Abstract
Ice formation is a prevalent issue that poses security risks across various sectors. Unwanted ice accumulation can lead to significant economic challenges and, in some instances, result in fatalities. Despite decades of research into anti-icing solutions, there is still a need for environmentally friendly, cost-effective, and efficient strategies. One promising approach is the use of anti-icing agents, which effectively inhibit ice nucleation. Natural deep eutectic solvents (NADES) represent a low-toxicity and efficient method for addressing this issue. Under cold conditions, NADES create a robust hydrogen bond network that lowers the melting point below freezing, effectively preventing ice growth. NADES are derived from plant metabolites, including sugars, amino acids, and organic acids. Their application as extraction solvents holds potential across various industries, such as pharmaceuticals, food, cosmetics, and aviation.
In this study, we introduce a machine learning (ML/QSAR) framework designed to identify the key features influencing the performance of NADES as anti-icing agents. We developed three predictive models for Melting Temperature, Glass Transition Temperature, and Enthalpy of Melting, utilizing a combination of genetic algorithm (GA), multiple-linear regression (MLR) and non-linear methods. The resulting ML/QSAR models demonstrated excellent performance, showcasing high predictive capability and strong correlation coefficients. The models can be further applied to develop new NADES for various applications, including as potential additives for coatings with anti-ice properties.

Figure 1. Workflow for ML models assembled and validation in this study

Gerardo M. Casanola-Martin

Appearances