Machine Learning-Guided Design of Binary Ionic Liquid Mixtures for Spacecraft Thermal Control

This study investigates the use of binary ionic liquid (IL) mixtures as advanced heat transfer fluids, with the aim of identifying combinations that exhibit improved thermophysical properties compared to their pure IL components. The focus is on predicting viscosity using machine learning (ML) models enhanced with physics-based descriptors. Thermophysical property data were collected for 698 binary mixtures, alongside extensive pure IL datasets used for reference. Multiple feature-generation strategies were explored to represent binary systems, including ideal mixing rules for conventional molecular descriptors to capture composition-dependent interactions, as well as quantum chemical surface charge densities and chemical potentials to incorporate structure-dependent realistic interactions. Initial analysis shows non-ideal behavior in viscosity with respect to both temperature and mole fraction, supporting the potential of binary systems to offer more favorable properties. Deep neural networks were developed for pure ILs and mixtures, and feature reduction techniques were applied to identify the most informative structural and energetic descriptors. This work provides a foundational step toward a generalizable ML framework for predicting binary IL behavior and supports future data-driven design of IL mixtures for single-phase thermal control loops, particularly for aerospace applications in extreme environments.