Understanding Nanoscale Topology-Adhesion Relationships Via Support Vector Regression

Adhesion is an important property related to interfacial failure and is crucial in nanofabrication, nanodevices, and biomedicine. Nanoscale roughness significantly reduces the adhesion predicted by classical mechanical models. To quantify he adhesion, an intrinsic traction-separation relation is required. Previous investigations establish traction separation laws with pre-assumed forms and use statistical topological parameters to summarize details of roughness. Here, support vector regression (SVR) is performed on nanoscale topology-adhesion correlated atomic force microscopy (AFM) data to establish a data-driven traction-separation relation. Instead of using statistical parameters, full details of the roughness are used as input features. Over 200 000 topological data sets are analyzed and an accurate prediction of adhesion with R-2 more than 0.98 is achieved. More importantly, a traction-separation relation is derived from an SVR-based machine learning process, followed by validation via finite element analysis. This work presents an AFM-SVR integrated approach to characterize nanoscale adhesion that can be generalized for different materials and interfaces.

Automated summary

This study used support vector regression on atomic force microscopy data to establish a data-driven traction-separation relation for nanoscale adhesion. By incorporating full details of roughness as input features, over 200,000 topological data sets were analyzed, achieving an accurate prediction of adhesion with R-2 above 0.98. The derived traction-separation relation was validated through finite element analysis, presenting an integrated approach for characterizing nanoscale adhesion.