Stiction detection in control valves using a support vector machine with a generalized statistical variable.

Presence of nonlinear factors, such as stiction, hysteresis, and deadband, in a control valve imposes limitations on the performance of the control loop. The importance of assessing the nonlinear conditions of a control valve and detecting its possible defects is a key issue in monitoring performance. Because, this may severely impair the performance of the loop and have a direct impact on the product quality. Valve stiction is one of the most common problems of control valves that cause oscillations in the control loops. Many methods have been proposed to detect this nonlinear phenomenon based on the process model; however, failure to accurately identify the system and the variability of the system parameters make the detection of this factor important independent of the process model. Despite the uncertainty of the measured signals, statistical methods have not applied widely due to their complexities. This paper presents a method for detecting valve stiction occurrence unknown process dynamics by extending a support vector machine (SVM) with a generalized statistical variable.