Computational Design of Low-Volatility Lubricants for Space Using Interpretable Machine Learning

TL;DR

This work tackles the volatility challenge of space lubricants by building interpretable ML models to predict vapor pressure for quick screening. It combines experimental data with MD-derived descriptors in Gaussian Process Regression and Graph Neural Networks to enable accurate, interpretable predictions across diverse chemistries. The results show strong predictive performance, with SHAP and substructure attribution revealing how temperature, density, and functional groups influence volatility, and virtual screening uncovers thousands of candidate molecules beyond traditional space lubricants. The approach provides a generalizable framework for designing space-qualified materials under extreme conditions.

Abstract

The function and lifetime of moving mechanical assemblies (MMAs) in space depend on the properties of lubricants. MMAs that experience high speeds or high cycles require liquid based lubricants due to their ability to reflow to the point of contact. However, only a few liquid-based lubricants have vapor pressures low enough for the vacuum conditions of space, each of which has limitations that add constraints to MMA designs. This work introduces a data-driven machine learning (ML) approach to predicting vapor pressure, enabling virtual screening and discovery of new space-suitable liquid lubricants. The ML models are trained with data from both high-throughput molecular dynamics simulations and experimental databases. The models are designed to prioritize interpretability, enabling the relationships between chemical structure and vapor pressure to be identified. Based on these insights, several candidate molecules are proposed that may have promise for future space lubricant applications in MMAs.

Figures (9)

• Figure 1: Stacked distribution of molecular weights for H/C, H/C/O, H/C/F, C/F, and H/C/O/F elemental compositions.
• Figure 2: Representative (a) MSE vs $\lambda$ hyperparameter and (b) coefficient vs $\lambda$ hyperparameter from LASSO. In (a), the error bars are calculated from the inner 10-fold CV. The green dashed vertical line indicates the $\lambda$ hyperparameter value that resulted in the best (minimum) mean MSE. The dashed blue vertical line is one standard error (1SE) greater than the best MSE. In (b), feature coefficients are shown as a function of the $\lambda$ hyperparameter, with the corresponding number of non-zero coefficients on the top axis.
• Figure 3: Parity plot of log vapor pressure ($P_{vap}$) of the temperature-dependent dataset containing 20 instances per molecule. The symbols and error bars represent the mean and standard deviation as predicted by the ten GPR models in the final evaluation.
• Figure 4: Histogram of all features in the top ten four-feature GPR models. $Temperature$, ${getawayHATSIP2}_{min}$, and ${Density}_{mean}$ appear in nearly all GPR models.
• Figure 5: (a) Representative swarm of SHAP values and (b) mean absolute SHAP values. Each row corresponds to a feature of the model. In (a), color indicates the feature value of the instance, with green colors indicating higher feature values and blue colors indicating lower ones. Negative SHAP values correspond to a decrease in the vapor pressure prediction. In (b), the mean of absolute SHAP values indicate the overall influence of the feature. In both (a) and (b), the rows are descending in order of importance, where the most important features are listed first.