Project outline
Large numerical simulations dominate marine and aerospace industries when performing structural analysis on engineering platforms. These models are computationally intensive, and model updating often requires complete re-simulation. To enable practical real-time sensing applications for noise and vibration, sufficient sensor coverage is needed to acquire the data which characterises the dynamical behaviour for quantifying both structure-borne noise and radiated air-borne noise.
However, the dynamics and scale of these structures present challenges for physical based measurements, such as satisfying a sufficient spatial resolution for system-wide analysis, real world time constraints and complex multiphysics systems that cannot be easily quantified.
As part of the Sustainable Sound Futures CDT, this proposed research aims to improve the resolution of measurement datasets by using virtual sensors to improve the spatial resolution, improving the ability for future real-time sensing applications to better detect harmful noise and vibration to both people and the wider environment.
Utilising an energy-based system model framework, this research will seek to ascertain the ideal construction of an appropriate Physics Informed Machine Learning (PIML) based virtual sensor, leveraging a data driven approach with physics domain knowledge to compare against traditional virtual sensing approaches.
The predictions will be validated through laboratory experimentation in a hybrid modelling approach and compared to current methods such as Kalman filters and numerical based hybrid modelling techniques for experimental data sets.
Project Partner
This project is hosted at the University of Salford and is supported by Qinetiq.
Student
Oscar Carter