Executive Summary : | Physics informed machine learning, has transformed the data driven methods in engineering analysis to a greater extent. The inherent ability to use the information of known physics of the system, improves the accuracy of these models. Among various structural assessment methods, the modal domain analysis has been used by many researchers and practitioners to understand and determine dynamic behavior of structures under different conditions. The analytical solutions of differential equations for longitudinal vibration of beams with corresponding boundary conditions are usually derived by linear combinations of mode shapes based on a principle known as mode superposition technique. Higher order mode shapes or the derivatives of the mode shapes have also been better indicators of presence of damage or cracks in the system (Whalen, 2008). Recently ModalPINN (Raynaud et al., 2022), an extension of regular Physics Informed Neural Networks (PINN), has proved that the use of Modal domain information in the network architecture improves the ac- curacy of PINNs in systems having flow with periodic patterns. Thus a well trained PINN with modal information of structures can be used as a surrogate for analytical solution of the vibrating beam with corresponding loading and boundary conditions. Also, PINN by its nature, predicts the output as a continuous field, enabling more accurate localization of damage in the system. This project aims to study the use of modal information of structures to build efficient physics informed ML/DL models for efficient real time smart structural health monitoring. Due to the easy deployment capabilities of such a trained PINN model, It can be used as a hybrid Digital twin setup running parallel simulations of real time structures enabling continuous structural health monitoring. |