Cross-scale graphene / aluminum composite material machine learning design method

Abstract

The invention discloses a cross-scale graphene / aluminum composite material machine learning design method, and belongs to the technical field of metal material property prediction. In order to inhibit serious interface reaction of graphene and a metal matrix at an interface, a method for accelerating a first principle by machine learning is utilized, alloy elements suitable for segregation towards a graphene / aluminum interface are selected in a periodic table range of elements, and elements suitable for being doped in the aluminum matrix and having no negative influence on mechanical properties of the matrix are selected; the influence of graphene distribution and temperature on the mechanical property of the cross-scale graphene / aluminum composite material is determined through a molecular dynamics simulation method, and the preparation process range of the cross-scale graphene / aluminum composite material is determined; alloy elements suitable for segregation to a graphene / aluminum interface and elements suitable for being doped in an aluminum matrix and having no negative influence on mechanical properties of the matrix are compared to obtain an intersection, doped elements are determined, and then experimental verification is performed based on a preparation process of the cross-scale graphene / aluminum composite material.

Description

Technical Field

[0001] The present invention belongs to the technical field of metal material property prediction, and in particular relates to a cross-scale graphene / aluminum composite material machine learning design method. Background Art

[0002] Graphene is considered to be an ideal reinforcement for metal matrix composites due to its excellent mechanical properties. Aluminum alloy has a low density, good plasticity, and excellent corrosion resistance. It is the most widely used type of non-ferrous metal material in modern industry. Adding graphene to aluminum alloy to prepare graphene-reinforced aluminum matrix composites has attracted widespread attention. Studies have found that adding graphene to aluminum matrix is ​​expected to improve its comprehensive performance. Graphene-reinforced aluminum matrix composites have broad development and application prospects in the fields of electronics, aviation, and automobiles.

[0003] However, due to the poor dispersion and wettability of graphene with alloys, when the content in the aluminum matrix is ​​not appropriate, it is easy to agglomerate and cannot form a strong interface bond. The uniaxial tensile deformation is most obvious along the direction of the reinforcement layer, but because the dispersion of graphene in the metal matrix is ​​random, it is difficult for traditional methods to accurately achieve uniaxial stretching along the direction of the graphene layer. In the process of synthesizing graphene / aluminum composites, interfacial reactions at high temperatures will occur, resulting in the formation of a brittle phase Al4C3. The presence of this brittle phase will hinder the effective transfer of loads, thereby adversely affecting the overall performance of the material.

[0004] In the preparation of composite materials, the dispersion of graphene is extremely important for the performance of graphene. The dispersion of graphene in composite materials will also have a great influence on the properties of the interface between graphene and the matrix. Good interface design can not only promote the uniform distribution of graphene, but also is a core factor in enhancing the performance of graphene / metal composite materials. Studies have shown that in the process of synthesizing graphene / aluminum composite materials, interfacial reactions at high temperatures may occur, resulting in the formation of brittle phase Al4C3. The presence of this brittle phase will hinder the effective transfer of loads, which will in turn have an adverse effect on the overall performance of the material, resulting in a significant reduction in the mechanical properties of the composite material.

[0005] Related research results show that the addition of microalloying elements during the preparation of composite materials can reduce the interface reaction between graphene and the metal matrix to a certain extent, thereby achieving the effect of improving the performance of the material. Therefore, exploring ways to inhibit the interface reaction of graphene / aluminum composite materials and selecting appropriate microalloying elements to achieve the effect of inhibiting the interface reaction are of great significance for exploring the excellent performance of aluminum alloy composites and promoting the development of new material design.

[0006] As a computational method based on quantum mechanics, first-principles calculations based on density functional theory are often used to calculate the intrinsic properties of. This method has accelerated the design process of new materials to a certain extent. However, with the increasing scale and complexity of the computing system, more complex and effective methods are needed to process huge amounts of data. How to quickly develop new materials is still a challenge for scientific researchers. There is currently a lack of literature reports on multi-scale design methods for the study of graphene / aluminum composites. In view of the "scale effect" in the field of materials, there is still a gap in the field of multi-scale design of new materials. In recent years, the rapid development of computers has provided impetus for the design of new materials, and the development of machine learning and artificial intelligence algorithms has accelerated the prediction process of new materials. Combining machine learning with computational simulation can further broaden the direction of graphene / aluminum-based composite design. At present, there are few reports on machine learning accelerated computational simulation of graphene / aluminum composites. Summary of the invention

[0007] The problem to be solved by the present invention is to suppress the serious interface reaction between graphene and metal matrix at the interface, and propose a cross-scale machine learning design method for graphene / aluminum composite materials.

[0008] To achieve the above object, the present invention is implemented through the following technical solutions:

[0009] A cross-scale machine learning design method for graphene / aluminum composite materials, comprising the following steps:

[0010] S1. Use machine learning to accelerate the first principles method to select alloy elements suitable for segregation to the graphene / aluminum interface within the periodic table;

[0011] S2. Use machine learning to accelerate the first-principles method to select elements within the periodic table that are suitable for doping in the aluminum matrix and do not have a negative impact on the mechanical properties of the matrix;

[0012] S3. Using molecular dynamics simulation methods, determine the effects of graphene distribution and temperature on the mechanical properties of cross-scale graphene / aluminum composites, and determine the preparation process range of cross-scale graphene / aluminum composites;

[0013] S4. Compare and intersect the alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1 and the elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix obtained in step S2, determine the doping elements, and then conduct experimental verification based on the preparation process of the cross-scale graphene / aluminum composite material obtained in step S3.

[0014] Furthermore, the specific implementation method of step S1 includes the following steps:

[0015] S1.1. Use the Visualizer module in Materialsstudio, a first-principles ab initio quantum mechanics software based on density functional theory, to build models of graphene and aluminum substrates. Use the CASTEP module to set calculation parameters including exchange-correlation function, self-consistent field convergence criterion, force acting on each atom in the crystal, maximum atomic displacement, internal stress, and average atomic energy.

