Aluminum alloy design method based on machine learning

Abstract

The invention belongs to the technical field of artificial intelligence and metal alloy design, and particularly relates to an aluminum alloy design method based on machine learning, which comprises the following steps: step 1, constructing a database, and obtaining element parameters for describing element characteristics and second phase parameters for describing second phase characteristics in aluminum alloy; 2, screening the element parameters and the second phase parameters to obtain machine learning model input parameters; 3, inputting parameters based on the machine learning model, and respectively constructing a thermal conductivity prediction model and a tensile strength prediction model; and 4, aluminum alloys with different components are generated randomly, the corresponding heat conductivity and tensile strength are predicted through the heat conductivity prediction model and the tensile strength prediction model respectively till the heat conductivity and the tensile strength reach design target values, and the components of the new aluminum alloy are obtained. The actual performance of the aluminum alloy designed through the method is highly matched with a predicted value.

Description

Technical Field

[0001] The present application relates to the technical field of artificial intelligence and metal alloy design, and specifically to an aluminum alloy design method based on machine learning. Background Art

[0002] Aluminum alloys are widely used in the automotive, aerospace, and shipbuilding industries due to their high specific strength, excellent thermal conductivity, and recyclability. However, the increased strength of most aluminum alloys is often accompanied by a decrease in thermal conductivity. The reason for this is that the lattice distortion and second phases introduced during the alloying process not only produce solid solution strengthening and precipitation strengthening effects, but also significantly scatter electrons and phonons, the latter of which directly reduces the thermal conductivity of the aluminum alloy. In addition, the high cost of experimentally verifying the performance of aluminum alloys has limited the development of aluminum alloys that can balance thermal conductivity and tensile strength.

[0003] At the same time, machine learning can predict the properties of aluminum alloys based on existing data, eliminating the need for experiments during the design phase. However, there are few reports on the application of machine learning in aluminum alloy design.

[0004] To this end, there is a continuous need in this field to develop an aluminum alloy design method based on machine learning. Summary of the Invention

[0005] This application aims to provide a machine learning-based aluminum alloy design method. By selecting specific machine learning input parameters, particularly the innovative introduction of second-phase parameters to characterize the second phase in aluminum alloys, the accuracy of model predictions is improved. Experimental preparation and testing have verified that the alloy's performance indicators closely match the model's predictions. Microstructural analysis reveals that a fragmented and spheroidized silicon phase structure is a key factor in the performance improvement.

[0006] In order to solve the above technical problems, this application provides the following technical solutions.

[0007] In a first aspect, the present application provides an aluminum alloy design method based on machine learning, the method comprising the following steps:

[0008] Step 1: Build a database to obtain element parameters used to describe element characteristics and second phase parameters used to describe the second phase characteristics in aluminum alloys;

[0009] Step 2: Screening the element parameters and the second phase parameters to obtain machine learning model input parameters;

[0010] Step 3: Based on the machine learning model input parameters, a thermal conductivity prediction model and a tensile strength prediction model are constructed respectively;

[0011] Step 4: Randomly generate aluminum alloys with different compositions, and predict the corresponding thermal conductivity and tensile strength respectively using the thermal conductivity prediction model and the tensile strength prediction model until the thermal conductivity and tensile strength reach the design target values, thereby obtaining the composition of the new aluminum alloy.

[0012] In one embodiment of the first aspect, the elemental parameters and the second phase parameters include a total of 123 parameters, of which 22 parameters are composition parameters, 96 parameters are physical descriptors, and 5 parameters are processing parameters.

[0013] In one embodiment of the first aspect, in step 2, the element parameters and the second phase parameters are screened based on two feature selection methods: Lasso regression and Gini impurity index, to select machine learning model input parameters for constructing the thermal conductivity prediction model and the tensile strength prediction model.

[0014] In one embodiment of the first aspect, in step 3, a thermal conductivity prediction model is constructed using an Xgboost algorithm, and a tensile strength prediction model is constructed using an SVM algorithm.

[0015] In one embodiment of the first aspect, before constructing the tensile strength prediction model, the original data is standardized.

[0016] In one embodiment of the first aspect, the design target value of thermal conductivity is greater than or equal to 190 W / (m·K), and the design target value of tensile strength is greater than or equal to 220 MPa.

[0017] In one embodiment of the first aspect, the composition of the new aluminum alloy is Al-2.64Si-0.43Mg-0.10Zn-0.03Cu.

