Intelligent Parameter Identification for Dynamical Systems in Complex Environments

ABSTRACT

Accurate parameter identification is essential for reliable modelling and control of dynamical systems operating in complex environments, such as spacecraft, robots, and unmanned aerial vehicles. In practice, parameter estimation is challenged by non-Gaussian measurement noise, unknown disturbances, and time-varying parameter changes caused by payload variation or structural reconfiguration. This seminar presents a series of intelligent parameter identification methods based on deep learning and physics-informed modelling. For time-invariant systems, a direct parameter mapping approach using deep neural networks is developed for accurate parameter identification, further enhanced through ensemble learning and physical constraints. For time-varying systems, an LSTM-based framework is introduced to track rapid parameter changes. In addition, a meta-learning-based method is presented for external disturbance force identification of unmanned aerial vehicles under time-varying wind conditions. Numerical results demonstrate strong robustness, high accuracy, and improved generalization compared with traditional identification methods.

Regenerative Cooling of Disk-Shaped Pressure Gain Combustor with Nitrous Oxide

ABSTRACT

Although rotating detonation combustors (RDCs) offer potential for improved efficiency, their extreme thermal environment remains a primary barrier to practical application. This seminar presents a multi-stage experimental study investigating regenerative cooling as a viable thermal management strategy. We first characterise the heat transfer properties using a water-cooled RDC. Using the empirical data, we develop a predictive heat transfer model tailored to this RDC. Finally, we validate the model through a hot-fire test with nitrous oxide in a partial regenerative cooling setup. The results demonstrate the feasibility of nitrous oxide for regenerative cooling and regenerative cooling as a thermal management strategy toward sustained, flight-weight RDC operation.