Predictive Models for the Binary Diffusion Coefficient at Infinite Dilution in Polar and Nonpolar Fluids.

Area of Science:

• Chemical Engineering
• Physical Chemistry
• Computational Chemistry

Background:

• Experimental diffusivity data is scarce, hindering accurate modeling of rate-controlled processes.
• Diffusivity is a critical parameter in chemical kinetics and transport phenomena.
• Accurate prediction of diffusivities is essential for process design and optimization.

Purpose of the Study:

• To develop and evaluate machine learning models for estimating solute diffusivities in polar and nonpolar solvents.
• To compare the performance of various machine learning algorithms, including gradient boosting, for diffusivity prediction.
• To provide accurate and accessible computational tools for diffusivity estimation.

Main Methods:

• Trained machine learning models on extensive databases of polar (1431 points) and nonpolar (1129 points) systems.
• Evaluated five algorithms: multilinear regression, k-nearest neighbors, decision tree, random forest, and gradient boosted.
• Identified key predictive parameters including temperature, viscosity, molar mass, critical pressure, and Lennard-Jones energy.

Main Results:

• The gradient boosted algorithm achieved the best performance for both polar (5.07% AARD) and nonpolar (5.86% AARD) systems.
• Developed models require minimal input parameters and demonstrate superior accuracy compared to classic models like Wilke-Chang.
• The polar model utilizes 6 parameters, while the nonpolar model uses 5, offering practical applicability.

Conclusions:

• Machine learning, particularly gradient boosting, offers a powerful approach for predicting solute diffusivities.
• The developed models provide accurate and efficient alternatives to experimental measurements and traditional correlations.
• The models are provided as a command-line program for ease of use in research and industry.