[0018] Compared with the existing technology, the positive effect of the present invention is that the present invention establishes a machine learning prediction model for the thermal conductivity (TC) and tensile strength (UTS) of cast aluminum alloy. By performing feature expansion and screening through Lasso regression and Gini impurity, and comprehensively considering multiple physical and chemical parameters, the prediction accuracy is significantly improved. The TC and UTS prediction R of the test set 2 The values exceed 0.9 and 0.8, respectively. The Al-2.64Cu-0.43Mg-0.10Zn-0.03Si alloy, developed using a virtual alloy design method, exhibits thermal conductivity >190 W / (m·K) and tensile strength >220 MPa. Test results of the experimentally prepared alloy show that the actual properties closely match the predicted values. Microstructural analysis reveals that the fragmented and spheroidized Si phase effectively reduces electron scattering in the aluminum matrix and at the Al / Si interface, thereby improving thermal conductivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Showing the aluminum alloy design process based on machine learning.

[0020] Figure 2 Displays the distribution of ultimate tensile strength (UTS) and thermal conductivity (TC) in the dataset.

[0021] Figure 3 Shows the trend of the number of features with non-zero coefficients as the α value changes in (a) TC dataset and (b) UTS dataset.

[0022] Figure 4 Showing the importance ranking of (a) TC features and (b) UTS features.

[0023] Figure 5 Comparison between the TC predictions and measured values based on the Xgboost algorithm is shown. The input features are: (a) composition information; (b) composition + processing parameters; (c) feature-filtered composition + processing parameters + elemental physical parameters; (d) feature-filtered composition + processing parameters + elemental physical parameters + second phase information.

[0024] Figure 6 Comparison between the predicted and measured UTS values based on the SVM algorithm is shown. The input features are: (a) composition information; (b) composition + processing parameters; (c) feature-screened composition + processing parameters + elemental physical parameters; (d) feature-screened composition + processing parameters + elemental physical parameters + second phase information.

[0025] Figure 7 Showing the determination coefficient (R) of (a) TC and (b) UTS predictions under different input characteristics 2 ).

[0026] Figure 8 Shows the performance distribution of 400 virtual samples generated by the two prediction models.

[0027] Figure 9 Shows the comparison of thermal conductivity and tensile strength of Al-2.64Si-0.43Mg-0.10Zn-0.03Cu alloy with those reported in the literature;

[0028] Figure 10 Shows SEM images. DETAILED DESCRIPTION

[0029] Unless otherwise defined, technical or scientific terms used in this specification and claims shall have the same general meaning as understood by persons having ordinary skills in the technical field to which the present invention belongs.

[0030] The technical solution of this application will be clearly and completely described below in conjunction with the accompanying drawings.

[0031] As mentioned above, the thermal conductivity and tensile strength of aluminum alloys are mutually exclusive properties: good thermal conductivity results in poor tensile strength, and vice versa. Existing research indicates that by regulating the content of elements such as Si, Cu, Mg, and Zn, a synergistic effect of solid solution strengthening and precipitation strengthening can be achieved. Compared to elements such as Ti, Mn, and Cr, these elements have relatively less negative impact on thermal conductivity. However, a systematic quantitative model for the influence of single elements on strength and thermal conductivity is still lacking. From a physical mechanism perspective, the atomic radius, valence state, and solid solubility of solute elements all significantly influence alloy properties. For example, the difference in radius between solute atoms and the Al matrix (atomic radius 0.143 nm) disrupts the lattice periodicity and exacerbates electron scattering. Introducing Ce (0.183 nm), a large atomic radius, can effectively alleviate the lattice distortion caused by Fe (0.127 nm) and Si (0.134 nm), thereby improving thermal conductivity by reducing electron transition resistance. Furthermore, the volume fraction and morphology of the second phase play a significant role in regulating thermal conductivity: a continuously distributed second phase significantly impedes electron transport, shortening the mean free path of electron migration between silicon phases; and compared to fibrous eutectic silicon, spherical eutectic silicon effectively reduces the probability of free electron scattering at the Al / Si interface. A thermal conductivity model developed by Chen et al. indicates that the overall thermal conductivity of the alloy can be characterized as the harmonic mean of the thermal conductivities of the aluminum matrix and the precipitated phase. Given the multi-component nature of commercial aluminum alloys, the development of high-strength, high-conductivity aluminum alloys using a bottom-up design strategy remains a significant challenge.

