Predicting the structures and properties of interfaces in nanomaterials by coupling computational simulation and machine learning technique

Nanomaterials exhibit many unique properties compared to traditional bulk materials, interfaces play a more important role in nanoscale systems by significantly influencing the mechanical performance. In this thesis, we focus on an intricate exploration of various interfaces, ranging from simple GBs in bicrystal models to intricate GB networks within polycrystalline structures and interfaces within nanocomposite materials. Various computational methodologies, including MD, DFT, and advanced machine learning algorithms, were employed to simulate and predict the mechanical properties of interfaces with microstructural complexity.Firstly, utilizing MC/MD simulations, we established a distinct correlation between GB motion in the Cantor alloy and the Cr concentration within the GBs. A formulation is calculated to link the GB mobility with the Cr concentration. Subsequently, DFT simulations highlight that vacancies in Tungsten GBs prefer to appear in the layer adjacent to the GB plane rather than the GB plane itself. These vacancies, as the findings suggest, cause the strength to decrease under tensile loading. Then, to expedite the prediction of interfacial properties, a cGAN model was developed to predict GB network evolution in polycrystalline samples based on the training data of MD simulation results. Finally, two modified deep learning models are introduced including the CNN-Prob and FNN-Prob, to predict the yield stress of a composite material, Cu-Cu/Zr. These models encompassed dual components for predicting both mean values and associated standard deviations.