CN115061435A - Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof - Google Patents
Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof Download PDFInfo
- Publication number
- CN115061435A CN115061435A CN202210635591.4A CN202210635591A CN115061435A CN 115061435 A CN115061435 A CN 115061435A CN 202210635591 A CN202210635591 A CN 202210635591A CN 115061435 A CN115061435 A CN 115061435A
- Authority
- CN
- China
- Prior art keywords
- hardness
- entropy alloy
- machine learning
- model
- features
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000000956 alloy Substances 0.000 title claims abstract description 96
- 229910045601 alloy Inorganic materials 0.000 title claims abstract description 94
- 238000010801 machine learning Methods 0.000 title claims abstract description 29
- 238000002360 preparation method Methods 0.000 title abstract description 7
- 238000000034 method Methods 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 230000009467 reduction Effects 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 9
- 230000008030 elimination Effects 0.000 claims abstract description 7
- 238000003379 elimination reaction Methods 0.000 claims abstract description 7
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000012795 verification Methods 0.000 claims abstract description 4
- 238000003723 Smelting Methods 0.000 claims description 24
- 239000000463 material Substances 0.000 claims description 21
- XKRFYHLGVUSROY-UHFFFAOYSA-N Argon Chemical compound [Ar] XKRFYHLGVUSROY-UHFFFAOYSA-N 0.000 claims description 12
- RTAQQCXQSZGOHL-UHFFFAOYSA-N Titanium Chemical compound [Ti] RTAQQCXQSZGOHL-UHFFFAOYSA-N 0.000 claims description 12
- 239000010936 titanium Substances 0.000 claims description 12
- 229910052719 titanium Inorganic materials 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 229910003286 Ni-Mn Inorganic materials 0.000 claims description 9
- 238000007637 random forest analysis Methods 0.000 claims description 9
- 229910052782 aluminium Inorganic materials 0.000 claims description 8
- 230000002596 correlated effect Effects 0.000 claims description 8
- 238000002844 melting Methods 0.000 claims description 8
- 230000008018 melting Effects 0.000 claims description 8
- 229910052804 chromium Inorganic materials 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 7
- 229910052748 manganese Inorganic materials 0.000 claims description 7
- 229910052759 nickel Inorganic materials 0.000 claims description 7
- CSCPPACGZOOCGX-UHFFFAOYSA-N Acetone Chemical compound CC(C)=O CSCPPACGZOOCGX-UHFFFAOYSA-N 0.000 claims description 6
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 claims description 6
- 229910052786 argon Inorganic materials 0.000 claims description 6
- 238000001816 cooling Methods 0.000 claims description 6
- 239000010949 copper Substances 0.000 claims description 6
- 238000005336 cracking Methods 0.000 claims description 6
- 238000002156 mixing Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 229910052802 copper Inorganic materials 0.000 claims description 5
- 230000000875 corresponding effect Effects 0.000 claims description 5
- 238000004506 ultrasonic cleaning Methods 0.000 claims description 5
- 238000007664 blowing Methods 0.000 claims description 4
- 238000001035 drying Methods 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 4
- 239000002923 metal particle Substances 0.000 claims description 4
- 238000005303 weighing Methods 0.000 claims description 4
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 claims description 3
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 3
- 230000001276 controlling effect Effects 0.000 claims description 3
- 238000010891 electric arc Methods 0.000 claims description 3
- 235000019441 ethanol Nutrition 0.000 claims description 3
- 238000011049 filling Methods 0.000 claims description 3
- 238000000227 grinding Methods 0.000 claims description 3
- 238000003760 magnetic stirring Methods 0.000 claims description 3
- 239000001301 oxygen Substances 0.000 claims description 3
- 229910052760 oxygen Inorganic materials 0.000 claims description 3
- 238000005086 pumping Methods 0.000 claims description 3
- 230000000630 rising effect Effects 0.000 claims description 3
- 230000008646 thermal stress Effects 0.000 claims description 3
- 238000005406 washing Methods 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 2
- 229910052742 iron Inorganic materials 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims description 2
- 238000010219 correlation analysis Methods 0.000 abstract 1
- 210000001787 dendrite Anatomy 0.000 description 7
- 239000000203 mixture Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 238000002790 cross-validation Methods 0.000 description 3
- 238000005192 partition Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000002441 X-ray diffraction Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 229910001325 element alloy Inorganic materials 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 229910052739 hydrogen Inorganic materials 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 229910000765 intermetallic Inorganic materials 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 239000006104 solid solution Substances 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000001757 thermogravimetry curve Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41865—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by job scheduling, process planning, material flow
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32252—Scheduling production, machining, job shop
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Investigating And Analyzing Materials By Characteristic Methods (AREA)
Abstract
The invention discloses a machine learning method for rapidly predicting the hardness of a high-entropy alloy and a preparation process thereof, which are used for solving the problems of time and labor consumption and inaccurate result of the traditional trial and error method for designing high-entropy alloy components. The method comprises the steps of collecting a high-entropy alloy hardness database as a training set, and calculating a feature descriptor according to alloy components; selecting a proper model according to RMSE under different test set division ratios; performing dimension reduction processing on the obtained features to obtain the most important feature subset; training and optimizing the selected model using the most important feature subset; predicting the hardness of the high-entropy alloy with unknown components by using the trained model; and carrying out test verification on the predicted alloy components. The characteristic dimension reduction method comprises correlation analysis, a recursive elimination method and an exhaustion method, 3 most important characteristics are obtained after dimension reduction treatment, and the hardness of the alloy can be conveniently and accurately predicted according to the three characteristics.
