Method for predicting tensile strength of multi-component platinum-based alloys based on machine learning

Abstract

The present application relates to a kind of methods for predicting tensile strength of multi-component platinum-based alloy based on machine learning, comprising collecting platinum-based alloy tensile strength experimental data at room temperature, using domain knowledge to classify and process elements, constructing feature pool by the method of first-principle density functional theory calculation, with ridge regression prediction ten-fold cross-validation accuracy in machine learning as objective function, iteration learning according to feature importance order to filter optimal sub-feature combination, with optimal sub-feature as input, to construct machine learning prediction model of platinum alloy tensile strength at room temperature based on ridge regression algorithm.Compared with previous machine learning methods using element dose ratio and simple intrinsic characteristics as input, the present application first combines first-principle calculation features based on expert domain knowledge with machine learning modeling, and the calculation features established for this alloy system can more accurately distinguish alloy samples, compensate for the prediction error caused by insufficient data, and improve prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of precious metal platinum-based alloy tensile strength, and in particular to a method for predicting the tensile strength of multi-component platinum-based alloy based on machine learning. BACKGROUND

[0002] Platinum-based alloys have excellent mechanical and oxidation resistance, and can be used as a bonding layer for turbine blades of aero-engine and gas turbine, playing a role of bonding the blade alloy base and thermal barrier coating. In addition, it can also be used in other extreme high temperature environments, such as rocket tail baffle, etc., which are all due to the fact that platinum-based alloys have high tensile strength while having strong high-temperature oxidation resistance.

[0003] Tensile strength is a basic mechanical property of an alloy, which is mainly determined by the constituent elements of the alloy in a solid solution alloy. It is generally believed that the room temperature tensile strength of an alloy is linearly positively correlated with the high temperature tensile strength, and the alloy material with high high-temperature tensile strength can be indirectly obtained by screening the room temperature tensile strength.

[0004] The existing technology mainly relies on experimental trial and error to obtain high tensile strength of multi-component platinum-based alloy. Platinum-based alloys are very expensive, and the experimental cost is extremely high. SUMMARY

[0005] The purpose of the present application is to provide a method for predicting the tensile strength of multi-component platinum-based alloy based on machine learning, aiming to construct a high-precision machine learning model with a small amount of data and realize rapid prediction of the tensile strength of multi-component platinum-based alloy.

[0006] The method for predicting the tensile strength of multi-component platinum-based alloy based on machine learning comprises the following steps:

[0007] Step 1, collecting platinum-based alloy material data and corresponding room temperature tensile strength data of platinum-based alloy materials to form a data set, wherein each platinum-based alloy material is composed of two or more elements;

[0008] Step 2, all elements contained in all platinum-based alloy materials in step 1 are divided into four types of matrix A, rare and precious metal B, difficult-to-dissolve element C and rare earth element D according to the specific role of the elements in the material system, wherein the matrix A is platinum Pt;

[0009] Step 3, based on the elements in the data set, the binary solid solution properties and element single properties of Pt 31 M are calculated by the first-principle density functional method, respectively, to obtain the characteristics of each binary solid solution and the characteristics of each element single, wherein M is a placeholder, referring to other elements in each platinum-based alloy material except the Pt element;

[0010] Step 4, the weighted average operation is carried out on the characteristics of each binary solid solution and the characteristics of each element single substance corresponding to the four elements A, B, C and D in the element dose ratio as weight to obtain the weighted average characteristics, and then the addition and subtraction operation is carried out on the weighted average characteristics to obtain the operation characteristics, and the weighted average characteristics and the operation characteristics are taken as initial characteristics;

[0011] Step 5, the Pearson correlation coefficient between any two characteristics in the initial characteristics is calculated, if the Pearson correlation coefficient is greater than 0.9, the redundant characteristics are removed, only one kind of characteristics is retained, and a characteristic pool is constructed; the importance of each characteristic in the characteristic pool is sorted;

[0012] Step 6, the machine learning ridge regression algorithm accuracy is taken as the objective function, and the sub-feature set corresponding to the highest machine learning ridge regression model accuracy is screened out from the characteristic pool through sub-feature iteration screening;

[0013] Step 7, the sub-feature set screened out in step 6 is taken as the input independent variable, the normal temperature tensile strength of the platinum-based alloy material is taken as the output dependent variable, and the ridge regression machine learning algorithm is used for modeling to obtain a machine learning model for predicting the normal temperature tensile strength of the platinum-based alloy.

