WAE-based lead-free solder design method and system
Patent Information
- Application Number
- CN202510722253.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2025-10-17
Smart Images
Figure CN120808989A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of lead-free solder, in particular to a lead-free solder design method and system based on WAE. BACKGROUND
[0002] Currently, Sn-Ag-Cu (SAC) lead-free solder is widely used in the electronic industry due to its environmental protection, fatigue resistance and other properties. Among them, the ultimate tensile strength (UTS) and elongation (EL) as key mechanical performance indicators directly determine the reliability and service life of the solder joint. However, in the material development process, there is a natural contradiction between improving UTS and EL - enhancing strength often sacrifices material toughness, which becomes the core problem of SAC solder performance optimization.
[0003] To break this dilemma, researchers try to add trace alloying elements to SAC solder to balance UTS and EL. However, the diversity of alloy composition forms a huge design space, and traditional trial-and-error method not only needs to consume a lot of time and cost, but also is difficult to systematically explore potential high-quality formula.
[0004] High-throughput virtual screening technology provides a new path for this, which can accelerate the material development process by generating virtual samples and predicting performance.
[0005] However, when dealing with discrete variables such as SAC solder alloy composition, the element distribution presents sparse and non-continuous characteristics. If traditional random sampling or enumeration method is used, not only a large candidate sample library needs to be generated to cover the potential combination space, but also a large number of data points deviating from the original distribution will be generated. Such "out-of-distribution" samples lack consistency with real data, which will force the machine learning model to extrapolate in the area where the training data is insufficient, resulting in a significant decline in performance, and thus affecting the effectiveness and reliability of material design. SUMMARY
[0006] The purpose of the present application is to overcome the defects of the prior art and provide a lead-free solder design method and system based on WAE.
[0007] The purpose of the present application can be achieved by the following technical solutions:
[0008] According to one aspect of the present application, a lead-free solder design method based on WAE is provided, and the method steps include:
[0009] S1, collect alloy element composition information, UTS value and EL value of the sample SAC solder alloy, and construct a feature parameter based on the alloy element composition information; based on the feature parameter, the alloy element composition information, the UTS value and the EL value of the sample SAC solder alloy, a data set is constructed;
[0010] S2, after preprocessing the data set, two best feature sets are output by using genetic algorithm combined with support vector regression and lasso regression respectively to search the preprocessed data set;
[0011] S3, taking the UTS value and the EL value in the data set as target variables and the best feature set as independent variables, a quantitative prediction model of ultimate tensile strength and a quantitative prediction model of elongation are constructed;
[0012] S4, according to the composition distribution of the SAC solder alloy in the data set, a WAE model is trained, and a virtual sample set is generated by using the trained WAE model;
[0013] S5, the established quantitative prediction model of ultimate tensile strength and the quantitative prediction model of elongation are used respectively to predict each virtual sample in the virtual sample set, and the corresponding UTS prediction value and EL prediction value of each virtual sample are obtained;
[0014] S6, the virtual samples whose UTS prediction value and EL prediction value both meet the preset requirements are selected from the virtual sample set, and the lead-free solder design is completed.
[0015] As a preferred technical solution, the feature parameter in S1 is used to describe the material properties of the sample SAC solder alloy, and the feature parameter includes stoichiometric features, element attribute statistical features and electronic structure related features.
[0016] As a preferred technical solution, the preprocessing process in S2 includes: first, eliminating constant features in the data set, then evaluating the correlation between features based on the Pearson correlation coefficient, removing redundant features to reduce the dimension of features; the Pearson correlation coefficient is set to 0.90.
[0017] As a preferred technical solution, the specific process of training the WAE model in S4 is: using the encoder of the WAE model to map the composition distribution of the SAC solder alloy in the data set to the latent space of the WAE model, and using the maximum mean difference regularization term to constrain the latent distribution, so that the latent distribution is similar to the preset prior distribution;
[0018] The process of generating a virtual sample set by using the trained WAE model is specifically: using the decoder of the WAE model to generate a virtual sample set from the latent space.
[0019] 5. The WAE-based lead-free solder design method according to claim 1, wherein the data set is randomly stratified in a 4:1 ratio to divide into a training set and a test set.