Description
Technical Field
The invention relates to the field of metal material performance prediction, in particular to a method for predicting the hardness performance of a high-entropy alloy by using a machine learning method and verifying a prediction result through a test.
Background
The high-entropy alloy is a new alloy system, and is different from the design concept of the traditional alloy material, the high-entropy alloy consists of five or more than five main elements, and the content of each element is about 5 at.% to 35 at.%. The high configurational entropy phenomenon in high entropy alloys results in alloys that tend to form simple random solid solutions rather than intermetallic compounds such as FCC, BCC or HCP phases. In addition, high entropy alloys exhibit four unique effects: the high-entropy alloy has the advantages of high entropy effect, lattice distortion, slow diffusion and cocktail effect, the effects bring more excellent and abundant performances for the high-entropy alloy, such as high strength and hardness, high-temperature oxidation resistance, wear resistance, corrosion resistance, hydrogen embrittlement resistance and the like, and the high-entropy alloy has wide application prospects.
In the traditional high-entropy alloy design, the components or the process are improved by methods such as theoretical calculation, experimental verification and the like, a trial-and-error method is usually adopted, the methods are time-consuming and labor-consuming, and the result is influenced by various factors. Particularly for materials with huge composition space, such as high-entropy alloys, it is very difficult to find high-entropy alloys with excellent performance by the traditional method. Along with the wide application of the AI technology, the method is widely applied to the fields of automatic driving, image recognition and the like, and particularly along with the development and progress of the technology, the computing capability of a computer is greatly improved, so that the method for searching for materials with excellent performance by applying a machine learning method becomes possible.
The steps of predicting material properties based on machine learning methods are often divided into the following steps: establishing a data set, selecting a model, reducing dimensions of characteristics, training and optimizing the model, predicting a result and verifying a test. The feature selection determines the prediction performance of the model to a great extent, the number of features is dozens or even hundreds, and if the features are considered completely, the calculation amount of a computer is increased greatly. Fortunately, a large number of these features are redundant and are not relevant to the desired result. Therefore, how to select the proper feature combination is particularly important in the process of machine learning and predicting hardness.
Disclosure of Invention
In view of the above problems, the invention provides a machine learning method for rapidly predicting the hardness of a high-entropy alloy and a preparation process thereof, so as to solve the problems of time and labor consumption in designing a high-performance high-entropy alloy by using a traditional method.
The machine learning method for rapidly predicting the hardness of the high-entropy alloy comprises the following steps,
step one, collecting and sorting a high-entropy alloy hardness database, calculating feature descriptors according to alloy components, and constructing a feature data set;
step two, selecting a proper model according to RMSE under different test set division ratios;
step three, performing dimensionality reduction on the obtained features to obtain the most important feature subset;
training the selected model by using the obtained most important feature subset and optimizing the hyper-parameters of the model;
fifthly, the trained model can be used for predicting the hardness of the high-entropy alloy with unknown components;
step six, testing and verifying the hardness of the predicted components;
further, the high-entropy alloy in the step one is an Al-Co-Cr-Cu-Fe-Ni-Mn seven-element alloy system.
Further, the feature descriptor in the step one includes 20 features, which are an average atomic radius, an atomic radius difference, a work function, an average melting point, a mixing enthalpy, a mixing entropy, an Ω parameter, a valence electron concentration, an average electronegativity, an electronegativity difference, a Λ parameter, a number of mobile electrons, an electron affinity, a density, a first ionization energy, an average young modulus, a lattice distortion energy, an cohesive energy, a local size mismatch, and an average shear modulus.
Further, the machine learning model in the second step includes linear regression, random forest regression, support vector machine-radial basis kernel function, support vector machine-sigmoid kernel function, support vector machine-polynomial kernel, K-nearest neighbor and ridge regression.
Further, the test set partition in step two accounts for 15%, 20%, 25%, 30%, 35%, 40% of the total data set, respectively.
