Estimation with neural networks of the water content in imidazolium-based ionic liquids using their experimental density and viscosity values.

A multilayer perceptron neural network (NN) model has been created for the estimation of the water content present in the following ionic liquids (ILs): 1-butyl-3-methylimidazolium tetrafluoroborate, 1-butyl-3-methylimidazolium methylsulfate, 1,3-dimethylimidazolium methylsulfate and 1-ethyl-3-methylimidazolium ethylsulfate. To achieve this goal, their density and viscosity values were used. The experimental values of these physicochemical properties, employed to design the NN model, were measured and registered at 298.15K. They were determined at different relative humidity values ranging from 11.1 to 84.3%. The estimated results were then compared with the experimental measurements of the water content, which were carried out by the Karl Fischer technique, and the difference between the real and estimated values was less than 0.05 and 3.1% in the verification and validation processes, respectively. In addition, an external validation process was developed using four bibliographical references. In this case, the mean prediction error was less than 6.3%. In light of these results, the NN model shows an acceptable goodness of fit, sufficient robustness, and an adequate estimative capacity to determine the water content inside the studied range of the ILs analyzed.