[0020] As a preferred technical solution, the ultimate tensile strength quantitative prediction model in S3 is established based on the training set and the test set by a support vector regression algorithm; and the elongation quantitative prediction model is established based on the training set and the test set by a lasso regression algorithm.
[0021] As a preferred technical solution, the specific process of constructing the ultimate tensile strength quantitative prediction model in S3 is as follows: first, a radial basis function is selected as an SVR kernel function, and hyperparameters are initialized; then, based on the training set, the model is repeatedly trained and the hyperparameters are adjusted to minimize the regression error; then, based on the test set, a leave-one-out cross-validation is used to find an optimal hyperparameter combination by minimizing the root mean square error as an objective, so as to construct the ultimate tensile strength quantitative prediction model from the optimal hyperparameter combination.
[0022] As a preferred technical solution, the specific process of constructing the elongation quantitative prediction model in S3 is as follows:
[0023] First, the feature weights are initialized, then the lasso regression algorithm is used to impose sparsity constraints on the feature weights by L1 regularization, and then based on the sparsity constraints, the coordinate descent method is used to iteratively train and solve based on the training set, and the model is evaluated based on the test set, with the objective of minimizing the root mean square error, and finally the optimal weights are output, so as to construct the elongation quantitative prediction model from the optimal weights.
[0024] As a preferred technical solution, the preset requirement in S6 includes that the UTS predicted value is greater than 70 MPa and the EL predicted value is greater than 30%.
[0025] According to another aspect of the present application, a WAE-based lead-free solder design system is provided, which comprises a data set construction module, a feature optimization module, a prediction model construction module, a virtual sample set generation module and a lead-free solder design module.
[0026] The data set construction module is used to collect alloy element composition information, UTS values and EL values of sample SAC solder alloys, and construct feature parameters based on the alloy element composition information of the sample SAC solder alloys; and construct a data set based on the feature parameters, the alloy element composition information of the sample SAC solder alloys, the UTS values and the EL values.
[0027] The feature optimization module is used for feature preprocessing and screening, and after preprocessing of the data set, two best feature sets are output by using genetic algorithm combined with support vector regression and lasso regression respectively on the preprocessed data set;
[0028] The prediction model construction module is used for constructing the ultimate tensile strength quantitative prediction model and the elongation quantitative prediction model by taking the UTS value and the EL value in the data set as target variables and the best feature set as independent variables;
[0029] The virtual sample set generation module is used for training the WAE model according to the component distribution of the SAC solder alloy in the data set, and generating the virtual sample set by using the trained WAE model;
[0030] The lead-free solder design module is used for predicting each virtual sample in the virtual sample set by using the established ultimate tensile strength quantitative prediction model and the elongation quantitative prediction model respectively, obtaining the UTS prediction value and the EL prediction value corresponding to each virtual sample, and screening the virtual sample whose UTS prediction value and EL prediction value both meet the preset requirements from the virtual sample set, to complete the lead-free solder design.
[0031] Compared with the prior art, the present application has the following beneficial effects:
[0032] 1、In the present application, the virtual sample set is generated by using the trained WAE model, each virtual sample in the virtual sample set is predicted by using the established ultimate tensile strength quantitative prediction model and the elongation quantitative prediction model, the UTS prediction value and the EL prediction value corresponding to each virtual sample are obtained, and finally the virtual sample whose UTS prediction value and EL prediction value both meet the preset requirements is screened from the virtual sample set to complete the lead-free solder design. By using the WAE model combined with high-throughput screening, high-quality virtual samples are generated, the extrapolation risk is reduced, the high-strength and high-toughness SAC lead-free solder meeting the performance requirements is selected from the high-quality virtual samples, the multi-objective optimization design is realized, and the effectiveness and reliability of the lead-free solder design are improved.
[0033] 2、In the present application, the best feature set is generated by using genetic algorithm combined with support vector regression and lasso regression respectively on the preprocessed data set, the ultimate tensile strength quantitative prediction model is established by support vector regression algorithm by taking the UTS value and the EL value in the data set as target variables and the best feature set as independent variables, the elongation quantitative prediction model is established by lasso regression algorithm, the feature data obtained is imported into the prediction model of UTS and EL, and thus the UTS and EL of the sample SAC solder alloy are simply and quickly predicted, and the prediction result can be obtained only in a few seconds, thereby improving the overall efficiency.