Further, the suitable machine learning model in the step two is a random forest regression algorithm model.
Further, the dimension reduction processing process for the feature data set in the third step includes removing features with large correlation by using a pearson correlation coefficient, only retaining one feature in the feature subset with large correlation, obtaining a feature subset with the best model fitting effect by using a recursive elimination method, and further reducing dimensions of the remaining features by using a permutation and combination mode by using an exhaustion method until obtaining the most important feature subset.
Further, the pearson correlation coefficient measures two features that are highly correlated based on the pearson correlation coefficient value being greater than or equal to 0.95.
Further, the basis for removing one of the two highly correlated features is the importance to the model, with the highly important features being preserved.
Further, the optimal feature combination is 3 features of work function, valence electron concentration and Young modulus.
Further, the random forest regression algorithm in the fourth step utilizes a grid search method to perform hyper-parameter tuning.
Further, the step five of predicting the hardness of the alloy with unknown composition is realized by calculating 3 important characteristics in the step three.
A method of preparing an Al-Co-Cr-Cu-Fe-Ni-Mn high entropy alloy as described above, comprising the steps of:
step one, ultrasonic cleaning and material weighing: respectively putting the high-purity metal particles Al, Co, Cr, Cu, Fe, Ni and Mn of the Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy into a container, sequentially cleaning the high-purity metal particles in an ultrasonic cleaning device for 10-20 minutes by using acetone and absolute ethyl alcohol, and finally drying the material by cold air for later use; converting the Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy into corresponding weight according to atomic percent and weighing; sequentially putting the weighed materials into copper crucibles of a WK-II type vacuum arc furnace in sequence according to the sequence of rising melting points, simultaneously grinding off a layer of the surface of a pure titanium ingot by 400-mesh abrasive paper, washing by alcohol, drying by blowing, putting into the middle crucible position of a vacuum furnace, closing and screwing a furnace door for locking;
step two, vacuumizing: after the step is finished, using machinery and a molecular pump to pump vacuum to 1-5 multiplied by 10 -4 Pa, then slowly filling argon to make the vacuum degree reach 1-5 multiplied by 10 4 Pa, pumping the vacuum degree of the electric arc furnace to 1-5 multiplied by 10 -4 Pa; the process is repeated for 3 times, so that no oxygen exists in the vacuum chamber; finally, argon is slowly introduced until the pressure is between-0.02 and-0.05 Mpa.
Step three: smelting: firstly, smelting a pure titanium ingot for 3-5 times, controlling the current to be 300A and the time to be 30-150s, and turning over the pure titanium ingot after each smelting and then smelting the pure titanium ingot for the next time; then smelting the high-entropy alloy, wherein the current is slowly increased, so that the cracking of a sample caused by large thermal stress generated by too fast temperature rise is avoided, the maximum arc current is increased to 300A, and the duration is 50-200 s; slightly rotating the arc rod in the smelting process to uniformly heat the material; when the current is closed, the current is also slowly reduced, and the cracking caused by rapid cooling is also avoided; magnetic stirring with an external magnetic field is used during smelting, and the current is set to be 1-3 mA so as to fully ensure the uniformity of components;
and step four, repeatedly smelting for 4-7 times according to the requirements of the step two to ensure that the components are uniform, and cooling for 10-30min to obtain the high-hardness Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy.
The invention has the beneficial technical effects that:
the invention firstly determines the number of the characteristics in the characteristic combination, then screens the characteristics with large correlation by using the Pearson correlation coefficient, and then screens by using the genetic algorithm to obtain the optimal characteristic combination.
Al predicted by the invention 51 Co 22 Cr 19 Ni 4 Mn 4 The hardness of the high-entropy alloy is 805.3 +/-1.9 HV.
Al obtained by testing of the invention 51 Co 22 Cr 19 Ni 4 Mn 4 The hardness of the high-entropy alloy is 785.9 +/-9.9 HV.
Al predicted by the invention 51 Co 22 Cr 19 Ni 4 Mn 4 The cast structure of the high-entropy alloy is a typical dendrite, and mainly contains a large amount of B2 and BCC phase and a trace amount of FCC phase.
Al obtained by the test of the invention 51 Co 22 Cr 19 Ni 4 Mn 4 The dendrite is rich in Al, Co and Ni, and the dendrite is rich in Al, Cr and Mn among dendrites. This is because Al is easily bonded to other elements. Wherein the Cr-rich region is mainly BCC phase, and the Ni-and Co-rich region is mainly B2 phase.