[0014] The first principle calculation is a kind of computational chemistry method for obtaining the relationship between atoms and electrons by approximately solving the Schrodinger equation, and the basic thermodynamic properties of some materials can be quickly calculated. These calculation results can reflect the macroscopic properties of the material to a certain extent. Therefore, the binary solid solution properties and the element single substance properties are used to obtain the characteristics of each binary solid solution and the characteristics of each element single substance, and the mapping relationship between the properties of the platinum-based alloy material and the tensile strength is established.

[0015] The decision tree algorithm is a classic algorithm in the field of machine learning, which is a model for determining and predicting in the form of tree structure (including binary tree and multi-tree). The independent variables are classified and determined by information entropy, and the prediction results are obtained at the end of the tree. At the same time, the decision tree algorithm can calculate the influence degree (importance relationship) of each independent variable feature on the output variable, so that the importance of the independent variable feature can be sorted by using the decision tree algorithm, so as to screen the main features and eliminate the redundant features.

[0016] The ridge regression algorithm is an algorithm for processing regression problems in machine learning, and the algorithm has great advantages in processing linear problems and ill-conditioned data fitting. Since the tensile strength of the alloy and the modulus and other typical characteristics have strong linear correlation, and the ridge regression has advantages in processing ill-conditioned data, so it is suitable for small data, therefore, the ridge regression is applied as the machine learning algorithm for iteration screening of the sub-feature set.

[0017] The preferred embodiment of the present application is that in step 1, the elements of all platinum-based alloy materials collected include platinum Pt, rhodium Rh, iridium Ir, gold Au, zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo, nickel Ni, yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb, samarium Sm.

[0018] The preferred embodiment of the present application is that in step 2, the rare and precious metal B includes rhodium Rh, iridium Ir, gold Au, the insoluble element C includes zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo, nickel Ni, and the rare earth element D includes yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb, samarium Sm.

[0019] The preferred embodiment of the present application is that in step 3, the binary solid solution characteristics include binary stacking fault energy, binary charge density, binary bulk modulus, binary shear modulus, binary Young's modulus, binary hardness, binary Poisson's ratio, binary electron work function, and the element single characteristics include single stacking fault energy, single charge density, single bulk modulus, single shear modulus, single Young's modulus, single hardness, single Poisson's ratio, and single electron work function.

[0020] The preferred embodiment of the present application is that in step 6, according to the importance ranking of the characteristics in step 5, the last feature is deleted each time, and the remaining features are retained for machine learning ridge regression model training, and the corresponding sub-feature set when the machine learning ridge regression model has the highest accuracy is screened out.

[0021] The preferred embodiment of the present application is that in step 6, as another way to screen the sub-feature set, the machine learning ridge regression algorithm accuracy is taken as the objective function, the ten-fold cross-validation method is adopted, the root mean square error and the correlation coefficient of the predicted platinum-based alloy room temperature tensile strength value and the collected platinum-based alloy room temperature tensile strength value are taken as the judgment standard, and the feature with the smallest root mean square error and the largest correlation coefficient is selected as the sub-feature set for screening.

[0022] The preferred embodiment of the present application is that in step 7, the sub-feature set obtained in step 6 is taken as the input, the hyperparameters required by the machine learning ridge regression model are set, the machine learning ridge regression model program is started for training, the hyperparameters are continuously adjusted during the training process, the root mean square error of the machine learning ridge regression predicted platinum-based alloy room temperature tensile strength value and the collected platinum-based alloy room temperature tensile strength value is minimized, the hyperparameters at this time are retained, and the construction of the machine learning model for predicting the platinum-based alloy room temperature tensile strength is completed.

