Automatic monitoring of states evolution in dinamic scene supervision

A neural network model is presented for monitoring events coded in image sequences. The image sequences define the sampling of high frequency space/time changeable phenomena where both the morphological and densitometric aspects of the scene are taken into account. In this frame, a model for implementing a multilevel neural network architecture is proposed. This model is tested in the field of power production for monitoring the combustion instability degree in power plant gas combustors. The main goal of this study is both to provide a support for preventing the oscillation states in the combustor's flame front and to characterise the instability itself. The work has been developed within a collaboration with ENEL Production and Research S.p.A. (Italian National Department for Electric Power), that supplied the study cases and the technical support for conducting the experiments. The preliminary results obtained show the effectiveness of the approach proposed.