Virtual Screening of Novel Eco-Friendly Gaseous Dielectrics through Dimensionless Bond Decomposition and Machine...

Area of Science:

• Materials Science
• Chemical Engineering
• Environmental Science

Background:

• Sulfur hexafluoride (SF₆) is a potent greenhouse gas, driving the urgent need for environmentally friendly alternatives in the high-voltage electrical industry.
• Developing SF₆-free alternatives faces challenges due to conflicting requirements like high dielectric strength, low global warming potential, and safety.

Purpose of the Study:

• To establish a predictive modeling framework for identifying novel gaseous dielectrics with improved environmental and performance characteristics.
• To utilize chemical bond descriptors for efficient virtual screening of potential SF₆ replacements.

Main Methods:

• Developed a machine learning approach using an automatic bond decomposition mechanism on dimensionless SMILES formulas.
• Optimized artificial neural networks to correlate experimental data with theoretical predictions for eight key insulation gas properties.
• Applied the bond-based model to screen 3727 compounds from PubChem for dielectric properties.

Main Results:

• Achieved excellent correlations between experimental and theoretical predictions for insulation gas properties using bond descriptors.
• The bond-based machine learning algorithm demonstrated stability and reliability within defined applicability domains.
• Virtual screening identified a shortlist of promising gaseous dielectric candidates with balanced performance, though none surpassed SF₆ in all aspects.

Conclusions:

• Chemical bonds serve as universal descriptors for predicting gaseous dielectric properties, enabling efficient virtual screening.
• The developed methodology provides guidelines for the rational design of novel dielectric compounds.
• The identified candidates warrant further experimental investigation for practical application as SF₆ replacements.