[0016] S1.2. Use the plane wave cutoff energy and Brillouin zone K point in the calculation parameters obtained in step S1 to perform a convergence test to obtain the cutoff energy and Brillouin zone K point used for calculation;

[0017] S1.3. Randomly select 30-50 groups of elements in the periodic table, introduce them into the model of graphene and aluminum matrix established in step S1.1, and use the transition state search method to calculate the element diffusion migration energy;

[0018] S1.4. Set the atomic radius, covalent radius, atomic volume, relative atomic mass, first ionization energy and group number of the element as input feature quantities, and use the calculated element diffusion migration energy as output to establish a data set;

[0019] S1.5. Use the CatBoost algorithm to establish a machine learning model for the segregation properties of elements at the graphene / aluminum interface, use the data set obtained in step S1.4 to train the machine learning model for the diffusion properties of elements at the graphene / aluminum interface, and then use the trained machine learning model for the diffusion properties of elements at the graphene / aluminum interface to predict all elements in the periodic table to obtain alloy elements suitable for segregation at the graphene / aluminum interface.

[0020] Furthermore, in the CatBoost algorithm of step S1.5, the number of training iterations is set to 60, the learning rate learning_rate is set to 0.02, and the regularization coefficient L2 is set to 3.

[0021] Furthermore, the alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1.5 include Sc, Zr, Mg, Si, Cu, Ti, Ni, Zn, Mo, V, Nb, W, and Ta.

[0022] Furthermore, the specific implementation method of step S2 includes the following steps:

[0023] S2.1. Use the Visualizer module in Materialsstudio, a first-principles ab initio quantum mechanics software based on density functional theory, to construct an atomically doped aluminum matrix and microscopic tensile model. The model structure contains 71 aluminum atoms and 1 doping atom. Use CASTEP to set the calculation parameters including the exchange correlation function, the self-consistent field convergence criterion, the force acting on each atom in the crystal, the maximum displacement of the atom, the internal stress, and the average energy of the atom;

[0024] S2.2. Perform a convergence test on the plane wave cutoff energy and the Brillouin zone K point in the calculation parameters obtained in step S2.1 to obtain the cutoff energy and the Brillouin zone K point used for calculation;

[0025] S2.3. Randomly select 20-30 elements in the periodic table as doping elements in the atomic doped aluminum matrix and micro tensile model established in step S2.1, and calculate the atomic doping energy and theoretical stress;

[0026] S2.4. Set the ionic radius, third ionization energy, covalent radius, electron affinity, second ionization energy, first ionization energy and electronic configuration of the element as the input characteristic quantities of atomic doping energy, take the atomic doping energy as the output quantity, and establish the first data set; set the atomic volume, ionic radius, group number, second ionization energy, atomic number, first ionization energy and atomic radius of the element as the input characteristic quantities of theoretical stress, take the theoretical stress as the output quantity, and establish the second data set;

[0027] S2.5. Use the CatBoost algorithm to establish a machine learning model of the atomic doping energy properties of element-doped aluminum matrix and a machine learning model of the theoretical stress properties of element-doped aluminum matrix. Then use the first data set and the second data set obtained in step S2.4 to train the machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix, respectively. Then use the trained machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix to predict all elements in the periodic table and obtain elements that are suitable for doping in the aluminum matrix and have no negative impact on the mechanical properties of the matrix.

[0028] Furthermore, step S2.5 obtains elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix, including Sc, Zr, Cr, Ni, Cu, Mg, Si, Mn, W, Ti, Hf, Ta, B, Nb, and V.

[0029] Furthermore, the molecular dynamics simulation of step S3 is implemented using the open source molecular massively parallel simulator LAMMPS, using the EAM potential function between metals, the Airebo potential function between C, and the LJ potential between metal and C, and using periodic boundary conditions to simulate infinitely extended substrates and graphene layers;

[0030] S3.1. Effect of graphene distribution on mechanical properties of graphene / aluminum composites across scales:

[0031] Establishing matrix models of 10 layers of aluminum atoms respectively, which contain graphene of different sizes or angles, specifically including: pure matrix, a structural model in which graphene accounts for 100% of the matrix, a structural model in which graphene accounts for 50% of the matrix in the x direction, a structural model in which graphene accounts for 50% of the matrix in the y direction, a structural model in which graphene accounts for 25% of the matrix in the x direction, a structural model in which graphene accounts for 25% of the matrix in the y direction, and a composite structural model in which graphene forms angles of 30°, 45°, and 60° with the stretching direction;

[0032] Uniaxial stretching is applied to the constructed model, and the stress-strain curve in the system is recorded by progressive stretching. The calculation method of strain ∈ and stress σ is as follows:

[0033]

[0034] Where L0 is the initial length and L is the length after stretching;

[0035]

[0036] Where F is the tensile force applied to the system, and A is the cross-sectional area under force;

[0037] The calculation results show that the preparation process of cross-scale graphene / aluminum composite materials is to use an extrusion process during the material preparation process to achieve the maximum reinforcement effect of the graphene arrangement direction along the force direction;

[0038] S3.2. Effect of temperature on the structural stability of graphene / aluminum composites across scales:

[0039] A graphene / aluminum composite structure model was established, with a graphene sheet in the middle and 8 layers of aluminum atoms above and below. The effects of different temperatures on the interface area and binding energy were calculated. The temperature range was 200K to 1400K, the calculated temperature interval was 50K, and the temperature control was carried out using the Nosé-Hoover thermostat method.

[0040] Binding Energy E binding Calculated as follows:

[0041] E binding =E total(E graphene +E aluminum

[0042] Among them, E total is the total energy of the graphene and aluminum matrix composite system, E graphene and E aluminum are the energies of the independent graphene sheet and the aluminum matrix, respectively;

[0043] The calculation results show that the temperature-binding energy curve of the graphene / aluminum composite model shows that before 500K, the absolute value of the interface binding energy of the system tends to increase rapidly. In the range of 500K to 1100K, the absolute value of the interface binding energy increases gradually and slowly. After the temperature reaches 1100K, the curve is basically stable, and the preparation temperature is between 1100K and 1300K.

[0044] Furthermore, after experimental verification of step S4, it was concluded that by adding Sc, Zr, Mg, Si, Cu, and Ti elements, the interface reaction of graphene / aluminum is suppressed without weakening the performance of the matrix, thereby suppressing the generation of brittle phase; during the preparation process, the extrusion process is used to arrange the graphene along the force direction, so that the graphene can exert the maximum reinforcement effect; the preparation temperature is set to 850°C, so that the combination of graphene and aluminum is the most stable.

