Automatic detection and quantification of stiction in control valves☆
Introduction
A typical chemical plant has hundreds of control loops. Control performance is important to ensure tight product quality and low cost of the product in such plants. The presence of oscillation in a control loop increases the variability of the process variables thus causing inferior quality products, larger rejection rates, increased energy consumption, reduced average throughput and profitability. The only moving part in a control loop is the control valve. Control valves frequently suffer from problems such as stiction, leaks, tight packing, and hysteresis. Bialkowski (1992) reported that about 30% of the loops are oscillatory due to control valve problems. In a recent work Desborough and Miller (2001) reported that control valve problems account for about one-third of the 32% of controllers classified as ‘poor’ or ‘fair’ in an industrial survey (Desborough, Miller, & Nordh, 2000). If the control valve contains nonlinearities, e.g., stiction, backlash, and deadband, the valve output may be oscillatory which in turn can cause oscillations in the process output. Among the many types of nonlinearities in control valves, stiction is the most common and one of the long-standing problems in the process industry. It hinders proper movement of the valve stem and consequently affects control loop performance. Stiction can easily be detected using invasive methods such as the valve travel or bump test. But to apply such invasive methods across an entire plant site is neither feasible nor cost-effective because of their manpower, cost and time intensive nature.
Although many invasive tests/methods have been suggested (Aubrun et al., 1995, Gerry and Ruel, 2001, McMillan, 1995, Ruel, 2000, Sharif and Grosvenor, 1998, Taha et al., 1996, Wallén, 1997) for analysis and performance of control valves, only a few non-invasive studies or methods (Horch, 1999, Rengaswamy et al., 2001, Singhal and Salsbury, 2005, Stenman et al., 2003, Yamashita, 2005) have appeared in the literature. Horch's method is successful mainly in detecting valve stiction in flow control loops. It cannot be applied for loops involving an integrator or those carrying compressible fluids. However, Horch and Isaksson (2001) suggested the ‘camel method’ based on the distribution of the second derivatives of the controlled variable to detect valve stiction in an integrating plant. The methods described in Singhal and Salsbury (2005), Rengaswamy et al. (2001), Yamashita (2005) depend on the qualitative shape of the time trends of the data which is often distorted by the presence of noise and disturbances. Also, in real life the shape of the time trends of data is heavily affected by the process and controller dynamics. Stenman et al. (2003) described a model-based segmentation method to detect stiction in control valves. This method requires the model of the process and some tuning parameters. To obtain the closed loop model of the process from routine operation data is generally non-trivial. Moreover, all these methods can only detect stiction but cannot quantify it. As pointed out rightly in Desborough and Miller (2001), ‘a passive or non-invasive method that can reliably and automatically classify valve performance in closed loop is desperately needed in process industry’, a non-invasive method capable of detecting and quantifying stiction will be useful in the process industry to identify valves that need maintenance or repair.
An effective non-intrusive data-based monitoring method could reduce the cost of control loop performance maintenance by screening and short-listing those loops and/or valves that need maintenance. This paper describes a data-based model free non-invasive method that can automatically detect and quantify stiction present in control valves. The main contributions of this paper are:
- •
A model free method for detecting and quantifying stiction in control valves from routine operating data has been developed. The method does not require the performance of any additional valve travel test or commonly known bump test of the control loop.
- •
The novel feature of the method is that it can detect and quantify stiction using controlled variable , controller output and set point data. It does not require valve positioner data. If data is available it is very easy to detect and quantify stiction from the mapping of and . But this is not the case, when only , , and data are available because the mapping of and is often confounded by the loop dynamics and disturbances. To the best knowledge of the authors, there are no available methods in the literature that can detect and quantify stiction from only and data.
- •
Finally, the algorithm has been fully automated.
- •
The method is useful in short-listing the valves suffering from stiction from hundreds or thousands of control valves used in chemical plants or elsewhere. Thus it contributes to reduce the plant maintenance cost and increases the overall profitability of the plant.
The paper has been organized as follows: a formal definition of the stiction is first given, followed by a description and summary of the method used to detect nonlinearity in control loops that the authors have proposed in a separate paper (Choudhury, Shah, & Thornhill, 2003). The main contributions of this paper are the new results on an automated detection scheme for ‘sticky’ valves and the quantification of the amount of stiction present in the valve. Several industrial case studies of the proposed method are presented to demonstrate the practicality and applicability of the proposed methods.
Section snippets
What is stiction?
Different people or organizations have defined stiction in different ways. Some of these definitions have been presented in Choudhury, Thornhill, & Shah (2005b). Based on careful investigation of real process data a new definition of stiction has been proposed by the authors (Choudhury, Thornhill, & Shah, 2005a) and is summarized as follows.
