Prediction of self-diffusion coefficients via a hybrid PCP-SAFT + ANN model incorporating COSMO-SAC sigma-profile...

Area of Science:

• Physical Chemistry
• Chemical Engineering
• Computational Chemistry

Background:

• Accurate estimation of the self-diffusion coefficient is crucial for understanding mass transport in fluids.
• Predicting self-diffusion coefficients is challenging due to the significant impact of thermodynamic conditions and molecular properties.

Purpose of the Study:

• To develop a hybrid predictive model for estimating self-diffusion coefficients under diverse conditions.
• To combine the PCP-SAFT equation of state with an artificial neural network (ANN) for enhanced prediction accuracy.

Main Methods:

• A hybrid model integrating the PCP-SAFT equation of state and an ANN was developed.
• The model utilized a dataset of 2263 experimental measurements for 67 compounds.
• Thermodynamic inputs (density, residual entropy) from PCP-SAFT and molecular descriptors from COSMO-SAC were incorporated.

Main Results:

• The hybrid model achieved high predictive performance, with R² values of 0.9937 (training) and 0.9763 (testing).
• Average absolute relative deviation (AARD) was 8.89% for training and 15.89% for testing.
• The model demonstrated accuracy across a wide range of temperatures (93.0–973.2 K) and pressures (up to 3036 bar).

Conclusions:

• The proposed hybrid framework offers a unified and reliable approach for predicting fluid diffusion behavior.
• This model effectively captures the influence of thermodynamic conditions and molecular characteristics on self-diffusion.