Static analysis of two-side supported 2-ply laminated glass panes through physics-informed neural networks

This paper utilizes an emerging machine learning technique, namely physics-informed neural networks (PINNs), to analyze the static mechanical behavior of two-side supported 2-ply laminated glass panes. The PINN model utilizes prior physical information in the form of partial differential equations during model training, eliminating the need for tremendous amounts of data required by traditional data-driven machine learning methods. The governing equations for the structural behavior of two-side supported 2-ply laminated glass panes are derived, based on which a PINN framework encompassing the neural network architecture and the loss functions is constructed. The training stability is evidently improved after the linearly inducing process, with the relative error maintained within a reasonable range. Extensive validation examples and case studies are provided to demonstrate the accuracy and effectiveness of the proposed method. Composite action analysis results reveal that below 70 % in the design is a conservative estimate for thinner laminated glass. Furthermore, the PINN method achieves exceptional computational efficiency, reducing computational times from six hours to just a few minutes for the uncertainty evaluation of laminated glass, indicating its potential in engineering practice.