Microstructure analysis and stress predictions in Al thin foils: a CNN coupled ANN study on the tensile and shear deformations

In this study, a CNN (convolutional neural network) coupled with an ANN (artificial neural network) deep machine learning model (DML) is used to predict the magnitude of normal and shear stresses by analyzing the microstructural images of deformed Al thin foils. At first, molecular dynamics (MD) simulations of tensile and shear deformations are carried out on poly-crystalline Al thin foils. The specimens are subjected to various MD variables, including strain rates (1010 s−1, 5 × 109 s−1), temperatures (300, 500 K), and boundary conditions (periodic and non-periodic). Voids of different sizes and at different positions are inserted to study their effects on deformation behaviour and mechanisms. The stress data and microstructural images generated from the MD simulations are used to train the DML model for predictions. The MD results show that the presence of voids lowers the strength (UTSperfect = 1.7 GPa; UTSvoids = 0.8 GPa−1.2 GPa) of the Al thin foils. Also, the position and size of the voids significantly affect the deformation behaviour. The absolute difference between the MD stress magnitudes and those predicted by the DML model is less than 5%, indicating that the model accurately predicts stress magnitudes from the recognised microstructures.