[0045] Beneficial effects of the present invention:

[0046] The cross-scale machine learning design method for graphene / aluminum composite materials described in the present invention is based on a method of accelerating first-principles calculations by machine learning. It realizes the prediction of the entire periodic table by performing first-principles calculations of only some elements, thereby greatly reducing the computing cost.

[0047] The cross-scale machine learning design method for graphene / aluminum composite materials described in the present invention is based on the prediction results of the full periodic table of alloying elements and graphene and the interface interaction of graphene / aluminum-based composite materials, and optimizes the alloying elements (Sc, Zr, Ti, V, Si and other elements) that modify graphene without negative effects on the mechanical properties of the matrix, the alloying elements (Sc, Ti, Cu, Si, Ni and other elements) that are easy to segregate to the interface, and the alloying elements (Mg, Si, Cu, Sc, Zr elements) that inhibit interface reactions, thereby guiding the micro-alloying design of special matrix for composite materials.

[0048] The cross-scale graphene / aluminum composite material machine learning design method described in the present invention determines the influence of graphene size, angle and temperature on the graphene / aluminum composite model through molecular dynamics simulation calculation method, thereby assisting in determining the processing technology.

[0049] The cross-scale graphene / aluminum composite material machine learning design method described in the present invention guides the design of a new graphene / 6061 aluminum composite material. Compared with the existing graphene / 6061 aluminum composite material, the microalloyed graphene / 6061 aluminum composite material prepared by this patent has an elongation at break increased from 5% to 7.5% while maintaining a strength of more than 300 MPa, an increase of nearly 50%, and the plasticity of the composite material is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a cross-scale graphene / aluminum composite machine learning design method according to the present invention;

[0051] Figure 2 The crystal structure diagram of the diffusion migration of atoms in the aluminum matrix and the graphene / aluminum interface of the present invention, wherein (a), (b), and (c) are respectively the initial state structure, intermediate state structure, and final state structure of the atomic diffusion structure model in the aluminum matrix, and (d), (e), and (f) are respectively the initial state structure, intermediate state structure, and final state structure of the atomic diffusion segregation structure model to the graphene / aluminum interface;

[0052] Figure 3 This is a prediction diagram of the full periodic table data of the migration of atoms to the graphene / aluminum interface using the machine learning model established by the CatBoost algorithm in Example 1;

[0053] Figure 4 This is a full periodic table data prediction diagram of Example 1 using the CatBoost algorithm to predict the doping energy results of atoms in an aluminum matrix;

[0054] Figure 5 This is a full periodic table data prediction diagram for Example 1 using the CatBoost algorithm to predict the theoretical stress results of atoms in an aluminum matrix. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0056] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0057] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 -Attached Figure 5 The detailed instructions are as follows:

[0058] Embodiment 1:

[0059] A cross-scale machine learning design method for graphene / aluminum composite materials, comprising the following steps:

[0060] S1. Use machine learning to accelerate the first principles method to select alloy elements suitable for segregation to the graphene / aluminum interface within the periodic table;

[0061] Furthermore, the graphene is selected to be a single-layer graphene;

[0062] Furthermore, the specific implementation method of step S1 includes the following steps:

[0063] S1.1. Use the Visualizer module in Materialsstudio, a first-principles ab initio quantum mechanics software based on density functional theory, to build models of graphene and aluminum substrates. Use the CASTEP module to set calculation parameters including cutoff energy, Brillouin zone K point, exchange correlation function, self-consistent field convergence criterion, force acting on each atom in the crystal, maximum atomic displacement, internal stress, and average atomic energy.

[0064] Furthermore, the exchange correlation functional uses the GGA-PBE approximation (generalized gradient approximation), and the self-consistent field convergence criterion is 2.0×10 -6 eV / atom, the force acting on each atom in the crystal is no greater than The maximum atomic displacement is less than The internal stress is less than 0.01GPa, and the average atomic energy is less than 2.0×10 -5 eV / atom.

[0065] In the first-principles method of the CASTEP module, the ground state energy of the system can be obtained by solving the Kohn-Sham equation, which is expressed as:

[0066]

[0067] in, is the reduced Planck constant, m is the mass of the electron, is the Laplace operator, V eff (r) is the effective potential, ψ i (r) is the Kohn-Sham wave function, ∈ i is the electron energy level;

[0068] S1.2. Perform a convergence test on the plane wave cutoff energy and the Brillouin zone K point using the calculation parameters in step S1 to obtain the cutoff energy and the Brillouin zone K point used for calculation;

[0069] Furthermore, the convergence test refers to the calculation test for selecting appropriate parameters to obtain the combination of cutoff energy and K-point density for subsequent calculations, with less computing resources and sufficient computing accuracy. The specific steps are to test the plane wave cutoff energy, starting from a low value (such as 250eV) and gradually increasing to a sufficiently high value (such as 600eV), with an increment of 50eV each time. At each plane wave cutoff energy level, calculate the total energy of the system and observe the trend of the total energy changing with the cutoff energy. When the change in total energy becomes less than a certain threshold between two consecutive plane wave cutoff energy tests, it is considered to be converged. K-point density test: Using the same model, test the K-point density in the Brillouin zone. Start with a coarse grid (such as 2×2×2) and gradually increase to a finer grid (such as 10×10×10). At each K-point density, calculate the total energy of the system and record it. Determine the optimal value of the K-point density so that the change in total energy with the increase in K-point density is lower than a certain threshold. ) It was found that as the plane wave cutoff energy increased, the total energy of the system showed a downward trend. When the plane wave cutoff energy reached 350eV, the system energy fluctuations were basically stable; the energy change was large before the K point was 8, and the total energy fluctuation of the system was small after reaching 8. Therefore, a plane wave basis set with a cutoff energy of 350eV was used to expand the electronic wave function. The Brillouin zone K point used an 8×8×8 grid;

[0070] S1.3. Randomly select 30-50 groups of elements in the periodic table, introduce them into the model of graphene and aluminum matrix established in step S1.1, and use the transition state search method to calculate the element diffusion migration energy;

[0071] Furthermore, in the model of graphene and aluminum matrix, the graphene layer is stacked vertically with the aluminum matrix to simulate the interface in the actual composite material. Other atoms are introduced into the aluminum matrix region near the interface as the initial state, and the atoms are moved to the interface as the final state. mig Determined by transition state search, the expression is:

[0072] E mig =max R[E(R)]-E initial

[0073] Where E(R) is the energy along the diffusion path, R is the coordinate of the atom; E initial is the total energy of the atom in its initial stable position;

[0074] The transition state search method is to define the diffusion path from the initial position and the final position, and add 20 configurations between the initial stable state and the final stable state as state interpolation points. Solve the energy of these states to find the transition state position with the highest energy, that is, minimize the following energy function, expressed as:

[0075]

[0076] Where k is a constant used to apply force between adjacent images to prevent them from overlapping. By minimizing the above equation, the maximum energy position can be obtained as the transition state, thus determining the migration energy E mig .