[0032] Machine learning (ML) has been widely used in the field of materials science research, especially in the performance prediction and alloy design of aluminum alloys. By applying ML technology, researchers can optimize the prediction process of key material properties, including fracture toughness, corrosion resistance and wear characteristics, with unprecedented accuracy and efficiency. Multi-objective optimization has been proven to be a key strategy for improving the comprehensive properties of cast aluminum alloys (such as hardness, strength and modulus). This method can achieve the simultaneous optimization of multiple performance indicators that are often in a competing relationship, thereby breaking through the limitations of traditional alloy design and developing customized materials that meet specific application requirements. However, the collaborative optimization of mutually exclusive performance indicators such as strength and thermal conductivity remains challenging. Although some studies have attempted to integrate composition and process parameters as input features into ML models to predict alloy properties and guide design, these models have significant deficiencies in interpretability, especially the lack of second phase information that has a decisive influence on the properties of aluminum alloys.

[0033] While thermodynamic calculations are currently widely used to obtain second-phase information in aluminum alloys, their integration with machine learning remains rare. Existing research often uses thermodynamic calculations as an auxiliary tool for phase structure prediction, rather than specifically for calculating alloy second-phase information. Compared to the alloying element information widely used in existing machine learning research, second-phase information has a stronger correlation with alloy macroscopic properties and microstructure, significantly improving the accuracy of machine learning predictions.

[0034] This study aimed to design a cast aluminum alloy with both high thermal conductivity (TC) and ultimate tensile strength (UTS) through machine learning. By incorporating compositional physical descriptors and second-phase information obtained through Pandat into the input feature set and identifying the optimal feature combination through feature engineering, ML prediction models for UTS and thermal conductivity were constructed. Guided by this model, a new aluminum alloy was successfully designed, fabricated, and tested, with strength and thermal conductivity properties that closely matched the predicted values. Microstructural characterization also revealed its strengthening mechanism.

[0035] Machine Learning Process

[0036] like Figure 1 As shown in the figure, the aluminum alloy development process described in this article covers four major modules: data collection and feature calculation, feature screening, prediction model construction, and new alloy design and verification.

[0037] This application first collects data on 248 cast aluminum alloys from public literature to construct an initial data set, which includes key information such as alloy composition, solution aging treatment process parameters (temperature and time), thermal conductivity and UTS value (Table 1 is part of the sample data). The alloy composition covers 22 elements (Si, Fe, Cu, B, Bi, Pb, Zn, Mn, Mg, Sn, Ti, V, Mo, Ni, Ce, Co, Cr, La, Sc, Sr, Zr, Al), and the content is expressed in mass fraction (wt.%); the process parameters include solution temperature, solution time, quenching temperature, aging temperature and aging time. Missing values in the data set are uniformly marked as -1. Figure 2 Displays the distribution characteristics of UTS and thermal conductivity (TC): UTS values range from 50-350 MPa, and TC values range from 100-220 W / (m·K). Note that some data only contain a single performance indicator of UTS or TC.

[0038] Table 1: Sample data in the database

[0039]

[0040] In order to deeply analyze the composition information, it is necessary to expand the feature set by introducing physical descriptors to enrich the feature dimension. This application extracted 48 physical parameters from the Materials Project database to construct a feature system. The specific parameters are shown in Table 2. At the same time, Pandat software was used to calculate the second phase type and volume fraction of each alloy at room temperature. Combining the Pandat calculation results with the published literature data, 30 second phase contents and their corresponding lattice constants were systematically sorted out, as shown in Table 3. Finally, the average value p of each parameter was used. mean and standard deviation p std As input features:

[0041]

[0042] Where c i Indicates the element mass percentage content (wt.%), p i is the physical parameter of the corresponding element, i represents the physical parameter number, and n represents the total number of physical parameters. The initial feature set contains a total of 123 feature items, including 22 compositional features, 96 physical descriptors, and 5 processing features.

[0043] Table 2: Physical parameters of elements and secondary phases

[0044]

[0045]

[0046]

[0047] Table 3: Second phase types and their lattice parameters

[0048]

[0049]

[0050]

[0051]

[0052] Subsequently, two feature selection methods based on Lasso regression and Gini impurity were used to screen the optimal feature combination for building a machine learning model. Thermal conductivity (TC) and ultimate tensile strength (UTS) were predicted using a tree ensemble model and support vector machine algorithm. The dataset was divided into training and test sets with a ratio of 85%:15%.

