The development and investigation analysis of an arx-based generalized likelihood ratio (GLR) stiction detection method

Control valve stiction is one of the main sources of nonlinearity which can result in many deleterious effects on the control loop performance of a process. The study of stiction detection methods has now becoming one of the essential research areas in process control. In this present work, an ARX-based Generalized Likelihood Ratio (GLR) stiction detection method is proposed and its effectiveness is analyzed. The implementation of the proposed method involves three main stages; 1) ARX model identification, 2) GLR test, and 3) statistical hypothesis testing. The proposed detection method was applied to two benchmark simulated case studies. Results showed that the method effectively detect stiction. The presence of stiction is declared if the GLR test statistics, L(R) exceeds the decision threshold limit, h(α)=3.841, and the null hypothesis is rejected at 5 significance level. On the other hand, if L(R) value lies below h(α)=3.841, the null hypothesis is accepted and the absence of stiction is confirmed. In addition, it is also observed that the proposed method is reasonably insensitive and robust to the changes in the process gain, K and time constant, � as it generally allows up to ±10 changes in the two parameters for both case studies. © 2018 Penerbit UTM Press. All rights reserved.