Developing a feed forward neural network multilayer model for prediction of binary diffusion coefficient in liquids

In the present study, a feed-forward artificial neural network (ANN) was developed to estimate the Fick diffusion coefficient in binary liquid systems. It was found as a function of the mole fraction of one component, diffusion coefficient at infinite dilution, viscosity, and molar volume of each component. These values are easily accessible from literatures and handbooks. Data from 54 systems consisting of 851 data points were collected and the ANN was trained with one and two hidden layers using various numbers of neurons. After selection of the best ANN, the results were compared with other models. The results show that this model has a superior performance on estimating the Fick diffusion coefficient. In addition, a new ANN was developed to predict the Maxwell–Stefan (MS) diffusivity, based on which the Fick diffusion coefficient was calculated. The results show that direct prediction of the Fick diffusivity leads to a higher accuracy compared with that of the indirect prediction. Moreover, in the direct calculation, there is no need for the thermodynamic correction factor, which can be treated as a second advantage of this method. The accuracy of the method was evaluated through data points of other systems, which were not previously introduced to the developed ANN.

► The diffusion coefficient in liquids was found by collection data of 54 systems consist of 851 data points.
► The diffusivity in solution was found using properties of substances at pure and infinite dilution.
► The developed ANN can predict the diffusivity with a higher precision compared with the conventional formulas.
► The developed ANN can predict the diffusivity in solutions which was not previously introduced to the network.
► The proposed method can predict the diffusivity values without calculating the thermodynamics factor.