LeCoPro - Learning control for production machines

From 01-09-2009 to 31-08-2013

Description

Up till now, only a few implementations of learning controllers have been realized for mechatronic systems. Moreover, only model-based techniques were adapted and only implementations were done on small, relatively simple mechatronic subsystems. In a first track of the LeCoPro-project, the potential of different learning control algorithms for more complex mechatronic (sub)systems will be investigated. The advantages and drawbacks of model-based learning control methods (Iterative Learning Control – ILC; Model Predictive Control - MPC), suitable for the control of systems of limited complexity, and non-model-based methods (Machine Learning), suitable for the control of more complex systems, will be compared. State-of-the-art techniques in these research domains will be extended and adapted to address the specific features of mechatronic systems (fast non-linear dynamics, uncertainty of system parameters, stringent stability and robustness requirements, etc.).

Many production machines consist of strongly interacting subsystems, each responsible for a different mechatronic subtask. Most of these systems are controlled in a decentralized way, because of flexibility, modularity and scalability reasons. To address such control configurations, in a second track of the project global learning controllers will be developed for machines with interacting subsystems that are controlled in a decentralized manner. Nowadays controller parameters of subsystems in such production machines are tuned independently. This often results in an unsatisfactory behaviour of the global system, since this approach does not take the interaction between the subsystems into account. The objective of the LeCoPro-project in this track is to develop global learning controllers, which learn the optimal coordination of the decentralised control actions for each individual subsystem such that a global optimum is obtained. Similar to the track on subsystem controllers, also in this track model-based as well as non-model-based (global) learning control techniques will be developed.

Since the practical applicability of all developed methodologies is a prerequisite of the LeCoPro-project, a significant part of the planned work is dedicated to practical validation of these methodologies. At least one generic test benches will be designed and built within the scope of the project, on which the practical performance of the different learning control techniques will be evaluated. On this test bench, the effectiveness of different learning controllers with respect to improvement of the system productivity in comparison with the actual situation and efficiency of implementation will be extensively demonstrated. This will not only allow a systematic comparison of different techniques to be carried out, but moreover provide a direct evidence for the Flemish machine-building companies of the industrial potential of the developed learning strategies.

Team

• Moritz Diehl, Co-promoter

Financing

Funding: IWT - Agentschap voor Innovatie door Wetenschap en Technologie 

Program/Grant Type: IWT SBO - IWT Projects Strategic Basic Research