Prediction of the Liquid Viscosities of Pure Components and Mixtures Using Neural Network and ASOG Group Contribution Methods

Abstract

This study proposes a simple method for predicting liquid viscosities of pure components and mixtures from chemical structures only. Initially, the constants B and T ⁰ used in a modified Andrade equation for pure liquid viscosity were predicted using a three-layer neural network method with an error back-propagation learning algorithm. The network was trained using 194 data sets and information concerning 29 chemical groups was used as descriptors of input. The components covered were paraffines, olefins, alkynes, aromatic hydrocarbons, chlorides, bromides, alcohols, ketones, esters, ethers, aldehydes and organic acids. The temperature range is approximately the melting point to the bubble point of the compounds and the average and maximum deviations of viscosity are 9.5 and 14.3%, respectively.

The viscosities of binary systems were predicted using the ASOG-VISCO group contribution method. This study covers mixtures composed of CH², ArCH, CyCH, OH, H²O, CO and COO groups with a temperature range of 293.15–303.15 K. and an average deviation of viscosities of 4.5%.

Keywords:

• Liquid viscosity
• Neural network
• ASOG-VISCO group method
• Pure component viscosity

Acknowledgements

This work was partially supported by a Grant-in-Aid for Scientific Research awarded by the Ministry of Education, Sports and Culture, Japan (Grant No. 14550744).