Typically, the engineers who perform inspections in structures are not the same as those that design them. Consequently, the model-based approaches are limited for Structural Health Monitoring (SHM) due to the high cost associated with obtaining an adequate model. On the other hand, it is fundamental to have a model to reach a higher hierarchy in SHM procedures. Thus, this Ph.D. project intends to propose a practical strategy for a helpful model for SHM applications utilizing a digital twin viewpoint. The key idea is to involve data-driven models combined with a Physics-Informed Neural Network (PINN) as a deep learning framework. As an industrial example, we aim to investigate the direct application of this procedure for remote monitoring of structures with bolted joints with different geometries and assemblies. A first FE model can be obtained using images by photogrammetry of the existing system to be monitored. A PINNs can combine this model to solve a forward and inverse problem to get information about the contact pressure, torque load in the bolted joints, and hysteretical dissipation by supervised learning. These effects are unusually modeled using the traditional FE model, but we could incorporate this information in a data-driven model for prediction purposes. This project presents this study's motivation, goals, work plan, and context, engaging the scientific and technological contribution.
News published in Agência FAPESP Newsletter about the scholarship: