Machine Learning Enabled Prediction of Mechanical Properties of Tungsten Disulfide Monolayer.

One of two-dimensional transition metal dichalcogenide materials, tungsten disulfide (WS), has aroused much research interest, and its mechanical properties play an important role in a practical application. Here the mechanical properties of h-WS and t-WS monolayers in the armchair and zigzag directions are evaluated by utilizing the molecular dynamics (MD) simulations and machine learning (ML) technique. We mainly focus on the effects of chirality, system size, temperature, strain rate, and random vacancy defect on mechanical properties, including fracture strain, fracture strength, and Young's modulus. We find that the mechanical properties of h-WS surpass those of t-WS due to the different coordination spheres of the transition metal atoms. It can also be observed that the fracture strain, fracture strength, and Young's modulus decrease when temperature and vacancy defect ratio are enhanced. The random forest (RF) supervised ML algorithm is employed to model the correlations between different impact factors and target outputs. A total number of 3600 MD simulations are performed to generate the training and testing dataset for the ML model. The mechanical properties of WS (i.e., target outputs) can be predicted using the trained model with the knowledge of different input features, such as WS type, chirality, temperature, strain rate, and defect ratio. The mean square errors of ML predictions for the mechanical properties are orders of magnitude smaller than the actual values of each property, indicating good training results of the RF model.