[0034] 3、The preprocessing process of the application first removes constant features in the data set, then evaluates the correlation between features based on the Pearson correlation coefficient, removes redundant features to reduce the dimensionality of the features, reduces the amount of calculation, improves the model training speed, avoids the multicollinearity problem caused by strong correlation between features, and optimizes the model accuracy.
[0035] 4、In the application, for WAE model training, the encoder of the WAE model is used to map the component distribution of the SAC solder alloy in the data set to the latent space of the WAE model, and the maximum mean difference regular term is used to constrain the latent distribution, so that the latent distribution is similar to the preset prior distribution; the virtual samples generated by the WAE model have more practical reference value, and the sample space of lead-free solder design is expanded.
[0036] 5、In the application, when constructing the quantitative prediction model of ultimate tensile strength, the hyperparameters are optimized by grid search or random search combined with leave-one-out cross-validation, so that the optimal parameter combination can be found, the model can better fit the data, and the accuracy of predicting the ultimate tensile strength can be improved. When constructing the quantitative prediction model of elongation, L1 regularization of lasso regression algorithm and coordinate descent method are used to impose sparsity constraint on feature weight and iteratively solve, focus on key features, exclude secondary factor interference, and improve elongation prediction accuracy.
[0037] 6、The lead-free solder design system based on WAE constructed in the application includes a data set construction module, a feature optimization module, a prediction model construction module, a virtual sample set generation module and a lead-free solder design module, so that the entire lead-free solder design process can be realized by computer, and no experiments and chemical drugs are involved in the entire process. The method is low-cost and pollution-free. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1 A WAE-based lead-free solder design method in the application is shown in the flowchart;
[0039] Figure 2 An implementation flowchart of a WAE-based lead-free solder design method in the embodiment is shown in the flowchart;
[0040] Figure 3a A leave-one-out cross-validation result graph of SAC lead-free solder UTS in the embodiment is shown in the graph;
[0041] Figure 3b A training set and test set result graph of SAC lead-free solder UTS in the embodiment is shown in the graph;
[0042] Figure 3c A leave-one-out cross-validation result graph of SAC lead-free solder EL in the embodiment is shown in the graph;
[0043] Figure 3dA result chart of the SAC lead-free solder EL training set and test set in the embodiment;
[0044] Figure 4a A sample distribution schematic diagram of virtual samples generated by the WAE model in the SAC105 system samples in the embodiment;
[0045] Figure 4b A sample distribution schematic diagram of virtual samples generated by the WAE model in the SAC387 system samples in the embodiment. DETAILED DESCRIPTION
[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor should fall within the protection scope of the present application.
[0047] In material design, high-throughput virtual screening is an effective strategy. By generating virtual samples and predicting their performance, the screening of high-quality alloy formulations can be accelerated. However, when the composition is a discrete variable, the elements show sparse and non-continuous distribution. If the traditional random sampling or enumeration method is directly used, not only a large candidate sample library needs to be generated to cover the potential combination space, but also a large number of data points deviating from the original distribution will be generated. Such 'out-of-distribution' samples lack consistency with real data, which may force the machine learning model to extrapolate the prediction in the area where the training data is insufficient, resulting in a significant decrease in performance, and thus affecting the effectiveness and reliability of material design. Therefore, the present application proposes a method of using WAE combined with high-throughput screening for material design.