Drawings
FIG. 1 shows RMSE of 7 different models of the present invention under different test set partitions;
FIG. 2 is a heat map of the Pearson correlation coefficient distribution among 20 features of the present invention;
FIG. 3 is a flow chart of feature selection according to Pearson's correlation coefficient according to the present invention;
FIG. 4 is a diagram of the results of the recursive elimination method of the present invention;
FIG. 5 is a graph of the results of the exhaustive characteristic elimination of the present invention;
FIG. 6 is a graph showing the fitting performance of the model established by the present invention on a training set and a test set, respectively;
FIG. 7 shows Al predicted by the model established in the present invention 51 Co 22 Cr 19 Ni 4 Mn 4 XRD pattern of high entropy alloy;
FIG. 8 shows Al of the present invention 51 Co 22 Cr 19 Ni 4 Mn 4 The structure appearance of the high-entropy alloy and the surface distribution map of Al, Co, Cr, Mn and Ni elements.
Detailed Description
In the following, the machine learning method for rapidly predicting hardness of high-entropy alloy and the manufacturing process thereof according to the present invention will be further described with reference to the following examples and drawings, in order to clearly and concisely describe the embodiments, not all features in the description are described, and only the device structure and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and other details not closely related to the present invention are omitted. The scope of protection of the invention is not limited to the contents of the examples.
The machine learning method for rapidly predicting the hardness of the high-entropy alloy and the preparation process thereof comprise the following steps,
step one, collecting and sorting a high-entropy alloy hardness database, calculating a characteristic descriptor according to alloy components, and constructing a characteristic data set.
According to the embodiment of the invention, the data set used by the seven-element high-entropy alloy system Al-Co-Cr-Cu-Fe-Ni-Mn in the invention comprises 178 components, and the components of the alloy are expressed as Al a Co b Cr c Cu d Fe e Ni f Mn g Wherein a is more than or equal to 0 and less than or equal to 46.2, b is more than or equal to 0 and less than or equal to 42.9, c is more than or equal to 0 and less than or equal to 36.8, d is more than or equal to 0 and less than or equal to 29, e is more than or equal to 0 and less than or equal to 50, f is more than or equal to 0 and less than or equal to 50, and g is more than or equal to 10 and less than or equal to 28 (atomic percent), the average value of the hardness in the data set is 402.9HV, and the standard deviation is 177.0 HV.
Although machine learning based on alloy composition can be used to predict the properties of the alloy, it is phenomenological and does not reveal the physical and chemical characteristics of the elements, the interactions and reactions between the elements, the types and proportions of the constituent phases and other physical and metallurgical mechanisms that affect the properties of the alloy, and this approach is very limited in screening applications of new alloys. Then calculating the physicochemical characteristics of the material based on the alloy composition can solve this problem to some extent. The feature descriptor comprises 20 features, namely average atomic radius (r), atomic radius difference (delta r), work function (W), average melting point (T), mixing enthalpy (delta H), mixing entropy (delta S), omega parameter (omega), Valence Electron Concentration (VEC), average electronegativity (chi), electronegativity difference (chi), lambda parameter (lambda), flowing electron number (E/a), Electron Affinity (EAE), density (rho), First Ionization Energy (FIE), Young modulus (E), lattice distortion energy (mu), cohesive energy (Ec), local size mismatch (D.r) and shear modulus (G), and a feature data set is constructed.
Before training the model, it is necessary to identify a generally accepted fact that variations in the magnitude and range of the input vectors affect the performance of the ML algorithm. This is because features with higher amplitudes typically have greater comparable weights than features with lower amplitudes, which can mislead the machine learning model. To avoid this, it is necessary to ensure that all features are normalized, the formula is as follows:
wherein X norm Denotes the normalized value of the ith feature, X i Represents the value of the ith feature, μ represents the mean of the feature, σ represents the standard deviation, and the processed feature value will be [ -1,1]In between.
Step two, selecting a proper model according to RMSE under different test set division ratios;
according to the No free lunch theorem (No free lunch), at least one machine learning model is needed to ensure the accuracy of the model, so the model selected by the invention comprises Linear Regression (LR), Random Forest Regression (RFR), support vector machine-radial basis kernel function (SVR-r), support vector machine-sigmoid kernel function (SVR-s), support vector machine-polynomial kernel (SVR-p), K-nearest neighbor (KNN) and Ridge regression (Ridge).
In order to explore the influence of different test set divisions on the model performance, the data set is divided randomly, and the test set division ratio columns respectively account for 15%, 20%, 25%, 30%, 35% and 40% of the total data set.
The prediction accuracy of the model is calculated by the root mean square error, and the calculation formula is as follows:
wherein RMSE represents a root mean square error value; n represents the total number of samples; y is i Representing the true value of the ith sample;representing the predicted value of the ith sample.
