Bridging finite element and deep learning: High-resolution stress distribution prediction in structural components

Abstract Finite-element analysis (FEA) for structures has been broadly used to conduct stress of various civil and mechanical engineering structures. Conventional methods, such as FEA, provide high fidelity results but require the solution large linear systems that can be computationally intensive. Instead, Deep Learning (DL) techniques generate significantly faster than conventional run-time analysis. This prove extremely valuable in real-time structural assessment applications. Our proposed method uses deep neural networks form convolutional (CNN) bypass FEA predict high-resolution distributions on loaded steel plates with variable loading boundary conditions. The CNN was designed trained use geometry, conditions, load input contours. technique’s performance compared finite-element simulations using a partial differential equation (PDE) solver. DL model mean absolute error 0.9% an peak 0.46% von Mises distribution. study shows feasibility potential