[0077] The transition state method is used to calculate the diffusion migration of 40 atoms at the graphene / aluminum interface. The 40 elements include Li, B, N, Na, Mg, Al, Si, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Y, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Cd, In, Sn, La, Ta, W, Re, Os, Ir, Pt, Au, Hg, and Tl. The diffusion migration energy (E) of different atoms to the graphene / aluminum interface is mig The first principle calculation results of ) are shown in Table 1:

[0078] Table 1 Calculation results of atomic diffusion migration energy to graphene / aluminum interface

[0079]

[0080] S1.4. Set the atomic radius, covalent radius, atomic volume, relative atomic mass, first ionization energy and group number of the element as input feature quantities, and use the calculated element diffusion migration energy as output quantity to establish a data set;

[0081] Furthermore, the diffusion migration energy is used as the output quantity, and the basic information of the atom (atomic number, atomic radius, ionic radius, atomic volume, electron affinity, covalent radius, atomic volume, relative atomic mass, period, number of electrons in s, p, d, f orbits, first ionization energy, second ionization energy, third ionization energy) is used as the input feature to establish a machine learning data set. The feature elimination method is used to remove unimportant features to reduce the complexity of the model. The number of input features is determined to be 6, and the final input descriptors are atomic radius, covalent radius, atomic volume, relative atomic mass, first ionization energy and family number.

[0082] S1.5. Use the CatBoost algorithm to establish a machine learning model for the diffusion properties of elements at the graphene / aluminum interface. Use the data set obtained in step S1.4 to train the machine learning model for the diffusion properties of elements at the graphene / aluminum interface. Then use the trained machine learning model for the segregation properties of elements at the graphene / aluminum interface to predict all elements in the periodic table to obtain alloy elements suitable for segregation to the graphene / aluminum interface.

[0083] Furthermore, in the CatBoost algorithm of step S1.5, the number of training iterations is set to 60, the learning rate learning_rate is set to 0.02, and the regularization coefficient L2 is set to 3;

[0084] Furthermore, Random Forest and Adaptive Boost Regressor were used to establish a machine learning model of the segregation properties of elements at the graphene / aluminum interface for comparison. In the Random Forest algorithm, n_estimators: the number of trees in the forest is 70, the minimum number of split samples is 2, and the minimum number of samples of leaf nodes is 2; in the Adaptive Boost Regressor algorithm, the number of n_estimators basic models is 80, the learning_rate learning rate is 0.02, and the loss function type is 'linear'. 5-fold cross validation was used in the performance evaluation of the model, and the evaluation parameters of the final model were the average of 5 training processes. The mean absolute error (MAE), root mean square error (RMSE), and determination coefficient (R 2 ) and mean relative error (MAPE) as evaluation indicators, the expression is:

[0085]

[0086]

[0087]

[0088]

[0089] in, Represents the predicted value of the machine learning model, y i Represents the true value.

[0090] The obtained index comparison results are shown in Table 2:

[0091] Table 2 Comparison of performance indicators of different algorithms

[0092]

[0093] As can be seen from the table, among the selected algorithms, the model built by the CatBoost algorithm performs best, achieving the lowest MAE and RMSE. 2 And MAPE is also excellent.

[0094] The machine learning model established using the CatBoost algorithm predicts the diffusion migration energy of all atoms in the periodic table to the graphene / aluminum interface: Ni, Cu, Co, etc. are atoms with diffusion migration energy below 0.2eV; Li, Si, N, Fe, Zn, etc. have diffusion migration energy between 0.2 and 0.5eV. The diffusion migration barrier of aluminum atoms to the graphene / aluminum interface is 0.53eV. Therefore, compared with aluminum atoms, atoms that are easier to diffuse and migrate to the graphene / aluminum interface and thus segregate include Sc, Zr, Mg, Si, Cu, Ti, Ni, Zn, Mo, V, Nb, W, Ta, etc.

[0095] Furthermore, the alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1.5 include Sc, Zr, Mg, Si, Cu, Ti, Ni, Zn, Mo, V, Nb, W, and Ta.

[0096] S2. Use machine learning to accelerate the first-principles method to select elements within the periodic table that are suitable for doping in the aluminum matrix and do not have a negative impact on the mechanical properties of the matrix;

[0097] Furthermore, the specific implementation method of step S2 includes the following steps:

[0098] S2.1. The Visualizer module in Materialsstudio, a first-principles ab initio quantum mechanics software based on density functional theory, was used to construct an atomically doped aluminum matrix and microscopic tensile model. The model structure contained 71 aluminum atoms and 1 doping atom. The calculation parameters were set using CASTEP, including the exchange correlation function, the self-consistent field convergence criterion, the force acting on each atom in the crystal, the maximum atomic displacement, the internal stress, and the average atomic energy.

[0099] Furthermore, the exchange correlation functional uses the GGA-PBE approximation (generalized gradient approximation), and the self-consistent field convergence criterion is 2.0×10 -6 eV / atom, the force acting on each atom in the crystal is no greater than The maximum atomic displacement is less than The internal stress is less than 0.01GPa, and the average atomic energy is less than 2.0×10 -5 eV / atom.

