| Abstract Scope |
A variety of deep learning tools and techniques for analyzing and understanding complex systems have been applied in physics. Our application of deep learning in physics is in the characterization and prediction of the state of DoD material systems. We will discuss our efforts, results, and lessons learned to leverage neural networks for early detection and localization of cracks and damage in metallic plates from a guided wave signal. On the experimental side, a novel robotic system automates the generation of large physical machine learning training datasets that cover a wide range of possible cracking and surface damage states in metal plates. A key aspect of our model design is the incorporation of physical principles, namely, symmetries associated with both time and the square arrangement of sensors. An investigation into the combined use of computational and experimental training data, along with an evaluation of various ML methods, will be discussed. |