Tungsten-nickel-iron alloy strength and toughness dual-objective optimization method and system based on machine learning

Abstract

The invention discloses a tungsten-nickel-iron alloy strength and toughness dual-objective optimization method and system based on machine learning, and the method comprises the following steps: obtaining original data of the components, sintering process, processing process, heat treatment process, strength and toughness of a tungsten-nickel-iron alloy, and carrying out the integration and summarization, so as to obtain an initial data set; preprocessing the initial data set to obtain a processed data set; constructing a plurality of candidate models based on a machine learning algorithm, screening the plurality of candidate models, and training the screened optimal candidate model by using the processed data set to obtain a prediction model; performing strength and toughness prediction on the target tungsten-nickel-iron alloy by using the prediction model to obtain a prediction result; and on the basis of the prediction result, candidate alloys meeting the application requirements are screened out, a dual-objective optimization framework is constructed by using a Pareto optimal criterion to optimize the candidate alloys, and the alloy with the strength and toughness cooperatively optimized is obtained.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of metal material performance prediction and optimization, and particularly relates to a tungsten-nickel-iron alloy strength and toughness double-target optimization method and system based on machine learning. BACKGROUND

[0002] Tungsten-nickel-iron alloy has excellent performance and has been widely used in advanced weapon equipment such as large-caliber kinetic energy penetrator cores and weapon gyro outer rotor bodies. The typical microstructure of the alloy is composed of tungsten particles and a matrix phase composed of elements such as Ni, Fe, Mo, Cu and Co, showing a unique two-phase structure and having high melting point and high density.

[0003] The most significant mechanical performance advantage of tungsten-nickel-iron alloy is its high tensile strength and yield strength, which enables it to maintain structural integrity under extreme conditions. However, the main application bottleneck of the alloy is the balance between high strength and high toughness. Specifically, when the strength of the alloy is increased to more than 1300 MPa, the elongation rate is usually reduced to less than 10%. This indicates that in the pursuit of high strength, a certain toughness may be sacrificed, thereby limiting its use in certain application scenarios that require high toughness.

[0004] Currently, the composition and process design of tungsten-nickel-iron alloy mainly rely on experimental methods, and there is a lack of reliable models describing the quantitative relationship between composition, process and performance, which cannot accurately and quickly carry out alloy design according to the required performance. Machine learning, as a typical data mining and analysis method, can reveal the intrinsic information of materials and establish a non-linear relationship through analysis and mining of a large amount of data, and effectively explore the unknown composition and process space. This method can significantly improve the efficiency of new material research and development and save alloy design costs. In recent years, researchers have successfully predicted the hardness and sintering density of tungsten-nickel-iron alloy through machine learning models, but have not made breakthroughs in predicting strength and toughness, and the double-target optimization problem of strength and toughness has not been solved.

[0005] Therefore, using the tungsten-nickel-iron alloy data set, using machine learning method to carry out double-target optimization of strength and toughness has important guiding significance for realizing rapid design of tungsten-nickel-iron alloy and expanding its application field. SUMMARY

[0006] The application aims to solve the problems of the prior art and provides the following solutions:

[0007] A tungsten-nickel-iron alloy strength and toughness double-target optimization method based on machine learning, comprising the following steps:

[0008] Obtain original data of composition, sintering process, processing technology, heat treatment process, strength and toughness of tungsten-nickel-iron alloy, and integrate and summarize to obtain an initial data set;

[0009] Preprocess the initial data set to obtain a processed data set;

[0010] Construct a plurality of candidate models based on a machine learning algorithm, screen the plurality of candidate models, and train the optimal candidate model screened out using the processed data set to obtain a prediction model;

[0011] Use the prediction model to predict the strength and toughness of a target tungsten-nickel-iron alloy to obtain a prediction result;

[0012] Based on the prediction result, a candidate alloy meeting the application requirement is screened out, and a dual-objective optimization framework is constructed using the Pareto optimal criterion to optimize the candidate alloy, thereby obtaining an alloy with synergistically optimized strength and toughness.

