利用机器学习分类含缺陷的单壁碳纳米管轴向应变
Classifying the Axial Strain of Single-Walled Carbon Nanotubes Containing Defects Using Machine Learning

单壁碳纳米管具有规律、简单、均匀的结构，而且还具有极好的稳定性。因此单壁碳纳米管在多个领域具有潜在应用。在现实生产和制作过程当中，可能不会完美地制备出单壁碳纳米管，发现管壁中存在一些缺陷是不可避免的现象。这些缺陷对单壁碳纳米管的应变造成一定的影响。利用大规模原子分子并行模拟器可以计算含缺陷的单壁碳纳米管的应变值。但是大规模计算时间、资源等成本很高。本文是选取一个较好的特征作为机器学习的输入，然后利用机器学习训练数据并得到学习好的模型，用学习好的模型去分类新的缺陷原子位置不同的单壁碳纳米管的应变类别。结果表明，首先，一个较好的特征作为机器学习的输入至关重要。其次，利用机器学习训练好的模型去预测的方式，对含缺陷的单壁碳纳米管的计算时间、资源等成本极大降低。
Single-walled carbon nanotubes have a regular, simple, and homogeneous structure, and they also have excellent stability. Therefore, single-walled carbon nanotubes have potential applications in several fields. In the reality of production and fabrication, single-walled carbon nanotubes may not be perfectly prepared, and it is inevitable that some defects are found in the walls of the tubes. These defects have an impact on the strain of single-walled carbon nanotubes. The strain values of single-walled carbon nanotubes with defects can be calculated using a large-scale atomic-molecular parallel simulator. However, the cost of large-scale computation time and resources is high. In this paper, a better feature is selected as an input for machine learning, and then the machine learning is used to train the data and obtain a learned model, which is used to classify the strain classes of new single-walled carbon nanotubes with different defective atomic positions. The results show that, firstly, a better feature as an input to machine learning is crucial. Second, the cost in terms of computational time and resources for single-walled carbon nanotubes containing defects is greatly reduced by using the machine learning trained model to predict.