Predicting the mechanical properties of pristine and defective carbon nanotubes using a random forest model

Area of Science:

• Materials Science
• Computational Materials Science
• Nanotechnology

Background:

• Conventional methods like molecular dynamics and density functional theory for material property computation are time-consuming.
• Data-driven models offer a computationally efficient alternative for predicting material characteristics.

Purpose of the Study:

• To develop and validate a random forest (RF) model for predicting the mechanical properties of carbon nanotubes (CNTs).
• To assess the model's predictive accuracy and robustness for both pristine and defective CNTs across various strain conditions.

Main Methods:

• Classical molecular dynamics simulations were used to calculate stress and Poisson's ratio for CNTs.
• A random forest model was trained using fitting parameters and ultimate tensile strength, with features selected via principal component analysis.
• Comparative analysis of machine learning algorithms identified the RF model as the superior performer.

Main Results:

• The RF model achieved high predictive accuracy, with RMSE values of 0.013 for pristine and 0.0143 for defective CNTs.
• Correlation coefficients exceeded 0.99, demonstrating excellent predictive power for CNTs within the trained diameter range (0.4-2 nm).
• The model successfully predicted properties for CNTs with diameters larger than the training set (>2 nm), indicating robustness.

Conclusions:

• The developed RF model serves as a robust and efficient substitute for computationally intensive molecular dynamics simulations.
• The findings provide valuable insights for advancing research in nanocomposites, nanoelectronics, and nanomechanical systems utilizing CNTs.