Control Valve Stiction: Experimentation, Modeling, Model Validation and Detection with Convolution Neural Network

Abstract

The controller and the control valve are the workhorses of the process industry. The profitability, the reduction in energy consumption and raw material usage along with the increase in product quality are maintained by the process control hardware and software. However, control loops can suffer from poor performance due to ill tuned controllers or mostly due to problems associated with the pneumatic control valves as they are the only moving parts in the control loops. These oscillations will lead to increase energy consumption and increased wear and tear of equipment along with poor product quality. This paper proposes discrete data-driven models to simulate the stiction and oscillation of a control valve based on first order dynamics. The model is validated through experimental results obtained from a sticky valve test bed. Furthermore, a Convolution Neural Network is utilized successfully to identify the control valve stiction. Libraries for VP (Valve Position) vs. CO (Controller Output) plots were utilized to train the convolution neural network.