Neural network-based controllers for an electrothermal furnace system

Neural network schemes are applied in this thesis to a temperature control system problem. The electrothermal furnace is a very popular instrument for applications in material testing area. In this work feedforward neural networks are trained for both identification and control problems of the electrothermal furnace system. The thesis demonstrates that neural networks can be used effectively for this application problem, which is a highly nonlinear dynamical system. The first emphasis is on the electrothermal furnace model identification and the second emphasis is on the design of neural network based PID and internal model control strategies. Both static and dynamic back-propagation methods are discussed. In the electrothermal furnace models that are introduced, multi-layer feedforward networks are interconnected in novel configurations. A novel technique based on the internal model control for nonlinear systems using neural networks is proposed. The control structure proposed directly incorporates a model of the plant that was identified by a neural network and its inverse as part of the control strategy. The potential utilizations of the proposed methods are illustrated through experimental and numerical simulations of an electrothermal furnace system