(186h) Leveraging Machine Learning Techniques to Identify Ionic Liquids Possessing High Ionic Conductivity

Ionic liquids are a class of salts that are liquid at room temperature with several unique and desirable properties such as negligible vapor pressure, precise designer tunability, high thermal and electrochemical stability. Because of these favorable properties, ionic liquids are currently being considered as a potential replacement for conventional organic electrolytes in Li-ion batteries. However, a major limitation of ionic liquids is their sluggish transport properties that are limiting their widespread application in battery technology. This drawback can be tackled by choosing the right cation-anion combination based on chemical intuition that could improve transport properties. The vast chemical space of available cations and anions for such manipulation could be challenging and unfeasible. The increase in the availability of experimental data in the open literature for ionic liquids provides an excellent opportunity to apply machine learning methods for correlating the ionic liquid properties and using the knowledge to discover new ionic liquids with desired properties.

Much of the data found in the literature is primarily focused on imidazolium-based cations because of their low viscosity and high ionic conductivity characteristics. However, imidazolium-based cations are associated with low electrochemical stability because of the presence of an acidic proton in the imidazolium ring. Alternates to the imidazolium-based ionic liquid family are classes of cations derived from groups such as phosphonium, piperidinium, and pyrrolidinium, which are not known to have high electrochemical stability, but are often characterized by very low ionic conductivity and high viscosity.

In this presentation, we will discuss our efforts developing a machine learning model based on the ionic conductivity data covering several different cation families obtained from the NIST ILThermo Database. A feed-forward artificial neural network (FFAAN) comprised of an input layer, a hidden layer, and an output layer is obtained from the data on all ionic liquid types. We will demonstrate that the model can capture ionic conductivity with very high accuracy for both the training and test cases. We will present our approach on using data mining techniques to identify patterns governing ionic conductivity trends within the data. This insight is fed into a classification model that is used to identify high ionic conductivity ionic liquids with implications in advancing the field of ionic liquids as next generation electrolytes for Li-ion, Na-ion, and K-ion batteries.