performance parameters, through the simulation and prediction of unknown structures, their performance can be quickly evaluated, and the machine learning model can be used to adjust the material structure to meet higher performance requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of composite material structures, and particularly to a design method for multi-layer composite structures based on machine learning. Background Art

[0002] Composite materials are composed of a combination of two or more different materials and have excellent mechanical properties, chemical stability, and thermal properties. Therefore, they are widely used in many engineering applications. However, traditional composite material design and preparation methods usually rely on experience and trial-and-error, which limits the further improvement and innovation of material properties.

[0003] With the rapid development of machine learning and artificial intelligence technologies, scientists have begun to apply them to the field of material research to accelerate the process of material design and optimization. Machine learning can establish an association model between material structures and properties by analyzing a large amount of material data, structural features, and performance parameters, and use these models to predict and optimize the properties of new materials. The material design method based on machine learning can accelerate the material R & D process, reduce the trial-and-error cost, and achieve precise control and optimization of material properties. This method has broad application prospects and can promote the development and application of new materials in fields such as aerospace, automotive engineering, and electronic devices. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a design method for multi-layer composite structures based on machine learning.

[0005] To solve the above technical problem, the present invention provides a design method for multi-layer composite structures based on machine learning. The composite structure is composed of I-shaped unit cells, and a single-layer composite structure is composed of at least two of the composite structures made of materials with different properties. The method includes the following steps:

[0006] Step 1: Obtain the size of the single-layer composite structure, and obtain the size, planar shape, and material properties of the I-shaped unit cell.

[0007] Step 2: According to the parameters in Step 1, use a two-dimensional data matrix to generate a random structure of the single-layer composite structure, denoted as dataset t, and divide the dataset t into a training set v and a test set u.

[0008] Step 3: Conduct finite element analysis on the training set v to obtain the target mechanical properties of the training set v.

[0009] Step 4: According to the characteristics of the training set v, establish a machine learning model for the single-layer multi-component composite structure to obtain the mechanical properties of the training set v, and use the target mechanical properties of the training set v to optimize the mapping relationship between the characteristics and mechanical properties of the training set v.

[0010] Step 5: Perform finite element analysis on the test set u to obtain the target mechanical properties of the test set u. Input the features of the test set u into the machine learning model of the single-layer multi-component composite structure, compare the mechanical properties of the test set u with the target mechanical properties, and verify the generalization ability of the machine learning model of the single-layer multi-component composite structure;

[0011] Step 6: Generate a dataset s of one-dimensional data matrices for the multi-layer composite structure according to the physical properties of the single-layer composite structure. Divide the dataset s into a training set m and a test set n, and perform finite element analysis on the training set m and the test set n respectively to obtain the target mechanical properties of the training set m and the test set n;

[0012] Step 7: Establish a machine learning model for the multi-layer multi-component composite structure according to the dataset s. Input the one-dimensional data matrix of the training set m into the machine learning model of the multi-layer multi-component composite structure to obtain the mechanical properties of the training set m. Use the target mechanical properties of the training set m to optimize the mapping relationship between the one-dimensional data matrix and the mechanical properties of the training set m, and then use the test set n to verify the generalization ability of the machine learning model of the multi-layer multi-component composite structure, thus completing the design of the multi-layer multi-component composite structure.

[0013] Preferably, the dimensions of the I-shaped unit cell include the width, height, web thickness, and flange thickness of the I-shaped unit cell. The planar shape of the I-shaped unit cell includes a straight shape and a serrated shape, and the serrated shape includes a rectangular serrated shape, a triangular serrated shape, and a corrugated serrated shape.

[0014] Preferably, the aspect ratio of the single-layer composite structure is not less than 5:1.

[0015] Preferably, the ratio of the training set v to the test set u and the ratio of the training set m to the test set n are both 4:1.

[0016] Preferably, the dataset t is not less than 5000.

[0017] Preferably, the mechanical properties and the target mechanical properties include toughness, compressive strength, bending strength, and shear strength.

[0018] Preferably, the features include material properties and data matrices.

[0019] Preferably, the number of layers of the multi-layer composite structure is not less than 2 layers. [[ID=…]]

[0020] Preferably, both the machine learning model of the single-layer multi-component composite structure and the machine learning model of the multi-layer multi-component composite structure are supervised machine learning.

