Investigation on Heat Transfer in a Semicrystalline Polymer by Combining Molecular Simulation and Machine Learning

Abstract

In this work, a molecular dynamics simulation is firstly utilized to explore the thermal conductivity of a semicrystalline polymer by manipulating the crystallinity and stretching. The thermal conductivity exhibits a continuous rise with increasing crystallinity, which mainly comes from crystalline phase rather than amorphous phase by distinguishing their respective contribution. A high crystallinity improves the concentration and order degree of crystalline phase and lowers the interface ratio, which thus improves thermal conductivity. Conversely, the thermal conductivity of amorphous phase is nearly unchanged. By decomposition of heat flux into different transfer modes, the bonded interaction is the main contribution to heat transfer. By analyzing the vibrational density of states, the heights of three characteristic peaks are improved simultaneously with increasing crystallinity. In addition, the increases in specific heat, intrinsic mean free path and group velocity of phonons can rationalize the high thermal conductivity of a semicrystalline polymer. Secondly, stretching improves the thermal conductivity parallel to the stretching direction, which is also contributed primarily by bonded interaction. Meanwhile, the enhancement factor in thermal conductivity decreases with increasing initial crystallinity, which is determined by crystalline phase. Crystallinity increases significantly for the low initial crystallinity. And the enhanced orientation degree of chains improves the thermal conductivity of crystalline phase. However, crystallinity firstly decreases and then rises for the high initial crystallinity. Moreover, the enhanced thermal conductivity of amorphous phase is comparable for different initial crystallinities, which is due to the similar orientation degree of chains. Finally, the eXtreme Gradient Boosting regression model yields a more excellent prediction of thermal conductivity parallel to the stretching direction than the linear or polynomial regression model. Meanwhile, both crystallinity and orientation degree of chains exhibit the similar feature importance in determining the thermal conductivity of a semicrystalline polymer. In summary, this work provides a deep understanding into how crystallinity and stretching regulate thermal conductivity of a semicrystalline polymer at molecular scale.