[0053] The coefficient of determination (R 2 ) and root mean square error (RMSE) to evaluate the prediction accuracy of the machine learning model:

[0054]

[0055] Where y i Indicates the measured value; represents the predicted value; The arithmetic mean of the measured values, i is the number of actual observations or predicted values, and n is the total number of actual observations or predicted values. The better the prediction effect, the lower the RMSE value and the higher the R 2 Finally, a new alloy with optimized performance is designed based on the prediction model, and its performance is verified through experiments.

[0056] Feature Engineering and Machine Learning

[0057] Selecting appropriate features is crucial for building a reliable machine learning model, especially when the dataset contains only 248 samples. Eliminating insignificant features can effectively reduce computational complexity and improve prediction accuracy. The Lasso algorithm and Gini impurity screening are common methods for feature selection. The Lasso algorithm implements feature selection by introducing an L1 regularization term into the linear regression loss function:

[0058] (5)

[0060] Where y i represents the actual observed value; is the predicted value; α is the penalty coefficient; |w| represents the sum of the absolute values of the feature coefficients, i represents the actual observed value or the predicted value, and n represents the total number of actual observed values or predicted values. The larger the absolute value of the feature coefficient, the more important the feature is. In the process of minimizing the loss function, the coefficients of irrelevant features will gradually shrink to zero. As the penalty coefficient α increases, the coefficients of more features will return to zero to reduce the |w| value. This will achieve the elimination of unimportant features. Figure 3 As shown in Figure 2, as the value of α increases, the number of features with non-zero coefficients (i.e., the number of important features) decreases. Finally, 62 features with non-zero coefficients are retained for subsequent screening.

[0061] Gini impurity measures the degree of uncertainty reduction after a tree model branches and is a commonly used metric for evaluating feature importance in tree algorithms. If a feature significantly reduces prediction uncertainty after application, it indicates high feature importance. This study used three tree algorithms: random forest (RF), Xgboost, and gradient boosted decision tree (GBDT) to calculate the Gini impurity of each feature. Figure 4 The mean ranking of feature importance scores obtained by the three algorithms is shown, and the features with the highest importance are retained as input to the performance prediction model.

[0062] Normalization is performed before UTS prediction, but not before TC prediction.

[0063]

[0064] Where, Denotes standardized data, where X represents the original data, μ represents the sample mean, and σ represents the standard deviation. For tree ensemble algorithms, standardization does not affect the division of branch nodes because the information gain is independent of the data dimension. However, in support vector machine (SVM) algorithms, dimensional differences between variables can significantly affect the calculation results, thereby reducing prediction accuracy.

[0065] To verify the necessity of introducing element and second phase information, this study compared the prediction accuracy of different feature combination models. Xgboost algorithm and SVM algorithm were used to predict thermal conductivity (TC) and tensile strength (UTS), respectively, with a total of four types of input features. Figure 5-7 As shown in the figure, when only the composition information containing 22 elements is used as the model input, the prediction accuracy is the lowest: the error of multiple data points in the TC prediction exceeds 10%, and some errors in the UTS prediction exceed 15%. In view of the key role of heat treatment in regulating microstructure, it is not reliable to rely solely on composition information for performance prediction. After introducing processing parameters, the prediction accuracy of both TC and UTS models is significantly improved, among which the root mean square error (RMSE) of UTS prediction shows a clear downward trend. However, due to the significant differences between different alloying elements, the proportion of zero values in the composition data is still high, and these invalid data are difficult to provide effective information for the model. Further feature expansion is performed by calculating the standard deviation and mean of the elements and second phase physical parameters, and the structural characteristics and physical properties are successfully integrated into the input data. After feature expansion, the TC prediction error is significantly reduced, Figure 6 (c) shows that the test set data points are closer to the zero error line, but the RMSE value of the UTS prediction increases. It is worth noting that when the second phase information is introduced, both the TC and UTS prediction models achieve the best performance and obtain the highest coefficient of determination (R 2 ) and the lowest RMSE value, among which the relative errors of TC predictions are all controlled within 10%, and the relative errors of UTS predictions are mostly lower than 15%.