[0021] Preferably, the machine learning model of the single-layer multi-element composite structure and the machine learning model of the multi-layer multi-element composite structure include a convolutional neural network, a deep neural network, and a recurrent neural network.

[0022] Implementing the present invention has the following beneficial effects:

[0023] The present invention can quickly and accurately predict material properties and optimize material structures. By obtaining a large amount of material data and performance parameters, using feature extraction and selection methods, an association model between material properties and features is established. Then, by simulating and predicting unknown structures, their performance can be quickly evaluated, and the machine learning model can be used to adjust the material structure to meet higher performance requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flowchart of the method according to an embodiment of the present invention;

[0025] Figure 2 is a schematic diagram of the continuous combination of unit components according to an embodiment of the present invention;

[0026] Figure 3 is a schematic diagram of the straight I-shaped unit cell and the zigzag I-shaped unit cell structures according to an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of the dimensions of the straight I-shaped unit component and its 90° rotation according to an embodiment of the present invention;

[0028] Figure 5 is a schematic diagram of the dimensions of the zigzag I-shaped unit component and its 90° rotation according to an embodiment of the present invention;

[0029] Figure 6 is a schematic diagram of the dimensions of the straight I-shaped unit cell and the zigzag I-shaped unit cell according to an embodiment of the present invention;

[0030] Figure 7 is a schematic cross-sectional view of a single-layer composite structure according to an embodiment of the present invention;

[0031] Figure 8 is Figure 7 a schematic diagram of the transformed two-dimensional data matrix;

[0032] Figure 9 is a flowchart of the machine learning model of the single-layer multi-element composite structure according to an embodiment of the present invention;

[0033] Figure 10 is a schematic diagram of a composite structure with high compressive performance according to an embodiment of the present invention;

[0034] Figure 11 [[ID=`54]]is a schematic diagram of the hybrid stacked multi-layer composite structure according to an embodiment of the present invention;

[0035] Figure 12 For Figure 11 Schematic diagram of the transformed one-dimensional data matrix. Detailed implementation manners

[0036] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0037] As Figure 1 shown, a design method for a multi-layer composite structure based on machine learning, the composite structure is composed of I-shaped unit cells, and the single-layer composite structure is composed of the composite structures of at least two materials with different properties, including the following steps:

[0038] Step 1: Obtain the size of the single-layer composite structure, and obtain the size, planar shape and material properties of the I-shaped unit cell.

[0039] As Figure 2 shown, the composite structure is composed of a hard material and a soft material. Preferably, the soft material uses TPU with an elastic modulus of 1000 Mpa, and the hard material uses PLA with an elastic modulus of 3000 Mpa. Each independent composite structure is a unit component, and the hard material and the soft material at the edge of the composite structure can be connected to the same type of material in adjacent unit components. The toughness, strength and stiffness of the composite structure in the plane depend on the size of the I-shaped unit cell of the composite structure, the planar shape of the I-shaped unit cell and the material properties. Among them, the size includes width, height, waist thickness and web thickness, and the planar shape includes straight shape, rectangular serrated shape, triangular serrated shape, corrugated serrated shape, etc.

[0040] Step 2: According to the parameters in Step 1, use a two-dimensional data matrix to generate a random structure of the single-layer composite structure, denoted as dataset t, and divide dataset t into a training set v and a test set u.

[0041] As Figures 3 to 7 shown, four unit components can be randomly combined into any single-layer composite structure, and the unit components are continuously connected. The aspect ratio of the single-layer composite structure is not less than 5:1. Preferably, the size of the single-layer composite structure is 64 mm * 64 mm * 2 mm. The single-layer composite structure can be arranged and combined by four unit components in an 8 * 8 manner to generate 5000 random composite structures. Among them, 4000 composite structures are used as the training set, and the remaining 1000 composite structures are used as the training set.

[0042] As Figure 8 shown, use 1, 2, 3, and 4 respectively as Figure 3 and Figure 4Labels for four types of unit components are used, and a two-dimensional data matrix of 1, 2, 3, 4 is used to exhaust every combination of the single-layer composite structure, forming a data matrix of the single-layer composite structure.

[0043] Step 3: Perform finite element analysis on the training set v to obtain the target mechanical properties of the training set v.

[0044] Based on the training set v, perform finite element simulation analysis on 4,000 single-layer composite structures of the training set to obtain the target mechanical properties. The target mechanical properties include toughness, compressive strength, bending strength, and shear strength. The obtained target mechanical properties are used as supervision signals to help the neural network model learn the mapping relationship between the input features and the output targets, and thereby improve the accuracy of the prediction results.

