CS Forum Talk: Machine Learning for Lubricant Optimisation

Speaker: Dr. Filippo Federici Canova, Aalto University.

When: Monday 27/04/205 at 13:15-14:00

Where: Lecture hall T2, CS Building

Host: Petteri Kaski

Abstract:

Theoretical models are crucial to correctly interpret experiments,
especially when complex systems such as molecules, clusters, and self-assembled monolayers are considered. While isolated molecules can be modelled in great detail using quantum mechanical methods, the high
computational cost makes it impossible to simulate their dynamics.  By
using these high-accuracy results, however, one can model interatomic
interactions with simpler approximations, and study nucleation and
growth dynamics of large systems for a longer time using classical
molecular dynamics (MD). Unfortunately, to represent the interatomic
interactions present in such a complex system, large numbers of
parameters need to be optimised, and this is often a limiting factor
for multi-scale modelling.

In this work, we focused on multi-scale modelling of
1,4-bis(4'-cyanophenyl)-2,5-bis-(decycloxy)benzene (CDB) dyamics on
KCl(001). First we simulated the properties of a CDB molecule on the
surface using quantum density functional theory.  We used Genetic
Algorithms (GA) to obtain a set of classical interaction parameters
that best mimic the interactions obtained from DFT. The classical
model is then used for MD simulations of larger systems, to study how
the CDB molecules nucleate on the surface, and how films are grown.
Ultimately, we are able to use this information and efficiently
generate virtual AFM (vAFM) images using a dipole tip model, and
directly compare the simulations to experimental data.

About the speaker:

Dr.  Filippo Federici Canova is newly appointed Aalto Science Fellow,
in the School of Science.  After obtaining a PhD in computational
physics from Tampere University of Technology, he moved to Tohoku
University (Japan) where he studied the frictional dynamics of nano
confined ionic liquids.  Currently he is working on the development of
molecular and fluid mixture descriptors in order to machine-learn the
relationships between atomic structure of lubricants and their
performance.