A Challenge at the Intersection of Tribology and AI

Friction plays a crucial role in many technological systems, directly impacting their energy efficiency, durability, and performance. Despite centuries of research, precisely controlling friction remains a major challenge due to the multiscale nature of tribological phenomena and their sensitivity to environmental conditions. Recent work in the fields of tribotronics and metainterfaces has paved the way for new approaches to modify and control the frictional properties (slip/adhesion) of an interface. Currently, these approaches are limited to static interface configurations (i.e., the tribological properties of the interface are fixed over time). Nevertheless, smart materials offer promising prospects for overcoming this limitation and enabling dynamic control of the interface, thereby allowing it to adapt in real time to changes in its loading conditions. This thesis project thus proposes a conceptual breakthrough that will involve designing adaptive tribological surfaces capable of dynamically adjusting their frictional properties in real time, using artificial intelligence and, in particular, reinforcement learning.

Thesis Objectives

1. Development of an innovative digital twin:

Establishment of a single-scale modeling framework inspired by meta-interfaces. This study will draw on various contact models and the theory of Greenwood and Williamson (1966) to predict the tribological behavior of a model interface based on its topography (typically, the statistical distribution of the heights of the surface asperities that compose it can be used). This digital twin will generate the virtual environment necessary for training reinforcement learning algorithms.

2. Implementation of advanced reinforcement learning algorithms:

Development of optimal control strategies for surface topography to adapt to “simple” non-stationary environments. In particular, this will involve exploring certain RL methods (model-based RL, meta-learning), deep neural network architectures adapted to physical systems, and techniques for managing the exploration-exploitation trade-off.

3.    Robustness in the Face of Complex Behaviors:
Development of advanced adaptive control strategies for surface topography capable of maintaining optimal tribological performance to address more complex scenarios involving surface heterogeneity and/or loading conditions. In particular, this will involve developing feedback loops to ensure robustness in the face of disturbances. 4. Transfer to a physical prototype: Development of sim-to-real methods to adapt the strategies developed in simulation to the real world, taking into account uncertainties and unmodeled dynamics.