This project aims at needs that are experienced by a wide range of industries when they attempt to (i) find good controllers, and (ii) ensure these controllers remain performant under variable conditions and throughout their entire lifetime.
For the first need, optimal solutions have to be found in a variety of complex situations (non-linear systems, constraints, ...). This should further be done as efficiently as possible, keeping the required human tuning effort low, as well as the effort spent building accurate system models (cheaper and faster deployment). The same targets hold for the second need, in that we want to have optimal performance under all conditions and not end up with controllers that are acceptable under all conditions but often yield poor system performance. Like before, the less human effort for tuning or modelling that is required, the better the solution.
While those needs are relevant for a single system they become even more relevant for fleets since in that case the effort increases linearly with the total number of systems. All controllers must then be tuned such that each system in the fleet performs satisfactorily. This could be considered as several controllers that need to be tuned separately. We expect however that considering them at the same time gives us the opportunity to significantly increase the efficiency and effectiveness of the controller tuning. This is becoming possible thanks to the increasing interconnectivity of these systems and the resulting possibility to share information between them. Since the systems we consider consist of mechanical/electronic/pneumatic/hydraulic/... subsystems, a (set of) controllers, connection(s) to the cloud or to other systems, they are further denoted as cyber physical systems (CPSs). The problem we address in this project is to optimally tune controllers for a fleet of CPSs.
The project aims to develop a series of software tools to allow multi-system learning control as well as a methodology on how these tools have to be used. We further aim to demonstrate these developments on a variety of industrially relevant test cases.