[0045] Step 4: According to the features of the training set v, establish a machine learning model for the single-layer multi-component composite structure to obtain the mechanical properties of the training set v, and use the target mechanical properties of the training set v to optimize the mapping relationship between the features and the mechanical properties of the training set v.

[0046] As Figure 9 shown, select a machine learning algorithm. The machine learning algorithms include neural network, linear regression, and support vector machine. Preferably, the neural network is a convolutional neural network, a deep neural network, or a recurrent neural network. Use the material properties of the hard material and the soft material and the two-dimensional data matrix of the training set v in Step 2 as the input features of the CNN model, and use the mechanical properties in Step 3 as the learning objective. Initialize the parameters of the machine learning algorithm, define the loss function, select the optimization algorithm, and perform backpropagation. Conduct multiple rounds of training and optimize the mapping relationship between the input features and the mechanical properties of the CNN model to obtain a machine learning model for the single-layer multi-component composite structure. In the machine learning model of the single-layer multi-component composite structure, the input features first undergo a convolution operation through the first convolutional layer, and then are sampled through the first pooling layer to extract the feature map. Next, the feature map is processed through the second convolutional layer and the second pooling layer. Subsequently, all feature maps are processed into a one-dimensional vector through the global average layer. Finally, this one-dimensional vector is input into the fully connected layer to perform classification, regression, or other tasks.

[0047] Step 5: Perform finite element analysis on the test set u to obtain the target mechanical properties of the test set u. Input the features of the test set u into the machine learning model of the single-layer multi-component composite structure, compare the mechanical properties of the test set u with the target mechanical properties, and verify the generalization ability of the machine learning model of the single-layer multi-component composite structure.

[0048] Perform finite element analysis on 1,000 composite structures in the test set u to calculate the mechanical properties of this part. At the same time, input the material properties of the hard and soft materials of the single-layer composite structure in the test set and the two-dimensional data matrix into the machine learning model of the single-layer multi-component composite structure in Step 4 to predict the target mechanical properties. Compare the mechanical properties predicted by the machine learning model of the single-layer multi-component composite structure with the target mechanical properties calculated by finite element analysis to evaluate the performance of the machine learning model of the single-layer multi-component composite structure, and predict unknown data to verify the generalization ability of the machine learning model of the single-layer multi-component composite structure.

[0049] Step 6: Generate a data set s of one-dimensional data matrices for the multi-layer composite structure according to the physical properties of the single-layer composite structure. Divide the data set s into a training set m and a test set n, and perform finite element analysis on the training set m and the test set n to obtain the target mechanical properties of the training set m and the test set n respectively.

[0050] Through the machine learning model of the single-layer multi-component composite structure, four single-layer composite structures with the highest properties in terms of toughness, compressive strength, bending strength, and shear strength can be obtained. These four single-layer composite structures have good performance in terms of fracture resistance, compression resistance, bending resistance, and shear resistance respectively.

[0051] Such as Figure 10 and 11 As shown, arrange N layers of the high-toughness single-layer composite structure and the high-compressive-strength single-layer composite structure at intervals. Preferably, N is 21, to obtain a composite structure with high strength in both toughness and compressive strength. Similarly, replace the high-compressive-strength single-layer composite structure with a high-bending-strength or high-shear-strength single-layer composite structure, then a composite structure with high properties in both toughness and bending strength, or a composite structure with high properties in both toughness and shear strength can be obtained. The three composite structures have very high compression resistance, bending resistance, and shear resistance respectively. At the same time, these three composite structures all have very high fracture resistance, achieving the purpose of designing a multi-component composite structure.

[0052] Step 7: Establish a machine learning model for the multi-layer multi-component composite structure according to the data set s. Input the one-dimensional data matrix of the training set m into the machine learning model of the multi-layer multi-component composite structure to obtain the mechanical properties of the training set m. Use the target mechanical properties of the training set m to optimize the mapping relationship between the one-dimensional data matrix and the mechanical properties of the training set m, and then use the test set n to verify the generalization ability of the machine learning model of the multi-layer multi-component composite structure to complete the design of the multi-layer composite structure.

