Temporal Forecasting of High-Rate Dynamic Using Physics-Informed Machine Learning and Hardware-Software Co-Design

Due to aging, fatigue, corrosion, and even natural disasters; the health of the structure is prone to degradation throughout its service life. The explosively-fast growing efforts on Structural health monitoring (SHM) always try to exploit different aspects of the automation of damage detection, localization, and prognosis tasks. One of the main challenges is the hardware and software co-design to implement the model in real-life situations. On the other hand, the fast-advancing artificial intelligence draws the researchers' attention to adopt different data-driven approaches in this field. This brings other challenges like domain-specific model adaptation, data bias, data scarcity, model validation by physics rules, etc. To find a common ground between computational model approaches and data-driven approaches; new interest is growing in physics-informed machine learning (ML). To solve the above issues, this dissertation aims to achieve hardware-software design for SHM in two aspects: (1) data-driven ML and (2) physics-informed ML. The design approaches also want to adopt both co-design and sequential paradigms based on domain-specific problems. As a byproduct of this dissertation, this research also wants to contribute to the data generation process. The research community will be benefited from the generated data as well as the methodologies involved to generate the data because it will give them the flexibility to generate more data based on their needs.