Investigation on Statistical-based Detection Techniques For Control Valve Stiction Detection

Control valve stiction is one of the main causes that can affect the performance of a
control loop. As the final control element, it can cause disruptions towards the
operations especially on the plant production of oil and gas industry. An initiative
had been made in 1989 where a stiction detection method is first developed to detect
stiction in control valve. Since then, many methods are produced and redeveloped
but only few uses the statistical-based methods. Hence, this project will cover
statistical-based methods which had once been used for fault detection and will be
tested for the effectiveness in detecting control valve stiction. Two case studies are
chosen which are simulation case study involving stiction and non-stiction conditions
namely, well-tuned controller (Base case), tightly-tuned controller (Case 1), presence
of disturbances controller (Case 2) and presence of stiction controller (Case 3).
Another case study is the real industrial data from a chemical plant. The process
output (pv) and controller output (op) for each case are generated based on nonlinear
principle component analysis (NLPCA) method. Two statistical-based methods are
chosen to be tested which are generalized likelihood ratio (GLR) test as well as error
testing method. Based on the results of testing for method 1, there are some
limitations for GLR test method to detect between the presence of stiction and the
presence of disturbances in the system. However, for method 2, it can be seen that
the error testing method focusing at the root mean squared error (RMSE) calculation
is an effective tool and method to detect stiction and manage to differentiate stiction
and non-stiction system for the simulation and real industrial case studies chosen.