Computational design of low-volatility lubricants for space using interpretable machine learning.

Area of Science:

• Materials Science
• Tribology
• Computational Chemistry

Background:

• Lubricant properties are critical for the function and lifespan of moving mechanical assemblies (MMAs) in space.
• Liquid lubricants are necessary for high-speed/high-cycle MMAs but few possess the low vapor pressure required for space vacuum conditions.
• Existing lubricants for space applications have limitations, constraining MMA designs.

Purpose of the Study:

• To develop a data-driven machine learning (ML) approach for predicting vapor pressure of liquid lubricants.
• To enable virtual screening and discovery of novel space-suitable liquid lubricants.
• To create interpretable ML models that identify structure-volatility relationships.

Main Methods:

• Trained ML models using data from high-throughput molecular dynamics simulations and experimental databases.
• Focused on developing interpretable models to understand chemical structure's influence on vapor pressure.
• Validated models for accuracy in the ultra-low-volatility regime.

Main Results:

• Developed accurate, interpretable ML models for predicting vapor pressure in low-volatility molecules.
• Identified key chemical structural features that govern volatility.
• Successfully screened candidate molecules for potential use as space lubricants.

Conclusions:

• The developed ML framework accurately predicts vapor pressure for space lubricant discovery, outperforming existing methods in extreme low-volatility conditions.
• Interpretable models provide insights into molecular design for reduced volatility.
• The approach facilitates the discovery of promising new lubricants for MMAs in space and can be applied to materials discovery in other harsh environments.