[0013] Preferably, the preprocessing method comprises:

[0014] In the initial data set, delete original feature variables with a missing value ratio exceeding 50% to perform data cleaning to obtain a cleaned data set;

[0015] In the cleaned data set, for performance data with the same number of original feature variable values, perform mean fusion to obtain a fused data set;

[0016] Use K-means clustering analysis to remove outlier data points in the fused data set to perform data balancing to obtain the processed data set.

[0017] Preferably, the method for obtaining the prediction model comprises:

[0018] Perform feature extraction and selection on the processed data set to screen out feature variables having an influence on strength and toughness;

[0019] Construct a plurality of candidate models based on a machine learning algorithm, use five-fold cross-validation to evaluate the plurality of candidate models, and screen out the optimal candidate model;

[0020] Divide the feature variables into a training set and a test set according to a ratio of 4:1, train the optimal candidate model using the training set, and test the performance of the trained model using the test set to obtain the prediction model.

[0021] Preferably, the method for optimizing the candidate alloy using the Pareto optimal criterion to construct a dual-objective optimization framework comprises:

[0022] Adopting the principle of sequential screening, constraint conditions are set according to the design requirements of the alloy:

[0023] f TS = Q 0.8 (Y TS )

[0024] f EL = Q 0.8 (Y EL )

[0025] p t >= 17.5 g / cm 3

[0026] Wherein, f TS represents the predicted tensile strength of the alloy, Y TS represents the tensile strength in the data set, f EL represents the predicted elongation of the alloy, Y EL represents the elongation in the data set, p t represents the theoretical density of the alloy, Q 0.8 represents the 80th percentile in the data set;

[0027] Based on the prediction results and the constraint conditions, candidate alloys meeting the application requirements are screened out, and a double-objective optimization problem is constructed:

[0028] max x∈χ [f TS (x), f EL (x)]

[0029] Wherein, x represents the input variable, and x represents the design space formed after screening;

[0030] The objective function numerical vector is constructed:

[0031] F(x) = [f TS (x), f EL (x)]

[0032] The double-objective optimization analysis of the objective function numerical vector is carried out by using the Pareto optimal criterion:

[0033] f TS (x1) >= f TS (x2)

[0034] f EL (x1) >= f EL (x2)

[0035] If at least one of the inequalities in the analysis result is true, it is judged that x2 is dominated by x1, and all solutions not dominated by other points are constructed into the Pareto front solution set;

[0036] The Pareto front solution set is hierarchically sorted by using fast non-dominated sorting, and the Pareto optimal set of the first layer is extracted as an output, so as to obtain the alloy after strength and toughness collaborative optimization.

[0037] The application also provides a machine learning-based tungsten-nickel-iron alloy strength and toughness double-target optimization system, which applies the method in any of the above aspects and comprises a data acquisition module, a data processing module, a model construction module, a prediction module and an optimization module.

[0038] The data acquisition module is used to acquire and integrate and aggregate the original data of the composition, sintering process, processing process, heat treatment process, strength and toughness of the tungsten-nickel-iron alloy, so as to obtain an initial data set.

[0039] The data processing module is used to pre-process the initial data set, so as to obtain a processed data set.

[0040] The model construction module constructs a plurality of candidate models based on a machine learning algorithm, screens the plurality of candidate models, and trains the optimal candidate model screened out by using the processed data set, so as to obtain a prediction model.

[0041] The prediction module uses the prediction model to predict the strength and toughness of a target tungsten-nickel-iron alloy, so as to obtain a prediction result.

[0042] The optimization module screens out a candidate alloy meeting the application requirement based on the prediction result, and uses a Pareto optimal criterion to construct a double-target optimization framework to optimize the candidate alloy, so as to obtain the alloy after strength and toughness collaborative optimization.

[0043] Preferably, the working process of the data processing module comprises:

[0044] In the initial data set, the original characteristic variables with a missing value ratio exceeding 50% are deleted, data cleaning is performed, and a cleaned data set is obtained.

[0045] In the cleaned data set, the performance data with the same original characteristic variable values are subjected to mean fusion, and a fused data set is obtained.

[0046] The outlying data points in the fused data set are removed by using K-means clustering analysis, data balancing is performed, and the processed data set is obtained.