[0100] S2.2. Perform a convergence test on the plane wave cutoff energy and the Brillouin zone K point in the calculation parameters obtained in step S2.1 to obtain the cutoff energy and the Brillouin zone K point used for calculation;

[0101] Furthermore, the cutoff energy was determined to be 350 eV and the Brillouin zone K point to be an 8 × 8 × 8 grid;

[0102] S2.3. Randomly select 20-30 elements in the periodic table as doping elements in the atomic doped aluminum matrix and micro tensile model established in step S2.1, and calculate the atomic doping energy and theoretical stress;

[0103] Furthermore, in order to balance the time and efficiency of the simulation, the number of selected Al atomic layers was tested. It was found that the surface energy of the system showed a downward trend with the increase of the number of Al layers. When the number of layers was greater than 7, the surface energy was 0 J / m 2 The following explains that Al can exist stably when the number of layers is greater than 7. Therefore, an 8-layer aluminum substrate is selected as the research object, and 8 layers of aluminum atoms are established as the aluminum substrate crystal structure, with the doping position at the center of the crystal structure.

[0104] The doping energy and theoretical stress of the atomically doped aluminum matrix are calculated using the first principle method. The atoms involved are Li, B, Na, Mg, Al, Si, K, Ca, Ti, V, Cr, Mn, Ni, Cu, Zn, Y, Zr, Nb, Mo, In, Sn, La, W, Tl. 71 The X1 model calculates the doping of alloy atoms. The structure contains 71 aluminum atoms and 1 doping atom, and the doping ratio is 1.38 at.%. The doping energy E Fom The calculation formula is as follows:

[0105] E Fom =E dopad -E pure +μ Al -μ dopad

[0106] Among them, E doped is the total energy of the system after atomic doping (eV), E pure is the total energy of the system before atomic doping (eV), μ Al is the energy of a single aluminum atom (eV), μ doped is the energy of a single dopant atom (eV).

[0107] In the theoretical stress calculation, the atomic doped aluminum matrix model Al is also selected. 71X1 was taken as the research object. In the stretching step, a uniform step stretching of 0.05nm was adopted, and the interface separation distance was set to 0.05nm, 0.10nm, 0.15nm...0.70nm. When the interface was about to break, a uniform stretching of 0.01nm was adopted.

[0108] The calculation formula of theoretical stress G is:

[0109]

[0110]

[0111] Among them, E total is the total energy of the system before stretching (eV), E sg is the total energy of the system after stretching (eV), λ is the stretching distance (m), S is the interface area (m 2 ), E frac is the fracture energy (eV), x is the interface separation distance (m), and e is a natural constant.

[0112] The above formula is used to calculate the doping energy (E Fom ) and the theoretical stress (G) of the atomically doped aluminum matrix are calculated, and the results are shown in Table 3:

[0113] Table 3 Doping energy of atoms in aluminum matrix (E Fom ) and theoretical stress (G)

[0114]

[0115] S2.4. Set the ionic radius, third ionization energy, covalent radius, electron affinity, second ionization energy, first ionization energy and electronic configuration of the element as the input characteristic quantities of atomic doping energy, take the atomic doping energy as the output quantity, and establish the first data set; set the atomic volume, ionic radius, group number, second ionization energy, atomic number, first ionization energy and atomic radius of the element as the input characteristic quantities of theoretical stress, take the theoretical stress as the output quantity, and establish the second data set;

[0116] Furthermore, the above first-principles calculation results are used as outputs, and similarly, the basic information of 16 atoms (including atomic number, atomic radius, ionic radius, covalent radius, atomic volume, relative atomic mass, period, number of s orbital electrons, number of p orbital electrons, number of d orbital electrons, number of f orbital electrons, etc.) are used as input features to establish a data set. The feature elimination method is also used to reduce the redundancy between features, and the doping energy input features are finally determined to be ionic radius, third ionization energy, covalent radius, electron affinity, second ionization energy and first ionization energy. The input features of theoretical stress are atomic volume, ionic radius, family number, second ionization energy, atomic number, first ionization energy and atomic radius.

[0117] S2.5. Use the CatBoost algorithm to establish a machine learning model of the atomic doping energy properties of element-doped aluminum matrix and a machine learning model of the theoretical stress properties of element-doped aluminum matrix. Then use the first data set and the second data set obtained in step S2.4 to train the machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix, respectively. Then use the trained machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix to predict all elements in the periodic table and obtain elements that are suitable for doping in the aluminum matrix and have no negative impact on the mechanical properties of the matrix.

[0118] Furthermore, the KNN algorithm was used as a comparison to construct and screen the machine learning model for the atomic doped aluminum matrix formation energy and maximum stress input and output data sets. In the CatBoost algorithm, the number of iterations training rounds is 50, the learning_rate learning rate is 0.02, and the L2 regularization coefficient is 3; in the KNN algorithm, the number of neighbors K is 4. KNN performs regression prediction based on distance, and the predicted target value is the weighted average of the K nearest neighbor samples:

[0119]

[0120] Among them, y i is the target value of the i-th nearest neighbor sample.

[0121] CatBoost uses gradient boosted decision trees for regression to minimize the loss function, and the loss function L is expressed as:

[0122]

[0123] The root mean square error (RMSE), mean absolute percentage error (MAPE) and mean square error (MSE) are introduced as evaluation indicators to assess the effectiveness of various models. The results are shown in Table 4.

[0124] Table 4 compares the MSE, RMSE, and MAPE indicators of different algorithms

[0125]

[0126] The results show that Catboost has significant advantages, so the Catboost model is used to predict doping energy and theoretical stress. The machine learning model established using the CatBoost algorithm predicts the full periodic table data of the doping energy of atoms in the aluminum matrix and the theoretical stress of the atomic doped aluminum matrix.

[0127] The results show that the elements with relatively low doping energy in the aluminum matrix include Sc, W, Ta, etc., while alloy atoms such as K, Na, Y, and Tl are difficult to dissolve into the aluminum matrix. From the theoretical stress results of atoms in the aluminum matrix, it can be seen that alloy atoms such as Sc, Cu, B, Zr, Ni, Nb, V, Cr, Mn, and W contribute to the strengthening of aluminum alloys.

[0128] Furthermore, step S2.5 obtains elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix, including Sc, Zr, Cr, Ni, Cu, Mg, Si, Mn, W, Ti, Hf, Ta, B, Nb, and V.

