Application of deep learning method for time reduction and precision improvement in displacement measurement during in-situ SEM tensile test

Abstract

In this study, a novel material deformation measurement method is suggested to evaluate the mechanical properties of thin films. It is important to identify the mechanical properties of thin films in order to do a reliable design of these products. One of the methods to get the mechanical properties of thin films is to acquire images during an in-situ SEM (Scanning Electron Microscope) tensile test and analyze them. However, SEM image acquisition is mostly affected by the time varying motion of pixel positions in the consecutive images, a phenomenon called drift and distortion. In this study, some images of low resolution during in-situ SEM tensile test of the thin film are acquired in short time and restored to high resolution images using the deep learning model called Super Resolution Residual Network (SRResNet) to calculate the deformation of thin films. The proposed method increased the precision of measurement by 5 times, improved the accuracy of measurement by 20%, and reduced the measurement time by 19 times.