[0023] The preferred embodiment of the present application is that in step 4, it further includes the step of normalizing the initial characteristics according to the following formula,

[0024]

[0025] wherein, x normalization is the normalized initial feature, x is the initial feature before normalization, mu is the standard deviation of x, and sigma is the mean of x.

[0026] The preferred embodiment of the present application is that the training set is divided into ten parts by using the ten-fold cross-validation method, and the model accuracy is represented by the average accuracy of ten times of training, each time taking nine parts as the training set and the remaining one part as the test set, and the hyperparameters are adjusted according to the average accuracy during the training process, so that the root mean square error of the machine learning ridge regression prediction of the tensile strength value of the platinum-based alloy at room temperature and the collected tensile strength value of the platinum-based alloy at room temperature is minimized.

[0027] Compared with the prior art, the present application has the following significant advantages:

[0028] 1. The present application first introduces the machine learning prediction method in the prediction method of the tensile strength of the multi-component platinum-based alloy, which can quickly predict and screen potential multi-component platinum-based alloys with high tensile strength, bypassing the blindness of experimental trial and error, and can greatly save experimental cost and experimental time.

[0029] 2. The application of existing machine learning technology in material property prediction and screening ignores the classification process based on the role of elements in materials, making the physical meaning of the established model unclear and affecting the model accuracy and physical interpretation of the model. The present application classifies elements according to their effects in materials, and then constructs features based on the classified elements, making the physical meaning of the finally constructed model more clear.

[0030] 3. The application of existing machine learning technology in material property prediction and screening uses element dose ratio or other intrinsic features corresponding to elements as model input, which has various disadvantages, such as general feature information without more specific information for the material system, which will affect the accuracy of the model. The present application uses first-principle calculation method to calculate and construct features for specific multi-component platinum-based alloy material system, and the obtained features are more representative in this material system and contain more specific information, making the prediction results of the model more reliable. Moreover, since the features in the present application contain more specific information of the system, a high-precision model can be constructed with less data, which is an excellent solution to the difficulty of machine learning modeling with small data sets. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings are used to Figure 1 is the flowchart of the method for predicting the tensile strength of multi-component platinum-based alloy based on machine learning;

[0032] Figure 2 Figure 2 Figure 3

[0033] Figure 4 Figure 3 Figure 5 DETAILED DESCRIPTION

[0034] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, it should be understood that the preferred embodiments described below are only used to explain the present application and will not limit the scope of protection of the present application.

[0035] Example 1

[0036] As shown in Figure 1: the method for predicting the tensile strength of multi-component platinum-based alloy based on machine learning disclosed in this embodiment comprises the following steps: Figure 1

[0037] Step 1, collect platinum-based alloy material data and corresponding normal temperature tensile strength data of platinum-based alloy material to form a data set, for example: Pt 0.8 Rh 0.15 Ta 0.03 La 0.01 Ce 0.01 , the corresponding normal temperature tensile strength is 400Mpa, and there are 100 such data. Among them, each platinum-based alloy material is composed of two or more elements. All the elements of the collected platinum-based alloy materials include platinum Pt, rhodium Rh, iridium Ir, gold Au, zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo, nickel Ni, yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb, samarium Sm, and 24 elements.

[0038] Step 2, according to the specific role of all elements contained in all platinum-based alloy materials in the material system, the elements are classified by expert field knowledge in this embodiment, which are divided into four types: matrix A, precious metal B, difficultly soluble element C, and rare earth element D. Among them, the matrix A is platinum Pt, the precious metal B includes rhodium Rh, iridium Ir, and gold Au, the difficultly soluble element C includes zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo, and nickel Ni, and the rare earth element D includes yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb, and samarium Sm. Each data collected can also be considered as one or a combination of two or more of the four types of elements in order to construct a high-accuracy feature pool subsequently.

[0039] Step 3, based on the elements in the data set, the first-principle density functional method is used to calculate the Pt 31 ​The binary solid solution properties and the element single property of M are obtained, and each binary solid solution characteristic and each element single characteristic are obtained, wherein M is a placeholder, which refers to other elements in each platinum-based alloy material except the Pt element.

