Quantifying Closed Loop Performance Based on on-Line Performance Indices

Abstract

This article aims to construct an "inference model" (IM ) that assesses the closed loop performance and robustness for SISO controllers, with no need of intrusive tests (i.e. set-point changes or open-loop tests). The input variables are indexes that can be easily calculated on-line (minimal variance, perceptual error, standard deviation, among others). The IM is generated for a large set of plants and tuning parameters. The possible inputs for the IM are 9 standard assessment measurements (e.g., Harris Index, standard deviation, etc) on-line available, commonly present in commercial tools . The target for IM is the closed loop open loop rise time ratio (Rt R ), Gain Margin (GM), and normalized (ISE). These values are obtained by intrusive tests. Four different classes of inferential models (i.e., Neural Networks, Neuro Fuzzy, PLS, and QPLS) are compared. The best results are obtained by Neural Network IM. The results obtained show that the methodology is very promising.