Ultrasonic Guided Waves (UGWs) are particular elastic waves that propagate in media with specific geometries, typical of many aerospace, civil and biomedical structures. Due to their properties of wide range coverage and interaction with the entire cross-section, UGWs are a very advantageous a Non-Destructive Evaluation (NDE) method technique that can be used to inspect defects in solid material. There are various methods to simulate UGWs such as the Global-Local (GL) method approach which can be used to predict scattering in complex structures. GL Most methods, though, rely on finite difference or finite elements schemes which can be computationally expensive, especially to simulate UGWs propagation over a broad frequency range and interaction with complex damages. In this research project we explore a new method to model UGWs in solid media with Physics Informed Neural Networks (PINN). PINNs are a new type of machine learning method which solve Partial Differential Equations (PDEs) through stochastic processes. The PINN model can be used as a continuous time model which creates a data-efficient spatio-temporal approximator for the UGW propagation in solid media. The fundamental framework and challenges of its implementation are presented, as well as preliminary results/simulation. In the future, PINNs UGWs can be used to efficiently simulate inspections in complex structures and generate data for Structural Health Monitoring and prognosis.