FIG. 1 shows the RMSE of 7 different models selected by the present invention under different test set partitions, and it can be seen from the figure that the RMSE of the Random Forest Regression (RFR) model is the smallest under different test set proportions, so the present invention selects the random forest regression model as the prediction model.
Step three, performing dimensionality reduction on the obtained features to obtain the most important feature subset;
the feature data set includes a plurality of features related to the predicted performance, but not all of the features affect the predicted result of the target performance, and the features can be divided into useful features, useless features and redundant features for predicting the physical property of the material. Therefore, the dimension reduction of the original characteristic data set can ensure the precision, reduce the complexity of the model, improve the prediction efficiency of the model and have stronger generalization capability in predicting the high-entropy alloy of unknown components.
Firstly, removing features with high correlation degree by using a Pearson correlation coefficient for all feature data sets; the Pearson Correlation Coefficient (PCC) is used to measure the correlation between two quantities, with values between-1 and +1, -1 representing a completely negative correlation and +1 representing a completely positive correlation, with absolute values closer to 1 indicating a greater correlation between the two features, and one of them can be substituted for the other, which is referred to as a redundant feature. The calculation formula of the pearson correlation coefficient is as follows:
wherein r is xy Denotes the Pearson correlation coefficient between the features x, y, x i ,y i Each representing two values, x m 、y m Represents the average of two features.
Fig. 2 shows a pearson correlation coefficient heat map between different features. Lighter colors represent more negative correlation of two features, whereas darker colors are more positive correlation, and features with a correlation greater than 0.95 are highly correlated. According to pearson correlation thermogram analysis, there is a high correlation between characteristic atomic radius, atomic radius difference, local size mismatch, lattice distortion energy, density, between characteristic valence electron concentration and number of mobile electrons, and between melting point, young's modulus and shear modulus.
The highly correlated feature removal is according to the flow chart shown in fig. 3. To determine the features to be retained, the importance of a certain feature to the model is evaluated, and the importance of each feature is ranked, as shown in table 1. Finally, 6 features that are highly correlated and of low importance, including the atomic radius (r), local size mismatch (D.r), lattice distortion energy (μ), number of mobile electrons (e/a), melting point (T), and shear modulus (G) features, were removed, leaving 14 features.
TABLE 1
In order to further screen the best performing features, recursive elimination of features is performed. We take turns to randomly select one feature (14 features after excluding the highly correlated ones) and the remaining n-1 features are used as input vectors to build the RFR model. In such a process, the alloy features corresponding to the smallest importance are then excluded, leaving n-1 alloy features, which are then used to reconstruct the model. We recursively perform the elimination until no features remain, and then obtain several feature subsets with different numbers of features. We use a cross-validation approach to test the scores of the model using various feature subsets, during which the model itself remains unchanged.
As can be seen in fig. 4, when the number of features is less than 7, the cross-validation score increases as the number of selected features increases within 7 and reaches a maximum at 7. We have obtained a new subset of features comprising 7 key features, specifically δ r, W, VEC, EAE, ρ, FIE and E, which closely influence the stiffness of the HEAs.
The last step of feature dimension reduction is exhaustive feature screening. We created a series of RFR models by using an exhaustive combination of all remaining 7 alloy features as input vectors. By comparing the RMSE values of the model prediction results under different feature combinations, 3 key features with the largest influence on the alloy hardness are finally determined.
The features are further filtered by considering all possible combinations of these features to identify the subset with the lowest model error. FIG. 5 illustrates RMSE values for RFR models based on different subsets combined with 1-7 features. As can be seen, the minimum RMSE (i.e., ". major:" mark "in the figure) corresponds to a number of features of 3, including W, VEC and E.
Based on the feature screening process described above, we constructed an RFR model using 80% of the data set and 3 additional features, with the remaining 20% of the data set being used to test the model. The predicted performance of our model is shown in fig. 6. As can be seen, the scatter point of the experimental values and the predicted values of the alloys in the training set and the test set is close to the diagonal line y ═ x, indicating that the model is suitable.
Training the selected model by using the obtained most important feature subset and optimizing the hyper-parameters of the model;
by using a 10-fold cross validation method, an over-parameter n _ estimators which has the largest influence on a random forest model is adjusted by drawing a learning curve, when the value is 31, the prediction performance of the model is the best, so the n _ estimators is set to 31, the subsequent parameter adjusting direction is the direction for reducing the complexity of the model, so that the lowest point of the generalization error is approached, the max _ depth which can reduce the complexity of the model and has the second influence degree is used, and when the max _ depth is 14, the model performance is the best.
Fifthly, the trained model can be used for predicting the hardness of the high-entropy alloy with unknown components; inputting the 3 important characteristics obtained in the third step of calculating the alloy with unknown components as a prediction set into a trained model, thereby obtaining a corresponding hardness value; the highest entropy alloy of predicted hardness is: al (Al) 51 Co 22 Cr 19 Ni 4 Mn 4 The hardness was found to be 805.3. + -. 1.9 HV.
And sixthly, alloy hardness test verification of the predicted components.
Preparation of Al 51 Co 22 Cr 19 Ni 4 Mn 4 High entropy alloys in which the atoms of the elements are hundredThe composition comprises Al 50-52%, Co 21-23%, Cr 18-20%, Ni 3-5%, and Mn 3-5%.
Step one, Al is added 51 Co 22 Cr 19 Ni 4 Mn 4 Preparing materials (Al material, Co material, Cr material, Ni material and Mn material) of the high-entropy alloy are respectively put into a glass beaker, are sequentially cleaned for 10-20 minutes in an ultrasonic cleaning device by using acetone and absolute ethyl alcohol, and are finally dried by cold air for standby. Adding the Al 51 Co 22 Cr 19 Ni 4 Mn 4 The high-entropy alloy material is converted into corresponding weight according to atomic percent, and each metal component is weighed. Sequentially putting the weighed materials into copper crucibles of a WK-II type vacuum arc furnace in sequence according to the sequence of rising melting points, simultaneously grinding off a layer of the surface of a pure titanium ingot by 400-mesh abrasive paper, washing by alcohol, drying by blowing, putting into the middle crucible position of a vacuum furnace, closing and screwing a furnace door for locking;
step two, vacuumizing: after the step is finished, using machinery and a molecular pump to pump vacuum to 1-5 multiplied by 10 -4 Pa, then slowly filling argon to make the vacuum degree reach 1-5 multiplied by 10 4 Pa, pumping the vacuum degree of the electric arc furnace to 1-5 multiplied by 10 -4 Pa; the process is repeated for 3 times, so that no oxygen exists in the vacuum chamber; finally, argon is slowly introduced until the pressure is between-0.02 and-0.05 Mpa.
Step three: smelting: firstly, smelting a pure titanium ingot for 3-5 times, controlling the current to be 300A and the time to be 30-150s, and turning over the pure titanium ingot after each smelting and then smelting the pure titanium ingot for the next time; then smelting the high-entropy alloy, wherein the current is slowly increased, so that the cracking of a sample caused by large thermal stress generated by too fast temperature rise is avoided, the maximum arc current is increased to 300A, and the duration is 50-200 s; slightly rotating the arc rod in the smelting process to uniformly heat the material; when the current is closed, the current is also slowly reduced, and the cracking caused by rapid cooling is also avoided; magnetic stirring with an external magnetic field is used during smelting, and the current is set to be 1-3 mA so as to fully ensure the uniformity of components;
step four, repeatedly smelting for 4-7 times according to the requirements of step two to ensure that the components are uniform, and cooling for 10-30min to obtain high-hardness Al 51 Co 22 Cr 19 Ni 4 Mn 4 High entropy alloy.
For prepared Al 51 Co 22 Cr 19 Ni 4 Mn 4 Phase structure analysis of the high entropy alloy, Al as shown in FIG. 7 51 Co 22 Cr 19 Ni 4 Mn 4 The XRD pattern of the high-entropy alloy shows that Al 51 Co 22 Cr 19 Ni 4 Mn 4 The high entropy alloy is mainly composed of B2, BCC phase, with a small amount of FCC phase present.
For Al prepared in example 1 51 Co 22 Cr 19 Ni 4 Mn 4 Hardness test was conducted on the high-entropy alloy, and it can be seen from FIG. 3 that Al is present 51 Co 22 Cr 19 Ni 4 Mn 4 Has better hardness, and the hardness value reaches 785.9 +/-9.9 HV.
For Al prepared in example 1 51 Co 22 Cr 19 Ni 4 Mn 4 The high entropy alloy is subjected to the determination of the structure morphology and the components, and FIG. 8 is Al of the invention 51 Co 22 Cr 19 Ni 4 Mn 4 The structure appearance of the high-entropy alloy and the surface distribution map of Al, Co, Cr, Mn and Ni elements.
The apparent dendrite morphology is shown in the surface distribution diagram of FIG. 8, and the dendrites are rich in Al, Co and Ni, and the dendrites are rich in Al, Cr and Mn. This is because Al is easily bonded to other elements. Wherein the Cr-rich region is mainly BCC phase, and the Ni-and Co-rich region is mainly B2 phase.