[0048] Embodiment 1
[0049] In this embodiment, a lead-free solder design method based on WAE is adopted, and the method steps are as shown in Figure 1 , and specifically include:
[0050] S1, collect the alloy element composition information, UTS value and EL value of the sample SAC solder alloy, and construct a feature parameter based on the alloy element composition information; based on the feature parameter, the alloy element composition information, the UTS value and the EL value of the sample SAC solder alloy, a data set is constructed;
[0051] S2, after preprocessing the data set, two best feature sets are output by using genetic algorithm combined with support vector regression and lasso regression respectively to search the preprocessed data set;
[0052] S3, taking the UTS value and the EL value in the data set as target variables, and taking the optimal feature set as independent variables, to construct a quantitative prediction model of the ultimate tensile strength and a quantitative prediction model of the elongation;
[0053] S4, training the WAE model according to the composition distribution of the SAC solder alloy in the data set, and generating a virtual sample set by using the trained WAE model;
[0054] S5, using the established quantitative prediction model of the ultimate tensile strength and the quantitative prediction model of the elongation respectively to predict each virtual sample in the virtual sample set, and obtaining the UTS prediction value and the EL prediction value corresponding to each virtual sample;
[0055] S6, screening the virtual sample whose UTS prediction value and EL prediction value both meet the preset requirements from the virtual sample set, to complete the lead-free solder design. The WAE is used to generate high-quality samples to reduce the risk of model extrapolation, and the process is realized by using computer technology, which avoids the consumption of human and material resources caused by chemical experiments and reduces the time cost.
[0056] In the embodiment, the feature parameters in S1 are used to describe the material properties of the sample SAC solder alloy, and the feature parameters include stoichiometric features, element attribute statistical features and electronic structure related features.
[0057] The preprocessing process in S2 includes: firstly, removing constant features in the data set, and then removing redundant features based on the Pearson correlation coefficient to reduce the dimension of the features; the Pearson correlation coefficient is set to 0.90.
[0058] WAE (Wasserstein Auto-Encoder) is a generative model combining the structure of auto-encoder and the Wasserstein distance, which is used to generate high-quality virtual samples. The model encodes the real data into latent variables, and constrains its distribution close to the preset prior (such as Gaussian distribution) through the maximum mean discrepancy (MMD); then, the latent space is sampled, and the reconstructed sample is generated through the decoder, and the model parameters are optimized through the reconstruction loss and the MMD loss, to realize the effective generation of high-quality virtual samples.
[0059] The WAE can also be applied to other properties of SAC alloys or different properties of different alloys in combination with high-throughput screening methods, and the method has high scalability.
[0060] The specific process of training the WAE model in S4 is: using the encoder of the WAE model, the component distribution of the SAC solder alloy in the data set is mapped to the latent space of the WAE model, and the maximum mean difference regularization term is used to constrain the latent distribution, so that the latent distribution is similar to the preset prior distribution;
[0061] The process of generating a virtual sample set using the trained WAE model is as follows: using the decoder of the WAE model, a virtual sample set is generated from the latent space.
[0062] In this embodiment, the data set is randomly stratified in a ratio of 4:1 to divide into a training set and a test set.
[0063] Support Vector Machine (SVM) is a classic supervised learning method, including Support Vector Regression (SVR) for regression problems and Support Vector Classification (SVC) for classification problems. SVR builds a regression function that fits the data as much as possible within the allowed error insensitive interval, and determines the optimal regression hyperplane by minimizing the total loss function. The support vectors on the boundary determine the generalization ability of the model; SVC finds the optimal classification hyperplane by maximizing the interval between different classes of data. The decision boundary is determined only by the support vectors on the edge, thereby reducing the complexity of the model and improving the classification performance.
[0064] The Least Absolute Shrinkage and Selection Operator (LASSO) algorithm is based on ordinary linear regression and introduces an L1 regularization term. By penalizing and constraining the model coefficients, feature selection and model parameter estimation are performed simultaneously to reduce model complexity and improve model generalization performance, especially for high-dimensional data analysis.
[0065] The ultimate tensile strength quantitative prediction model described in S3 is based on the training set and the test set, and is established by the support vector regression algorithm; the elongation quantitative prediction model is based on the training set and the test set, and is established by the LASSO algorithm.
[0066] The preset requirements in S6 include that the UTS predicted value is greater than 70 MPa and the EL predicted value is greater than 30%.
