Selection schemes and neural networks in adaptive real-time control : predictive simulation adaptive control

Sammanfattning: The vision of self-aware machines was the starting point of this work. The idea is a machine having knowledge about itself and its surrounding environment, being able to react to changes in the environment. To support this vision, a number of engineering disciplines need to be merged and the control strategy "predictive simulation adaptive control'', PSAC, developed.The focus of the thesis is the synthesis of a number of algorithms and ideas from different engineering disciplines and fields. The main disciplines that the work is based on are simulation techniques, selection schemes and neural networks; all of these combined with the constraints imposed by the real-time demands of control systems.Selection schemes, or optimisation algorithms, are introduced here and used directly for real-time control of the test system, a rod-less pneumatic cylinder. The selection process is primarily based on genetic algorithms and the outcome of numerous simulations of the system for different possible control signals.Neural networks in general, and the version used here in particular, the Kohonen self-organising map, is widely used for classification and storage of information. Here it is used first to approximate friction in a rod-less pneumatic cylinder, and later on, possible ways to utilise this technique for condition monitoring are briefly discussed.Real-time systems and programming are a necessity when designing modern control systems. From the real-time constraints, special demands are put on the implemented algorithms and ideas.By bringing all this together, piece by piece, the vision comes a little bit closer. One step on the path, is the PSAC control concept proposed here. The control concept is successfully implemented and tested on a position servo consisting of a pneumatic rod-less cylinder controlled by on/off-valves.