학술논문
高熵合金生成相的机器学习预测 / Prediction of High Entropy Alloys'Generated Phases Using Machine Learning
Document Type
Academic Journal
Author
Source
西安工业大学学报 / Journal of Xi'an Technological University. 44(1):69-78
Subject
Language
Chinese
ISSN
1673-9965
Abstract
针对传统高熵合金制备试验成本高、周期长、不可控因素等问题,文中通过机器学习模型进行高熵合金相阶段预测,进行了多分类和多标签分类,利用主流机器学习算法进行建模,并通过特征重要性分析及超参数调优进行比较,探究了不同算法对预测结果的影响.研究结果表明:多分类支持向量机(SVM)达到0.86的准确率,多标签随机森林(RF)达到了 0.94的准确率;机器学习可以提供可靠的合成特性特性,在合金合成时,电负性(△x)和价电子浓度(VEC)可作为主要考虑因素.高熵合金和机器学习的结合有望为改进合金系统的设计提供参考.
The preparation and test of traditional high-entropy alloys have the disadvantages of higher cost,longer cycle and some uncontrollable factors.To overcome these problems,the paper presents phase prediction of high-entropy alloys by machine learning models.Multi-classification and multi-label classification are done,and modeling is carried out by the commonly used machine learning algorithms.By analyzing the feature importance and comparing the hyper-parameter tuning,the influence of different algorithms on the prediction results are discussed.The results show that the accuracy of multi-classification support vector machine(SVM)was 0.86 and of multi-label random forest(RF)0.94.Machine learning can offer reliable synthetic property characterization,and electronegativity(△x)and valence electron concentration(VEC)can be the main factors to be considered in alloy synthesis.The combination of high-entropy alloys and machine learning is expected to provide references for improving the design of alloy systems.
The preparation and test of traditional high-entropy alloys have the disadvantages of higher cost,longer cycle and some uncontrollable factors.To overcome these problems,the paper presents phase prediction of high-entropy alloys by machine learning models.Multi-classification and multi-label classification are done,and modeling is carried out by the commonly used machine learning algorithms.By analyzing the feature importance and comparing the hyper-parameter tuning,the influence of different algorithms on the prediction results are discussed.The results show that the accuracy of multi-classification support vector machine(SVM)was 0.86 and of multi-label random forest(RF)0.94.Machine learning can offer reliable synthetic property characterization,and electronegativity(△x)and valence electron concentration(VEC)can be the main factors to be considered in alloy synthesis.The combination of high-entropy alloys and machine learning is expected to provide references for improving the design of alloy systems.