[0067] Table 1. Experimental values of the composition, UTS and EL of some lead-free solders
[0068]
[0069]
[0070] The method uses machine learning method to predict the UTS and EL of SAC lead-free solder, uses WAE combined with trained model for high-throughput screening design, and the implementation process is as shown in Figure 2 , and the specific implementation process is as follows:
[0071] (1) 109 UTS and 99 EL lead-free solder data were collected from literature, part of which is shown in Table 1:
[0072] (2) The generated Magpie features are used to describe the characteristic parameters of the sample, and the characteristic values corresponding to each high-entropy alloy sample are obtained, part of which is shown in Table 2.
[0073] Table 2. Part of the characteristic and its meaning table
[0074]
[0075] (3) The data set is randomly divided into training set and test set in the ratio of 4:1, wherein the sample numbers of the training set and the test set of the UTS data set are 82 and 27 respectively, and the sample numbers of the training set and the test set of the EL data set are 75 and 24 respectively.
[0076] (4) After deleting the constant features, the Pearson correlation coefficient is used to remove the redundant features to reduce the feature dimension, and the Pearson correlation coefficient is set to 0.90, UTS feature set retains 53 features, and EL feature set retains 52 features; Genetic algorithm combined with support vector regression and lasso regression is used to search for the best feature combination, and finally 7 and 8 best features are obtained, part of which is shown in Table 3.
[0077] (5) Taking the UTS and EL of SAC solder collected in the step (1) as the target variable, and taking the 7 and 8 best features in the step (4) as the independent variable, based on the training set divided in the step (3), the support vector regression algorithm is used to establish the quantitative prediction model of SAC solder UTS, and the lasso regression algorithm is used to establish the quantitative model of SAC solder EL.
[0078] (6) According to the composition information collected in step (1), the WAE model is trained, and high-quality virtual samples are generated. The prediction model of UTS and EL established in step (5) is used to predict the UTS and EL target value of the virtual sample.
[0079] (7) According to the virtual sample obtained in step (6), the sample with UTS greater than 70 MPa and EL greater than 30% is selected, that is, the high-strength and high-toughness SAC solder is used for experiment.
[0080] Table 3. Best feature set data table
[0081]
[0082] In this embodiment, based on the UTS prediction model established by 109 SAC lead-free solder alloy samples, the results of leave-one-out cross-validation are shown in Figure 3a The correlation coefficient R between the experimental value and the predicted value is 0.89, and the root mean square error RMSE is 7.29.
[0083] Embodiment 2
[0084] In this embodiment, a lead-free solder design system based on WAE is applied, which includes a data set construction module, a feature optimization module, a prediction model construction module, a virtual sample set generation module and a lead-free solder design module.
[0085] The data set construction module is used to collect the alloy element composition information, UTS value and EL value of the sample SAC solder alloy, and to construct the feature parameters based on the alloy element composition information of the sample SAC solder alloy. Based on the feature parameters, the alloy element composition information, the UTS value and the EL value of the sample SAC solder alloy, the data set is constructed.
[0086] The feature optimization module is used for feature preprocessing and screening. After preprocessing the data set, this module uses genetic algorithm combined with support vector regression and lasso regression to search the preprocessed data set, and then outputs two best feature sets.
[0087] The prediction model construction module is used to construct the ultimate tensile strength quantitative prediction model and the elongation quantitative prediction model by taking the UTS value and the EL value in the data set as the target variable and the best feature set as the independent variable.
[0088] The virtual sample set generation module is used to train the WAE model according to the composition distribution of the SAC solder alloy in the data set, and to generate a virtual sample set by using the trained WAE model.
[0089] The lead-free solder design module is used to predict each virtual sample in the virtual sample set by using the established ultimate tensile strength quantitative prediction model and elongation quantitative prediction model respectively, to obtain the UTS prediction value and the EL prediction value corresponding to each virtual sample, and to screen the virtual samples whose UTS prediction value and EL prediction value both meet the preset requirements from the virtual sample set, thereby completing the lead-free solder design.
