Estimation of density and viscosity of deep eutectic solvents: Experimental and machine learning approach

Abstract

Deep eutectic solvents (DESs) are increasingly recognized as sustainable alternatives suitable for a range of industrial applications. A precise comprehension of their properties is important for progress in science and engineering. In this study, we synthesized four novel ternary DESs using mandelic acid and measured their densities and viscosities at temperatures ranging from 298 to 353 K. Subsequently, an artificial neural network model was developed to predict DES density and viscosity based on temperature, critical properties, acentric factor, and molar ratio. The neural network parameters were optimized using experimental data from synthesized DESs and literature sources, both collectively over 500 data points for density and viscosity. Additionally, we investigated the influence of input parameters on model accuracy and assessed their significance. The results show that the average percentage relative error was 0.501 for density and 4.81 for viscosity. This research helps advance science and engineering applications of DESs.