A design method of a machine learning based speckle type penetration resistant composite structure and a turing speckle type composite structure

Abstract

The embodiment of the present application discloses a composite structure design method based on machine learning and a Turing pattern composite structure, which is formed by a plurality of single-layer plate shell structures spirally stacked with a certain rotation angle to form a three-dimensional overall structure, and the single-layer plate shell structures are connected by homogeneous thin layers, and the single-layer plate shell structures are embedded with two materials with different elastic modulus, i.e., different hardness, according to a certain geometric shape, a certain width, a certain Turing pattern model, and a certain Turing pattern diffusion time. The embodiment of the present application also discloses a design method for designing the above-mentioned composite material, which is based on machine learning to realize the prediction of material structure parameters based on different Turing patterns and parameter transformation of the single-layer structure composite process, and then realizes intelligent design of the composite material.

Description

Technical Field

[0001] This invention relates to the field of composite structure technology, and in particular to a design method for a pattern-based anti-penetration composite structure based on machine learning, as well as a Turing pattern-based composite structure. Background Technology

[0002] In daily production, with the improvement of safety awareness, the demand for the safety and protective performance of personnel, materials and facilities is also gradually increasing.

[0003] Penetration-resistant protective structures have found wider applications in people's daily lives and various engineering fields. These penetration-resistant structures are not only used to protect various types of equipment and infrastructure from damage caused by natural disasters and malicious attacks, but are also commonly used in people's daily lives, such as motorcycle helmets and car windshields.

[0004] Based on the above requirements, researchers have designed a variety of structures to improve the penetration resistance of structures, such as overlapping structures, stacked spiral structures, and layered structures. Among these composite structures, the structure with alternating soft and hard materials often has a more regular arrangement. When subjected to load impact, this structure consumes energy through crack propagation and interlayer shear to achieve the goal of impact resistance.

[0005] To improve the penetration resistance of structures, researchers have tried many methods, and Turing patterns are one of them. Turing patterns are a common nonlinear dynamic evolution phenomenon in nature. Their patterns are spatial patterns generated by chemical reactions or physical processes. These patterns are usually driven by local interactions and dissipation mechanisms rather than by external force fields. Using the complex geometric patterns of Turing patterns to design interlocking topologies between soft and hard materials can improve the geometric complexity of crack propagation paths in soft-hard composite structures, thereby increasing the overall toughness of the composite structure. This is of great significance for the design of penetration-resistant and impact-resistant composite structures.

[0006] However, since the patterns of Turing patterns are diverse, and the material stacking process is affected by various parameters such as geometry, width, thickness, and rotation angle, the process of exploring the combination of Turing patterns and single-layer materials into composite materials will produce completely different structural parameters under the influence of different Turing patterns and stacking parameters. For such a large number of samples, it is obviously unrealistic for researchers to conduct research through traditional experimental methods. For this reason, existing research on Turing patterns is often limited to single-layer structures, and existing research has not extended Turing patterns to the scope of multi-scale composite structures.

[0007] To overcome the above problems, this invention proposes a design method for a pattern-based anti-penetration composite structure based on machine learning, and a Turing pattern-based composite structure designed using this method. Summary of the Invention

[0008] The technical problem to be solved by the embodiments of the present invention is to provide a design method for a pattern-based anti-penetration composite structure based on machine learning, comprising the following steps:

[0009] Step S1: Generate Turing pattern images by adjusting three pattern models: BZ oscillation response, Gray Scott model, and LE model. Then, convert the generated Turing pattern images into binary high-precision pixel images, thereby generating binary matrix digital images representing the interior and exterior of the Turing pattern regions.

[0010] Step S2: Assign geometric dimensions to the binary matrix digital image generated in step S1 to create a three-dimensional representation, and assign soft and hard materials according to the values ​​of the binary matrix to construct a single-layer anti-penetration structure geometric entity containing two materials and the corresponding geometric data matrix. Based on the Turing pattern image described in step S1, divide the single-layer anti-penetration structure into a training set and a test set.

[0011] Step S3: Based on the training set in S2, perform finite element analysis on the pattern-type anti-penetration single-layer and multi-layer structures of the training set to construct a learning target data matrix. Use the mechanical properties of the target as a supervision signal to help the machine learning model learn the mapping relationship between the input features and the output target.

