Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 17: Poster 2
CPP 17.8: Poster
Dienstag, 6. September 2022, 11:00–13:00, P2
Nucleation patterns in polymer crystallization analyzed by machine learning — •Atmika Bhardwaj1, 2, Marco Werner1, and Jens-Uwe Sommer1, 2 — 1Leibniz-Institut für Polymerforschung Dresden e. V., Hohe Str. 6, Germany 01069 — 2Technische Universität Dresden (TUD)
Today, many efforts seek to link machine learning (ML) algorithms to the concepts of theoretical physics. Our work focuses on developing ML tools to derive meaningful interpretations from the data generated through molecular dynamics simulations. We aim to find and quantify the nucleation patterns in polymer crystallization. The transition dynamics occurring under-cooled polymer melt is a local environmental phenomenon rather than a property of individual particles (or monomers), and depends on subtle conformation patterns such as entanglements between the chains. Our first objective is to define a set of fingerprint parameters to capture the crucial information in the local conformation and a monomer's environment and to quantify the degree of crystallinity. We use self-supervised auto-encoders to contain those local fingerprints. The second objective is to recognize the precursors or nucleation sites that stimulate crystal growth before the occurrence of such growth. We are currently working on both convolutional neural networks and recurrent neural networks to investigate the spatial and temporal patterns of the precursors.