[0090] The specific implementation of this embodiment is the same as that of embodiment 1, and the difference lies in that in this embodiment, based on 109 SAC lead-free solder alloy samples, the training set and test set results of the established UTS prediction model are as shown in Figure 3bAs shown in the figure, the round points are the results of the training set, the star points are the results of the test set, the correlation coefficient R between the experimental value and the predicted value of the training set is 0.96, and the root mean square error RMSE is 4.44; the correlation coefficient R between the experimental value and the predicted value of the test set is 0.85, and the root mean square error RMSE is 13.32
[0091] Example 3
[0092] In this embodiment, a WAE-based lead-free solder design method is used, and the specific implementation is the same as that of Example 1, and the difference is that in this embodiment, the leave-one-out cross-validation result 3c of the EL prediction model established based on 99 SAC lead-free solder alloy samples is shown, the correlation coefficient R between the experimental value and the predicted value is 0.70, and the root mean square error RMSE is 5.16.
[0093] Example 4
[0094] In this embodiment, a WAE-based lead-free solder design method is used, and the specific implementation is the same as that of Example 1, and the difference is that in this embodiment, the training set and test set results of the EL prediction model established based on 99 SAC lead-free solder alloy samples are shown Figure 3d As shown in the figure, the round points are the results of the training set, the star points are the results of the test set, the correlation coefficient R between the experimental value and the predicted value of the training set is 0.77, and the root mean square error RMSE is 4.60; the correlation coefficient R between the experimental value and the predicted value of the test set is 0.70, and the root mean square error RMSE is 5.75.
[0095] Example 5
[0096] In this embodiment, a WAE-based lead-free solder design method is used, and the specific implementation is the same as that of Example 1, and the difference is that the WAE model is trained, and high-quality virtual samples are generated, and the sample distribution is visualized by PCA (principal component analysis) projection to a two-dimensional plane, as shown in Figure 4a and Figure 4b As shown in the figure, the round samples represent the virtual samples generated by WAE, and the diamonds represent the original samples, PC1 represents the first principal component, and PC2 represents the second principal component. As can be seen from the figure, the generated virtual samples are similar to the original data distribution, effectively reducing the extrapolation risk of the prediction model. One sample with UTS greater than 70 MPa and EL greater than 30% is selected for experimental verification, and the results are shown in Table 4.
[0097] Table 4. Predicted values and experimental values of high-strength and high-toughness lead-free solders based on WAE
[0098]
[0099] To sum up, the present scheme generates a virtual sample set by using the trained WAE model, predicts each virtual sample in the virtual sample set by using the established ultimate tensile strength quantitative prediction model and elongation quantitative prediction model, and obtains the UTS prediction value and the EL prediction value corresponding to each virtual sample; finally, the virtual sample whose UTS prediction value and EL prediction value both meet the preset requirements is screened out from the virtual sample set, and the lead-free solder design is completed. By using the WAE model combined with high-throughput screening, high-quality virtual samples are generated, the extrapolation risk is reduced, the high-strength and high-toughness SAC lead-free solder meeting the performance requirements is selected from the high-quality virtual samples, the multi-objective optimization design is realized, and the effectiveness and reliability of the lead-free solder design are improved.
[0100] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims
1. A lead-free solder design method based on WAE, characterized in that: The method steps include: S1. Collecting alloy element composition information, UTS values, and EL values of sample SAC solder alloys, and constructing characteristic parameters based on the alloy element composition information; constructing a data set based on the characteristic parameters, alloy element composition information, UTS values, and EL values of the sample SAC solder alloys; S2. After preprocessing the data set, use genetic algorithm combined with support vector regression and lasso regression to search the preprocessed data set and output two optimal feature sets; S3. Using the UTS value and EL value in the data set as target variables and the optimal feature set as independent variables, a quantitative prediction model for ultimate tensile strength and a quantitative prediction model for elongation are constructed; S4. Training the WAE model based on the composition distribution of the SAC solder alloy in the dataset, and generating a virtual sample set using the trained WAE model; S5. Use the established ultimate tensile strength quantitative prediction model and elongation quantitative prediction model to predict each virtual sample in the virtual sample set, and obtain the UTS prediction value and EL prediction value corresponding to each virtual sample; S6. Filter out virtual samples whose UTS prediction values and EL prediction values meet the preset requirements from the virtual sample set to complete the lead-free solder design.
2. A lead-free solder design method based on WAE according to claim 1, characterized in that: The characteristic parameters in S1 are used to describe the material properties of the sample SAC solder alloy, and the characteristic parameters include stoichiometric characteristics, elemental attribute statistical characteristics, and electronic structure-related characteristics.