[0129] S3. Using molecular dynamics simulation methods, determine the effects of graphene distribution and temperature on the mechanical properties of cross-scale graphene / aluminum composites, and determine the preparation process range of cross-scale graphene / aluminum composites;

[0130] Furthermore, the molecular dynamics simulation of step S3 is implemented using the open source molecular massively parallel simulator LAMMPS, using the EAM potential function between metals, the Airebo potential function between C, and the LJ potential between metal and C, and using periodic boundary conditions to simulate infinitely extended substrates and graphene layers;

[0131] S3.1. Effect of graphene distribution on mechanical properties of graphene / aluminum composites across scales:

[0132] Establishing matrix models of 10 layers of aluminum atoms respectively, which contain graphene of different sizes or angles, specifically including: pure matrix, a structural model in which graphene accounts for 100% of the matrix, a structural model in which graphene accounts for 50% of the matrix in the x direction, a structural model in which graphene accounts for 50% of the matrix in the y direction, a structural model in which graphene accounts for 25% of the matrix in the x direction, a structural model in which graphene accounts for 25% of the matrix in the y direction, and a composite structural model in which graphene forms angles of 30°, 45°, and 60° with the stretching direction;

[0133] Uniaxial stretching is applied to the constructed model, and the stress-strain curve in the system is recorded by progressive stretching. The calculation method of strain ∈ and stress σ is as follows:

[0134]

[0135] Where L0 is the initial length and L is the length after stretching;

[0136]

[0137] Where F is the tensile force applied to the system, and A is the cross-sectional area under force;

[0138] The calculation results show that the preparation process of cross-scale graphene / aluminum composite materials is to use an extrusion process during the material preparation process to achieve the maximum reinforcement effect of the graphene arrangement direction along the force direction;

[0139] Furthermore, the calculation results show that in the parallel stretching direction, the tensile stress increases with the increase of the graphene area percentage, and the tensile stress of the composite structure is greater than the tensile stress of pure aluminum, which shows that graphene has a strengthening effect. In the vertical stretching direction, the tensile stress decreases with the increase of the graphene area percentage, and within the angle range with the stretching, the tensile stress decreases with the increase of the angle between the graphene and the stretching direction. Therefore, in the material preparation process, it is necessary to use an extrusion process to make the graphene arrangement direction along the force direction, so that the graphene can play the greatest strengthening effect.

[0140] S3.2. Effect of temperature on the structural stability of graphene / aluminum composites across scales:

[0141] A graphene / aluminum composite structure model is established, with a graphene sheet in the middle and 8 layers of aluminum atoms above and below. The effects of different temperatures on the interface area and binding energy are calculated. The temperature range is 200K to 1400K, and the calculated temperature interval is 50K. The temperature control usually adopts the Nosé-Hoover thermostat method.

[0142] Binding Energy E binding Calculated as follows:

[0143] E binding =E total -(E graphene +E aluminum )

[0144] Among them, E total is the total energy of the graphene and aluminum matrix composite system, E graphene and E aluminum are the energies of the independent graphene sheet and the aluminum matrix, respectively;

[0145] The calculation results show that the temperature-binding energy curve of the graphene / aluminum composite model shows that before 500K, the absolute value of the interface binding energy of the system tends to increase rapidly. In the range of 500K to 1100K, the absolute value of the interface binding energy increases gradually and slowly. After the temperature reaches 1100K, the curve is basically stable, and the preparation temperature is between 1100K and 1300K.

[0146] S4. Compare and intersect the alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1 and the elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix obtained in step S2, determine the doping elements, and then conduct experimental verification based on the preparation process of the cross-scale graphene / aluminum composite material obtained in step S3.

[0147] Furthermore, after experimental verification of step S4, it was concluded that by adding Sc, Zr, Mg, Si, Cu, and Ti elements, the interface reaction of graphene / aluminum is suppressed without weakening the performance of the matrix, thereby suppressing the generation of brittle phase; during the preparation process, the extrusion process is used to arrange the graphene along the force direction, so that the graphene can exert the maximum reinforcement effect; the preparation temperature is set to 850°C, so that the combination of graphene and aluminum is the most stable.

[0148] Furthermore, the composition and preparation process were determined as follows: Sc, Zr, Mg, Si, Cu, and Ti elements inhibited the graphene / aluminum interface reaction, and the graphene / 6061 aluminum-based composite material was prepared by powder metallurgy combined with pressure infiltration. The temperature was set to 850°C, and the graphene was oriented by extrusion deformation. Mechanical property tests showed that the tensile strength of the composite material under the solution aging state reached 304.14MPa, and the elongation was 7.5%. While maintaining strength, the plasticity of the material was significantly improved.

[0149] Embodiment 2:

[0150] The difference between this embodiment and embodiment 1 is that the graphene is defective graphene;

[0151] Furthermore, compared with defect-free graphene, aluminum atoms are more easily adsorbed on the surface of defective graphene. Therefore, machine learning is used to accelerate the first-principle method to calculate the migration and segregation of atoms at the single-vacancy defect graphene / aluminum interface, and to find elements within the periodic table that can inhibit the interface reaction. The other steps remain unchanged. However, when calculating the diffusion migration of atoms at the graphene / aluminum interface, the diffusion migration of atoms at the defective graphene / aluminum interface is calculated instead.

[0152] Furthermore, in step S1, 8 layers of aluminum atoms are selected as the research object, graphene is a layer of atoms in the sandwich structure, and a carbon atom is removed from the graphene to simulate a single vacancy. The diffusion behavior of alloy atoms to the defective graphene / aluminum interface is also calculated using the first principle method.

[0153] Furthermore, in step S1, the diffusion migration energies of 40 atoms to the defective graphene / aluminum interface are calculated. The 40 elements include Li, B, N, Na, Mg, Al, Si, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Y, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Cd, In, Sn, La, Ta, W, Re, Os, Ir, Pt, Au, Hg, and Tl. The calculation results are shown in Table 5. E Fom Represents the potential barrier required for atoms to diffuse and migrate to the defective graphene / aluminum interface.

[0154] Table 5 Calculation results of atomic diffusion migration energy to defective graphene / aluminum interface

[0155]

[0156]

[0157] The potential barrier E required for the above atoms to diffuse and migrate to the defective graphene / aluminum interface is used. Fom As output, the basic information of atoms (atomic number, atomic radius, ionic radius, atomic volume, electron affinity, covalent radius, atomic volume, relative atomic mass, period, number of electrons in s, p, d, f orbits, first ionization energy, second ionization energy, third ionization energy) is used as input features to establish a data set. The feature elimination method is used to reduce the redundancy between features, and the input feature numbers corresponding to this data set are determined to be atomic radius, covalent radius, atomic volume, first ionization energy and family number.

