Machine Learning Techniques to Assess Rheological Similarity in Fluorinated Thermoplastics

Abstract

In this paper, we assess the rheological similarity between different lots of polyvinylidene fluoride-co-chlorotrifluoroethylene (PVDF-CTFE) using dimension reduction methods. Our goal was to identify the PFAS-free PVDF-CTFE polymer lot most rheologically similar to the legacy PFAS-based polymer from among six candidate lots. The target legacy lot is a PVDF-CTFE polymer that uses per- and polyfluoroalkyl substances (PFAS) as a surfactant in the emulsion polymerization. This study is motivated by the EPA’s regulation against the use of PFASs in synthesis. We utilized principal component analysis and linear discriminant analysis to reduce the dimensionality of our data sets. Our results showed that the synthesis temperature is the dominant factor that determines the rheological properties. Other synthesis parameters such as the initiator concentration and mixing speed only contribute at a secondary level. We also concluded that the proposed dimension reduction analysis of rheological similarity does not require a crossover point between the storage and loss modulus curves in the rheological data, which is more versatile compared to traditional analysis where key metrics such as zero-shear viscosity, crossover frequency, and modulus are used to assess similarity.