[0012] Step S4: Construct a machine learning model, using the single-layer and multi-layer structural parameters of the training set as inputs to the machine learning model, and the geometric data matrix set of the learning target in Step S3 as the target. Perform multiple rounds of training and optimize the mapping relationship between the input features of the machine learning model and the learning target to obtain a machine learning model for designing a pattern-type anti-penetration composite structure.

[0013] Step S5: Perform finite element analysis on the single-layer composite structure and multi-layer composite structure of the test set to calculate the target mechanical properties of the structure. At the same time, use the single-layer structure parameters and multi-layer structure parameters of the test set as input parameters of the machine learning model to predict its mechanical properties. Compare the target mechanical properties predicted by the machine learning model obtained in step S4 with the mechanical properties calculated by finite element analysis to verify the generalization ability of the machine learning model.

[0014] Step S6: Using the machine learning model described in Step 5, predict the mechanical properties of single-layer and multi-layer composite structures based on Turing patterns, and design pattern-type single-layer and multi-layer anti-penetration structures with higher anti-penetration capabilities based on the prediction results.

[0015] In step S1, the total number of Turing patch images generated is greater than 10,000.

[0016] In step S2, the ratio of the number of training sets to the number of test sets is 7:3.

[0017] The single-layer structure parameters mentioned in steps S4 and S5 include: the data matrix of the single-layer structure and the soft and hard material properties of the structure; the multi-layer structure parameters mentioned in steps S4 and S5 include: the data matrix of the multi-layer structure, the soft and hard material properties of the structure, and the stacking form.

[0018] The structural soft and hard material properties in the single-layer and multi-layer structural parameters include: density, elastic modulus, Poisson's ratio, and yield strength; the stacking form in the multi-layer structural parameters includes: number of stacked layers, thickness of each layer, and angle of each layer.

[0019] Furthermore, this invention also proposes a Turing pattern composite structure based on the above design method. The Turing pattern composite structure is formed by spirally stacking multiple single-layer shell-like structures. The multiple single-layer shell-like structures are interconnected by homogeneous thin layers. Each single-layer shell-like structure is formed by interlocking two materials with different elastic moduli, i.e., different hardness, according to the Turing pattern model.

[0020] The Turing pattern model type can be any one or more of the following: BZ oscillatory response, Gray Scott model, LE model, etc.

[0021] In each of the single-layer shell-like structures, the diffusion region of the Turing pattern is filled with soft material, and the undiffused region is filled with hard material, wherein the ratio of the elastic modulus of the hard material to that of the soft material is at least 2:1.

[0022] The homogeneous thin layer connecting the single-layer shell-like structures can be made of the soft material, or it can be made of a third material with an elastic modulus of 1 to 2 times that of the soft material, and its thickness can be 5% to 20% of the thickness of the single-layer shell-like structure in contact with the homogeneous thin layer.

[0023] The multi-layered shell structure is formed by stacking at least three single-layered shell structures.

[0024] Implementing the embodiments of the present invention has the following beneficial effects: ① The present invention provides a penetration-resistant composite structure. Due to the design of Turing patterns with different models and diffusion times and the filling with soft materials, when the structure is subjected to a large load and cracks are generated, the propagation path of the crack is greatly increased, thereby dissipating more energy. On the other hand, since the structure is a multi-layer rotational design, this structure increases the number of deflections when the crack propagates, thereby increasing the dissipated energy and greatly improving the overall penetration resistance of the structure.

[0025] ② This invention provides a machine learning-based design method for patterned anti-penetration composite structures. The design process features multiple adjustable structural parameters, allowing adjustment of both individual layer parameters and parameters between layers. This provides rich and flexible parameter adjustment dimensions, facilitating the investigation of the impact of different parameter adjustments on the structural parameters of single-layer and multi-layer structures. It can meet practical needs such as daily life protection, national defense explosion protection, and aerospace impact resistance. ③ This invention provides a design method for anti-penetration composite structures. By combining finite element analysis and machine learning algorithms when designing patterned single-layer structures, it not only predicts the mechanical properties of various patterned single-layer and multi-layer composite structures but also optimizes specified mechanical properties, thereby improving the impact resistance of single-layer and multi-layer composite structures. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of a patterned anti-penetration composite structure provided by the present invention;

[0027] Figure 2 These are the three types of Turing pattern models provided by this invention;

[0028] Figure 3 A schematic diagram of the anti-penetration composite structure of Gray Scott pattern-filled soft material under different parameters provided by the present invention;

[0029] Figure 4 A schematic diagram of an anti-penetration composite structure formed by rotating a single-layer plate shell structure 1 of uniform thickness 6 layers at a 30° angle, as provided in an embodiment of the present invention.

[0030] Figure 5 A schematic diagram of an anti-penetration composite structure consisting of five single-layer plate shells with gradually varying thicknesses from top to bottom, provided in an embodiment of the present invention;

[0031] Figure 6 This invention provides a machine learning-based method for designing pattern-type anti-penetration composite structures.

[0032] Figure 7 This is an example diagram of the Turing pattern and its transformed binary matrix from Embodiment 2 of the present invention;

[0033] Figure 8 This is a schematic diagram of the convolutional neural network structure in Embodiment 2 of the present invention. Detailed Implementation

[0034] 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.

[0035] Example 1

[0036] This embodiment mainly discloses the structure of a Turing pattern anti-penetration composite structure, such as... Figure 1 As shown in Figure a, the overall structure is formed by spirally stacking single-layer shell-like structures, with each single-layer shell-like structure connected to the others by a homogeneous thin layer. The upper and lower surfaces of the composite structure are covered with an outer shell.

[0037] like Figure 1 As shown in b, a single-layer plate structure based on a Gray Scott pattern is shown. The single-layer plate shell structure is composed of hard and soft materials interlocked with patterns generated by several Turing pattern models, wherein the soft material fills the diffused areas and the hard material fills the undiffused areas.

[0038] In this embodiment, the Turing pattern is generated by three models: the BZ oscillatory response, the Gray Scott model, and the LE model. Figure 2 These are Turing patterns generated by these three models under a certain parameter, among which... Figure 2 'a' is generated by the BZ oscillation reaction. Figure 2 b is generated by the Gray Scott model. Figure 2 c is generated by the LE model.

[0039] In this embodiment, three penetration-resistant composite structures filled with soft material using Gray Scott patterning are also provided, each with different parameters. These three patterns exhibit variations in average width and line density of the pattern curves at the same size. Figure 3 We present soft materials with patterned composite structures under three different grid numbers and diffusion time parameters.

[0040] In the single-layer plate structure, such as Figure 1 In the b-Turing pattern, the diffusion region is filled with a softer material, and the non-diffusion region is filled with a harder material. The elastic modulus of the soft and hard materials can be reasonably adjusted, provided that the elastic modulus of the harder material is at least 1.5 times that of the softer material.

[0041] For example, under this premise, TPU with a softer material of 800 MPa is preferred, and PLA with a harder material of 3000 MPa is preferred. These two materials are used to fill different areas of the Gray Scott pattern.

[0042] In this embodiment, in the patterned multilayer composite structure, each single-layer structure can be filled with a homogeneous thin layer. The homogeneous thin layer can be filled with a material with an elastic modulus similar to that of the soft material, or it can be filled directly with a material. The thickness of the homogeneous thin layer must be less than 1 / 5 of the thickness of the adjacent single-layer shell structure.

[0043] In this embodiment, each layer of the patterned multilayer composite structure can be spirally stacked at a certain angle. The spiral angle of a single layer can be arbitrary, and can be rotated at a fixed angle, at a uniformly varying angle, or at a linearly varying angle. Regardless of the rotation method, the patterned multilayer composite structure can achieve crack propagation and deflection when subjected to load, thereby improving the structure's impact resistance.

[0044] like Figure 4 The diagram shown is a schematic of an anti-penetration composite structure formed by rotating a single-layer shell structure of 6 uniform thicknesses at a 30° angle, as provided in an exemplary embodiment of the present invention.

[0045] In this embodiment, the number of layers in the patterned multilayer composite structure can be selected in various ways, but in order to achieve a better impact resistance, the composite structure must have at least three layers, which can be adjusted according to the actual situation.

[0046] Furthermore, the thickness of each layer can be the same or different. In the multilayer composite structure of this invention, the layer thickness can be the same, can be gradient decreasing, can be gradient increasing, or can be arranged in a cross-mixed manner, for example... Figure 5 The diagram shown is a schematic of a penetration-resistant composite structure consisting of five single-layer shell structures with gradually varying thicknesses from top to bottom, provided by an embodiment of the present invention.

[0047] In this embodiment, an outer shell can be attached to the outer surface of the multi-layer composite structure. The material and thickness of the shell can be varied, and specific options can be selected based on the circumstances to maximize the penetration resistance of the composite structure. For example, in high-temperature and high-pressure working environments, heat-insulating and pressure-resistant materials should be preferred for the shell.

