CN-115132293-B - Method and system for rapidly predicting creep stress index of tin-based solder alloy by using integrated model

CN115132293BCN 115132293 BCN115132293 BCN 115132293BCN-115132293-B

Abstract

The invention discloses a method and a system for rapidly predicting creep stress indexes of tin-based solder alloy by an integrated model, which are characterized in that element compositions, test temperatures and creep stress index values of the tin-based solder alloy are collected from a literature and are used as data set samples, the element compositions and the test temperatures of the tin-based solder alloy are arranged to be used as modeling characteristics, the data set is randomly divided into a training set and a testing set according to the ratio of 4:1, the collected creep stress index values of the tin-based solder alloy are used as target variables, the constructed characteristics are used as independent variables, the independent variables are scaled by RobustScaler based on the divided training set, three learners are trained and integrated to obtain an R-X-L integrated model, and the creep stress indexes of the testing set samples and 4 independent experimental samples are rapidly predicted by using the R-X-L integrated model. The model for forecasting the creep stress index of the tin-based solder alloy based on reliable literature data and modeling method has the advantages of simplicity, convenience, rapidness, low cost, no pollution and the like.

Inventors

ZHOU XIAN
LIU CHEN
WANG JIAJUN
CHEN HUIMIN
LI MINJIE
LU WENCONG
DONG ZIQIANG
CHEN YOUYANG
Peng jubo
CAI SHANSHAN
LUO XIAOBIN

Assignees

上海大学
上海大学
云南锡业集团（控股）有限责任公司研发中心
云南锡业集团（控股）有限责任公司研发中心

Dates

Publication Date: 20260421
Application Date: 20220601
Priority Date: 20220601

Claims (3)

1. A method for rapidly predicting creep stress index of a tin-based solder alloy by an integrated model, comprising the following steps: 1) Collecting element composition, test temperature and creep stress index values of the tin-based solder alloy from the literature, and adding new relevant experimental data to be used as a data set sample; 2) Finishing element composition and test temperature data of the tin-based solder alloy, and using the element composition and the test temperature data as modeling characteristics; 3) Randomly dividing a data set into a training set and a testing set according to the ratio of 4:1; 4) Taking the creep stress index value of the tin-based solder alloy collected in the step 1) as a target variable, taking the characteristics in the step 2) as independent variables, using RobustScaler scaling data for the independent variables based on the training set divided in the step 3), training three learners, and integrating to obtain an R-X-L integrated model; 5) Rapidly forecasting creep stress indexes of the test set sample and 4 independent experiment samples in the step 3) by using the R-X-L integrated model established in the step 4); In the step 4), the specific steps of training the R-X-L integrated model are as follows: 4-1) randomly extracting a plurality of subsamples from the total training sample set S by using a Bootstrap resampling method Using each When model training is carried out, a plurality of attribute values are randomly selected to carry out node splitting, a plurality of regression trees are generated to the maximum extent, pruning is not needed, and finally a multi-element nonlinear regression combined learner 1 is formed; 4-2) considering the limitation of the complexity of the tree model on the basis of the loss function, and combining the loss function and the complexity as an objective function, wherein the objective function of the mth decision tree can be written as: Wherein, the method comprises the steps of, For the observation of the i-th sample, N represents the number of samples of the data set, Is the predicted value of the first m decision trees of the sample, x is the input value of the sample, l is the deviation between the observed value and the predicted value of the sample calculated by the loss function, the first term of the objective function is the loss function of the original mth decision tree, and the second term For the model complexity of the mth decision tree, the weights can be learned smoothly to avoid overfitting, where T is the number of leaf nodes, For the leaf node weight, Representing the predicted value of the sample in the leaf node, wherein gamma and lambda are super parameters; continuously adding a decision tree on the basis of the original decision tree until an objective function is minimum, and stopping iterative training to obtain a learner 2; 4-3) given data set Wherein For inputting space The negative gradient of the loss function of the model output is recorded as in each iteration of the gradient lifting Arranging training samples in descending order of absolute value of gradient, wherein the first a% of samples with larger gradient are used as subset A, and the rest (1-a)% of samples with small gradient are used as subset And from B% of samples are randomly selected as a subset B; Calculation of Variance gain on Selecting To segment the sample nodes; The corresponding sample nodes are selected and corresponding output values are determined, the steps are repeated until the fitting accuracy threshold value is reached or the upper limit of the number of the trees is reached, and finally the learner 3 is obtained; 4-4) integrating the three learners obtained in the steps 4-1), 4-2) and 4-3) to obtain an R-X-L integrated model.
2. The method for rapid prediction of creep stress index of tin-based solder alloy by integrated model according to claim 1, wherein in said step 4-3), The definition is as follows: Wherein, the method comprises the steps of, , , , Coefficient of For normalizing the gradient sum over B to N is the number of samples of the data set, d is the sample node, For inputting space Is a vector of dimension s, A is a negative gradient of a loss function output by the model, A is a sample subset with a larger gradient of the first a% selected after training samples are arranged in descending order of absolute value of the gradient, A small gradient sample subset of (1-a)% remaining, B is the slave Randomly selecting the number of b% of samples; Represented as Is set for the number of sample nodes d, Is that Is set for the number of sample nodes d, For the j-th dimensional feature of the i-th sample, , 。
3. A system for rapidly predicting creep stress index of a tin-based solder alloy, comprising a memory for storing a computer program, input data and output data, and a processor for performing the method for rapidly predicting creep stress index of a tin-based solder alloy according to the integrated model of claim 1 or 2.

Description

Method and system for rapidly predicting creep stress index of tin-based solder alloy by using integrated model Technical Field The invention relates to the field of creep performance of tin-based solder alloy, in particular to a method for rapidly predicting creep stress index of tin-based solder alloy by using an integrated model. Technical Field Tin-based solder alloys are the most commonly used solder alloy systems for soldering, have excellent mechanical strength and hardness properties, and have been identified as a very promising material in the microelectronics industry. However, since the electronic device is severely creep deformed during operation or when the temperature is raised in the environment, the application of such an alloy is hindered, and thus it is required to design a tin-based solder alloy having more excellent creep resistance. The conventional trial-and-error method is time-consuming and labor-consuming to design tin-based solder alloys with high creep resistance. Therefore, machine learning methods such as RF (Random Forest), XGBoost (eXtreme Gradient Boosting), lightGBM (LIGHT GRADIENT Boosting Machine) have been widely used in predicting target properties of materials. The creep resistance of the tin-based solder alloy is researched by using the existing creep stress index, so that the material design speed can be greatly improved, and the method has important significance for the tin-based solder alloy with excellent design performance. Therefore, how to design an R-X-L integration model to rapidly predict the creep stress index of a tin-based solder alloy is a technical problem that needs to be solved. Disclosure of Invention The invention aims to overcome the defects of high creep performance test cost and low prediction accuracy of a tin-based solder alloy in the prior art, and provides a method for rapidly predicting the creep stress index of the tin-based solder alloy by using an integrated model with simplicity, convenience, rapidness, low cost and strong generalization performance. The integrated model of the invention completes the learning task by constructing and combining a plurality of learners. Compared with a single learner, the accuracy and stability of the prediction result can be improved, the over-fitting problem can be solved, and the selection of parameters can be improved. The invention uses the R-X-L integrated model integrated with three learners with superior performance to rapidly predict the creep stress index of the tin-based solder alloy, and remarkably improves the efficiency. In order to achieve the aim of the invention, the invention adopts the following technical scheme: a method for an integrated model to rapidly predict creep stress index of a tin-based solder alloy, comprising the steps of: 1) Collecting element composition, test temperature and creep stress index values of the tin-based solder alloy from the literature, and adding new relevant experimental data to be used as a data set sample; 2) Finishing element composition and test temperature data of the tin-based solder alloy, and using the element composition and the test temperature data as modeling characteristics; 3) Randomly dividing a data set into a training set and a testing set according to the ratio of 4:1; 4) Taking the creep stress index value of the tin-based solder alloy collected in the step 1) as a target variable, taking the characteristics in the step 2) as independent variables, using RobustScaler scaling data for the independent variables based on the training set divided in the step 3), training three learners, and integrating to obtain an R-X-L integrated model; 5) And (3) rapidly forecasting creep stress indexes of the test set sample and 4 independent experimental samples in the step (3) by using the R-X-L integrated model established in the step (4). Preferably, in the step 4), the specific steps of training the R-X-L integration model are as follows: 4-1) randomly extracting a plurality of subsamples T TS from a total training sample set S by using a Bootstrap resampling method, randomly selecting a plurality of attribute values to perform node splitting when each T TS is used for model training, generating a plurality of regression trees to the maximum extent without pruning, and finally forming a multi-element nonlinear regression combined learner 1; 4-2) considering the limitation of the complexity of the tree model on the basis of the loss function, and combining the loss function and the complexity as an objective function, wherein the objective function of the mth decision tree can be written as: where y i is the observation of the i-th sample, i=1,.. F m (x) is the predictive value of the first m decision trees of the sample, x is the input value of the sample, and l is the deviation of the observed value and the predicted value of the sample calculated by the loss function; the first term of the objective function is the loss function of the original mt