Towards standardized polymer solubility measurements using a parallel crystallizer

Recently developed machine learning approaches to understanding material properties provide new opportunities for studying and developing new knowledge about polymers. The vast application of polymers and their sensitivity to many parameters like molecular weight, temperature, morphology, etc. requires developing high throughput experimentation methods that lead to high quality data for machine learning predictions. Although polymer informatics has significantly improved in recent years using artificial intelligence and machine learning techniques, a significant hurdle in the polymer informatics area is the lack of complete and sufficiently detailed databases. One example where this is a challenge is in developing a dataset for polymer solubility and dissolution, where the classification of "soluble" or "insoluble" is sensitive to molecular weight, concentration, temperature, heating/cooling rate, and other factors that are not always reported or well-controlled. To overcome the challenges with the current dataset on polymer solubility, in this work, we will focus on applying turbidity measurements with precise temperature control using a Crystal16 parallel crystallizer. With this method, the experiments are possible to do in reproducible and time-effective ways with minimal human error while also collecting time-resolved dissolution data continuously throughout the experiment. In this work, we analyze the effect of experimental parameters on measured turbidity for representative polymer/solvent pairs and identify key parameters to control for producing accurate and high quality data for different types of polymers. This approach can lead to detailed results in a controllable manner and will allow researchers to develop information-rich data sets for predicting the polymer solubility using machine learning techniques.