Study on artificial neural networks and structure–activity relationship for constructing viscosity correlations of amine aqueous solutions based on chemical structure information

In this study, the viscosity of 12 alkanolamine or diamine aqueous solutions was measured at atmospheric pressure, with amine mass fractions ranging from 15 % to 100 % and temperatures ranging from 293.15 K to 353.15 K. An empirical model was used to correlate the viscosity experimental results of the 12 alkanolamine or diamine systems. Building upon the empirical model, the influence of chemical structure on viscosity was further explored, and two artificial neural networks with different data partitioning schemes, namely R-ANN and C-ANN, were developed. The mean absolute error (MAE) of the R-ANN and C-ANN models were 0.42 and 0.53, respectively. The evaluation results demonstrated that the R-ANN model effectively predicted the viscosity of the 12 alkanolamine and diamine systems. Additionally, the C-ANN model showed reliable predictive performance on a test set consisting of new amines—those not included in the training set. Finally, the structure–activity relationship between amine structure and viscosity was analyzed by calculating the electrostatic potential (ESP) of amine molecules. This research provides an effective theoretical framework and computational approach for predicting and understanding the viscosity of amine aqueous solutions.