[0048] Example 2

[0049] This embodiment mainly discloses a design method for anti-penetration structures based on machine learning, such as... Figure 6 As shown, the specific steps include:

[0050] Step S1: Generate a Turing map image of a certain size by adjusting the three map models: BZ oscillation response, Gray Scott model, and LE model. Convert the generated Turing map image into a binary high-precision pixel image, and then generate a binary matrix digital image representing the interior and exterior of the Turing map region.

[0051] Specifically, the total number of patch images generated in step S1 is greater than 10,000.

[0052] Step S2: Assign geometric dimensions to the binary matrix digital image generated in step S1 to create a three-dimensional representation. Based on the values ​​of the binary matrix, assign soft and hard materials respectively to construct a single-layer anti-penetration structure geometric entity containing two materials and the corresponding geometric data matrix. Based on the Turing pattern image described in step S1, divide the single-layer anti-penetration structure into a training set and a test set.

[0053] Specifically, the ratio of the training set to the test set in step S2 is 7:3.

[0054] Specifically, such as Figure 7 As shown, in the Turing pattern, the diffused regions are filled with soft material, and the non-diffused regions are filled with hard material. Each pixel in the pattern image is converted to either 0 or 1, where the soft material regions (diffused regions) are 1, and the hard material regions (diffused regions) are 0. This method transforms the pattern image into a penetration-resistant single-layer geometric data matrix. Figure 7 d represents a 20*20 matrix of 0 and 1 data.

[0055] Step S3: Based on the training set in S2, perform finite element simulation analysis on the Turing pattern anti-penetration single-layer structure and its superimposed multi-layer composite structure of the training set. Construct the learning target data matrix for the single-layer structure and the multi-layer structure respectively, and use the mechanical properties of the target as the supervision signal to help the machine learning model learn the mapping relationship between the input features and the output target.

[0056] Step S4: Construct a machine learning model. Use the data matrix of the single-layer structure and the material properties of the soft and hard materials of the structure in the training set as input to the machine learning model. Use the data matrix of the multi-layer structure, the material properties of the soft and hard materials of the structure, and the superposition form of the multi-layer structure as input to the machine learning model. Use the data matrix of the learning target of the single-layer structure and the multi-layer structure in Step S3 as the target. Perform multiple rounds of training and optimize the mapping relationship between the input features of the machine learning model and the learning target to obtain a machine learning model for designing a Turing pattern anti-penetration composite structure.

[0057] Specifically, the multi-layer structure stacking form includes: the number of stacked layers, the thickness of each layer, the stacking angle of each layer, etc.

[0058] like Figure 8The diagram shows a schematic of a convolutional neural network. In this network, the input first undergoes a convolution operation through the first convolutional layer, and then is sampled through the first pooling layer to extract feature maps. Next, the feature maps are processed through the second convolutional layer and the second pooling layer. Then, all feature maps are unfolded into a one-dimensional vector through a global averaging layer, and finally input into a fully connected layer for tasks such as classification or regression.

[0059] Step S5: Perform finite element analysis on the single-layer and multi-layer composite structures in the test set to calculate the target mechanical properties of these structures. Simultaneously, use the data matrix of the single-layer structure and the material properties of the structural soft and hard materials from the training set as input to the machine learning model. Use the data matrix of the multi-layer structure, the material properties of the structural soft and hard materials, and the stacking form of the multi-layer structure (number of layers, thickness of each layer, angle of each layer, etc.) as input to the machine learning model. Predict the mechanical properties of the single-layer and multi-layer structures. Compare the target mechanical properties predicted by the machine learning model obtained in Step S4 with the mechanical properties calculated by the finite element analysis to verify the generalization ability of the machine learning model.

[0060] Specifically, the structural soft and hard material properties in step 5 include the density, elastic modulus, Poisson's ratio, and yield strength of single-layer composite structures and multi-layer composite structures.

[0061] Step S6: Using the machine learning model described in Step 5, predict the mechanical properties of single-layer and multi-layer composite structures based on Turing patterns, and design pattern-type single-layer and multi-layer anti-penetration structures with higher anti-penetration capabilities based on the prediction results.

