A physics-informed neural network framework for laminated composite plates under bending

· Interpretable and scalable physics-informed neural networks are developed for laminate plates. · No actual data required, relying on points sampled within the solution domain. · Using energy-based loss function to avoid complex tuning and computation. · Boundary correction functions act as hard constraints to adjust network outputs. The use of machine learning in the field of structural engineering is becoming more common. However, the high dependence of traditional purely data-driven models on the size and quality of the database has posed challenges to the practical application of machine learning. Applying physics-informed machine learning can achieve accurate predictions while reducing the need for extensive input data. This study develops a Physics-Informed Neural Network (PINN) framework to predict the bending behaviors of laminated composite plates. In this framework, the Classical Laminated Plates Theory (CLPT) is incorporated as the physical constraint, and the loss function is formulated based on the energy method. The machine learning prediction results were validated with the CLPT analytical solutions and Finite Element Method (FEM) results, which were sourced from existing literature. These validations demonstrate that the PINN framework achieves satisfactory bending behavior predictions, potentially serving as a promising alternative.