A New Measure to Improve the Reliability of Stiction Detection Techniques

A variety of methods have been developed to identify the presence of stiction in linear closed-loop systems. One of the commonly used approaches is based on identification of a Hammerstein model (linear dynamic model preceded by static nonlinear element) between the controller output and process output. These techniques utilize the fact that control valve stiction introduces nonlinearities in the otherwise linear feedback system. However, the present work shows that these techniques could provide ambiguous results depending on the frequency response of the controller and the process of interest. Therefore, in this work, a reliability measure to validate the results from Hammerstein model-based stiction detection approaches is proposed. This measure of reliability is important from an industrial perspective because (i) it helps in reducing false alarms and (ii) it improves the computational speed by guiding the selection of search space in the linear model identification step. The applicability of this reliability measure in industrial setting is demonstrated using various simulation and industrial case studies.