Carbon Nanotube Growth Rate Regression using Support Vector Machines and Artificial Neural Networks.

摘要

Control of carbon nanotube growth rates is a challenging problem, thus limiting their use in a wide variety of applications. Carbon nanotubes demonstrating metallic or semiconducting properties allow for high strength materials and high current densities in smaller wires. Due to their simplicity and desirable properties, SWNTs are considered for chiral-selective growth experiments. A machine learning based approach for chiral selective growth of SWNTs using a laser-induced chemical vapor deposition growth system is introduced. Determination of SWNT growth rates is performed through in-situ Raman spectroscopy using a 532 nm excitation laser. A total of 450 experiments are performed and a subset of 121 experiments are used to train a SWNT vs. MWNT SVM classifier. The SVM classifier determines parameter values for 99% probability or greater of SWNT growth with an accuracy of 95.04%. This subset of synthesis parameters are evaluated using an ANN to predict SWNT growth rates and growth lengths. Analysis of the ANN growth rate model showed a peak in growth rate as a function of water concentration and growth temperature. The growth length model was trained using the same growth experiments as the growth rate model and showed a 80% reduction in validation errors. The growth length model also identified an optimal water/ethylene ratio for maximizing SWNT length.