Predicting static friction in a molecular dynamic system using machine learning

Friction is immensely important, whether the concern is moving or still objects, energy loss, heat transfer or wear and tear. Despite its obvious importance friction is not fully understood, due to small scale behaviours both effecting and being effected by large scale behaviours, making a scalable friction theory difficult to define. In this thesis we wish to study these small scale interactions at the junctions where objects meet, and how the placement of junctions relative to each other can tell us about their interactions. We study a molecular dynamical system of 3c-SiC crystals in the form of a rectangular block with dodecahedron asperities underneath , the shape of which is picked due to it being close to the equilibrium shape of SiC nanoparticles. This top plate meets another stationary bulk crystal plate on which it slides with the asperities forming the junctions. The choice of system is made as to reduce any effects that might conceal the effects of junction-junction interactions. Preliminary studies are done varying physical aspects of the system such as the sizes of various parts of the system, normal force, velocity of the top plate in the push phase etc, to find a physically and computationally feasible system before proceeding with large scale simulations varying the placement of the aforementioned asperities. The placement of the asperities form the input into a neural network predicting the maximum static friction and the slope of the load curves as experienced by the top plate. We found that the system we studied did not show an effect of the asperity configuration on the friction force. A large grid search over hyperparameters of the networks was done to thoroughly examine any network which might be able to learn from the simulations data. These methods were also applied to a dataset of random value to create a baseline to compare our actual data to. The networks were able to confirm the insensitivity of the friction to variations in configurations.