No abstract

In part 2 of this two-part series, an approach for diagnosis and quantification of stiction using a simple single-parameter model is proposed. The stiction model, in conjunction with an identified process model from routine operating data, is shown to successfully facilitate stiction diagnosis. An optimization approach is used to jointly identify the process model and the stiction parameter. This approach is based on the identification of a Hammerstein model of the system comprising the sticky valve and the process. In this work, a new identification procedure for Hammerstein systems that supports stiction diagnosis is proposed. Industrial and simulation case studies are shown to demonstrate the application of the proposed approach for diagnosing stiction.

A spate of industrial surveys over the past decade indicate that only about one-third of industrial controllers provide acceptable performance. Since significant commercial benefits exist in diagnosing and improving the remaining two-thirds of the industrial controllers, the past few years have seen an emergence of control loop performance monitoring techniques using routine operating data. About 20−30% of all control loops oscillate due to valve problems caused by static friction or hysteresis. In the first of this two-part paper, a qualitative pattern recognition approach is described for stiction diagnosis. Stiction in control valves leave distinct qualitative shapes in the controller output (OP) and controlled process variable (PV) data. These shapes can be generally categorized as being square, triangular, and saw-toothed. To classify the patterns that evolve due to stiction, a pattern recognition approach using dynamic time warping (DTW) technique is proposed. The success of our proposed approach is built on a new technique for detection and time characterization of oscillations. A robust method for generating a stiction template pattern for each oscillating cycle as opposed to a global pattern for the whole data set is proposed. The qualitative approach was tested on data sets of varying complexity that include nonconstant behavior, intermittent stiction, and external disturbances, and results for eight data sets are presented.

No abstract

No abstract