▎ 摘　　要

Accurately predicting the mechanical properties of graphene-reinforcedmetal matrix composites is of utmost importance due to its criticalrole in the design and utilization of nanocomposite materials. Theconventional approach of employing molecular dynamics (MD) simulationsfor this purpose faces a substantial increase in computational costswhen considering the combined effects of multiple factors. In contrast,machine learning (ML) models offer a rapid and efficient alternativeby swiftly comprehending and predicting material properties followingadequate training. In this paper, we employed a long short-term memory(LSTM) model, based on MD calculation data, to accurately predictthe mechanical response and key mechanical properties of nickel-graphenecomposite nanomaterials. Specifically, we thoroughly investigatedthe comprehensive impact of temperature, graphene orientation angle,and graphene volume fraction on the mechanical properties. Our verificationprocess revealed that high graphene volume and high orientation anglesled to increased dislocation absorption, consequently weakening thecomposite material. To assess the hardness prediction capabilities,we conducted a comparative analysis between the LSTM model and classicalmultilayer perceptron (MLP) neural networks, as well as the traditionalnonlinear regression method, support vector machine (SVM). The obtainedresults demonstrated that the LSTM models exhibited a remarkable abilityto accurately predict the mechanical properties of nickel-graphenecomposite nanomaterials, showcasing Pearson correlation coefficientsexceeding 0.95 when compared to the calculation data. Moreover, theLSTM model effectively comprehends and predicts the complete indentationdepth-force curve, thus providing enhanced predictions of materialproperties. This study proposes an innovative combination of MD simulationsand ML models, which holds significant application potential in predictingand designing the performance of graphene-reinforced metal matrixcomposite materials.