3. A lead-free solder design method based on WAE according to claim 1, characterized in that: The preprocessing process in S2 includes: first, removing constant features in the data set, then evaluating the correlation between features based on the Pearson correlation coefficient, removing redundant features, and reducing the dimension of the features; the Pearson correlation coefficient is set to 0.
90.
4. A lead-free solder design method based on WAE according to claim 1, characterized in that: The specific process of training the WAE model in S4 is as follows: using the encoder of the WAE model, mapping the composition distribution of the SAC solder alloy in the dataset to the latent space of the WAE model, and using the maximum mean difference regularization term to constrain the latent distribution so that its latent distribution is close to the preset prior distribution; The process of generating a virtual sample set using the trained WAE model is specifically as follows: using the decoder of the WAE model to generate a virtual sample set from the latent space.
5. The lead-free solder design method based on WAE according to claim 1, characterized in that: The dataset was randomly stratified in a 4:1 ratio and divided into a training set and a test set.
6. A lead-free solder design method based on WAE according to claim 5, characterized in that: The ultimate tensile strength quantitative prediction model described in S3 is based on the training set and the test set and is established by the support vector regression algorithm; the elongation quantitative prediction model is based on the training set and the test set and is established by the lasso regression algorithm.
7. A lead-free solder design method based on WAE according to claim 6, characterized in that: The specific process of constructing the ultimate tensile strength quantitative prediction model in S3 is as follows: first, a radial basis function is selected as the SVR kernel function and the hyperparameters are initialized; then, based on the training set, the model is repeatedly trained and the hyperparameters are adjusted to minimize the regression error; then, a grid search or random search is used, based on the test set, using the leave-one-out cross-validation method, with the goal of minimizing the root mean square error, to find the optimal hyperparameter combination, and thus the ultimate tensile strength quantitative prediction model is constructed by the optimal hyperparameter combination.
8. The lead-free solder design method based on WAE according to claim 6, characterized in that: The specific process of constructing the elongation quantitative prediction model in S3 is as follows: First, the feature weights are initialized. Then, the lasso regression algorithm is used to impose sparsity constraints on the feature weights through L1 regularization. Under the sparsity constraint, the coordinate descent method is used to iteratively train and solve the problem based on the training set. The model is evaluated on the test set with the goal of minimizing the root mean square error. Finally, the optimal weights are output, and a quantitative prediction model for elongation is constructed based on the optimal weights.
9. The lead-free solder design method based on WAE according to claim 1, characterized in that: The preset requirements in S6 include: a UTS predicted value greater than 70 MPa and an EL predicted value greater than 30%.
10. A lead-free solder design system based on WAE, characterized in that: The system applies a lead-free solder design method based on WAE according to any one of claims 1 to 9, and the system includes a data set construction module, a feature optimization module, a prediction model construction module, a virtual sample set generation module and a lead-free solder design module; The data set construction module is used to collect alloy element composition information, UTS value and EL value of the sample SAC solder alloy, and construct characteristic parameters based on the alloy element composition information of the sample SAC solder alloy; A data set was constructed based on characteristic parameters, alloy element composition information, UTS values, and EL values of sample SAC solder alloys. The feature optimization module is used for feature preprocessing and screening. After preprocessing the data set, the module uses a genetic algorithm combined with support vector regression and lasso regression to search the preprocessed data set and output two optimal feature sets. The prediction model construction module is used to construct a quantitative prediction model for ultimate tensile strength and a quantitative prediction model for elongation using the UTS value and EL value in the data set as target variables and the optimal feature set as independent variables; The virtual sample set generation module is used to train the WAE model according to the composition distribution of the SAC solder alloy in the data set, and generate a virtual sample set using the trained WAE model; The lead-free solder design module is used to respectively use the established ultimate tensile strength quantitative prediction model and elongation quantitative prediction model to predict each virtual sample in the virtual sample set, and obtain the UTS prediction value and EL prediction value corresponding to each virtual sample; And virtual samples whose UTS prediction values and EL prediction values meet the preset requirements are screened out from the virtual sample set to complete the lead-free solder design.