[0053] Such as Figure 9As shown, a machine learning model with a multi-layer and multi-element composite structure is established, which has the same feature extraction method, the same model architecture, and the same optimization algorithm as the machine learning model with a single-layer and multi-element composite structure in Step 4, and uses a one-dimensional data matrix of the multi-layer composite structure as the input parameter. As Figure 12 As shown, the single-layer composite structures with high toughness, high compressive strength, high bending strength, and high shear strength are respectively marked with a, b, c, and d to establish a one-dimensional data matrix of the multi-layer composite structure. The target mechanical properties of the data set s are analyzed by finite element simulation to obtain the mechanical properties of the training set m. The mapping relationship between the one-dimensional data matrix and the mechanical properties of the training set m is optimized using the target mechanical properties of the training set m, and then the generalization ability of the machine learning model with the multi-layer and multi-element composite structure is verified using the test set n to obtain the design scheme of the multi-layer and multi-element composite structure with the best comprehensive performance.

[0054] The foregoing disclosure is only a preferred embodiment of the present invention, and of course it cannot be used to limit the scope of the rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for designing a multi-layer composite structure based on machine learning, wherein the composite structure is composed of an I-shaped unit cell, and the single-layer composite structure is composed of the composite structure of at least two materials with different properties, characterized in that: The following steps are involved: Step 1: Obtain the dimensions of the single-layer composite structure, and obtain the dimensions, planar shape, and material properties of the I-shaped unit cell; the dimensions of the I-shaped unit cell include the width, height, and waist thickness of the I-shaped unit cell; the planar shape of the I-shaped unit cell includes a straight shape and a zigzag shape; the zigzag shape includes a rectangular zigzag shape, a triangular zigzag shape, and a corrugated zigzag shape; Step 2: Based on the parameters of step 1, a two-dimensional data matrix is used to generate a random structure of the single-layer composite structure, which is set as a data set t, and the data set t is divided into a training set v and a test set u; Step 3: performing finite element analysis on the training set v to obtain target mechanical properties of the training set v; the mechanical properties and the target mechanical properties include toughness, compressive strength, bending strength, and shear strength; Step 4: Based on the characteristics of the training set v, a machine learning model of a single-layer multi-component composite structure is established to obtain the mechanical properties of the training set v, and the target mechanical properties of the training set v are used to optimize the mapping relationship between the characteristics of the training set v and the mechanical properties; Step 5: Perform finite element analysis on the test set u to obtain target mechanical properties of the test set u, input the features of the test set u into the machine learning model of the single-layer multi-component composite structure, compare the mechanical properties of the test set u with the target mechanical properties, and verify the generalization ability of the machine learning model of the single-layer multi-component composite structure; Step 6: Generate a one-dimensional data matrix dataset s of the multi-layer composite structure based on the physical properties of the single-layer composite structure, divide the dataset s into a training set m and a test set n, perform finite element analysis on the training set m and the test set n, and obtain target mechanical properties of the training set m and the test set n, respectively; Step 7: Based on the data set s, a machine learning model of a multi-layer multi-component composite structure is established, and the one-dimensional data matrix of the training set m is input into the machine learning model of the multi-layer multi-component composite structure to obtain the mechanical properties of the training set m. The mapping relationship between the one-dimensional data matrix and the mechanical properties of the training set m is optimized using the target mechanical properties of the training set m. The generalization ability of the machine learning model of the multi-layer multi-component composite structure is then verified using the test set n to complete the design of the multi-layer composite structure.

2. The multi-layer composite structure design method based on machine learning according to claim 1, characterized in that: The length-to-thickness ratio of the single-layer composite structure is not less than 5:

1.

3. The multi-layer composite structure design method based on machine learning according to claim 1, characterized in that: The ratio of the training set v to the test set u, and the ratio of the training set m to the test set n are both 4:

1.

4. The multi-layer composite structure design method based on machine learning according to claim 1, characterized in that: The data set t is no less than 5000.

5. The multi-layer composite structure design method based on machine learning according to claim 1, characterized in that: The multi-layer composite structure has no less than 2 layers.

6. The method for designing a multi-layer composite structure based on machine learning according to claim 1, wherein: The machine learning model of the single-layer multi-component composite structure and the machine learning model of the multi-layer multi-component composite structure are both supervised machine learning.

7. The multi-layer composite structure design method based on machine learning according to claim 1, characterized in that: The machine learning model of the single-layer multi-component composite structure and the machine learning model of the multi-layer multi-component composite structure include convolutional neural networks, deep neural networks, and recursive neural networks.

Images