Physics-Informed Neural Networks for Predicting Internal Forces and Deformations of Structural Frames in a Single-Span Agricultural Greenhouse

The increasing frequency of climate change and extreme weather events has highlighted the urgency of structural safety in agricultural facilities. Accurate predictions of internal forces and deformations in frame members are essential for ensuring structural integrity. However, the conventional finite element method (FEM) analysis is computationally expensive and requires a complete reanalysis when structural configuration or loading conditions are modified. This study proposes a novel approach using physics-informed neural networks (PINNs) based on FEM principles to predict internal forces and deformations in single-span greenhouse frames. The proposed model integrates equilibrium equations, strain-displacement relationships, and material constitutive laws into the loss function. The accuracy of this approach was evaluated using a single-span agricultural greenhouse subjected to snow loads. The results show that PINN achieves remarkable accuracy, with maximum errors of 1.24% and deformation errors of 0.1 mm, showing excellent agreement with FEM analysis results. These findings validate the effectiveness of the proposed FEM-based PINN for analyzing complex structural systems, presenting a significant methodological advancement applicable to structural analyses of agricultural facilities.