In many mechatronic applications optimal control, i.e. optimizing the system input directly subject to input, state and output constraints, is preferred over feedforward control in enhancing the tracking or positioning performance. Computing the fastest point-to-point motion trajectory given actuator bounds for instance amounts to an optimal control problem, as well as computing the most energy efficient way to execute a give task within a prefixed time. Unfortunately, in many industrial applications the value of optimal control is severely compromised by uncertainty in the system model or its environment, as this causes the computed input to be far from optimal and generally not even acceptable or feasible.  Two classes of approaches exist to deal with this uncertainty: (i) model-based approaches that adapt the model parameters based on past input-output measurements; and (ii) signal-based approaches that directly estimate the disturbance signals. In previous research [1-2] we developed an optimization-based learning strategy that allows combining both approaches. In this research project you will apply and validate this learning strategy on industrial mechatronic test cases. In addition you will further extend the strategy to meet the learning speed and robustness requirements of these applications.

[1] Volckaert, M., Diehl, M., Swevers, J. (2013). Generalizationof norm optimal ILC for nonlinear systems with constraints. MechanicalSystems and Signal Processing, 39 (1-2), 280-296.

http://www.sciencedirect.com/science/article/pii/S088832701300109X

[2] Volckaert, M. (2012). NonlinearIterative Learning Control Through Dynamic Optimization, PhD Thesis KU Leuven, Department Mechanical Engineering.

https://lirias.kuleuven.be/handle/123456789/359680