AI predicts how graphene keeps machines cool

Graphene-enhanced grease is at the forefront of a new study utilizing a hybrid machine learning framework to predict its thermal behavior, wear characteristics, and lubrication regime transitions. The research, published in Results in Engineering by Ethan Stefan-Henningsen and Amirkianoosh Kiani, addresses a core issue in tribology—the science of friction, wear, and lubrication. The framework predicts how graphene-infused grease heats up and wears down, providing insights that could influence lubricant design for electric vehicles and wind turbines.

The study's experimental setup involved a block-on-ring sliding configuration to test greases with varying graphene concentrations. These greases were formulated by dispersing graphene nanoplatelets in a mineral oil-lithium grease matrix. The results showed that a 0.5 percent graphene concentration reduced peak temperatures by about 30 percent compared to control grease, due to graphene's thermal conductivity and lamellar structure.

The machine learning framework integrates temperature histories as a shared state variable, using nonparametric temperature prediction and ridge-regularized regression models to estimate wear and friction. The study demonstrated a significant reduction in wear scar area and depth with optimal graphene concentrations, achieving high validation metrics for temperature, wear, and friction predictions.

The research highlights graphene's role in maintaining lubrication even as the hydrodynamic film decays. Graphene nanoplatelets form tribofilms that cushion asperity interactions, reducing frictional power loss and enhancing wear resistance. The framework's energy analysis ties improved thermal control directly to reduced wear rates, offering a comprehensive tool for lubricant formulation.

Despite its promising results, the framework's generalization is limited by the specific conditions and graphene concentrations tested. Future research aims to expand testing across different conditions and grease chemistries. The study underscores the potential of combining machine learning with classical lubrication physics to create physically interpretable predictive models.