Accelerated discoveries of mechanical properties of graphene using machine learning and high-throughput computation

Machine learning (ML) has been vastly used in various fields, but its application in engineering science remains in infancy. In this work, for the first time, different machine learning algorithms and artificial neural network (ANN) structures are used to predict the mechanical properties of single-layer graphene under various impact factors of system temperature, strain rate, vacancy defect and chirality. The predictions include fracture strain, fracture strength and Young's modulus. High throughput computation (HTC) combined with classical molecular dynamics (MD) simulation is used to generate the training dataset for the ML models. It was discovered that both temperature and vacancy defect have negative effects on the predicted properties while strain rate has positive correlations with the prediction results. The stochastic gradient descent (SGD) method could not properly capture the effects of the different impact factors on the mechanical properties of graphene, while k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT) and ANN provided desirable prediction results. Discoveries in this work provide new perspectives on the study of mechanical properties using state-of-the-art computational methods.