[0066] Based on the established performance prediction model, the composition optimization of cast aluminum alloy was studied. In view of the large number of missing processing parameters in the data set, the heat treatment parameters were set to the process conditions with the highest frequency in the data set: 500℃ solution treatment for 6 hours + 250℃ aging treatment for 4 hours. Considering the excellent performance and wide application of Al-Si-Mg-Zn-Cu alloys, 400 groups of virtual alloy compositions with different element ratios were randomly generated and input into the model for prediction. Figure 8As shown in the figure, among the performance distribution of 200 virtual samples, the alloy with the composition of Al-2.64Cu-0.43Mg-0.10Zn-0.03Si exhibits a good comprehensive performance of TC and UTS.

[0067] Experimental verification

[0068] The alloy was prepared using an optimized composition consisting of commercially pure Al (99.95%), Mg (99.95%), Zn (99.95%), Mg-10wt.% Si, and a Mg-10wt.% Cu master alloy. The alloy was melted in a resistance furnace at 720°C, manually stirred, and allowed to stand for homogenization before being cast into 60mm diameter cylindrical ingots. The ingots were solution treated at 500°C for 6 hours and aged at 250°C for 4 hours.

[0069] Three tensile specimens were cut from the ingot with dimensions of 18 (L) mm × 3.4 (W) mm × 2 (T) mm. A Zwick-100 kN testing machine equipped with a BT2-EXMACWD device was used to test the tensile strength of the ingot at a constant strain rate of 1.0 × 10 -4 s -1 Conduct a tensile test. The calculation formula for thermal conductivity λ is λ=ραC p

[57] Where ρ is the density of aluminum (2.7 g / cm 3 ); α is the thermal diffusion coefficient; C p The constant-pressure specific heat capacity of aluminum is 0.88 kJ / (kg··K) and the thermal diffusivity α of a Ф12.7×2.0 mm sample was measured three times at room temperature (25°C) using a laser flash thermal conductivity meter (Netzsch LFA467HT).

[0070] Table 4 lists the predicted and measured values of the strength and thermal conductivity tests. The relative errors of the predictions of thermal conductivity (TC) and tensile strength (UTS) are 8.9% and 6.8%, respectively. Figure 9 As shown, compared with other cast alloys reported in the literature,

[0071] The Al-2.64Cu-0.43Mg-0.10Zn-0.03Si alloy exhibits a thermal conductivity of >190W / (m·K) and a tensile strength of >220MPa, showing significant advantages in application scenarios that require both high strength and high thermal conductivity.

[0072] Table 4: Predicted and measured thermal conductivity and tensile strength data for the Al-2.64Si-0.43Mg-0.10Zn-0.03Cu alloy

[0073]

[0074] X-ray diffraction (XRD) phase analysis of the Al-2.64Si-0.43Mg-0.10Zn-0.03Cu alloy revealed that, in addition to diffraction peaks from the aluminum matrix, only characteristic peaks of the silicon phase were detected. Energy dispersive spectroscopy (EDS) analysis was consistent with the XRD characterization, indicating that silicon exists as a secondary phase. Figure 10 This is a scanning electron microscope (SEM) image of the alloy. The microstructure shows the morphology of coexistence of primary α-Al cellular crystals and dark gray spherical silicon phases.

[0075] The formation of silicon phase effectively reduces the content of solid solution silicon in the aluminum lattice. Al =118pm) and silicon atoms (R Si =143pm), the atomic radius is significantly different, and the solid solution silicon will cause serious lattice distortion, which will have a significant negative impact on the thermal conductivity of aluminum alloy.

[62] The precipitation of silicon phase effectively alleviates the lattice distortion in the aluminum matrix, thereby improving the thermal conductivity of the material

[63] .

[0076] The size and morphology of the silicon phase have a significant impact on thermal conductivity: the continuous distribution of the α-Al / Si phase interface significantly shortens the mean free path of electrons, and this effect increases with the size of the second phase. The discontinuous interface formed by the spherical silicon phase facilitates long-range electron transport in the aluminum matrix, thereby improving thermal conductivity. The refinement and fragmentation of the eutectic silicon phase not only enhances the precipitation strengthening effect, but also effectively fills microdefects and reduces stress concentration through solid solution strengthening mechanisms and the elimination of needle-like second phases, thereby improving alloy strength.

