|
| 基于机器学习的钛合金弹性模量预测方法研究 |
| Prediction Method of Elastic Modulus of Titanium Alloy Based on Machine Learning |
| Received:September 06, 2023 |
| DOI：10.3969/j.issn.1674-6457.2024.01.004 |
| 中文关键词: 钛合金 第一性原理 机器学习 遗传算法 力学性能 |
| 英文关键词: titanium alloy first principles machine learning genetic algorithm mechanical property |
| 基金项目:国家自然科学基金(52075386);天津市自然科学基金多投入重点项目(22JCZDJC00650);中国博士后科学基金第67项研究基金(2020M672309);陕西省高性能精密成形技术与装备重点实验室项目(PETE2019KF02) |
| Author Name | Affiliation | | WANG Yuanyuan | National-local Joint Engineering Laboratory of Intelligent Manufacturing Oriented Automobile Die &
Mould, Tianjin University of Technology and Education, Tianjin 300222, China | | WU Chuan | National-local Joint Engineering Laboratory of Intelligent Manufacturing Oriented Automobile Die &
Mould, Tianjin University of Technology and Education, Tianjin 300222, China | | PENG Zhiwei | National-local Joint Engineering Laboratory of Intelligent Manufacturing Oriented Automobile Die &
Mould, Tianjin University of Technology and Education, Tianjin 300222, China | | SHI Wencai | National-local Joint Engineering Laboratory of Intelligent Manufacturing Oriented Automobile Die &
Mould, Tianjin University of Technology and Education, Tianjin 300222, China |
|
| Hits: 1543 |
| Download times: 776 |
| 中文摘要: |
| 目的 探索一种高效可行的预测方法以提高钛合金弹性模量的预测精度，采用第一性原理计算方法与机器学习相结合的方式建立高精度的预测模型。方法 通过数据挖掘获取材料数据库中钛合金的力学性质微观结构参数，结合第一性原理计算方法构建初始数据集，并对其进行预处理，包括噪音消除、归一化及标准化，以得到高质量的数据集。同时，采用随机森林特征重要性分析法对输入参数进行筛选，去除弱相关变量以降低预测模型的复杂度。在此基础上，构建随机森林模型、支持向量机模型、BP神经网络模型及优化后的GA-BP神经网络模型，综合对比各模型的回归能力，分析误差后选出最优的算法模型。结果 最终建立了钛合金弹性模量预测模型，其中随机森林模型、支持向量机模型、BP神经网络模型、GA-BP神经网络模型的预测相关系数R分别为0.836、0.943、0.917、0.986。结论 GA-BP模型对弹性模量的预测误差基本保持在5%~7%。遗传算法可以优化BP神经网络的权值和阈值，使预测精度大幅提升。说明通过该方法可以实现钛合金弹性模量的预测，大大节省研发和实验成本，加快高性能材料的筛选。 |
| 英文摘要: |
| The work aims to improve the prediction accuracy of elastic modulus of titanium alloy through an efficient and feasible prediction method, and establish a high-precision prediction model which combines first-principle calculation and machine learning. Through data mining, the microstructure parameters of mechanical properties of titanium alloy in the material database were obtained, and the initial data set was calculated and constructed based on the first principle, which was pretreated, including noise elimination, normalization and standardization, so as to obtain a high-quality data set. At the same time, the random forest characteristic importance analysis method was used to screen the input parameters and remove the weakly correlated variables to reduce the complexity of the prediction model. On this basis, a random forest model, a support vector machine model, a BP neural network model and an optimized GA-BP neural network model were constructed, and the optimal algorithm model was selected after comprehensive comparison of regression capacity of each model and error rate analysis. Finally, a prediction model for elastic modulus of titanium alloy was established, in which the correlation coefficient R of the random forest model, the support vector machine model, the BP neural network model and the optimized GA-BP neural network model was 0.836, 0.943, 0.917, and 0.986. Through comparative analysis, the prediction error of elastic modulus of GA-BP models is basically kept at 5%-7%, showing high prediction accuracy. It is found that genetic algorithm can optimize the weight and threshold of the BP neural network, so as to give higher prediction accuracy. This method can realize the prediction of elastic modulus of titanium alloy, greatly save the research and development and experimental costs, and is applicable to the selection of high-performance materials. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
渝公网安备 50010702501719号