SCNet: A Stress Calculation Neural Network in Digital Photoelastic Technique

In digital photoelastic analysis, accurately calculating stress components is crucial. Typically, this is achieved by solving equilibrium differential equations after determining isochromatic and isobaric parameters. However, reducing noise-induced errors and improving accuracy remains a pressing issue. To address this, a stress calculation method based on deep learning is proposed. The method employs neural networks to learn the entire process from phase unwrapping to stress separation, simplifying the otherwise complex inverse process. In addition, the method not only successfully avoids the cumbersome phase unwrapping process, but also omits the complicated image post-processing steps, which further improves the efficiency and accuracy of the calculation. Through the simulation and experimental verification of the radially compressed disk, we find that the method has excellent noise resistance and can effectively reduce the noise-induced stress calculation errors.