Applying neuro-fuzzy model dFasArt in control systems

This paper presents the neuro-fuzzy dFasArt (dynamic FasArt) architecture as an extension of the FasArt model including a dynamic algorithm formulation. This allows dFasArt to deal with identification and clustering problems using the temporal information of the signals. The focus is placed on the application of dFasArt to the control systems field for monitoring the controller performance. It is presented through two selected experiments covering some interesting control issues. The first one shows the use of dFasArt to decide when the parameters adaption is needed in a classic adaptive control scheme. The second one analyzes the behaviour of closed-loop controlled systems to establish a classification of the system operational states, starting from the measured data. Digital signal processing is used to represent the temporal signals with spatial patterns and dFasArt is proposed to classify these patterns on-line. Real scale plants have been used to carry out several experiments with good results. This shows dFasArt as a feasible tool to deal with control loop performance monitoring and controller performance assessment in industrial processes.