[0158] Furthermore, the performance of the machine learning model established in step S1 is shown in Table 6.

[0159] Table 6 MAE, MSE, RMSE, R of machine learning models 2 ,MAPE

[0160]

[0161] Furthermore, the established machine learning model was used to predict the migration of atoms to the defective graphene / aluminum interface. The results showed that the atoms that are more likely to diffuse and migrate to the defective graphene / aluminum interface than aluminum atoms are Sc, B, C, N, Si, Cu and other elements; compared with the diffusion and migration of atoms to the defect-free graphene / aluminum interface, the atoms that are more likely to gather on the defective graphene surface are Sc, O, N, B and other atoms. The elements that are easily segregated to the graphene / aluminum interface, especially the graphene / aluminum defect interface, are selected as follows: Sc, Zr, Mg, Si, Cu, Ti, Ni, Zn, Mn, Li, Mo, V, Cr, Nb, W.

[0162] Embodiment 3:

[0163] The difference between this embodiment and embodiment 1 is that first-principles calculations are newly performed to verify the inhibitory effects of the above-mentioned finally selected elements Sc, Zr, Mg, Si, and Cu on the graphene / aluminum interface reaction.

[0164] Specifically, the diffusion migration behavior of atoms in the form of vacancies at the Al4C3 / aluminum interface is calculated, and a diffusion migration model of aluminum atoms to the Al4C3 / aluminum interface region and a diffusion migration model of X atoms (Sc, Zr, Mg, Si, Cu) to the Al4C3 / aluminum interface region are established, and the diffusion migration energies are calculated respectively.

[0165] The size of the migration energy represents the difficulty of atoms diffusing and segregating to the interface. The lower the migration energy, the lower the potential barrier that needs to be crossed for diffusion, and the easier it is for atoms to segregate to the interface. The comparison of the migration energy of five atoms to the Al4C3 / aluminum interface and the migration energy of aluminum atoms to the Al4C3 / aluminum interface shows that the migration energy of the five selected atoms (Sc, Zr, Mg, Si, Cu) to the Al4C3 / aluminum interface is lower than that of aluminum atoms, proving the inhibitory effect of the above-mentioned screened elements on the graphene / aluminum interface reaction.

[0166] Embodiment 4:

[0167] The difference between this embodiment and embodiment 1 is that molecular dynamics calculation is newly used to verify the influence of Mg, Ti, Zr, and Cu on the microscopic stretching of the aluminum matrix.

[0168] First, a single-atom doped aluminum matrix model was established. The model was a 9-layer aluminum atom structure, each layer was arranged in 25*9, and four alloy atoms were uniformly doped. The doping atoms were Mg, Ti, Zr, and Cu. Similarly, the model was stretched along the x direction to obtain the stress-strain curve of the doping system.

[0169] The results show that the tensile strength of aluminum matrix doped with Mg, Ti, Zr and Cu atoms is 5.8 GPa, 7.3 GPa, 6.7 GPa and 7.5 GPa respectively. Compared with the undoped system, the tensile strength of the Mg and Zr atom doped system is reduced, while the tensile strength of the Ti and Cu atom doped system is improved to a certain extent.

[0170] Mg, Ti, Zr, and Cu atoms are added to the aluminum matrix to explore the synergistic effect of multiple elements on the mechanical properties of the aluminum matrix. The established molecular dynamics model is a 9-layer aluminum atomic structure, each layer is arranged in 25*9, and two atoms of each of the four alloys are evenly distributed in the matrix.

[0171] The stress-strain curve of the aluminum matrix co-doped with Mg, Ti, Zr, and Cu atoms was calculated. The tensile strength of the co-doped system is 11.5 GPa. Compared with single-atom doping, the mechanical properties of the co-doped system are most significantly improved.

[0172] The above calculation results verify the positive influence of Mg, Ti, Zr, and Cu atoms on the micromechanical properties of the matrix.

[0173] It should be noted that relational terms such as "first" and "second" are used only 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 terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0174] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A cross-scale machine learning design method for graphene / aluminum composite materials, characterized in that: The steps include: S1. Use machine learning to accelerate the first principles method to select alloy elements suitable for segregation to the graphene / aluminum interface within the periodic table; S2. Use machine learning to accelerate the first-principles method to select elements within the periodic table that are suitable for doping in the aluminum matrix and do not have a negative impact on the mechanical properties of the matrix; S3. Using molecular dynamics simulation methods, determine the effects of graphene distribution and temperature on the mechanical properties of cross-scale graphene / aluminum composites, and determine the preparation process range of cross-scale graphene / aluminum composites; S4. Compare and intersect the alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1 and the elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix obtained in step S2, determine the doping elements, and then conduct experimental verification based on the preparation process of the cross-scale graphene / aluminum composite material obtained in step S3.

2. A cross-scale graphene / aluminum composite material machine learning design method according to claim 1, characterized in that: The specific implementation method of step S1 includes the following steps: S1.

1. Use the Visualizer module in Materials studio, a first-principles ab initio quantum mechanics software based on density functional theory, to build models of graphene and aluminum substrates, and use the CASTEP module to set calculation parameters, including exchange-correlation function, self-consistent field convergence criterion, force acting on each atom in the crystal, maximum atomic displacement, internal stress, and average atomic energy; S1.

2. Perform a convergence test on the plane wave cutoff energy and the Brillouin zone K point using the calculation parameters in step S1 to obtain the cutoff energy and the Brillouin zone K point used for calculation; S1.

3. Randomly select 30-50 groups of elements in the periodic table, introduce them into the model of graphene and aluminum matrix established in step S1.1, and use the transition state search method to calculate the element diffusion migration energy; S1.

4. Set the atomic radius, covalent radius, atomic volume, relative atomic mass, first ionization energy and group number of the element as input feature quantities, and use the calculated element diffusion migration energy as output quantity to establish a data set; S1.