[0062] Of course, the above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be used to limit the scope of protection of the present invention. All modifications made according to the spirit and essence of the main technical solution of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A design method for a Turing pattern anti-penetration composite structure based on machine learning, characterized in that, Specifically, the following steps are included: Step S1: Generate Turing pattern images by adjusting three pattern models: BZ oscillation response, Gray Scott model, and LE model. Then, convert the generated Turing pattern images into binary high-precision pixel images, thereby generating binary matrix digital images representing the interior and exterior of the Turing pattern regions. Step S2: Assign geometric dimensions to the binary matrix digital image generated in step S1 to create a three-dimensional representation, and assign soft and hard materials according to the values ​​of the binary matrix to construct a single-layer anti-penetration structure geometric entity containing two materials and the corresponding geometric data matrix. Based on the Turing pattern image described in step S1, divide the single-layer anti-penetration structure into a training set and a test set. Step S3: Based on the training set in S2, perform finite element analysis on the pattern-type anti-penetration single-layer and multi-layer structures of the training set to construct a learning target data matrix. Use the mechanical properties of the target as a supervision signal to help the machine learning model learn the mapping relationship between the input features and the output target. Step S4: Construct a machine learning model, using the single-layer and multi-layer structural parameters of the training set as inputs to the machine learning model, and the geometric data matrix set of the learning target in Step S3 as the target. Perform multiple rounds of training and optimize the mapping relationship between the input features of the machine learning model and the learning target to obtain a machine learning model for designing a pattern-type anti-penetration composite structure. Step S5: Perform finite element analysis on the single-layer composite structure and multi-layer composite structure of the test set to calculate the target mechanical properties of the structure. At the same time, use the single-layer structure parameters and multi-layer structure parameters of the test set as input parameters of the machine learning model to predict its mechanical properties. Compare the target mechanical properties predicted by the machine learning model obtained in step S4 with the mechanical properties calculated by finite element analysis to verify the generalization ability of the machine learning model. Step S6: Using the machine learning model described in Step 5, predict the mechanical properties of single-layer and multi-layer composite structures based on Turing patterns, and design pattern-type single-layer and multi-layer anti-penetration structures with higher anti-penetration capabilities based on the prediction results.

2. The design method for a Turing pattern anti-penetration composite structure based on machine learning according to claim 1, characterized in that, The total number of Turing patch images generated in step S1 is greater than 10,000.

3. The design method for a Turing pattern anti-penetration composite structure based on machine learning according to claim 1, characterized in that, The ratio of the number of training sets to the number of test sets in step S2 is 7:

3.

4. The design method for a Turing pattern anti-penetration composite structure based on machine learning according to claim 1, characterized in that, The single-layer structure parameters mentioned in steps S4 and S5 include: the data matrix of the single-layer structure and the soft and hard material properties of the structure; The multi-layer structure parameters mentioned in steps S4 and S5 include: the data matrix of the multi-layer structure, the soft and hard material properties of the structure, and the stacking form.

5. The design method for a Turing pattern anti-penetration composite structure based on machine learning according to claim 4, characterized in that, The structural soft and hard material properties in the single-layer and multi-layer structural parameters include: density, elastic modulus, Poisson's ratio, and yield strength. The stacking form in the multi-layer structure parameters includes: the number of stacked layers, the thickness of each layer, and the angle of each layer.

6. A Turing pattern composite structure employing the design method for a machine learning-based Turing pattern anti-penetration composite structure as described in claim 1, characterized in that, The Turing spot-type composite structure is formed by spirally stacking multiple single-layer shell-like structures, which are interconnected by homogeneous thin layers. Each single-layer shell-like structure is formed by interlocking two materials with different elastic moduli according to one or more Turing spot diagram models.

7. A Turing pattern composite structure according to claim 6, characterized in that, The Turing pattern model type is any one or more of the following: BZ oscillatory response, Gray Scott model, and LE model.

8. A Turing pattern composite structure according to claim 6, characterized in that, The diffusion regions of the Turing patterns in each single-layer shell structure are filled with soft material, and the undiffused regions are filled with hard material, wherein the ratio of the elastic modulus of the hard material to that of the soft material is at least 2:

1.

9. A Turing pattern composite structure according to claim 8, characterized in that, The homogeneous thin layer connecting the single-layer shell-like structures is made of the soft material or a third material with an elastic modulus 1 to 2 times that of the soft material, and its thickness is 5% to 20% of the thickness of the single-layer shell-like structure in contact with the homogeneous thin layer.

10. A Turing pattern composite structure according to claim 6, characterized in that, The multi-layered shell structure is formed by stacking at least three single-layered shell structures.

Images