Active learning for nonparametric multiscale modeling of boundary lubrication

Area of Science:

• Multiscale physics
• Tribology
• Computational materials science

Background:

• Lubricated friction is a complex multiscale phenomenon.
• Existing models struggle under extreme conditions due to reliance on semiempirical constitutive relations.
• Molecular processes significantly influence macroscopic lubrication behavior.

Purpose of the Study:

• To develop a novel simulation framework for boundary lubrication.
• To overcome limitations of traditional models under extreme conditions.
• To accurately predict interfacial stresses without fixed constitutive laws.

Main Methods:

• Coupling of molecular dynamics and continuum models.
• Utilizing Gaussian process regression as surrogate models for interfacial stress prediction.
• Employing an active learning algorithm for adaptive model refinement.

Main Results:

• Demonstrated a simulation framework that seamlessly integrates molecular and continuum approaches.
• Successfully predicted interfacial shear and normal stress, adapting to complex scenarios like layering transitions.
• Validated the approach for nanoscale fluid flow over rough and heterogeneous surfaces.

Conclusions:

• The developed framework offers accurate boundary lubrication simulations beyond the reach of traditional methods.
• This approach enables reliable predictions at experimental length and timescales.
• It paves the way for advancements in understanding and engineering lubrication systems.