Prediction of Self-Diffusion in Binary Fluid Mixtures Using Artificial Neural Networks.

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Physical Chemistry

Background:

• Predicting self-diffusion constants in binary fluid mixtures is crucial for process design.
• Complex intermolecular interactions, such as hydrogen bonding, can lead to non-linear diffusive behavior, challenging traditional models.

Purpose of the Study:

• To develop accurate artificial neural network (ANN) models for predicting self-diffusion constants in binary fluid mixtures.
• To incorporate intermolecular interaction strengths, specifically self- and binary association energies, as input features for enhanced model performance.

Main Methods:

• Trained ANNs on an experimental database of 4328 self-diffusion constants from 131 mixtures.
• Calculated self- and binary association energies to quantify intermolecular interactions.
• Employed forward input feature selection to identify key predictive properties.

Main Results:

• Developed a generalized ANN model with an average absolute deviation of 4.1%.
• Demonstrated the importance of critical properties and self-association energies in predicting diffusion.
• Showcased accurate predictions for mixtures with strong hydrogen bonding and under extreme pressure changes.

Conclusions:

• ANNs provide a robust and accurate method for predicting self-diffusion constants in binary fluid mixtures.
• Intermolecular interaction energies are vital inputs for capturing complex diffusive behaviors.
• The developed ANN models show broad applicability across diverse fluid mixtures and conditions.