[0040] In step 3, the binary solid solution characteristics include binary stacking fault energy, binary charge density, binary bulk modulus, binary shear modulus, binary Young's modulus, binary hardness, binary Poisson's ratio, binary electron work function. The detailed calculation process is as follows: through the Materials studio material calculation software, taking the element platinum Pt as the matrix and the remaining other elements as the solute, the atomic ratio of 31:1 is used for cell modeling, and the stacking fault energy, charge density, bulk modulus, shear modulus, Young's modulus, hardness, Poisson's ratio, electron work function and other properties of all binary solid solutions existing in each sample are calculated respectively. The obtained characteristics are binary stacking fault energy, binary charge density, binary bulk modulus, binary shear modulus, binary Young's modulus, binary hardness, binary Poisson's ratio, and binary electron work function.

[0041] The element single property includes single stacking fault energy, single charge density, single bulk modulus, single shear modulus, single Young's modulus, single hardness, single Poisson's ratio, and single electron work function. Through the Materials studio material calculation software, the single cell of the existing elements in each sample is constructed, and the number of atoms in the cell is 32. The stacking fault energy, charge density, bulk modulus, shear modulus, Young's modulus, hardness, Poisson's ratio, and electron work function and other properties of all single elements existing in each sample are calculated respectively. The obtained characteristics are single stacking fault energy, single charge density, single bulk modulus, single shear modulus, single Young's modulus, single hardness, single Poisson's ratio, and single electron work function.

[0042] The specific calculation method is to construct Pt 31 M binary solid solution model, M elements have 23 kinds, so there are 23 kinds of Pt 31 M model. The calculation formula of the above binary solid solution properties and single properties is the same, and the difference is only the cell model. The basic properties such as cell energy, lattice constant, volume, and elastic constant are calculated through the Materials studio material calculation software, and then the model formula is calculated. The calculation formula is as follows:

[0043] Stacking fault energy calculation formula:

[0044]

[0045] Wherein E(b) is the stacking fault structure energy, E0 is the non-stacking fault structure energy, and A is the cell base plane area.

[0046] Bulk modulus calculation formula:

[0047]

[0048]

[0049] Where B is the bulk modulus, C 11 C 12 These are the elastic constants calculated using Materials Studio software.

[0050] Shear modulus calculation formula:

[0051]

[0052]

[0053]

[0054] Where G is the shear modulus, C 11 C 12 C 44 These are the elastic constants calculated using Materials Studio software.

[0055] The formula for calculating Young's modulus is:

[0056]

[0057] Where B is the bulk modulus and G is the shear modulus.

[0058] The formula for calculating Poisson's ratio is:

[0059]

[0060] Where B is the bulk modulus and G is the shear modulus.

[0061] The formula for calculating hardness is:

[0062] H V =2(k 2 G) 0.585 -3

[0063] Among them, H V denoted as hardness, k is a constant Pugh's ratio, and G is the shear modulus.

[0064] The formula for calculating the electron work function is:

[0065]

[0066]

[0067]

[0068] where Φ is the electron work function, a is a constant, M is the atomic mass, z is the valence, p is the density, a0 is Bohr's radius, E F is the Fermi level, r s is the effective electron radius.

[0069] The charge density calculation formula is:

[0070] n = B / V

[0071] where n is the charge density, B is the bulk modulus, and V is the unit cell volume.

[0072] For example: Pt 0.8 Rh 0.15 Ta 0.03 La 0.01 Ce 0.01 It contains four binary solid solutions of PtRh, PtTa, PtLa, and PtCe, so only the calculated properties of these four binary solid solutions are used as the corresponding features. It contains five elements, Pt, Rh, Ta, La, and Ce, so the calculated properties of the five elements are used as element features in addition to the binary solid solution features. According to this rule, each sample will contain 16 calculated features.

[0073] Step 4: Weighted average operation of each binary solid solution feature and each element feature of elements A, B, C, and D is performed using the element dose ratio as the weight, and the weighted average feature is obtained. Then, the weighted average features are added or subtracted to obtain the operation features. The weighted average features and the operation features are used as the initial features.

[0074] In step 4, it also includes the step of normalizing the initial features according to the following formula,

[0075]

[0076] where x normalization is the normalized initial feature, x is the initial feature before normalization, μ is the standard deviation of x, and σ is the mean of x.

[0077] This embodiment uses the above formula to normalize the initial features, which scales the features to the same range, avoiding the negative impact on model accuracy when the feature values differ greatly. Based on the normalized data, a machine learning prediction model is established, which can quickly predict the tensile strength of multi-component platinum-based alloys through the composition.

[0078] The calculation process of step 4 is: for example, the composition of platinum-based alloy material Pt 0.8 Rh 0.15 Ta 0.03 La 0.01 Ce 0.01, which contains the A-type element Pt, the B-type element Rh, the C-type element Ta, and the D-type element La and Ce. For the binary layer fault energy of step 3, in each type, the element dose ratio is multiplied by the binary layer fault energy corresponding to the element, and if multiple elements are included in the type, the element dose ratio is multiplied by the binary layer fault energy corresponding to the element and then added separately to complete the layer fault energy weighting tie-breaking calculation in each type. The final characteristics are represented as binary layer fault energy A, binary layer fault energy B, binary layer fault energy C, and binary layer fault energy D. The four weighted average characteristics are then mathematically operated using addition and subtraction operations to obtain binary layer fault energy A-B, binary layer fault energy A-C, binary layer fault energy A-D, and binary layer fault energy A+B+C+D, and binary layer fault energy A, binary layer fault energy B, binary layer fault energy C, and binary layer fault energy D, binary layer fault energy binary layer fault energy A-B, binary layer fault energy A-C, binary layer fault energy A-D, and binary layer fault energy A+B+C+D are used as initial characteristics, and other binary solid solution characteristics are mathematically operated in the manner of the binary layer fault energy characteristic. Finally, 128 new characteristics are generated.

[0079] Step 5, calculate the Pearson correlation coefficient between any two characteristics in the initial characteristics, if the Pearson correlation coefficient is greater than 0.9, remove the redundant characteristics, only keep one of the characteristics, build a feature pool, finally keep 31 characteristics. Then use the 31 characteristics as input for the decision tree algorithm in machine learning, calculate the importance index of each characteristic, the calculation principle is to calculate the information gain generated after increasing or decreasing each characteristic, which is used as the feature importance index standard. The feature importance ranking result is shown in Table 1. Figure 2

[0080] Step 6, take the machine learning ridge regression algorithm accuracy as the objective function, and perform sub-feature iteration screening to select the sub-feature set corresponding to the highest machine learning ridge regression model accuracy from the feature pool; in step 6, according to the feature importance ranking in step 5, delete the last feature each time, and keep the remaining features for machine learning ridge regression model training to select the sub-feature set corresponding to the highest machine learning ridge regression model accuracy. The sub-feature set is the first 14 features in Table 1. Figure 2

[0081] Step 7, take the sub-feature set selected in step 6 as the input independent variable, and the room temperature tensile strength of the platinum-based alloy material as the output dependent variable, and use the ridge regression machine learning algorithm to model to obtain a machine learning model for predicting the room temperature tensile strength of the platinum-based alloy.

[0082] ​​Further, in step 7, the sub-feature set obtained in step 6 is inputted, and the hyperparameters required for setting up the machine learning ridge regression model are set, as shown in Table 1, the machine learning ridge regression model program is started for training, and the hyperparameters are adjusted constantly during the training process, so that the root mean square error of the platinum-based alloy tensile strength value predicted by the machine learning ridge regression and the collected platinum-based alloy tensile strength value is minimized, and the hyperparameters at this time are retained, and the construction of the machine learning model for predicting the tensile strength of the platinum-based alloy at room temperature is completed. More specifically, using the ten-fold cross-validation method, the training set is divided into ten parts, and trained ten times, each time taking nine parts as the training set and the remaining one part as the test set, and the model accuracy is represented by the average accuracy of ten times of training, and the hyperparameters are adjusted constantly during the training process, so that the root mean square error of the platinum-based alloy tensile strength value predicted by the machine learning ridge regression and the collected platinum-based alloy tensile strength value is minimized.

