Accuracy improvement of demodulating the stress field with StressUnet in photoelasticity

Abstract

Evaluating the stress field based on photoelasticity is of vital significance in engineering fields. To achieve the goal of efficiently demodulating stress distribution and to overcome the limitations of conventional methods, it is essential to develop a deep learning method to simplify and accelerate the process of image acquisition and processing. A framework is proposed to enhance prediction accuracy. By adopting Resnet as the backbone, applying U-Net architecture, and adding a physical constraint module, our model recovers the stress field with higher structural similarity. Under different conditions, our model performs robustly despite complicated geometry and a large stress range. The results prove the universality and effectiveness of our model and offer an opportunity for instant stress detection.