Machine learning for detecting microscopic parameters characterizing mechanical properties of liquid crystal elastomers

Liquid crystal elastomers (LCE) are of great scientific and technical interest because of the potential of sensors and soft actuators. Molecular dynamics (MD) simulation is a promising means to clarify macroscopic deformation of LCE from a microscopic viewpoint, however, a detection of major parameters characterizing the deformation is difficult because there are many microscopic parameters. Therefore, a systematic analysis should be done for detecting the relation between microscopic parameters and mechanical properties. In this study, a machine learning (ML) approach is used to explore the relation between microscopic characteristics and mechanical parameters. With these models, we perform MD simulations and compute stress-strain curves. Then a regression analysis with random forest method to explain the difference of stress-strain curves is performed with 20 types of microscopic parameters. The ML results reveal the effective set of data descriptors that predict well the stress-strain curves. Therefore, ML technique has a capability to overcome the difficulty to manually explain the complex relation between microscopic parameters and macroscopic properties.