Valve Stiction Modeling: Data-Driven Vs. First Principles

Due to adverse impact of valve stiction in the process industry, both physical models and empirical data-driven models have been developed in the past decade to investigate the valve stiction behavior. To simulate valve stiction, both detailed physical models and empirical models have been developed in recent years. Physical models [1-3] describe the stiction phenomenon using force balances based on Newton's second law of motion. The main disadvantage of these models is that they require knowledge of several parameters such as the mass of the moving parts and different type of friction forces which cannot be easily measured and depend on the type of fluid and valve wear. On the other hand, empirical or data-driven models [1,3,4] use simple empirical relationships between valve input and output to describe valve stiction, with just a few parameters that can be determined from operating data. Due to their simplicity and easy implementation, data-driven models have gained tremendous research interest in recent years.

Compared to physical models, these data-driven models simplify the simulation of a sticky valve and has been used by several other researchers for valve stiction simulation. In this work, we compare the data-driven model that we proposed recently (shorted as He's two-parameter model) with other two data-driven models, namely Choudhury's model and Kano's model. We examine the fundamental difference between He's two-parameter model and Choudhury's/Kano's model. We point out that the different valve stiction behaviors produced by different data-driven models are rooted in the different model assumptions. To compare the three data-driven models, a well-established physical model is implemented. The differences among the three different data-driven models are revealed by comparing to the physical model. It is shown that He's two-parameter model can best reproduce the signature stick-slip behavior of a sticky valve which is simulated by the physical model and observed in actual control valves [5, 6]. Using the physical model, we further investigate the valve stiction behavior analytically, and derive a three-parameter model that more accurately reproduces the physical model response by including a constant gain K.  We show that the three-parameter model can accurately reproduce the physical model behavior without involving cumbersome numerical integration. We also show that K is insensitive to valve parameters and in general very close to 2. Therefore, in the case that valve parameters are not available, K can be set as a constant (e.g., 1.99) so the three-parameter model reduces to the modified two-parameter model.

The modified two-parameter model is tested using an industrial. It is shown that the model satisfactorily simulates the industrial case and captures the essential characteristics of the stiction. Therefore the proposed data-driven model should be a useful tool for combating valve stiction in process industry.

Key words: valve stiction model; control valve; data-driven modeling; first principles modeling

References:

1.       He QP,Wang J, PottmannM, Qin SJ (2007) A curve fitting method for detecting valve stiction in oscillating control loops. Industrial and Engineering Chemistry Research 46:4549–4560.

2.       Kayihan A, Doyle III FJ (2000) Friction compensation for a process control valve. Control Engineering Practice 8:799–812.

3.       Choudhury MAAS, Thornhill NF, Shah SL (2005) Modeling valve stiction. Control Engineering Practice 13:641–658.

4.       Kano, M., Maruta, H., Kugemoto, H., Shimizu, K. (2004). Practical model and detection algorithm for valve stiction. Proc IFAC DYCOPS, Cambridge, USA.

5.       Gerry J, Ruel M (2001) How to measure and combat valve stiction online. Proc ISA International Fall Conference, Houston, TX. http://www.expertune.com/articles/isa2001/StictionMR.htm.

6.       Ruel M (2000) Stiction: the hidden menace. Control Magazine 13:69–75. http://www.expertune.com/articles/RuelNov2000/stiction.html.