When two objects are rubbed together, they create friction, heat, and their surfaces might be worn away. However, the fundamental mechanisms behind these processes remain poorly understood. While empirical engineering models exist, they are tailored to specific systems and offer limited insight into the underlying physics. Any surface, when looked upon closely enough, is rough and full of asperities. As such, contact between surfaces does not happen through their whole area but through micro-contacts between their asperities. The friction and wear stem from processes at the scales of these micro-contacts. Understanding the mechanics of these micro-contacts is key to understanding tribological processes and such is the aim of this thesis. The investigations are done with Molecular Dynamics (MD) simulations with model potentials.

The focus is on the transition from ductile-to-brittle behavior of micro-contacts. For adhesive wear of homogeneous materials, this transition can be explained by a critical length scale $d^$ that depends on the material properties [AWM16]. However, completely homogeneous materials are rare in nature, and heterogeneities, such as porosity, grain structure, or inclusions, are often found. These heterogeneities create local fluctuations of mechanical properties and can greatly influence the global behavior. Departing from homogeneous material, the transition from ductile to brittle behavior should not be understood as a hard frontier but rather as a gradual switch with a transition zone where both behaviors can be expected. Investigation of a high-entropy composite reveals that the critical length scale $d^$ can be used to predict the bounds of the transition zone. This ductile-to-brittle transition is also observed for wear particle rolling between two surfaces. Using the material properties of the softer surface, $d^$ can be used to predict the transition. When sliding at higher velocities, more heat is generated and the temperature increases. The rise in temperature causes a brittle-to-ductile transition whose lower bound can be predicted by $d^$, but not the upper bound. Finally, increased velocity leads to the fragmentation of the wear debris, a brittle behavior, which the critical length scale can not explain. These results suggest that adding a plastic dissipation term and a kinetic energy term to $d^*$ might be necessary to get a more complete picture of the transition.

The last part of this work addresses the numerical limitations of MD simulations. MD simulations are expensive computationally and can only simulate small systems over short time scales. Two methods are tested to create a computationally efficient surrogate model: model-free Data Driven Computational Mechanics (DDCM) and neural networks. DDCM has been found to have energy conservation issues in dynamics. Remarkably, simple neural networks are able to learn some notion of the complex mechanics behind the wear process, notably the transition from ductile to brittle behavior. The current models suffer from technical issues such as loss of mass and roughness, but yield promising results.

The investigations of this thesis have shown that the concept of $d^$ can be applied farther and broader than initially thought. Brittleness, or ductility, is not only a material parameter; it is also a function of the length scale of the system, and $d^$ appears to dictate which behavior is favored.