(161j) Accelerating Large-Scale Screening of Tribological Properties and Chemistries

Gilmer, J., Vanderbilt University
Summers, A. Z., Vanderbilt University
Iacovella, C. R., Vanderbilt University
Cummings, P. T., Vanderbilt University
McCabe, C., Vanderbilt University
The tribological properties of monolayer films have been shown to be dependent on the chemistry of the films [1]. This provides a near infinite design space as a means of tuning performance of films. To enable more efficient evaluation of candidate film chemistries, a machine learning model was developed in recent work [2] based on the random forest algorithm. Using chemical descriptors determined from RDKit [3] as the fitting parameters, this model was developed by considering the screening of 420 distinct simulations using the Molecular Simulation and Design Framework (MoSDeF) [4]. Rather than developing a single model, five distinct random forest models were developed using different subsets of the data and performing standard testing; this allowed models that provided poor fitting to be discarded and bounds to be put on predicted values by calculating the mean and standard deviation of the predictions from each of the distinct models. Additionally, adversarial testing was performed on the test data in each model; the differences between testing the entire set of chemical descriptors, subsets of the descriptors that showed the greatest importance, and randomly varying these descriptors, were examined to determine appropriate heuristics for interrogating the models’ possible biases. Finally, we performed a series of simulations using chemistries not considered in the original screening to further validate the models. This evaluation suite is being developed as a workflow for the MoSDeF library, allowing best practices for the evaluation of monolayer tribology to be formalized and ultimately disseminated to the community. The overall goal of the work is to allow the a priori identification of films that might possess favorable tribological characteristics, reducing the computational expense required to evaluate the vast landscape of parameters.

  1. Brewer, N. J.; Beake, B. D.; Leggett, G. J. Friction Force Microscopy of Self-Assembled Monolayers: Influence of Adsorbate Alkyl Chain Length, Terminal Group Chemistry, and Scan Velocity. Langmuir 2001,17(6), 1970–1974.
  2. Summers, A.Z.; Gilmer, J.; Iacovella, C.R.; Cummings, P.T.; McCabe, C. Examining Chemistry-Property Relationships in Lubricating Monolayer Films through Molecular Dynamics Screening, In Preparation
  3. https://www.rdkit.org/
  4. http://github.com/mosdef-hub