Evaluating Rate Dependent Behavior of Silane Molecules with Molecular Dynamics Simulations and Machine Learning

Abstract

In this study, silane molecules of different functionalities and molecular weights are evaluated for their performance under different strain rates 109, 1010, and 1011 1/s in Mode I. loading. Molecular Dynamics simulations are performed over the systems made of a single silane molecule bound to the glass surface via one or two bonds and are simulated in the extended or relaxed position resting on the fiber surface. The silane molecules are categorized into three classes based on these three bond failure mechanism within the silane molecule: (1) Silicon-Oxygen/Silicon-Carbon, (2) Nitrogen-Carbon, and (3) Oxygen-Carbon. For all the configurations of silane molecules simulated, the peak strength demonstrated by the silanes which failed with the 1st bond failure mechanism was higher than those failing with the 2nd and 3rd mechanisms. For each failure mechanism, it is observed that the energy absorbed at failure increases with the chain length of the molecule. The non-bonded interactions between silane molecules and the glass surface contributed to the overall energy absorption. 16 to 29 % of the total energy absorbed was due to non-bonded interactions by different silane molecules. In the end, a methodology to develop and implement a Machine Learning model of Variational Auto-encoder is presented to utilize the MD simulations performed at different strain rates and configurations to predict new silanes with improved rate-dependent strength and energy absorption.