Model reduction techniques of soft tissue simulations.
Investigators: Prof. Haasdonk, Prof. Hunter, A/Prof. Besier
We will develop projection-based reduction techniques based on proper orthogonal decomposition (POD) and greedy approximation techniques with local basis generation procedures to satisfy the real-time constraint. In particular, we aim at in-situ control settings, where no cloud-computing service for the simulation can be requested, as is the case in robot control. Instead of only using state-information for basis construction, we include quantities that are more suitable for the control setting, e.g. low-rank factors of corresponding Riccati Equations. We will define and work with a hierarchy of muscle models and their respective model reduction. It can be foreseen, that more fundamental statements can be provided for the simple models, while the more complex models will mainly be made to work without analysis. Three levels of complexity consist of a linearized pure mechanical muscle model, a full nonlinear mechanical muscle model and a nonlinear model involving the electrophysiology. The resulting reduced simulation models will enable rapid, i.e. real-time, simulation for model predictive control (MPC) strategies as aimed in Focus Area B. The project will bridge between soft tissue modelling and control application settings.