The embodiments show that the invention provides the machine learning method for rapidly predicting the hardness of the high-entropy alloy and the preparation process thereof, the hardness of the alloy can be conveniently and accurately predicted by adopting the machine learning method by mining a large amount of existing component-performance data, and the problems of time consumption, labor consumption and inaccurate result of designing the high-entropy alloy by the traditional trial-and-error method can be solved to a certain extent
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this description, will appreciate that other embodiments can be devised which do not depart from the scope of the invention as disclosed herein. Furthermore, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the appended claims. The present invention has been disclosed with respect to the scope of the invention, which is to be considered as illustrative and not restrictive, and the scope of the invention is defined by the appended claims.
Claims (10)
1. A machine learning method for rapidly predicting the hardness of a high-entropy alloy is characterized by comprising the following steps,
step one, collecting and sorting a high-entropy alloy hardness database, calculating feature descriptors according to alloy components, and constructing a feature data set;
step two, selecting a proper model according to RMSE under different test set division ratios;
step three, performing dimensionality reduction on the obtained features to obtain the most important feature subset;
step four, training the selected model by using the obtained most important feature subset and optimizing the hyper-parameters of the model;
fifthly, the trained model can be used for predicting the hardness of the high-entropy alloy with unknown components;
and sixthly, alloy hardness test verification of the predicted components.
2. The machine learning method for rapidly predicting the hardness of the high-entropy alloy as claimed in claim 1, wherein the high-entropy alloy in the first step is a seven-element Al-Co-Cr-Cu-Fe-Ni-Mn alloy system.
3. The machine learning method for rapidly predicting the hardness of the high-entropy alloy according to claim 1, wherein the feature descriptors in the first step include 20 features, which are respectively an average atomic radius, an atomic radius difference, a work function, an average melting point, a mixing enthalpy, a mixing entropy, an omega parameter, a valence electron concentration, an average electronegativity, an electronegativity difference, a lambda parameter, a number of flowing electrons, an electron affinity, a density, a first ionization energy, a Young modulus, a lattice distortion energy, an cohesive energy, a local size mismatch, and an average shear modulus.
4. The machine learning method for rapidly predicting the hardness of the high-entropy alloy according to claim 1, wherein the machine learning model in the second step comprises linear regression, random forest regression, support vector machine-radial basis kernel function, support vector machine-sigmoid kernel function, support vector machine-polynomial kernel, K-nearest neighbor and ridge regression.
5. A machine learning method for rapid prediction of hardness of high entropy alloys according to claim 1, wherein the randomly partitioned test sets in step two account for 15%, 20%, 25%, 30%, 35%, 40% of the total data set, respectively.
6. The machine learning method for rapidly predicting the hardness of the high-entropy alloy according to claim 1, wherein the suitable machine learning model in the second step is a random forest regression algorithm model.
7. The machine learning method for rapidly predicting the hardness of the high-entropy alloy according to claim 1, wherein the step three of performing the dimension reduction processing on the feature data set comprises the following steps: the method comprises the steps of removing features with large correlation by using a Pearson correlation coefficient, only keeping one feature in a feature subset with large correlation, obtaining a feature subset with the best model fitting effect by using a recursive elimination method, and reducing the dimension of the rest features by using a permutation and combination mode by using an exhaustion method until obtaining the most important feature subset.
8. The machine learning method for rapidly predicting hardness of high-entropy alloy according to claim 7, wherein the Pearson correlation coefficient measures two highly correlated features based on the Pearson correlation coefficient value being 0.95 or more, and the removing of one of the two highly correlated features is based on the importance of the feature to the model, and the feature with high importance is retained.
9. The machine learning method for rapidly predicting the hardness of the high-entropy alloy according to claim 7, wherein the most important feature subset comprises 3 features of work function, valence electron concentration and Young modulus.
10. The method for preparing the Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy according to claim 2, characterized by comprising the following steps:
step one, ultrasonic cleaning and material weighing: respectively putting the high-purity metal particles Al, Co, Cr, Cu, Fe, Ni and Mn of the high-entropy alloy into a container, sequentially cleaning the high-purity metal particles in an ultrasonic cleaning device for 10-20 minutes by using acetone and absolute ethyl alcohol, and finally blowing the materials to dry for later use by using cold air; converting the high-entropy alloy into corresponding weight according to atomic percent and weighing; sequentially putting the weighed materials into copper crucibles of a WK-II type vacuum arc furnace in sequence according to the sequence of rising melting points, simultaneously grinding off a layer of the surface of a pure titanium ingot by 400-mesh abrasive paper, washing by alcohol, drying by blowing, putting into a middle crucible position of a vacuum furnace, closing and screwing a furnace door for locking;
step two, vacuumizing: after the step is finished, using a mechanical pump and a molecular pump to vacuumize to 1-5 multiplied by 10 -4 Pa, then slowly filling argon to make the vacuum degree reach 1-5 multiplied by 10 4 Pa, pumping the vacuum degree of the electric arc furnace to 1-5 multiplied by 10 -4 Pa; the process is repeated for 3 times, so that no oxygen exists in the vacuum chamber; finally, slowly introducing argon to-0.02 to-0.05 Mpa;
step three: smelting: firstly, smelting a pure titanium ingot for 3-5 times, controlling the current to be 300A and the time to be 30-150s, and turning over the pure titanium ingot after each smelting and then smelting the pure titanium ingot for the next time; then smelting the high-entropy alloy, wherein the current is slowly increased, so that the cracking of a sample caused by large thermal stress generated by too fast temperature rise is avoided, the maximum arc current is increased to 300A, and the duration is 50-200 s; slightly rotating the arc rod in the smelting process to uniformly heat the material; when the current is closed, the current is also slowly reduced, and the cracking caused by rapid cooling is also avoided; magnetic stirring with an external magnetic field is used during smelting, and the current is set to be 1-3 mA so as to fully ensure the uniformity of components;
and step four, repeatedly smelting for 4-7 times according to the requirements of the step two to ensure that the components are uniform, and cooling for 10-30min to obtain the high-hardness Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210635591.4A CN115061435A (en) | 2022-06-06 | 2022-06-06 | Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210635591.4A CN115061435A (en) | 2022-06-06 | 2022-06-06 | Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115061435A true CN115061435A (en) | 2022-09-16 |
Family
ID=83199558
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210635591.4A Pending CN115061435A (en) | 2022-06-06 | 2022-06-06 | Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115061435A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116227252A (en) * | 2023-05-10 | 2023-06-06 | 吉林大学 | Method for determining atomic distribution on surface of noble metal high-entropy alloy based on linear function |
-
2022
- 2022-06-06 CN CN202210635591.4A patent/CN115061435A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116227252A (en) * | 2023-05-10 | 2023-06-06 | 吉林大学 | Method for determining atomic distribution on surface of noble metal high-entropy alloy based on linear function |
CN116227252B (en) * | 2023-05-10 | 2023-07-04 | 吉林大学 | Method for determining atomic distribution on surface of noble metal high-entropy alloy based on linear function |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wen et al. | Machine learning assisted design of high entropy alloys with desired property | |
Yan et al. | Accelerated discovery of single-phase refractory high entropy alloys assisted by machine learning | |
Ohkubo et al. | Electron diffraction and high-resolution electron microscopy study of an amorphous Pd 82 Si 18 alloy with nanoscale phase separation | |
Chaudry et al. | Machine learning-aided design of aluminum alloys with high performance | |
CN115061435A (en) | Machine learning method for rapidly predicting hardness of high-entropy alloy and preparation process thereof | |
Hu et al. | Application of atom probe tomography in understanding high entropy alloys: 3D local chemical compositions in atomic scale analysis | |
CN111286599A (en) | Alloy heat treatment process optimization method based on machine learning | |
Xu et al. | Design of super-hard high-entropy ceramics coatings via machine learning | |
Ren et al. | Predicting single-phase solid solutions in as-sputtered high entropy alloys: High-throughput screening with machine-learning model | |
CN116312890A (en) | Method for screening high-hardness high-entropy alloy by aid of particle swarm optimization algorithm and machine learning | |
CN114678086A (en) | Low-activation high-entropy alloy and design method thereof based on machine learning | |
Liu et al. | Evaluating yield strength of Ni-based superalloys via high throughput experiment and machine learning | |
Dong et al. | Machine Learning Aided Prediction and Design for the Mechanical Properties of Magnesium Alloys | |
Kim et al. | Optimal machine learning feature selection for assessing the mechanical properties of a zeolite framework | |
Wu et al. | Machine learning-assisted efficient design of Cu-based shape memory alloy with specific phase transition temperature | |
CN117037962A (en) | Method for rapidly designing high-strength refractory multi-principal element alloy based on machine learning | |
CN113870957B (en) | Eutectic high-entropy alloy composition design method and device based on machine learning | |
CN116504315A (en) | Interpolation method for single-cell RNA sequencing data deletion based on improved BEGAN network | |
CN115579091A (en) | High-entropy alloy component design method based on machine learning and multi-performance collaborative optimization | |
Zhang et al. | Hardness-guided machine learning for tungsten alloy strength prediction | |
Cieślak et al. | Determination of the Debye temperature of the σ-phase Fe-Cr alloys | |
CN117766069A (en) | High-entropy alloy hardness prediction method based on machine learning and explanatory analysis | |
Krishna | Titanium-Based Alloys with High-Performance: Design and Development | |
Wang et al. | Searching for medium entropy alloys with desired mechanical property by adaptive design combined with experiments | |
Xiong et al. | Analysis of Phase Formations and Mechanical Properties in Complex Concentrated Alloys by Machine Learning Approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |