Quantification of Nonlinear Valve Stiction Model Using Compound Evolution Algorithms

The presence of oscillations in a control loop increases the deviations from the set point of the process variables, thus causing inferior products, larger rejection rates, increased energy consumption and reduced average throughput. There are several reasons for oscillations in control loops. They may be caused by excessively high controller gains, oscillating disturbances or interactions, but a very common reason for oscillations is friction in control valves. It is important to early detect valve stiction so that appropriate action can be taken to relieve the situation. In this paper a hybrid algorithm combines the fundamental elements of standard Genetic Algorithms with those proposed by Nelder and Mead in their Simplex algorithm. A nonlinear dynamic model consisting of a linear process and a nonlinear control valve with stiction is established. This approach involves obtaining easily measurable variables and using this information to estimate a set of unknown model parameters. By means of the hybrid algorithms proposed, the detailed procedure for the parameter identification with actual system’s input-output data are given. The effectiveness of identification is verified by the comparison between actual values of system and model in different situations.