Project at EPFL - Iterative Learning Control of a Linear Motor
During the last 6 months I have carried out my final project at EPFL in the automatic control laboratory. The title of the project was: Iterative learning control of a linera motor.
Iterative Learning Control deals with the set of repetitive processes and the notion that performance can be increased using the information from previous iterations to improve it. As opposed to traditional controllers that yield the same error each trial ILC aims to learn from previous iterations to reduce the error from one iteration to the next. It can be shown that the error converges despite plant modelling uncertainty and repeating disturbances
A common industrial application of linear motors is in production lines where they carry out repetitive tasks additionally as precision might be required it makes them ideal candidates for ILC.
A logical choice seems to be to combine ILC with optimisation techniques. However, most of the existing optimal algorithms are computationally complex requiring large calculation times between trials. Thus it is important to find optimal algorithms that keep the rapid convergence properties but at the same time are simple to implement and do not require extensive calculation. Owens and Feng introduced a parameterisation of an optimal ILC algorithm that has monotonic convergence and achieves zero tracking error if a positivity condition is satisfied.
Owens, Hatonen and Feng proposed a more general higher order version of the previous algorithm. The main contribution of this new version is the inclusion of “basis functions”, that allow convergence to zero error even if the plant is not positive. Inverse model type is an intuitive approach for ILC since in the ideal case of perfect knowledge of the system, using its inverse as a learning operator would lead to perfect tracking in one iteration. When the system is not perfectly modelled, as is always the case in practise, rapid convergence can still be achieved. Convergence conditions tacking into account model uncertainty have been established.
Combining inverse model-based and optimality, Harte, Hatonen and Owens presented an inverse type parameter optimal ILC, where similar convergence conditions are derived.
Here there is a picture of the entrance of the mechanical engineering section: