Exploring multi-scale convolutional neural network for modeling amorphous materials behavior: a comparative approach

Abstract

Glass transitions are a crucial phenomena in amorphous materials with potential for various applications. The Tool-Narayanaswamy-Moynihan model is a widely used empirical model that describes the enthalpy relaxation behavior of these materials. However, determining the appropriate values for its parameters can be challenging. To address the issue, this research explores the use of deep learning to describe the kinetics of enthalpy relaxation and glass transition. More specifically, a multi-scale convolutional neural network (MCNN) model is proposed, which outperforms traditional methods in extracting kinetic information from differential scanning calorimetry data. The MCNN model's robust architecture captures both low-frequency and high-frequency signal components, making it highly effective for complex pattern recognition in thermal analysis. The study's findings have significant implications for the utilization of amorphous materials in high-tech and classical engineering applications.