5. Use the CatBoost algorithm to establish a machine learning model for the diffusion properties of elements at the graphene / aluminum interface. Use the data set obtained in step S1.4 to train the machine learning model for the diffusion properties of elements at the graphene / aluminum interface. Then use the trained machine learning model for the segregation properties of elements at the graphene / aluminum interface to predict all elements in the periodic table to obtain alloy elements suitable for segregation to the graphene / aluminum interface.

3. A cross-scale graphene / aluminum composite machine learning design method according to claim 2, characterized in that: In the CatBoost algorithm of step S1.5, set the number of training iterations to 60, the learning rate learning_rate to 0.02, and the regularization coefficient L2 to 3.

4. A cross-scale graphene / aluminum composite material machine learning design method according to claim 3, characterized in that: The alloy elements suitable for segregation to the graphene / aluminum interface obtained in step S1.5 include Sc, Zr, Mg, Si, Cu, Ti, Ni, Zn, Mo, V, Nb, W, and Ta.

5. The cross-scale graphene / aluminum composite material machine learning design method according to claim 3, characterized in that: The specific implementation method of step S2 includes the following steps: S2.

1. The Visualizer module in Materials studio, a first-principles ab initio quantum mechanics software based on density functional theory, was used to construct an atomically doped aluminum matrix and microscopic tensile model. The model structure contained 71 aluminum atoms and 1 doping atom. The calculation parameters were set using CASTEP, including the exchange correlation function, the self-consistent field convergence criterion, the force acting on each atom in the crystal, the maximum atomic displacement, the internal stress, and the average atomic energy. S2.

2. Perform a convergence test on the plane wave cutoff energy and the Brillouin zone K point in the calculation parameters obtained in step S2.1 to obtain the cutoff energy and the Brillouin zone K point used for calculation; S2.

3. Randomly select 20-30 elements in the periodic table as doping elements in the atomic doped aluminum matrix and micro tensile model established in step S2.1, and calculate the atomic doping energy and theoretical stress; S2.

4. Set the ionic radius, third ionization energy, covalent radius, electron affinity, second ionization energy, first ionization energy and electronic configuration of the element as the input characteristic quantities of atomic doping energy, take the atomic doping energy as the output quantity, and establish the first data set; set the atomic volume, ionic radius, group number, second ionization energy, atomic number, first ionization energy and atomic radius of the element as the input characteristic quantities of theoretical stress, take the theoretical stress as the output quantity, and establish the second data set; S2.

5. Use the CatBoost algorithm to establish a machine learning model of the atomic doping energy properties of element-doped aluminum matrix and a machine learning model of the theoretical stress properties of element-doped aluminum matrix. Then use the first data set and the second data set obtained in step S2.4 to train the machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix, respectively. Then use the trained machine learning model of the atomic doping energy properties of element-doped aluminum matrix and the machine learning model of the theoretical stress properties of element-doped aluminum matrix to predict all elements in the periodic table and obtain elements that are suitable for doping in the aluminum matrix and have no negative impact on the mechanical properties of the matrix.

6. A cross-scale graphene / aluminum composite material machine learning design method according to claim 5, characterized in that: Step S2.5 obtains elements suitable for doping in the aluminum matrix and having no negative impact on the mechanical properties of the matrix, including Sc, Zr, Cr, Ni, Cu, Mg, Si, Mn, W, Ti, Hf, Ta, B, Nb, and V.

7. A cross-scale graphene / aluminum composite machine learning design method according to claim 6, characterized in that: The molecular dynamics simulation in step S3 is implemented using the open source molecular massively parallel simulator LAMMPS, using the EAM potential function between metals, the Airebo potential function between C, and the LJ potential between metal and C, and using periodic boundary conditions to simulate infinitely extended substrates and graphene layers; S3.

1. Effect of graphene distribution on mechanical properties of graphene / aluminum composites across scales: Establishing matrix models of 10 layers of aluminum atoms respectively, which contain graphene of different sizes or angles, specifically including: pure matrix, a structural model in which graphene accounts for 100% of the matrix, a structural model in which graphene accounts for 50% of the matrix in the x direction, a structural model in which graphene accounts for 50% of the matrix in the y direction, a structural model in which graphene accounts for 25% of the matrix in the x direction, a structural model in which graphene accounts for 25% of the matrix in the y direction, and a composite structural model in which graphene forms angles of 30°, 45°, and 60° with the stretching direction; Uniaxial stretching is applied to the constructed model, and the stress-strain curve in the system is recorded by progressive stretching. The calculation method of strain ∈ and stress σ is as follows: Where L0 is the initial length and L is the length after stretching; Where F is the tensile force applied to the system, and A is the cross-sectional area under force; The calculation results show that the preparation process of cross-scale graphene / aluminum composite materials is to use an extrusion process during the material preparation process to achieve the maximum reinforcement effect of the graphene arrangement direction along the force direction; S3.

2. Effect of temperature on the structural stability of graphene / aluminum composites across scales: A graphene / aluminum composite structure model is established, with a graphene sheet in the middle and 8 layers of aluminum atoms above and below. The effects of different temperatures on the interface area and binding energy are calculated. The temperature range is 200K to 1400K, and the calculated temperature interval is 50K. The temperature control usually adopts the Nosé-Hoover thermostat method. Binding Energy E binding Calculated as follows: AND binding =And total -(AND graphene +E aluminum ) Among them, E total is the total energy of the graphene and aluminum matrix composite system, E graphene and E aluminum are the energies of the independent graphene sheet and the aluminum matrix, respectively; The calculation results show that the temperature-binding energy curve of the graphene / aluminum composite model shows that before 500K, the absolute value of the interface binding energy of the system tends to increase rapidly. In the range of 500K to 1100K, the absolute value of the interface binding energy increases gradually and slowly. After the temperature reaches 1100K, the curve is basically stable, and the preparation temperature is between 1100K and 1300K.

8. The cross-scale graphene / aluminum composite material machine learning design method according to claim 7, characterized in that: After experimental verification of step S4, it is concluded that by adding Sc, Zr, Mg, Si, Cu, and Ti elements, the interface reaction of graphene / aluminum is suppressed without weakening the performance of the matrix, thereby suppressing the generation of brittle phase; during the preparation process, the extrusion process is used to arrange the graphene along the force direction, so that the graphene can exert the maximum reinforcement effect; the preparation temperature is set to 850°C, so that the combination of graphene and aluminum is the most stable.

Images