Control Loop Performance Assessment With a Dynamic Neuro-Fuzzy Model (dFasArt)

Abstract

Most of the industrial controllers have some kind of performance problem. This feature is becoming more difficult to supervise and assess because of the increasing number of control loops of the processes. A new method for monitoring and performance assessment by using a neuro-fuzzy architecture is proposed. This method is based on the dFasArt model, which allows for a self-organizing classification (nonsupervised) of dynamic signals and building categories that can be easily interpreted in terms of fuzzy theory. A new fuzzy performance index (FPI) is defined, leading to a straight online assessment of the control loops. A great advantage compared with other techniques is that the method can be also applied to find relationships between process variables and to establish propagation paths. Other advantages of this method are as follows: 1) it is not necessary to obtain the model of the plant; 2) it can be applied online, in parallel with the process, without any dedicated experiment; and 3) the results are clearly presented to plant operator to help the control engineer to decide how to improve the control performance.