Using Artificial Neural Networks to Refine and Analyze the Results of Atomic Force Microscopy of Dispersed Filled Elastomers (Contact and Semi-Contact Modes)

The paper proposes two new applications of artificial neural networks for decoding and analyzing results of the surface of elastomer nanocomposite scanning by means of atomic force microscopy (contact and semi-contact modes). The main advantage of this approach is that it enables studying local mechanical properties of the material not only on the sample surface but also in the near-surface layer. In case of the contact operation mode of the atomic force microscope, the artificial neural network was created and "trained" using the database with the pressing models of the probe of the atomic force microscope into a nonlinear hyperelastic medium with rigid spherical inclusions ("contact neural network"). This database contained the results of calculations of indentation curves for various values of the filler particle sizes and their localization in the near-surface layer of the material (depth and horizontal distance from the tip of the probe). The use of such a neural network allows speeding up the construction of indentation curves by several orders of magnitude compared to conventional methods based on the numerical solution of the corresponding boundary value problems for each specific case. As a result, computing time is also significantly reduced, therefore, with an already constructed and "trained" neural network, powerful and high-speed computers are not needed. In the semi-contact mode, the neural network was built using real scans of the relief and phase shift of the probe cantilever oscillations obtained on samples made of a dispersed filled elastomer ("semi-contact neural network"). It was shown that with its help it is possible to quite accurately predict what the results of the semi-contact scanning will look like with an increase in the maximum value of probe indentation (i.e. with a deeper study of the near-surface layer). Further, these modified scans are supposed to be used as a basis for analyzing near-surface layers using a contact neural network.