[0083] Parameter Parameter Meaning Parameter Value alpha Regularization coefficient 1 fit_intercept Whether to compute an intercept for this model True normalize Whether or not to standardize the individual features False solver Algorithm to use in the optimization problem auto max_iter Maximum number of iterations taken for the solver to None

[0084] Table 1

[0085] Example Two

[0086] In step 6, the machine learning ridge regression algorithm accuracy is taken as the objective function, the ten-fold cross-validation method is used, the root mean square error and the correlation coefficient of the predicted platinum-based alloy tensile strength value and the collected platinum-based alloy tensile strength value are taken as the evaluation criteria, and the features with the smallest root mean square error and the largest correlation coefficient are selected as the sub-feature set for screening, and the sub-feature screening result is shown in Table 1. Figure 3

[0087] Specifically, using the ten-fold cross-validation method, the training set is divided into ten parts, and trained ten times, each time taking nine parts as the training set and the remaining one part as the test set, and the model accuracy is represented by the root mean square error and the correlation coefficient of the platinum-based alloy tensile strength value predicted by the test set and the collected platinum-based alloy tensile strength value.

[0088] In this embodiment, the ten-fold cross-validation method is used for training the machine learning model in the sub-feature screening process, and the average root mean square error and the average correlation coefficient r value of ten times of training results are used to represent the accuracy, and the machine prediction model with the highest accuracy obtained by training has a root mean square error of 92.1 Mpa and a correlation coefficient of 0.90. The machine learning model constructed by using the optimal feature subset can quickly and accurately predict the tensile strength of the multi-component platinum-based alloy.

[0089] ​The preferred embodiments of the present application are described in detail above with reference to the drawings. The typical known structures and known common knowledge techniques are not described in detail herein, and the ordinary skilled in the art can perfect and implement the technical solutions of the present application based on the inspiration given by the embodiments, and some typical known structures, known methods or known common knowledge techniques should not be an obstacle for the ordinary skilled in the art to implement the present application.

[0090] The scope of protection of the present application should be subject to the content of its claims, and the content recorded in the summary, detailed description and drawings of the specification is used to explain the claims.

[0091] Within the technical concept of the present application, several modifications can also be made to the specific embodiments of the present application, and the specific embodiments after these modifications should also be considered within the protection scope of the present application.

Claims

1. A method for predicting tensile strength of a multi-component platinum-based alloy based on machine learning, characterized by, The method comprises the following steps: Step 1, collecting data of platinum-based alloy materials and corresponding tensile strength data of the platinum-based alloy materials at room temperature to form a data set, wherein each platinum-based alloy material is composed of two or more elements; Step 2, according to the specific role of each element in the material system, all elements contained in all platinum-based alloy materials in step 1 are divided into four types of matrix A, noble metal B, difficult-to-dissolve element C and rare earth element D, wherein the matrix A is platinum Pt; Step 3, based on the elements in the dataset, calculate Pt 31 M binary solid solution properties and element substance properties, get each binary solid solution characteristics and each element substance characteristics, wherein M is a placeholder, refers to other elements in each platinum-based alloy material except Pt element; Step 4, performing weighted average operation on the characteristics of each binary solid solution and each element substance corresponding to the four elements A, B, C and D by taking the element dose ratio as the weight to obtain weighted average characteristics, and then performing addition and subtraction operation on each weighted average characteristic to obtain operation characteristics, and taking the weighted average characteristics and the operation characteristics as initial characteristics; Step 5, calculating the Pearson correlation coefficient between any two characteristics in the initial characteristics, if the Pearson correlation coefficient is greater than 0.9, removing the redundant characteristics and only keeping one of the characteristics to construct a characteristic pool; and performing importance sorting on each characteristic in the characteristic pool; Step 6, taking the accuracy of the machine learning ridge regression algorithm as the objective function to perform sub-feature iteration screening to screen out a sub-feature set corresponding to the highest accuracy of the machine learning ridge regression model from the characteristic pool; Step 7, taking the sub-feature set screened out in step 6 as the input independent variable and the tensile strength of the platinum-based alloy material at room temperature as the output dependent variable, and using the ridge regression machine learning algorithm to model to obtain a machine learning model for predicting the tensile strength of the platinum-based alloy at room temperature.

2. The method of predicting tensile strength of a plurality of multicomponent platinum-based alloys based on machine learning of claim 1, wherein, In step 1, the elements of all platinum-based alloy materials collected contain platinum Pt, rhodium Rh, iridium Ir, gold Au, zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo, nickel Ni, yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb and samarium Sm.

3. The method of predicting tensile strength of a plurality of multicomponent platinum-based alloys based on machine learning of claim 2, wherein, In step 2, the noble metal B contains rhodium Rh, iridium Ir and gold Au, the difficult-to-dissolve element C contains zirconium Zr, ruthenium Ru, rhenium Re, tungsten W, tantalum Ta, titanium Ti, hafnium Hf, osmium Os, molybdenum Mo and nickel Ni, and the rare earth element D contains yttrium Y, lanthanum La, cerium Ce, praseodymium Pr, neodymium Nd, europium Eu, gadolinium Gd, erbium Er, ytterbium Yb and samarium Sm.

4. The method of predicting tensile strength of a plurality of multicomponent platinum- based alloys based on machine learning of claim 1, wherein, In step 3, the binary solid solution characteristics include binary stacking fault energy, binary charge density, binary bulk modulus, binary shear modulus, binary Young's modulus, binary hardness, binary Poisson's ratio and binary electron work function, and the element substance characteristics include substance stacking fault energy, substance charge density, substance bulk modulus, substance shear modulus, substance Young's modulus, substance hardness, substance Poisson's ratio and substance electron work function.

5. The method of predicting tensile strength of a plurality of multicomponent platinum- based alloys based on machine learning of claim 1, wherein, In step 6, according to the importance sorting of the characteristics in step 5, the last characteristic is deleted each time, and the remaining characteristics are retained for machine learning ridge regression model training to screen out a sub-feature set corresponding to the highest accuracy of the machine learning ridge regression model.

6. The method of predicting tensile strength of multi-component platinum group alloy based on machine learning of claim 1, wherein, In step 6, the root mean square error and the correlation coefficient of the predicted platinum-based alloy tensile strength value and the collected platinum-based alloy tensile strength value are used as the evaluation criteria for the machine learning ridge regression algorithm accuracy objective function, and the feature set with the minimum root mean square error and the maximum correlation coefficient is selected as the sub-feature set for screening.

7. The method of predicting tensile strength of multi-component platinum group alloy based on machine learning of claim 1, wherein, In step 7, the sub-feature set obtained in step 6 is input, and the hyperparameters required by the machine learning ridge regression model are set, and the machine learning ridge regression model program is started for training. During the training process, the hyperparameters are continuously adjusted so that the root mean square error of the machine learning ridge regression predicted platinum-based alloy tensile strength value and the collected platinum-based alloy tensile strength value is minimized, and the hyperparameters at this time are retained, completing the construction of the machine learning model for predicting the platinum-based alloy tensile strength at room temperature.

8. The method of predicting tensile strength of multi-component platinum group alloy based on machine learning of claim 1, wherein, In step 4, the initial features are also normalized according to the following formula, where x normalization is the normalized initial feature, x is the initial feature before normalization, μ is the standard deviation of x, and σ is the mean of x.

9. The method of predicting tensile strength of multi-component platinum group alloy based on machine learning of claim 7, wherein, Using the ten-fold cross-validation method, the training set is divided into ten parts, and trained ten times. In each training, nine parts are taken as the training set, and the remaining one part is taken as the test set. The model accuracy is represented by the average accuracy of ten times of training. During the training process, the hyperparameters are continuously adjusted according to the average accuracy so that the root mean square error of the machine learning ridge regression predicted platinum-based alloy tensile strength value and the collected platinum-based alloy tensile strength value is minimized.

Images