Detection and identification of stiction in control valves based on fuzzy clustering method
Loading...
Date
2016-08-01
Authors
Muhammad Amin Daneshwar
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The presence of static friction (stiction) in control valves deviates the valve position from its origin and therefore produces a nonlinear behavior (dead band plus stick band and slip jump) in control loops. The nonlinearity forces control loops to oscillate. The oscillation results in poor product quality and increased energy consumption. The detection of stiction for flow control loops which form significant control loops in industry in a timely manner is of great importance. After the presence of stiction has been detected, in order to mitigate stiction problem, it is necessary to estimate stiction parameters (quantification) in the earlier methods. However, this estimation which requires huge investment of time and effort is a challenging task. In this study, in order to improve covariance estimation of fuzzy clustering, linearly correlation of data is detected. Then a matrix (which contains a sequence of serially uncorrelated random numbers with zero mean and finite variance) is added to covariance matrix. This modification prevents the fuzzy clustering algorithm from turning into numerical problem. A method, which gain benefits from the idea that in the presence of stiction, the cluster centers of main regions of flow control loops are deviated from their origin, is proposed to detect the deviation (detection). Furthermore, based on the idea that, the slopes of the lines obtained from successive cluster centers, share some properties (in the presence of stiction), a new performance index which collects these properties to distinguish cause of oscillation (diagnosis) is proposed. Finally as an alternative to stiction quantification, by configuring a fuzzy identifier, an appropriate model of process with control valve stiction is identified (identification). The identified model is able to capture (identify) all relevant dynamics of the process with control valve stiction. The number of correct detections is now 85%. Not only has the identification time been decreased to less than a second (i.e. average is 0.4505 seconds), the performance of the proposed methods of stiction detection, diagnosis and identification has also been confirmed by both simulation and industrial data.