The phase plot of the input–output behavior of a valve ‘suffering from stiction’ can be described as shown in Fig. 1. It consists of four components:
Detection of stiction in control valves
In a control loop, a nonlinearity may be present either in the process itself or in the control valve. For our current analysis, we are assuming that the process nonlinearity is negligible in the vicinity of the operating point where the data has been collected. This is a reasonable assumption because the method works with routine operating data of a control loop under regulatory control. In general, when processes are fairly well regulated at standard operating conditions, the plant can be
Quantifying stiction
Strictly speaking, all valves are sticky to some extent. A detection and diagnosis algorithm can identify stiction in a large number of control valves. Some of them may be sticky by an acceptably small amount for the current application in hand while others may suffer from severe stiction and need immediate maintenance of the valve. Therefore, it is important to be able to quantify stiction so that a list of sticky valves in order of their maintenance priority can be prepared. It is well known (
An illustrative example
The objective of this section is to explain the sequence of steps of the proposed method with a detailed presentation of an industrial example. This example represents a level control loop in a power plant, which controls the level in a condenser located at the outlet of a turbine by manipulating the flow rate of the liquid condensate from the condenser. In total 8640 samples for each tag were collected at a sampling rate of 5 s. Fig. 4(a) shows the time trends for level (), set point and
Automatic detection and quantification of stiction
A typical chemical process has hundreds or thousands of control loops. In order to apply any diagnosis method to such a large number of industrial control loops, it must be automated. Figs. 2 and 3 describe the automation steps of the proposed method. The summary is given as follows:
Step 1: Detection of Nonlinearity. Calculate NGI and NLI for the control error signal (–). If both of the indices are greater than the threshold values, the loop is detected as nonlinear. Otherwise, the poor
Practical implementation issues
For any data analysis, a considerable amount of time is spent on data preprocessing to make the data suitable for analysis. The following sections describe some useful information when analyzing data using the method described here.
Diagnosis of an external disturbance
The purpose of this section is to demonstrate the efficacy of the proposed method for the detection and quantification of valve stiction through a simulated study under a controlled environment, where some other methods may fail to make a correct diagnosis.
Sometime an unmeasured oscillatory external disturbance, for example a sinusoidal disturbance, can enter a control loop and produces oscillatory controlled and manipulated variables. This can often be misdiagnosed as a valve problem. This
Industrial case studies
The objective of this section is to evaluate the proposed method on a number of selected control loop data obtained from different types of process industries. For each loop, the set point , controlled output () and controller output data were available. Unless otherwise stated, a data length of 4096 was used for the squared bicoherence calculation for each case. The time trends of these variables, the squared bicoherence plot, the c-means clustering plot, and the fitted ellipse plot
Conclusions
A non-invasive method for detecting and quantifying stiction in control valve has been presented in this paper. The method first detects nonlinearity in a control loop by the use of the sensitivity of the normalized bispectrum or bicoherence to the nonlinear interactions that may be present in the control error signal. If nonlinearity is detected, then and signals are filtered using frequency domain Wiener filter to obtain filtered and signals. If an ellipse can be fitted suitably
Acknowledgements
Financial support to the first author in the form of a Canadian International Development Agency (CIDA) scholarship is gratefully acknowledged. The project has also been supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), Matrikon Consulting Inc., and the Alberta Science and Research Authority (ASRA) in the form of an NSERC-Matrikon-ASRA Industrial Research Chair Program at the University of Alberta. The authors are also grateful to BP Amoco Oil Ltd., Celanese
References (36)
- et al.
Fault detection in control loops
Control Engineering Practice
(1995) A control loop performance monitor
Control Engineering Practice
(1995)A simple method for detection of stiction in control valves
Control Engineering Practice
(1999)- et al.
A qualitative shape analysis formalism for monitoring control loop performance
Engineering Applications of Artificial Intelligence
(2001) - et al.
A simple method for detecting valve stiction in oscillating control loops
Journal of Process Control
(2005) - et al.
Detection of multiple oscillations in control loops
Journal of Process Control
(2003) Pattern recognition with fuzzy objective function algorithms
(1981)- Bialkowski, W. L. (1992). Dreams vs. reality: A view from both sides of the gap. In Control systems (pp. 283–294)....
- Choudhury, M. A. A. S. (2004). Detection and diagnosis of control loop nonlinearities, valve stiction and data...
- Choudhury, M. A. A. S., Shah, S. L., & Thornhill, N. F. (2002). Detection and diagnosis of system nonlinearities using...
Diagnosis of poor control loop performance using higher order statistics
Automatica
Cited by (0)
- ☆
The methodology described in this paper has been submitted for an international patent. A preliminary version of this paper was presented at DYCOPS 2004, July 5–7, 2004. Cambridge, MA, USA.