Summary
Force reconstruction in Atomic Force Microscopy (AFM) data represents a significant challenge due to the complex dynamic interactions between the AFM tip and the sample, which are both highly nonlinear and non-smooth. Various methods have been proposed to retrieve this information, yet traditional approaches often fail to accurately represent these non-smooth interactions. Recent advancements demonstrate that machine learning could be an influential factor in extracting detailed information from data. However, conventional machine learning methods tend to smooth the force-distance curves, which misrepresents the tip-sample interactions. This study employs recent advancements in machine learning to preserve and retrieve the non-smooth characteristics of these dynamics. To achieve this, we combine nearest neighbor data-clustering machine learning and Sparse Identification of Nonlinear Dynamics (SINDy) to accurately reconstruct the force-distance relations from synthetic data. This methodology allows for the recovery of the non-smooth equations of motion for cantilever beams in various AFM models, such as DMT and JKR. Furthermore, the algorithm precisely estimates intermolecular distances, demonstrating its effectiveness in predicting the transition between attractive and repulsive forces. The results highlight the potential of this method to significantly improve force reconstruction fidelity, offering a robust tool for advanced nanomechanical studies.
