Designing sulfonated polyimide-based fuel cell polymer electrolyte membranes using machine learning approaches

Fuel cells are the efficient electrochemical energy conversion devices with wide-ranging applications. Polymer Electrolyte Membrane (PEM) is the primary component of a PEM fuel cell whose proton conductivity majorly determines the performance of the fuel cells. Due to the high cost and limited range of operating parameters, alternatives of perfluorinated ionomers based commercial PEMs are urgently required. Sulfonated polyimides (SPIs) based hydrocarbon PEMs, have exhibited better proton conductivity even at low hydration levels and high temperatures, making them possible candidates for replacing commercial PEMs. However, finding alternative SPI PEMs is a critical polymer discovery problem that requires enormous experimental efforts where Machine learning (ML) approaches can help to reduce such efforts. To this end, both supervised and unsupervised ML approaches are developed to predict the proton conductivity of SPIs. A hybrid dataset of 81 unique SPIs is generated that consists of collected chemical structure–properties data from reported literature and calculated quantitative structure–property and semi-empirical quantum chemical descriptors. Using simple and interpretable Decision Trees, rules that lead to a low or high class of proton conductivity labels with high accuracy are identified. The trained decision tree model can accurately predict the proton conductivity class labels with a prediction accuracy of 88% and a kappa statistic of 0.77. The random forest regression (RFR) model, on the other hand, identified additional set of features that can predict proton conductivity with reasonable error. Thus, high information-gain features have been identified and their correlation with the proton conductivity class labels have been explored. These findings are key to designing novel SPI PEMs while correlating proton transport at the ionomer level with factors such as the morphology of the microstructure and inter-chain interactions.