Constraining Gaussian processes for physics-informed acoustic emission mapping

The automated localisation of damage in structures is a challenging but critical ingredient the path towards predictive or condition-based maintenance high value structures. use acoustic emission time arrival mapping promising approach to this challenge, severely hindered by need collect dense set artificial measurements across structure, resulting lengthy and often impractical data acquisition process. In paper, we consider physics-informed Gaussian processes for learning these maps alleviate problem. approach, process constrained physical domain such that information relating geometry boundary conditions structure are embedded directly into process, returning model guarantees any predictions made satisfy physically-consistent behaviour at boundary. A number scenarios arise when training measurement limited, including where sparse, also limited coverage over interest. Using complex plate-like as an experimental case study, show our significantly reduces burden collection, it seen incorporation condition knowledge improves accuracy observations reduced, particularly not available all parts structure.