[0047] Preferably, in the model construction module, the method for constructing the prediction model comprises:

[0048] The processed data set is subjected to feature extraction and selection, and the characteristic variables having an influence on the strength and toughness are screened out.

[0049] constructing several candidate models based on machine learning algorithm, evaluating the several candidate models by five-fold cross-validation, and screening the optimal candidate model;

[0050] dividing the feature variables into a training set and a test set according to a ratio of 4:1, training the optimal candidate model by using the training set, and testing the performance of the model after training by using the test set, to obtain the prediction model.

[0051] Preferably, the workflow of the optimization module comprises:

[0052] adopting sequential screening principle, and setting constraint conditions according to the design requirements of the alloy:

[0053] f TS = Q 0.8 (Y TS )

[0054] f EL = Q 0.8 (Y EL )

[0055] p t ≥ 17.5 g / cm 3

[0056] wherein f TS represents the predicted tensile strength of the alloy, Y TS represents the tensile strength in the data set, f EL represents the predicted elongation of the alloy, Y EL represents the elongation in the data set, p t represents the theoretical density of the alloy, Q 0.8 represents the 80th percentile in the data set;

[0057] based on the prediction results and the constraint conditions, screening candidate alloys meeting the application requirements, and constructing a double-objective optimization problem:

[0058] max x∈χ [f TS (x), f EL (x)]

[0059] wherein x represents an input variable, and x represents a design space formed after screening;

[0060] constructing a target function numerical vector:

[0061] F(x) = [f TS (x), f EL (x)]

[0062] The target function numerical vector is subjected to double-target optimization analysis by using a Pareto optimization criterion:

[0063] f TS (x1)≥f TS (x2)

[0064] f EL (x1)≥f EL (x2)

[0065] If at least one of the inequalities in the analysis result is true, it is judged that x2 is dominated by x1, and all solutions not dominated by other points are constructed into a Pareto front solution set;

[0066] The Pareto front solution set is subjected to hierarchical sorting by using fast non-dominated sorting, and the first layer of the Pareto optimal set is extracted as an output, so as to obtain an alloy after strength and toughness collaborative optimization.

[0067] Compared with the prior art, the beneficial effects of the present application are:

[0068] The present application can quickly design a tungsten-nickel-iron alloy with excellent comprehensive performance without a large number of experimental processes. Through the prediction and optimization of the machine learning model, the number of experiments and the cost are significantly reduced, and the development process of the new tungsten-nickel-iron alloy is accelerated. The optimized tungsten-nickel-iron alloy exhibits better comprehensive performance than the existing results in the data set in terms of strength and toughness, meeting the application requirements of high strength and high toughness. By defining the composition-technology search space and conducting virtual alloy design, the present application can provide customized alloy design schemes for different application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0070] Figure 1 The method flowchart of the embodiment of the present application;

[0071] Figure 2 The schematic diagram of data balancing by K-means clustering in the embodiment of the present application;

[0072] Figure 3Schematic diagram of feature engineering in an embodiment of the present invention, where a is the Pearson coefficient between strength features, b is the Pearson coefficient between toughness features, c is the exhaustive screening result of strength feature combinations, d is the exhaustive screening result of toughness feature combinations, e is the SHAP importance ranking of strength features, and f is the SHAP importance ranking of toughness features;

[0073] Figure 4 The results of the model algorithm screening in the embodiment of the present invention, where a is the strength result and b is the toughness result;

[0074] Figure 5 is the training result of the XGBoost model in the embodiment of the present invention, where a is the strength result and b is the toughness result;

[0075] Figure 6 These are the steps and results of dual-objective optimization in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0078] Example 1

[0079] In this embodiment, if Figure 1 As shown, a dual-objective optimization method for strength and toughness of tungsten-nickel-iron alloy based on machine learning includes the following steps:

[0080] S1. Obtain and integrate the original data on the composition, sintering process, processing technology, heat treatment process, mechanical properties, strength, and toughness of the tungsten-nickel-iron alloy to obtain an initial data set.

[0081] In the embodiment, the alloy composition, heat treatment condition, deformation condition and corresponding performance data of tungsten-nickel-iron series, tungsten-nickel-copper series and tungsten-nickel-cobalt series alloys reported in literature are collected as characteristic original data, specifically including the content of Ni element, Fe element, Co element, Cu element, Re element, Mo element, Mn element, Cr element, Ta element, Hf element, La element, sintering temperature, sintering time, heat treatment temperature, heat treatment time, forging amount before heat treatment, forging amount after heat treatment, strength (tensile strength) and toughness (elongation). The collected initial sample set is the initial data set. The data in the initial data set is obtained by using the same powder metallurgy process of mixing, pressing and forging, and the obtained alloy data set structure is shown in Table 1:

[0082] Table 1

[0083] Number Mean Standard Deviation Minimum 25th Percentile 50th Percentile 75th Percentile Maximum X_Co 324 0.4 0.8 0.0 0.0 0.0 0.3 3.2 X_Cu 324 0.1 0.5 0.0 0.0 0.0 0.0 3.0 X_Re 324 0.2 1.0 0.0 0.0 0.0 0.0 13.5 X_Mo 324 0.8 2.9 0.0 0.0 0.0 0.0 17.1 X_Mn 324 0.1 0.6 0.0 0.0 0.0 0.0 4.0 X_Cr 324 0.0 0.2 0.0 0.0 0.0 0.0 1.5 X_Ta 324 0.2 0.8 0.0 0.0 0.0 0.0 5.0 X_Hf 324 0.0 0.1 0.0 0.0 0.0 0.0 1.5 X_La 324 0.0 0.1 0.0 0.0 0.0 0.0 0.8 X_NFC 324 9.5 9.3 2.3 5.6 7.0 10.0 50.0 X_NFR 324 2.5 2.4 0.0 2.0 2.3 2.3 28.0 X_ST 324 1472.7 56.6 1150.0 1470.0 1490.0 1500.0 1560.0 X_STT 324 5.1 8.5 0.1 1.0 1.5 2.0 24.0 X_PD 324 1.6 5.9 0.0 0.0 0.0 0.0 35.0 X_HT 324 572.9 559.7 25.0 25.0 350.0 1150.0 1300.0 X_HTT 324 1.9 2.8 0.0 0.0 0.7 2.0 9.0 X_D 324 3.2 11.5 0.0 0.0 0.0 0.0 84.8 Y_TS 324 990.0 181.4 450.0 898.8 951.0 1053.3 1646.0 Y_EL 324 16.4 9.1 1.0 9.1 15.6 24.0 37.1 .

[0084] S2. The initial data set is preprocessed to obtain a processed data set.

[0085] The preprocessing method includes: in the initial data set, deleting original feature variables with a missing value ratio of more than 50%, performing data cleaning to obtain a cleaned data set; in the cleaned data set, for performance data with the same original feature variable value, performing mean fusion to obtain a fused data set; using K-means clustering analysis to remove outlier data points in the fused data set, balancing the data, and obtaining a processed data set.

[0086] In the embodiment, the initial data set is preprocessed. First, the features with a missing value ratio of more than 50% in the characteristic original data are directly deleted to ensure the integrity and reliability of the data; for the missing values in the reserved variables, reasonable default values are used to fill in to ensure the integrity of the data. Then, for performance data with the same feature value, the data is fused by taking the mean value to reduce data redundancy and improve data quality. On this basis, K-means clustering analysis is used to remove outlier data points, as shown in Figure 2 , the turning point of the total error sum of squares and the point with relatively high contour coefficient are selected as the best clustering number, and then the 50 data points with the closest Euclidean distance to the cluster center in the cluster are selected. The finally determined best clustering number is 4, and after the above processing, the data amount in the finally balanced data set obtained is 200.

[0087] S3. A plurality of candidate models are constructed based on machine learning algorithms, and the plurality of candidate models are screened, and the processed data set is used to train the optimal candidate model screened to obtain a prediction model.

[0088] The method for obtaining a prediction model includes: extracting and selecting features from a processed data set to screen out characteristic variables that affect strength and toughness; constructing several candidate models based on a machine learning algorithm, evaluating the candidate models using five-fold cross-validation, and screening out the optimal candidate model; dividing the characteristic variables into a training set and a test set in a 4:1 ratio, training the optimal candidate model using the training set, and performing a performance test on the trained model using the test set to obtain a prediction model.

[0089] In this embodiment, feature engineering is used to extract and select features from the processed data set. The specific feature engineering is as follows: Figure 3 Calculate the Pearson correlation coefficient between any two features in the dataset, check the correlation between any two feature data, and remove collinear features with |ρ|>0.95. The formula is as follows:

[0090]

[0091] Where ρ represents the Pearson correlation coefficient, x i and y i represents the factor value of the i-th sample, and represents the average value of the corresponding factor in all samples. n -1 non-empty feature combination for model training, and the feature combination with the best performance is selected as an alternative. More preferably, the SHAP algorithm based on game theory is used to sort the features in the best feature combination by importance, and the feature with the least importance is deleted to obtain the final feature subset. The SHAP value calculation formula is as follows:

[0092]

[0093] in, is a constant, M represents the number of input features, represents the SHAP contribution of the i-th input variable to the alloy property in the j-th sample. The method of screening the best candidate model includes: evaluating the prediction accuracy and generalization ability of the commonly used regression algorithms in strength and toughness prediction through five-fold cross validation, and the performance of each model is as follows: Figure 4 As shown, the extreme gradient boosting regression algorithm (XGBoost) was ultimately selected as the basic algorithm for predicting the strength and toughness of tungsten-nickel-iron alloy. The constructed machine learning model consists of multiple regression trees, each of which is used to predict the residual, continuously correct the predicted value, and finally output the weighted sum result. The model prediction value is as follows:

[0094]

[0095] where K denotes the number of total trees, f k denotes the kth tree, denotes the function space of all trees; the optimization objective function is:

[0096]

[0097]

[0098] where T denotes the number of leaves, ω j denotes the weight of the jth leaf, and γ and λ both denote regularization parameters. For the nonlinear mapping relationship between the composition-process and mechanical properties of tungsten-nickel-iron alloy, the key hyperparameters of the selected XGBoost regression model are optimized by using the hyperparameter optimization framework Optuna based on Bayesian optimization. The hyperparameter search space is defined as follows:

[0099] n_estimators∈[50,300]

[0100] max_depth∈[3,10]

[0101] learning_rate∈[0.01,0.3]

[0102] subsample∈[0.6,1.0]

[0103] colsample_bytree∈[0.6,1.0]

[0104] gamma∈[0,0.2]

[0105] lambda∈[1,3]

[0106] alpha∈[0,0.5]

[0107] min_child_weight∈[1,5]

[0108] The above hyperparameters are automatically optimized for more than 200 iterations to maximize the R 2 value and minimize the MAE value of the model in the training set cross-validation as the objective function:

[0109] min θ MAE(θ),max θ R 2 (θ)

[0110] wherein, θ represents the combination of hyperparameters of the XGBoost model. The selected XGBoost model is subjected to hyperparameter optimization for more than 200 iterations, and the best combination of hyperparameters obtained is shown in Table 2:

[0111] Table 2

[0112] Hyperparameters Intensity Model Toughness Model n_estimators 150 90 max_depth 6 8 learning_rate 0.07 0.14 subsample 0.86 0.80 colsample_bytree 0.85 0.88 gamma 0.16 0.19 lambda 1.4 2.4 alpha 0.002 0.32 min_child_weight 4 5

[0113] The selected features in the balanced data set after feature engineering are taken as input values, and the strength and toughness are taken as prediction target values. Based on the XGBoost model, the optimal combination of hyperparameters determined is used to construct a tungsten-nickel-iron alloy performance prediction model. Subsequently, the model is trained using the model training set and tested using the model test set to evaluate the prediction performance of the model. The model prediction performance is shown in Table 3. Figure 5 The R2 of the strength model test set obtained is 0.87, and the R2 of the toughness model test set is 0.81, which has good prediction accuracy.

[0114] S4. The strength and toughness of the target tungsten-nickel-iron alloy are predicted using the prediction model to obtain the prediction results.

