The study of mechanical properties in sheaf-structured spherulite semi-crystalline polymers using a data-driven micromechanical model

A low-cost prediction system for the mechanical properties evaluation of semi-crystalline polymers has been developed using machine learning algorithms and micromechanical simulations. In this system, a microstructure modeling approach is initially used to calculate the microstructure-sensitive mechanical response of single-spherulite. A multi-scale modeling approach coupled multi-spherulite modelling at the micro-scale and single-spherulite modelling consisting of crystalline-amorphous lamella at the nano-scale was then used to capture the local and overall mechanical responses. Based on this approach, uniaxial tensile simulations are employed to obtain the mechanical properties of multi-spherulite semi-crystalline polymers. The effects of microstructure features, such as crystal initial growth directions, sheaf structure sizes, and crystallinity levels, on the mechanical properties were highlighted in the calculation and the results were used for database construction. The random forest algorithm was trained to establish the correlations between the mechanical properties and microstructure features. The confidence and limitations of the proposed system were thoroughly analyzed. The results demonstrate that the prediction ability is capable of accurately predicting the mechanical properties of semi-crystalline polymers. In addition, the microstructure-property relationship was revealed by analyzing feature importance, which sheds light on tailoring mechanical properties in manufacturing not only for semi-crystalline polymers but also for materials with spherulite microstructure.