AFM Indentation and Machine Learning in Analysis of Filler Distribution in Rubber

ABSTRACT Filled rubber surface was studied by AFM indentation and the depth of filler inclusions beneath the polymeric surface was determined using machine learning (ML). Obtained information is essential in the analysis of the microstructure and distribution of the filler. Experimental data were processed using the finite element method (FEM) and two ML models. FEM was employed to generate simulated indentation curves for a soft hyperelastic layer on a rigid substrate (with variations of layer thickness and indenter radius). Finite element simulation curves were fitted to an analytical function characterized by two parameters. These two constants and the radius of the indenter were used to train the first ML model with the layer thickness as the target variable. Experimental force curves were categorized into two types: “good” and “poor.” The former are accurately approximated by the FEM simulation; for these, the first ML model was applied to estimate the local thickness of the polymer layer covering the inclusion. The “poor” curves have distortions and were processed using the second ML model trained on the results from “good” curves (input parameters: indentation depth and the work of adhesion).