Neural Network Supported Microscale In Situ Deformation Tracking: A Comparative Study of Testing Geometries

Abstract

Micro- and nanomechanical testing techniques have become an integral part of today’s materials research portfolio. Contrary to well-studied and majorly standardized nanoindentation testing, in situ testing of various geometries, such as pillar compression, dog bone tension, or cantilever bending, remains rather unique given differences in experimental equipment and sample processing route. The quantification of such experiments is oftentimes limited to load-displacement data, while the gathered in situ images are considered a qualitative information channel only. However, by utilizing modern computer-aided support in the form of the recently developed Segment Anything Model (SAM), quantitative mechanical information from images can be evaluated in a high-throughput manner and adds to the data fidelity and accuracy of every individual experiment. In the present work, we showcase image-assisted mechanical evaluation of compression, tension and bending experiments on micron-scaled resin specimens, produced via two-photon lithography. The present framework allows for a determination of an accurate sample strain, which further enables determination of quantities such as the elastic modulus, Poisson’s ratio or viscoelastic relaxation after fracture.