Reptation theory-similar deep learning model for polymer characterization from rheological measurement

The use of machine learning to predict rheological properties of polymers has great potential to facilitate the characterization of novel materials. Here, we have suggested the analogy between the double reptation (DR) and the deep neural network model. The double reptation model itself can be the special case of the deep learning method; linear activation function, and identical sets of weights for the two hidden layers are the characteristics of the double reptation model. The identical sets of weights in the double reptation model are related with the molecular weight distribution (MWD). We first generated ground truth data based on double reptation model. Then, we analyzed the dataset with reptation-guided deep neural network (RGDNN). We showed that the RGDNN model is available to determine entanglement molecular weight (plateau modulus), and monomeric friction factors from the simulated experimental rheological data (prepared using DR model) without any additional information. Overall, a noteworthy conceptual improvement in the determination of major factors that determine the rheological behavior of ultrahigh molecular weight polyethylene (UHMWPE) gels has been achieved.