[0115] In this embodiment, the composition-process search space is defined. Based on the trained machine learning model, virtual alloy design work is carried out, and 77760 composition-process combinations of tungsten-nickel-iron alloy are generated. The strength and toughness values corresponding to these combinations are predicted by means of the model obtained in S3.

[0116] S5. Based on the prediction results, candidate alloys that meet the application requirements are screened out, and a double-objective optimization framework is constructed using the Pareto optimal criterion to optimize the candidate alloys, and alloys with synergistic optimization of strength and toughness are obtained.

[0117] The method for optimizing the candidate alloys using the Pareto optimal criterion to construct a double-objective optimization framework includes: adopting the sequential screening principle, and setting constraint conditions according to the design requirements of the alloy:

[0118] f TS = Q 0.8 (Y TS )

[0119] f EL = Q 0.8 (Y EL )

[0120] p t ≥ 17.5 g / cm 3

[0121] wherein, f TS represents the predicted tensile strength of the alloy, Y TSrepresents tensile strength in the dataset, f EL represents predicted elongation of the alloy, Y EL represents elongation in the dataset, p t represents theoretical density of the alloy, Q 0.8 represents the 80th percentile in the dataset; based on the prediction results and the constraint conditions, candidate alloys meeting the application requirements are screened out, and a bi-objective optimization problem is constructed:

[0122] max x∈χ [f TS (x),f EL (x)]

[0123] wherein x represents an input variable, and x represents a design space formed after screening; a target function numerical vector is constructed:

[0124] F(x)=[f TS (x),f EL (x)]

[0125] Bi-objective optimization analysis is performed on the target function numerical vector using the Pareto optimality criterion:

[0126] f TS (x1)≥f TS (x2)

[0127] f EL (x1)≥f EL (x2)

[0128] If at least one of the inequalities in the analysis result is true, it is judged that x2 is dominated by x1, and all solutions not dominated by other points are constructed into a Pareto front solution set; the Pareto front solution set is hierarchically sorted using fast non-dominated sorting, and the first layer of the Pareto optimal set is extracted as the output, to obtain the alloy after strength and toughness collaborative optimization. The steps and results of bi-objective optimization in this embodiment can be seen in Figure 6 .

[0129] Example Two

[0130] In this embodiment, a tungsten-nickel-iron alloy strength and toughness bi-objective optimization system based on machine learning includes a data acquisition module, a data processing module, a model construction module, a prediction module, and an optimization module.

[0131] The data acquisition module is used to obtain and integrate the original data of the composition, sintering process, processing process, heat treatment process, strength and toughness of the tungsten-nickel-iron alloy, to obtain an initial data set.

[0132] The data processing module is used to preprocess the initial data set to obtain a processed data set.

[0133] The working process of the data processing module includes: in the initial data set, deleting original characteristic variables with a missing value ratio exceeding 50%, performing data cleaning, and obtaining a cleaned data set; in the cleaned data set, for performance data with the same original characteristic variable values, performing mean fusion, and obtaining a fused data set; using K-means clustering analysis to remove outlier data points in the fused data set, performing data balancing, and obtaining a processed data set.

[0134] The model construction module constructs a plurality of candidate models based on a machine learning algorithm, screens the plurality of candidate models, and trains the optimal candidate model screened out from the processed data set to obtain a prediction model.

[0135] In the model construction module, the method for constructing the prediction model includes: performing feature extraction and selection on the processed data set to screen out characteristic variables that have an influence on strength and toughness; constructing a plurality of candidate models based on a machine learning algorithm, evaluating the plurality of candidate models using five-fold cross-validation, and screening out an optimal candidate model; dividing the characteristic variables into a training set and a test set according to a 4:1 ratio, training the optimal candidate model using the training set, and testing the performance of the trained model using the test set to obtain a prediction model.

[0136] The prediction module uses the prediction model to predict the strength and toughness of the target tungsten-nickel-iron alloy to obtain a prediction result.

[0137] The optimization module screens out candidate alloys that meet the application requirements based on the prediction result, and constructs a dual-objective optimization framework using the Pareto optimal criterion to optimize the candidate alloys to obtain alloys with synergistically optimized strength and toughness.

