▎ 摘　　要

Nowadays, various machine learning (ML) approaches are widely used in different research areas. However, the need for a large training dataset has restricted the attractiveness of ML techniques for industrial applications, since the preparation of a large dataset is very costly and inefficient. To deal with this limitation, an efficient method is required to fill the gap between industry and research. For this purpose, in this study a transfer learning-based deep neural network (TL-DNN) model was developed to predict the mechanical properties of various graphene reinforced nanocomposites. In this respect, a hybrid multi-layer feedforward DNN was designed, containing one source network and one target network. The source DNN was trained to predict the mechanical properties of graphene/graphene oxide nanocomposites with various matrix types including Al, Cu, PMMA, Si3N4, Al2O3, etc. By transferring the acquired knowledge of the source DNN to the target DNN, the mechanical properties of another material (graphene/epoxy nanocomposite) were estimated with high accuracy level, even with limited number of data samples. It should be mentioned that the optimal values of the network hyperparameters were determined using genetic algorithm (GA), simulated annealing (SA) and particle swarm optimization (PSO) algorithms.