Identification techniques for stiction quantification in the presence of nonstationary disturbances

The paper presents a detailed comparison of different identification techniques applied to valve stiction quantification, possibly in the presence of nonstationary unknown disturbances. The control loop with sticky valve is modeled as a Hammerstein system, in which the nonlinearity is identified using enumeration of the parameters’ space. Five different techniques for identification of the linear model are compared in terms of achievable performance. In particular, the capability to cope with the presence of nonstationary disturbances is analyzed. The techniques allow one to estimate the unknown actual valve position (MV), without requiring any process knowledge, being based only on data which are usually recorded in industrial plants: controller output (OP) and controlled variable (PV). Simulations show that external perturbations can be tolerated, thus ensuring a reliable evaluation of stiction in practical situations where external disturbances are usually present. Models which incorporate a time varying additive nonstationary disturbance grant a better process identification and a more accurate stiction estimation in the case of disturbance acting simultaneously with valve stiction. However, simpler models are the best choice when stiction happens to be the only source of loop oscillation. Results are confirmed by application to real data: pilot plant data are used to corroborate the effectiveness of the techniques