(390g) Elucidating and Handling Valve-Induced Nonlinearity in Industrial Feedback Control Loops

In much of the process control literature, valves are treated as having either very fast dynamics such that the flowrate out of the valve is equal to the valve output set-point that is the control signal to the valve, or the valve dynamics are approximated with a simple linear model (e.g., a first-order process model [1]). However, the flowrate out of the valve may be related nonlinearly to the control signal received by the valve, through static nonlinearities between the valve position and outlet flow rate (e.g., an equal percentage valve characteristic [1]) or dynamic effects such as stiction and deadzone [2]-[3]. Nonlinearities in process control loops are known to cause closed-loop performance or even stability issues; for example, the valve stiction-induced nonlinear dynamic behavior is well-known for causing control loop oscillations [2], and as a result, a number of methods for improving process control by accounting for valve stiction have been investigated [4]-[5]. In general, accounting for valve nonlinearities in process control design can be expected to positively impact the closed-loop performance obtained.

In this work, we clarify the effects that static and dynamic nonlinearities can have on a closed-loop system under classical control methods and under model predictive control (MPC), both of which are control methods widely used in process industries. Using the insight of the results obtained, we discuss methods for improving industrial control-loop performance when such negative effects occur. For example, when control loop oscillations occur in a loop containing a controller with an integral term (e.g., proportional-integral control) and a sticky valve, we propose an anti-windup-inspired method for reducing the windup-like effect caused by stiction on the integrator. To improve set-point tracking when MPC is used in the presence of valve stiction, we propose a set of modifications of the MPC ranging from the inclusion of suitable, computationally-efficient models of the valve dynamics within the MPC to the incorporation of constraints on the variation of the control action within two successive MPC sampling times. We provide theoretical results establishing conditions under which our methods are guaranteed to work and demonstrate their applicability and performance using several chemical process examples.

[1] Coughanowr DR, LeBlanc SE. Process Systems Analysis and Control, 3rd ed. Boston, MA: McGraw-Hill, 2009.

[2] Brásio ASR, Romanenko A, Fernandes NCP. Modeling, detection and quantification, and compensation of stiction in control loops: The state of the art. Industrial & Engineering Chemistry Research. 2014;53:15020-15040.

[3] Choudhury MAAS, Thornhill NF, Shah SL. Modelling valve stiction. Control Engineering Practice. 2005;13:641-658.

[4] Srinivasan R, Rengaswamy R. Approaches for efficient stiction compensation in process control valves. Computers & Chemical Engineering. 2008;32:218-229.

[5] Hägglund T. A friction compensator for pneumatic control valves. Journal of Process Control. 2002;12:897-904.