PI-CNN: A physics-informed machine learning model for the structural analysis of deep beams

Abstract

Deep beams are critical structural elements widely used in various engineering applications, including bridges, high-rise buildings, and industrial structures. Their complex behaviour under loading such as the large shear forces and the nonlinear stress distribution present a notable challenge. This complexity causes the analysis and design process more difficult. Traditional methods such as the finite element method (FEM) have been widely used in the analysis of deep beams. However, these methods often require extensive computational resources and/or significant human effort. This study presents the development and application of a novel physics-informed convolutional neural network (PI-CNN) model for the structural analysis of deep beams. PI-CNN was developed by incorporating existing knowledge about deep beams in the form of constative laws into the training process. This ensures that PI-CNN produces physically consistent predictions. The performance of the PI-CNN was evaluated against cases from the literature, and parametric studies were used on the different parameters affecting its performance. This study found that the Incorporation of physics knowledge has enhanced the predictive capabilities of PI-CNN and reduced its dependency on large datasets. Furthermore, PI-CNN was able to accurately predict the structural response of deep beams faster than traditional methods while maintaining the same level of accuracy.