[0138] The working process of the optimization module includes: adopting a sequential screening principle, and setting constraint conditions according to the design requirements of the alloy:

[0139] f TS =Q 0.8 (Y TS )

[0140] f EL =Q 0.8 (Y EL )

[0141] ρ t ≥17.5g / cm 3

[0142] Wherein, f TS represents the predicted tensile strength of the alloy, Y TS represents the tensile strength in the data set, f EL represents the predicted elongation of the alloy, YEL denotes the elongation in the data set, p t denotes the theoretical density of the alloy, Q 0.8 denotes the 80th percentile in the data set; based on the prediction results and the constraints, candidate alloys meeting the application requirements are screened out, and a bi-objective optimization problem is constructed:

[0143] max x∈χ [f TS (x),f EL (x)]

[0144] wherein x represents an input variable, and x represents a design space formed after screening; a target function numerical vector is constructed:

[0145] F(x)=[f TS (x),f EL (x)]

[0146] Bi-objective optimization analysis is performed on the target function numerical vector using the Pareto optimal criterion:

[0147] f TS (x1)≥f TS (x2)

[0148] f EL (x1)≥f EL (x2)

[0149] If at least one of the inequalities in the analysis result is true, it is judged that x2 is dominated by x1, and all solutions not dominated by other points are constructed into a Pareto front solution set; the Pareto front solution set is hierarchically sorted using fast non-dominated sorting, and the Pareto optimal set of the first layer is extracted as the output, to obtain the alloy after strength and toughness collaborative optimization.

[0150] The above-described embodiments are only descriptions of the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A dual-objective optimization method for strength and toughness of tungsten-nickel-iron alloy based on machine learning, characterized in that: The following steps are involved: Obtain and integrate the raw data on the composition, sintering process, processing technology, heat treatment process, strength and toughness of tungsten-nickel-iron alloy to obtain the initial data set; Preprocessing the initial data set to obtain a processed data set; Constructing a number of candidate models based on a machine learning algorithm, screening the candidate models, and then using the processed data set to train the best candidate model to obtain a prediction model; Using the prediction model to predict the strength and toughness of the target tungsten-nickel-iron alloy, and obtaining a prediction result; Based on the prediction results, candidate alloys that meet the application requirements are screened, and a dual-objective optimization framework is constructed using the Pareto optimality criterion to optimize the candidate alloys to obtain alloys with synergistic optimization of strength and toughness.

2. A dual-objective optimization method for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 1, characterized in that The pretreatment method comprises: In the initial data set, original feature variables with a missing value ratio exceeding 50% are deleted to perform data cleaning to obtain a cleaned data set; In the cleaned data set, for the performance data with exactly the same original feature variable values, mean fusion is performed to obtain a fused data set; K-means cluster analysis is used to remove outlier data points in the fused data set, perform data balancing, and obtain the processed data set.

3. A dual-objective optimization method for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 1, characterized in that: The method for obtaining the prediction model includes: performing feature extraction and selection on the processed data set to screen out characteristic variables that have an impact on strength and toughness; Constructing several candidate models based on a machine learning algorithm, evaluating the candidate models using five-fold cross validation, and screening out the optimal candidate model; The feature variables are divided into a training set and a test set in a ratio of 4:1, the optimal candidate model is trained using the training set, and the performance of the trained model is tested using the test set to obtain the prediction model.

4. A dual-objective optimization method for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 1, characterized in that The method for optimizing the candidate alloy by constructing a dual-objective optimization framework using the Pareto optimality criterion includes: Adopt the principle of sequential screening and set constraints according to the design requirements of the alloy: f TS =Q 0.8 (Y TS ) f EL =Q 0.8 (Y EL ) r t ≥17.5g / cm 3 Among them, f TS represents the predicted tensile strength of the alloy, Y TS represents the tensile strength in the data set, f EL represents the predicted elongation of the alloy, Y EL represents the elongation rate in the dataset, ρ t Indicates the theoretical density of the alloy, Q 0.8 Represents the 80th percentile in the data set; Based on the prediction results and the constraints, candidate alloys that meet the application requirements are screened out, and a dual-objective optimization problem is constructed: max x∈χ [f TS (x),f EL (x)] Where x represents the input variable, and χ represents the design space formed after screening; Construct the objective function numerical vector: F(x)=[f TS (x),f EL (x)] The Pareto optimal criterion is used to perform a dual-objective optimization analysis on the objective function numerical vector: f TS (x1)≥f TS (x2) f EL (x1)≥f EL (x2) If at least one of the inequalities in the analysis result holds true, then x2 is determined to be dominated by x1, and all solutions that are not dominated by other points are constructed into the Pareto frontier solution set; The Pareto frontier solution set is hierarchically sorted using fast non-dominated sorting, and the Pareto optimal set of the first layer is extracted as output to obtain an alloy with synergistic optimization of strength and toughness.

