Physics-Informed Neural Network for Analyzing Elastic Beam Behavior

Paper i proceeding, 2023

This paper introduces a methodology that combines a physics-based model with observed data for accurately modeling the deflection of an elastic beam in the context of structural health monitoring. The challenges associated with physics-based and data-based methods such as computational time, simplifying assumptions, and seamless integration of sensor data with physics-based models are addressed. The presented method offers a promising approach by effectively fusing data with prior physical knowledge in a cost-effective manner. The proposed methodology is validated through comparisons with analytical and finite element analysis methods for beams with various irregularities such as point loads and supports. The results demonstrate the advantages of integrating sensor data into the model for faster convergence and improved accuracy.