Cyclic stress-strain behavior and microstructural features in copper-Cu50Zr50 metallic glass core-shell structures: Molecular dynamics and deep machine learning predictions

Abstract

In this paper, we report the strain-controlled cyclic deformation behavior of copper-Cu₅₀Zr₅₀ metallic glass (MG) core-shell structures using the molecular dynamics (MD) simulation. Specimens with different MG shell thicknesses (12.5 Å−50 Å) and crystalline copper core thickness (50 Å) are used for the investigation. The cyclic deformations are carried out at a temperature of 300 K and strain amplitudes in the range of 0.05–0.13. With increasing MG thickness, the fatigue properties of the core-shell specimens improve. The fatigue ductility exponent is −0.45, and the fatigue strength exponent is −0.13. The deep machine learning model bidirectional long short-term memory (Bi-LSTM) is used to predict the cyclic stress-strain response of core-shell structures using the MD data. For training the model, 16,800 data points are used, comprising forty-three data sets. The model accurately predicts the cyclic behavior at all the strain amplitudes on the trained data. The R² values are in the range of 0.947–0.998 on the test data, indicating the goodness of the fit. Hence, the model can be used to predict the fatigue behavior of materials, reducing the time required for experimentation.