Machine learning assisted quantification of graphitic surfaces exposure to defined environments

Abstract

We show that it is possible to submit the data obtained from physical phenomena as complex as the tip-surface interaction in atomic force microscopy to a specific question of interest and obtain the answer irrespective of the complexity or unknown factors underlying the phenomena. We showcase the power of the method by asking "how many hours has this graphite surface been exposed to ambient conditions?" In order to respond to this question and with the understanding that we have access to as many experimental data points as needed, we proceed to label the experimental data and produce a "library." Then, we submit new data points to the test and request the model contained in this library answers to the question. We show that even with a standard artificial neural network, we obtain enough resolution to distinguish between surfaces exposed for less than 1 h, up to 6 h, and 24 h. This methodology has potential to be extended to any number of questions of interest.