Combining Machine Learning with Physical Knowledge in Thermodynamic Modeling of Fluid Mixtures

Area of Science:

• Physical chemistry and chemical engineering, focusing on fluid mixtures.

Background:

• Thermophysical properties are essential for science and engineering applications.
• Experimental data for these properties are often limited, necessitating accurate prediction methods.

Purpose of the Study:

• To provide a structured overview of hybrid models integrating physical modeling and machine learning (ML).
• To explore the synergy between established prediction techniques and novel ML approaches.

Main Methods:

• Review of existing physical prediction methods (molecular models, equations of state, excess property models).
• Analysis of the integration of machine learning techniques with traditional physical models.
• Illustration of hybrid model concepts with recent research examples.

Main Results:

• Identified various strategies for combining physical modeling and ML.
• Demonstrated the potential of hybrid models to improve prediction accuracy for thermophysical properties.
• Highlighted the growing trend of ML adoption in physical property prediction.

Conclusions:

• Hybrid models offer a promising avenue for overcoming data scarcity in thermophysical property prediction.
• The combination of physical insights and ML capabilities can lead to more robust and accurate predictive tools.
• Future research should focus on further developing and validating these hybrid approaches.