[0077] Key factors affecting TC and UTS

[0078] Table 5: Top five key alloy factors affecting TC and UTS

[0079]

[0080]

[0081] Here, solid solution concentration (wt.%)_mean refers to the average value of solid solubility in aluminum, and solid solution concentration (wt.%)_standard deviation refers to the standard deviation of solid solubility in aluminum.

[0082] The weakening effect of solid solution atoms on thermal conductivity is significantly greater than that of the second phase. In addition to the differences in solid solubility of different elements, the size difference between the solid solution atoms and the matrix atoms has a significant impact on thermal conductivity. As the atomic size difference increases, the matrix lattice distortion increases, leading to a more significant scattering effect of free electrons in the alloy.

[0083] Vaporization enthalpy and atomization enthalpy are closely related to the material's bond strength. Adding highly stable elements can simultaneously hinder electron transport and dislocation motion, thereby reducing thermal conductivity and increasing material strength. Valence electrons are easily captured by other atoms to form chemical bonds. A higher number of valence electrons promotes the formation of more secondary phases, thereby enhancing alloy strength.

[0084] Electron affinity is defined as the change in energy released when a neutral atom acquires an electron, forming a negative ion. Atoms with more negative electron affinity are more likely to capture electrons. During heat conduction, atoms with greater negative electron affinity are more likely to capture free electrons, which act as charge carriers, thereby reducing the thermal conductivity of the material.

[0085] The above description of the embodiments is intended to facilitate understanding and application of the present application by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to these embodiments and apply the general principles described herein to other embodiments without expending any creative effort. Therefore, the present application is not limited to the embodiments described herein, and improvements and modifications made by those skilled in the art based on the disclosure of this application without departing from the scope and spirit of this application are within the scope of this application.

Claims

1. A machine learning-based aluminum alloy design method, characterized in that: The method comprises the following steps: Step 1: Build a database to obtain element parameters used to describe element characteristics and second phase parameters used to describe the second phase characteristics in aluminum alloys; Step 2: Screening the element parameters and the second phase parameters to obtain machine learning model input parameters; Step 3: Based on the machine learning model input parameters, a thermal conductivity prediction model and a tensile strength prediction model are constructed respectively; Step 4: Randomly generate aluminum alloys with different compositions, and predict the corresponding thermal conductivity and tensile strength respectively using the thermal conductivity prediction model and the tensile strength prediction model until the thermal conductivity and tensile strength reach the design target values, thereby obtaining the composition of the new aluminum alloy.

2. The aluminum alloy design method based on machine learning according to claim 1, characterized in that: The element parameters and the second phase parameters include 123 parameters in total, of which 22 parameters are composition parameters, 96 parameters are physical descriptors, and 5 parameters are processing parameters.

3. The aluminum alloy design method based on machine learning according to claim 2, characterized in that: In step 2, based on two feature selection methods of Lasso regression and Gini impurity, the element parameters and the second phase parameters are screened to select the machine learning model input parameters for constructing the thermal conductivity prediction model and the tensile strength prediction model.

4. The aluminum alloy design method based on machine learning according to claim 3, characterized in that: The machine learning model has 48 input parameters, including:

5. The aluminum alloy design method based on machine learning according to claim 4, characterized in that: The key parameters used to build the thermal conductivity prediction model are: Here, solid solution concentration (wt.%)_mean refers to the average value of solid solubility in aluminum, and solid solution concentration (wt.%)_standard deviation refers to the standard deviation of solid solubility in aluminum.

6. The aluminum alloy design method based on machine learning according to claim 4, characterized in that: The key parameters used to construct the tensile strength prediction model are:

7. The aluminum alloy design method based on machine learning according to claim 1, characterized in that: In step 3, a thermal conductivity prediction model is constructed using the Xgboost algorithm, and a tensile strength prediction model is constructed using the SVM algorithm.

8. The aluminum alloy design method based on machine learning according to claim 7, characterized in that: Before constructing the tensile strength prediction model, the raw data were standardized.

9. The aluminum alloy design method based on machine learning according to claim 1, characterized in that: The design target value of thermal conductivity is greater than or equal to 190 W / (m·K), and the design target value of tensile strength is greater than or equal to 220 MPa.

10. The aluminum alloy design method based on machine learning according to claim 9, characterized in that: The composition of the new aluminum alloy is Al-2.64Si-0.43Mg-0.10Zn-0.03Cu.

Images