engleski

Multivariate AI-based predictive model of nanoscale friction

As a major source of uncertainties in micro- and nanopositioning devices, friction, characterised by complex stochastic phenomena, is one of the major challenges in developing reliable predictive models od systems’ behaviour. Based on recently performed lateral force microscopy experimental measurements of thin films’ nanoscale friction, fundamental frictional mechanisms in atomic-scale single- asperity contacts are investigated in this work vs. the concurrent influence of multivariate process parameters. The hence proposed nanoscale friction models are developed using innovative artificial intelligence-based methods. In fact, although in a previous study the employment of conventional (black box) machine learning methods provided rather good predictive performances, the intrinsic nature of these models prevents their usage in most practical applications. The novel methodology proposed in this work allows, in turn, attaining an extremely simple mathematical formulation (i.e., a white box model) providing an immediate insight and a unique scientific perspective into the multidimensional dependence of nanoscale friction on the studied variable influencing parameters. What is more, this artificial intelligence-based approach allows achieving a high predictive performance with R2 values in the range of 0.75, while the simplicity of the obtained expressions makes future studies and possible practical applications (e.g. in the corresponding control algorithms) rather straightforward.

nanoscale friction ; artificial intelligence ; predictive white box model

nije evidentirano