基于机器学习的低弹性模量钛合金建模和预测的研究

Research on the modeling and predicting low elastic modulus titanium alloys based on machine learning

针对钛合金弹性模量快速预测的需要,采用合金设计公式对原始合金数据进行转换,利用转换所得的Mo当量、d-电子结合次数和d-电子结合能作为数据集;采用多层感知器、随机森林网络和卷积神经网络三种机器学习方法,基于数据驱动方式搭建钛合金成分与弹性模量的关系模型。结果表明,相比随机森林网络模型和卷积神经网络模型,多层感知器模型具有更优的预测性能和预测精度。此外,多层感知器模型的预测能力符合预期,其相关指数评分达到0.66,均方根误差为7.54 GPa;说明多层感知器适用于医用钛合金的数据挖掘和研发。

Aiming at the need for rapid prediction of elastic modulus of titanium alloy,the original alloy data were converted by the alloy design formulas,and the molybdenum equivalent,d-electron binding order and d-electron binding energy obtained by the conversions were used as the original dataset.Three machine learning methods,namely multilayer perceptron,random forest network,and convolutional neural network,were used to build the relationship model between titanium alloy compositions and the elastic moduli based on data-driven mode.The results show that compared with random forest network model and convolutional neural network model,multi-layer perceptron model has better prediction performance and accuracy.In addition,the prediction ability of the multi-layer perceptron model meets the expectation,its correlation index score reaches 0.66,and the root mean square error is 7.54 GPa,suggesting that multi-layer perceptron is suitable for the data mining and development of biomedical titanium alloys.