5. A dual-objective optimization system for strength and toughness of tungsten-nickel-iron alloy based on machine learning, wherein the system applies the method according to any one of claims 1 to 4, characterized in that: include: Data acquisition module, data processing module, model building module, prediction module and optimization module; The data acquisition module is used to obtain the original data of the composition, sintering process, processing technology, heat treatment process, strength and toughness of the tungsten-nickel-iron alloy and integrate and summarize them to obtain an initial data set; The data processing module is used to preprocess the initial data set to obtain a processed data set; The model building module builds several candidate models based on the machine learning algorithm, screens the candidate models, and then uses the processed data set to train the best candidate model to obtain a prediction model; The prediction module uses the prediction model to predict the strength and toughness of the target tungsten-nickel-iron alloy to obtain a prediction result; The optimization module selects candidate alloys that meet the application requirements based on the prediction results, and optimizes the candidate alloys by constructing a dual-objective optimization framework using the Pareto optimality criterion to obtain an alloy with synergistic optimization of strength and toughness.

6. A dual-objective optimization system for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 5, characterized in that: The workflow of the data processing module includes: In the initial data set, original feature variables with a missing value ratio exceeding 50% are deleted to perform data cleaning to obtain a cleaned data set; In the cleaned data set, for the performance data with exactly the same original feature variable values, mean fusion is performed to obtain a fused data set; K-means cluster analysis is used to remove outlier data points in the fused data set, perform data balancing, and obtain the processed data set.

7. A dual-objective optimization system for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 5, characterized in that: In the model building module, the method for building the prediction model includes: performing feature extraction and selection on the processed data set to screen out characteristic variables that have an impact on strength and toughness; Constructing several candidate models based on a machine learning algorithm, evaluating the candidate models using five-fold cross validation, and screening out the optimal candidate model; The feature variables are divided into a training set and a test set in a ratio of 4:1, the optimal candidate model is trained using the training set, and the performance of the trained model is tested using the test set to obtain the prediction model.

8. The dual-objective optimization system for strength and toughness of tungsten-nickel-iron alloy based on machine learning according to claim 5, characterized in that: The workflow of the optimization module includes: Adopt the principle of sequential screening and set constraints according to the design requirements of the alloy: f TS =Q 0.8 (Y TS ) f EL =Q 0.8 (Y EL ) r t ≥17.5g / cm 3 Among them, f TS represents the predicted tensile strength of the alloy, Y TS represents the tensile strength in the data set, f EL represents the predicted elongation of the alloy, Y EL represents the elongation rate in the dataset, ρ t Indicates the theoretical density of the alloy, Q 0.8 Represents the 80th percentile in the data set; Based on the prediction results and the constraints, candidate alloys that meet the application requirements are screened out, and a dual-objective optimization problem is constructed: max x∈χ [f TS (x),f EL (x)] Where x represents the input variable, and χ represents the design space formed after screening; Construct the objective function numerical vector: F(x)=[f TS (x),f EL (x)] The Pareto optimal criterion is used to perform a dual-objective optimization analysis on the objective function numerical vector: f TS (x1)≥f TS (x2) f EL (x1)≥f EL (x2) If at least one of the inequalities in the analysis result holds true, then x2 is determined to be dominated by x1, and all solutions that are not dominated by other points are constructed into the Pareto frontier solution set; The Pareto frontier solution set is hierarchically sorted using fast non-dominated sorting, and the Pareto optimal set of the first layer is extracted as output to obtain an alloy with synergistic optimization of strength and toughness.

Images