Dublin Institute of Technology

ARROW@DIT
Conference papers School of Electrical and Electronic Engineering

2004-01-01

Control loop performance assessment: a

classification of methods
Niall O'Connor
Dublin Institute of Technology

Aidan O'Dwyer
Dublin Institute of Technology, aidan.odwyer@dit.ie

Follow this and additional works at: http://arrow.dit.ie/engscheleart

Part of the Controls and Control Theory Commons

Recommended Citation
O'Connor, Niall and O'Dwyer, Aidan : Control loop performance assessment: a classification of methods. Proceedings of the Irish
Signals and Systems Conference, pp. 530-535, Queens University Belfast, July, 2004

This Conference Paper is brought to you for free and open access by the
School of Electrical and Electronic Engineering at ARROW@DIT. It has
been accepted for inclusion in Conference papers by an authorized
administrator of ARROW@DIT. For more information, please contact
yvonne.desmond@dit.ie, arrow.admin@dit.ie.

This work is licensed under a Creative Commons Attribution-

Noncommercial-Share Alike 3.0 License
 ISSC 2004, Belfast, June 30 - July 2

Control Loop Performance Assessment: A Classification of

Methods
Niall O’Connorφ and Aidan O’Dwyer*
φ
School of Control Systems and Electrical * School of Control Systems and Electrical
Engineering, Engineering,
Dublin Institute of Technology Dublin Institute of Technology
Kevin St., Dublin 8, IRELAND Kevin St., Dublin 8, IRELAND
E-mail: φ alf_fonzo@hotmail.com E-mail: *aidan.odwyer@dit.ie

__________________________________________________________________________________________

Abstract – This paper presents an overview of a number of

controller performance assessment techniques. The techniques
discussed are divided into five categories, namely, time domain
assessment, frequency domain assessment, minimum variance
control (MVC) as a benchmark, statistical analysis, and other
more ‘problem specific’ assessment techniques. Recent work, by
various authors, in each of the five categories is outlined.
__________________________________________________________________________________________
1. Time domain assessment,
2. Frequency domain assessment,
3. Minimum variance control (MVC) as a
I INTRODUCTION benchmark,
According to [1] monitoring of process variables is 4. Statistical analysis techniques, and
useful, not only for assessing the status of the process, 5. Other more ‘problem specific’ assessment
but also for controlling product quality. According to techniques
[2], in the testing of thousands of control loops in
hundreds of operating plants, Techmation Inc. and II TIME DOMAIN ASSESSMENT
others have found that more than 30% of the
automatic control loops actually increase variability
over manual control due to poor controller tuning.
One reason why so many control loops perform
poorly is that there is often numerous (more than a
thousand) loops in a large process plant and not
enough control engineers to maintain every loop.
In [3] Jamsa-Jounela et al. make the point
that in order to ensure highest product quality it is
essential to maintain the control system in an
adequate manner. In [4] Vishnubhotla et al. discuss
how the current standard practice for industrial
process control is to install DCS (Distributed Control
Systems) and PLC control system platforms. These
system platforms accumulate large volumes of
process data, but there are very few data mining tools.
It should be obvious, therefore, that there is a
strong need for automatic assessment and monitoring
of control loop performance. The goal of monitoring Figure (1). Typical transient response of a
should be to provide information that can be used to feedback control system to a step set point change
assess the current status of the existing controller and
to assist control engineers in deciding whether The dynamic response characteristics of a system
redesign is necessary [5]. When the controller may be accurately assessed using a number of useful
performance is determined to be inadequate, it is time domain measures. These measures include rise
important to ascertain whether an acceptable level of time, settling time and integral error measures, see
performance can be achieved with the existing control Figure (1). The rise time (Tr) is defined as the time
structure [6]. from the step change in the set point until the
controlled variable first reaches the new set point [7].
With these goals in mind, the next step is to A short rise time is usually desired. The settling time
review some of the existing loop performance (Ts) is defined as the time the system takes to attain a
assessment techniques. It was decided to divide the ‘nearly constant’ value, usually + or – 5 percent of its
assessment techniques into the following categories: final value [7]. This measure is related to the rise time
and decay ratio. A short settling time is usually (B/A), see Figure (1), is the ratio of neighbouring
desired. peaks in an underdamped controlled-variable
The integral error measures indicate the response. Usually, periodic behaviour with large
cumulative deviation of the controlled variable from amplitudes is avoided in process variables; therefore,
its set point during the transient response. The a small decay ratio is usually desired, and an
Integral of Absolute Error (IAE) criterion is overdamped response is sometimes desired [7]. The
determined from the sum of areas above and below manipulated variable overshoot (C/D), see Figure (1),
the setpoint. It is an appropriate measure of control is of concern because the manipulated variable is also
performance when the effect on control performance a process variable that influences performance. Some
is linear with the deviation magnitude. The Integral of large variations can cause long-term degradation in
Squared Error (ISE) criterion is appropriate when equipment performance. The overshoot is the
large deviations cause greater performance maximum amount that the manipulated variable
degradation than small deviations. The Integral of exceeds its final steady state value and is usually
Time multiplied by Absolute Error (ITAE) criterion expressed as a percentage of the change in
penalizes deviations that endure for a long time. The manipulated variable from its initial to its final value.
Integrated Error (IE) criterion is not normally used Some overshoot is acceptable in some cases [7]. The
because positive and negative errors cancel in the maximum deviation of the controlled variable from
integral, resulting in the possibility of large positive the set point is an important measure of the process
and negative errors giving a small IE [7]. The degradation experienced due to disturbances. Usually
formulae for calculating the integral error measures a small value is desirable so that the process variable
are given below: remains close to its set point [7]. In many cases the
disturbance is composed predominantly of one or a
∞ few sine waves. Therefore, the behaviour of the
control system in response to sine inputs is of great
IAE = ∫ SP (t ) − CV (t ) .dt
0
(1)
practical importance, because through this analysis
∞
the relationship between the frequency of the
disturbances and the control performance is deduced.
∫ [SP (t ) − CV (t ) ] .dt
2
ISE = (2) Control performance is assessed by measuring the
0 amplitude of the output sine wave; the metric is often
∞ expressed as the ratio of the output to input sine wave
ITAE = ∫ t . SP (t ) − CV (t ) .dt
0
( 3) amplitudes [7].
In [12] Stanfelji et al. present a method for
∞ monitoring and diagnosing the performance of single
IE = ∫ [SP (t ) − CV (t ) ].dt
0
(4) loop-control systems based primarily on normal
operating data. This method involves analysing the
autocorrelation and cross correlations of a time series
of control loop variables. In [13] Hagglund describes
In [3] Jamasa-Jounela et al. present a set of a procedure for the automatic detection of sluggish
performance indices appropriate to process control loops obtained from conservatively tuned
monitoring and assessment. These indices include controllers. The ‘idle index’ describes the relation
IAE, ITAE, rise time and settling time. In [8] Swanda between the times of positive and negative correlation
and Seborg have developed a new methodology to between the control and measurement signal
assess the performance of PI controllers from closed increments. From this index the sluggishness of the
loop response data for a setpoint step change. This control loop can be determined.
method is based on two new dimensionless
performance indices, the dimensionless settling time
and the dimensionless IAE. This methodology is also
applicable to PID controllers. In [9] Horch and Stattin III FREQUENCY DOMAIN ASSESSMENT
extend this method to analyse the settling time-
normalised by the apparent process time delay-of a According to [14] traditional measures such as
setpoint step response. In [10] Ruel discusses a overshoot, rise time, decay ratio, settling time and the
number of metrics used to assess loop performance. ISE are difficult to translate into an economic
These include IAE, setpoint crossing, and average measure so as to justify process or control system
error. In [11] Huang and Jeng assess a simple redesign. They state, however, that frequency domain
feedback system by analysing IAE and rise time measures can be used to provide a measure of
observed from the response of the system to a step performance that can be translated into an economic
setpoint change. Optimal IAE’s and associated rise measure. This section will review some of the more
times are computed. Comparing its current IAE to the common frequency domain assessment methods.
optimal IAE allows an assessment of the performance Three different types of plots are commonly used to
of the system. graphically illustrate the frequency response of a
Explained in more detail below, there are a controlled system, see Figure (2). These three plots
variety of other time domain measures that may be are the Nyquist, Bode and Nichols plots. Nyquist
used to assess a systems performance. These include plots, also called polar plots, may be obtained by
offset, decay ratio, manipulated variable overshoot, either plotting the real versus the imaginary part of
maximum deviation of the controlled variable, and the frequency domain transfer function, G(jw) (using
magnitude of the controlled variable in response to a rectangular coordinates), or by plotting the magnitude
sine disturbance. Offset is defined as the difference at a particular phase angle of G(jw) (using polar
between the final, steady state value of the set point coordinates). Bode plots require two curves to be
and of the controlled variable. In most cases, a zero plotted; these plots show how the magnitude ratio and
steady state offset is desired [7]. The decay ratio
phase angle vary with frequency. The Nichols plot is problems in the method presented in [17] i.e. too
a single curve in a coordinate system with phase many relay tests are required, the frequency search
angle as the abscissa and log modulus as the ordinate. range is confined to the third quadrant, and the
Frequency is a parameter along the curve [15]. identified value of Lcmax cannot be used on-line to
redesign the controller. In [14] Belanger and Luyben
propose a new test to locate the peak regulator log
modulus. The test involves the insertion of a relay
between the controlled variable and a given load
disturbance model, with the feedback controller on
automatic. This causes the plant to exhibit a sustained
oscillation at the frequency where the Lcmax curve
exhibits a peak. This test can be applied to both
simulated models as well as existing plants.

Figure (3). Plot illustrating the maximum closed

loop log modulus Lcmax.

The capacity based method for

quantifying controllability is a method used to
quantitatively incorporate the economics of control
into conventional steady-state design methods [15]. In
Figure (2) Nyquist, Bode and Nichols plots [18] Elliot and Luyben outline a generic methodology
called the capacity based economic approach that can
illustrating Gain and Phase margin. be used to compare or screen preliminary plant
designs by quantifying both steady-state economics
Phase margin and gain margin are two and dynamic controllability. In [19] Elliot et al.
commonly used assessment measures. Phase margin demonstrate that the capacity based economic
(PM) is defined as the angle between the negative real approach can be successfully applied to a large
axis and a radial line drawn from the origin to the industrial scale process. In [20] Elliot and Luyben
point where the open loop frequency domain transfer analyse the effectiveness of the capacity based
function intersects the unit circle. The bigger the economic approach when controlling a complex
phase margin the more stable the closed loop system. recycle system consisting of a reactor and two
Phase margins of 450 are often considered appropriate distillation columns.
[15]. The gain margin (GM) is defined as the In [21] Kendra and Cinar discuss a method
reciprocal of the intersection of the open loop used to estimate the closed loop transfer function of a
frequency domain transfer function polar plot on the system by exciting the reference input with a zero
negative real axis. The bigger the gain margin, the mean, pseudo random binary sequence and observing
more stable the system. Typically gain margin values the process output and error response. Performance
of about 2 are recommended [15]. In [16] Astrom and assessment is based on the comparison between the
Hagglund discuss a simple method for estimating the observed frequency response characteristics and the
critical gain of a controlled system, from which the design specifications.
gain margin may be deduced.
The maximum closed loop log modulus,
Lcmax, is another quantity used to assess performance IV MINIMUM VARIANCE CONTROL (MVC)
in the frequency domain, see Figure (3). While the AS A BENCHMARK
phase and gain margin specifications can sometimes
give poor results when the shape of the frequency According to [22] and [23], minimum variance
response curve is unusual, the maximum closed loop control is considered the optimal feedback control
log modulus does not have this problem since it provided that the process can be described by a linear
directly measures the closeness of the open loop transfer function with additive disturbance. In [24]
frequency domain transfer function to the (-1,0) point Spring states that minimum variance is a better
at all frequencies [15]. In [17] Chiou and Yu propose benchmark than zero variance for evaluating
a monitoring procedure that identifies the maximum controller performance. Control systems cannot
closed loop log modulus in two to three relay reduce the variance in product quality below the
feedback experiments. In [5] Ju and Chiu present a variance inherent in the process. On the basis of
monitoring procedure incorporating the FFT (Fast minimum variance, an investment in controller
Fourier Transform) technique to identify Lcmax on maintenance can be evaluated realistically.
line. This proposed method addresses some of the
 According to [4], this benchmark control Harris index are suggested. In [33] Horch and
may or may not be achievable in practice depending Isaksson discuss a modification to the index
on process invertibilty and other process physical introduced by Harris [23]. The modified index and
constraints. Also, it is worth noting that this technique the original index are then evaluated and compared
requires knowledge of the process time delay, which using data from industrial processes. In [34] Isaksson
may not always be available. However, as a discusses the MVC benchmarking technique and
benchmark, it provides useful information such as suggests a set of alternative indices. In [35] and [36]
how much ‘potential’ there is to improve controller Huang discusses some of the aspects associated with
performance. In [25] Thornhill et al. make the point the minimum variance control law for linear time
that minimum variance control may require variant processes. Alternative benchmarks that are
excessively vigorous action of the manipulated more suitable for time variant processes are
variable and, as a result, can lead to maintenance suggested. In [37] Venkatesan introduces a minimum
problems for the actuators. This section presents a variance feedback control algorithm (MVFCA) that
review of some of the papers available that discuss can be used to calculate a series of adjustments
some of these details. required at the input that minimises the variance of
A number of papers are recommended that the output variable. In [38] Kucera presents a tutorial
give an overview of the MVC method. In [23] Harris paper emphasising the contribution of V. Peterka to
discusses how an estimate of the best possible control the steady state minimum variance control problem.
can be obtained by fitting a univariate time series to In [39] Qin presents an overview of the current status
process data collected under routine control. In [26] of control performance monitoring using minimum
Harris et al. discuss some of the concepts associated variance principles.
with assessing the effectiveness of a control system.
Also discussed in this paper is how these concepts V STATISTICAL ANALYSIS
were initially developed using a performance
benchmark of minimum variance control for SISO
systems. In [25] Thornhill et al. examine some of the According to [1], the goal of statistical process
factors that influence the minimum variance monitoring (SPM) is to detect the existence,
performance measure of a SISO control loop. The magnitude and time of occurrence of changes that
authors show that, for an arbitrary controller, the cause a process to deviate from its desired operation.
calculated minimum variance benchmark is different A number of useful techniques for the monitoring of
for servo and regulator operation. In [27] Grimble process variables are discussed in this paper. These
discusses the use of the generalised minimum methods include Shewhart control charts, moving
variance control law for control loop performance average control charts, cumulative sum charts and
assessment and benchmarking. In [28] Huang and partial least squares methods.
Shah discuss, in detail, some of the theory behind the The likelihood method is a useful technique
MVC method. for assessing performance. According to [1], this
Based on MVC theory, a performance index method may be used to determine if the error
(the Harris index) was first introduced by Harris [23]. response characteristics are acceptable based on
This index compares the actual variance in the specified dynamic performance bounds. Dynamic
process variable to that of a minimum variance response characteristics such as overshoot or settling
controller. In [22] and [29] Desborough and Harris time can be extracted from the pulse response of a
present a normalised performance index used to fitted time series model of the output error. The pulse
characterise the performance of control systems. This response of the estimated output error can be
index provides a measure of the proximity of control compared to the pulse response of the desired
to minimum variance control. Time domain and response specification to determine if the output error
spectral interpretations of the index are discussed and characteristics are acceptable. In [40] Tyler and
a fast, simple on-line method for estimating the index Morari propose a framework in which acceptable
is given. In [30] Bezergianni and Georgakis introduce performance is expressed by constraints on the closed
a modified version of the Harris index in which the loop transfer function impulse response coefficients.
closed loop performance is compared with that Using likelihood methods, a hypothesis test is
obtained with the best theoretical control action outlined to determine if control deterioration has
(minimum variance control) and no control action. In occurred. In [41] Zhang and Ho propose the use of
[4] Vishnubhotla et al. discuss a method of the likelihood ratio method as a means of sensitivity
performance assessment based on the Harris index. analysis of stochastic system performance.
The resulting index, gives an indication of the level of In [42] Li et al. develop a monitor to
performance of the controller, and an indication of the automatically detect poor control performance. The
action required to improve performance. In [24] monitor provides a measure (Relative Performance
Spring discusses a performance index based on Index – RPI) of a control loop performance relative to
minimum variance control. In [31] Ko and Edgar a reference model of acceptable control. The
outline a scheme for the estimation of achievable PI reference model simulates the controlled variable
control performance, measured by output variance, in output of a user defined, acceptably tuned, control
linear processes with dead time when stochastic load loop. In [43] Zhong demonstrates how to improve the
disturbances are affecting the process. effectiveness of equipment monitoring and process
A number of papers have been written in induced defect control through properly selecting,
which modifications to the MVC benchmark have validating and using the hypothetical distribution
been made. In [32] Eriksson and Isaksson discuss models. In [44] Mosca and Agnoloni study the early
how this technique provides an inadequate measure of detection problem of stability losses or close-to-
performance if the aim is not control of statistically instability conditions in feedback control systems,
random disturbances. Some modifications to the where the plant dynamics are uncertain and possibly
time-varying.
 VI OTHER ASSESSMENT TECHNIQUES determination of the capability of the control system,
development of suitable statistics for monitoring the
This section contains a number of more ‘problem performance of the existing system, development of
specific’ assessment techniques, as opposed to the methods for diagnosing the underlying causes for
more general methods discussed in previous sections. changes in the performance of the control system, and
The focus of this section is on methods to both detect incorporation of these methods in an industrial
and diagnose oscillations in control loops. The setting.
techniques discussed here may well be considered The main advantage that MVC as a
special cases of the methods discussed in previous benchmark has over the other four categories
sections. discussed in the paper is that it not only gives an
The first step in dealing with an under indication as to the current level of performance of
performing control loop with suspected oscillation the controlled system under investigation, but it can
disturbances is the detection stage. In [45] Hagglund also determine whether or not current performance
presents a closed loop performance monitor (CLPM) can be improved by retuning the controller. In [4]
to detect oscillations in the control loop. The Vishnubhotla et al. highlight this point by stating that
procedure presented is automatic in the sense that no ‘as a benchmark (MVC) … provides useful
additional parameters, other than the normal information such as how well the current controller
controller parameters, have to be specified. In [46] was tuned compared to the minimum variance
Huang et al. discuss a method of determining the controller and how much ‘potential’ there is to
presence of oscillations in selected frequency ranges, improve controller performance’. For example, an
based on the regularity of the zero crossings of index (ratio of minimum achievable output variance
filtered auto-covariance functions. In [47] Chang et to actual variance) value of 1 indicates that current
al. present a system-wide dynamic performance performance cannot be improved by retuning the
monitoring system (DPMS), which includes special existing controller. However, an index value below 1
features such as oscillation detection. In [48] Stenman indicates retuning the controller will have an impact
et al. propose a model-based method for detecting on improving system performance.
static friction (stiction) in control valves. In contrast While time domain, frequency domain or
to existing methods, only limited process knowledge statistical analysis techniques may give an accurate
is needed and it is not required that the loop has indication as to the current level of performance of
oscillating behaviour. In [49] Wallen proposes an the controller, no indication is given as to whether or
integrated system for valve diagnostics and automatic not retuning will lead to improved performance.
PID tuning. The purpose of the method is to detect Simulations must be run and re-run with differently
non-linearities such as friction and hysteresis since tuned parameters in order to determine if improved
these may drastically decrease the control control is possible. This could prove to be an
performance. inefficient use of time if it was discovered, after
Once an oscillation has been detected, the numerous simulations had been run and analysed, that
next step is to determine its cause. In [50] Thornhill it is not possible to improve on the current control
and Hagglund present a set of ‘operational signatures’ performance using the current controller structure.
that indicate the cause of an oscillation. This method Therefore, these findings would suggest that
involves the offline analysis of ensembles of data whatever assessment techniques are used,
from control loops. In [51] Horch proposes a simple benchmarks specific to the controller under
method for the diagnosis of oscillations in process assessment must be used in order to determine
control loops based on the cross correlation between whether retuning or controller redesign is necessary.
control variable and loop output. This method is According to [10], continuous performance
shown to correctly identify the two most important monitoring requires benchmarking so that it may be
reasons for oscillations in control loops in the process observed how performance has changed with time.
industry, namely, external oscillating disturbances Also, this benchmark must be specific to the plant
and stiction in control valves. In [52] Taha et al. under investigation. Future work will focus on the
present an on line automatic procedure for the development of a method to calculate controller
diagnosis of oscillations in control loops. This specific benchmarks, in one of the assessment
method works without disturbing normal plant categories outlined in this paper, in order to provide a
operation. more efficient monitoring and assessment tool.

VII CONCLUSIONS REFERENCES

According to [53], minimum variance control (MVC) [1] Cinar, A., and Undey, C. (1999). ‘Statistical process and
as a benchmark (as discussed in [22]) or variants of it, controller performance monitoring: A tutorial on current
methods and future directions’. Proceedings of the American
is used in virtually all industrial controller assessment Control Conference, San Diego, California, pp. 2625-2639.
packages due to its theoretical and practical [2] Ender D. (1993). ‘Process control performance: Not as good as
advantages. In [53] Hugo lists some of these software you think’. Control Engineering, 40, (10), pp. 180-190.
packages as follows: Performance assessment tool-kit [3] Jamsa-Jounela, S-L, Poikonen, R., Georgiev, Z., Zuehlke,
U., Halmevaara, K. (2002). ‘Evaluation of control
[54]; loop scout [55]; Process Doc [56]; and Aspen performance: Methods and Applications’ Proceedings of the
Watch [57]. Software packages such as Probe [58] 2002 IEEE International Conference on Control applications,
and Plant Triage [59] also offer a number of useful Glasgow, Scotland. pp. 681-686.
routines and algorithms related to MVC and some of [4] Vishnubhotla, A., Huang, B., and Shah, S.L., Badmus, L.,
(1999). ‘Control loop performance assessment: An enterprise
the other assessment techniques mentioned asset management solution’. Proceedings of International
previously. In [26] Harris et al. state that a Conference and Exposition for Advancing Measurement,
comprehensive approach for assessing the Triangle Park, NC, USA, pp. 31-40.
effectiveness of control systems requires [5] Ju, Jun and Chiu, Min-Sen. (1998). ‘A Fast Fourier
Transform Approach for On-line Monitoring of the maximum
 Closed-Loop Log Modulus’ Industrial and Engineering [30] Bezergianni S, Georgakis C. (2000). ‘Controller performance
Chemistry Research, 37, pp. 1045-1050. assessment based on minimum and open-loop output
[6] Thyagarajan, T., Yu, C-C, Huang, H-P. (2003) ‘Assessment variance’. Control Engineering Practice, 8 (7), pp. 791-797.
of controller performance: a relay feedback approach’. [31] Ko, B-S. and Edgar, T.F. (1998). ‘Assessment of achievable
Chemical Engineering Science, 58, pp. 497-512. PI control performance for linear processes with dead time’.
[7] Marlin, T.E. (2000). ‘Process control: designing processes and Proceedings of the American Control Conference, Evanston,
control systems for dynamic performance’. McGraw – Hill; IL, USA, pp. 1548-52.
2nd Edition, 2000. [32] Eriksson, P-G., and Isaksson, A.J., (1994). ‘Some aspects of
[8] Swanda, A.P., and Seborg, D.E. (1999). ‘Controller control loop performance monitoring’. Proceedings of the 3rd
performance assessment based on setpoint response data’. IEEE Conference on Control Applications, Glasgow,
Proceedings of the American Control Conference, Scotland, pp. 1029-1034.
Piscataway, NJ, USA, pp. 3863-7. [33] Horch, A., and Isaksson, A.J. (1998). ‘A modified index for
[9] Horch A, Stattin A. (2002). A complete practical control performance assessment’. Proceedings of the
implementation of a method for step response performance American Control Conference, Philadelphia, Pennsylvania,
assessment. Proceedings of the IEE Seminar Control Loop pp. 475-83.
Performance Assessment. London, UK, pp. 4/1-5. [34] Isaksson AJ. (1996). ‘PID controller performance
[10] Ruel, M. (2002). ‘Learn How To Assess And Improve Control assessment’. Proceedings of Control Systems '96. (Preprints),
Loop Performance’, ISA, Chicago.TOP Control Montreal, Que., Canada, pp.163-70.
www.topcontrol.com. [35] Huang B. (1999). ‘Performance assessment of processes with
[11] Huang, H.P., and Jeng, J.C. (2002). ‘Monitoring and abrupt changes of disturbances’. Canadian Journal of
assessment of control performance for single loop systems’. Chemical Engineering, 77 (5), pp. 1044-1054.
Industrial and Engineering Chemistry Research, 41, pp. 1297- [36] Huang B. (2002). ‘Minimum variance control and
1309. performance assessment of time-variant processes’. Journal
[12] Stanfelji, N., Marlin T.E., MacGregor, J.F. (1993). of Process Control, 12 (6), pp. 707-719.
‘Monitoring and diagnosing process control performance: [37] Venkatesan G. (2002). ‘An algorithm for minimum variance
The single loop case’. Industrial Engineering Chemistry control of a dynamic time-delay system to reduce product
Research, 1993, (32), pp. 301-314. variability’. Computers and Electrical Engineering, 28 (3),
[13] Hagglund, T., (1999). ‘Automatic detection of sluggish pp. 229-239.
control loops’. Control Engineering Practice, 7, pp. 1505- [38] Kucera V. (1999). ‘Minimum variance control: A homage to
1511. Peterka’.International Journal of Adaptive Control, 13 (6), pp.
[14] Belanger, P.W. and Luyben, W.L. (1996) ‘A new test for 433-449.
the evaluation of the regulatory performance of controlled [39] Qin, S.J. (1998). ‘ Control performance monitoring – a review
processes’. Industrial and Engineering Chemistry Research, and assessment’. Computers and Chemical Engineering, 23,
35, pp. 3447-3457. pp.173-186.
[15] Luyben, W.L. and Luyben, M.L. (1996). ‘Essentials of [40] Tyler, M.L., and Morari, M., (1995). ‘Performance
process control’. McGraw-Hill Science/Engineering/Math. Monitoring of Control Systems using Likelihood Methods’.
[16] Astrom, K.J., Hagglund, T. (1984) ‘Automatic tuning of Proceedings of the American Control Conference, Evanston,
simple regulators with specifications on phase and amplitude IL, USA, pp. 1245-9.
margins’. Automatica, 20 (5), pp. 645-651. [41] Zhang, B. and Ho, Y-C. (1989). ‘Variance reduction for
[17] Chiou, R-C. and Yu, C-C. (1993) ‘Monitoring procedure for likelihood-ratio method’. Proceedings of the 28th IEEE
intelligent control: on-line identification of maximum closed Conference on Decision and Control, 1, pp. 145 –150.
loop log modulus’. Industrial and Engineering Chemistry [42] Li Q, Whiteley JR, Rhinehart RR . (2003). ‘A relative
Research, 32, pp. 90-99. performance monitor for process controllers’. International
[18] Elliot, T.R., and Luyben, W.L. (1995). ‘Capacity-based Journal of Adaptive Control, 17 (7-9), pp. 685-708.
economic-approach for the quantitative assessment of process [43] Zhong, Lei. (2003). ‘Defect distribution model validation and
controllability during the conceptual design stage’. Industrial effective process control’.Proceedings of Spie - the
and Engineering Chemistry Research, 34 (11), pp. 3907- International Society for Optical Engineering, USA, pp.31-8.
3915. [44] Mosca, E., and Agnoloni, T. (2003). ‘Closed loop
[19] Elliott TR, Luyben WL, Luyben ML. (1997). ‘Application monitoring for early detection of performance losses in
of the capacity-based economic approach to an industrial- feedback control systems’. Automatica, 39, pp.2071-2084.
scale process’. Industrial and Engineering Chemistry [45] Hagglund, T. (1995). ‘A control loop performance monitor’.
Research, 36 (5), pp. 1727-1737. Control Engineering Practice, 3, pp. 1543-1551.
[20] Elliott TR, Luyben WL. (1996). ‘Quantitative assessment of [46] Huang, B., Thornhill, N.F., and Zhang, H., (2001).
controllability during the design of a ternary system with two ‘Detection of multiple oscillations in control loops’. Journal
recycle streams’. Industrial and Engineering Chemistry of Process Control, accepted for publication.
Research, 35 (10), pp. 3470-3479. [47] Chang, C.L., Liu, A.S., Chen, Y.T. (1993). ‘Dynamic
[21] Kendra, S.J., and Cinar, A. (1997). ‘Controller performance performance monitoring-system of taiwan-power-company’.
assessment by frequency domain techniques’. Journal of IEEE Transactions on Power Systems, 8 (3), pp. 815-822.
Process Control, 7(3), pp. 181-194. [48] Stenman A, Gustafsson F, Forsman K. (2003). ‘A
[22] Desborough, L. and Harris, T.J. (1992). ‘Performance segmentation-based method for detection of stiction in
assessment measures for univariate feedback control’. control valves’. International Journal of Adaptive Control, 17
Canadian Journal of Chemical Engineering, 70, pp. 1186- (7-9), pp. 625-634
1197. [49] Wallen, A., (1997). ‘Valve diagnostics and automatic tuning’.
[23] Harris, T.J. (1989). ‘Assessment of control loop Proceedings of the American Control Conference,
performance’. Canadian Journal of Chemical Engineering, Albuquerque, New Mexico, pp. 2930-2934.
67, pp. 856-861. [50] Thornhill, N.F., and Hagglund, T. (1997). ‘Detection and
[24] Spring R. (1996). ‘Minimum variance as a benchmark for diagnosis of oscillation in control loops’. Control Engineering
feedback control’. Tappi Journal 79 (8), pp. 97-101. Practice, 5, pp. 1343-1354.
[25] Thornhill, N.F., Shah, S.L., and Huang, B. (2000) [51] Horch, A., (1999). ‘A simple method for oscillation diagnosis
‘Controller performance assessment in setpoint tracking and in process control loops’. Proceedings of the 1999 IEEE
regulatory control’. International Journal of Adaptive Control International Conference on Control Applications,
and Signal Processing, 17, pp. 709-727. Piscataway, NJ, USA, pp. 807-12.
[26] Harris, T.J., Seppala, C.T., and Desborough, L.D., (1999). [52] Taha, O., Dumont, G.A., and Davies, M.S., (1996).
‘A review of performance monitoring and assessment ‘Detection and diagnosis of oscillations in control loops’.
techniques for univariate and multivariate control systems’. Proceedings of the 35th IEEE Conference on Decision and
Journal of Process Control, 9 (1), pp. 1-17. Control, Kobe, Japan, pp. 2432-2437.
[27] Grimble, M.J. (2002). ‘Controller performance benchmarking [53] Hugo, A.J., (2001) ‘Process controller performance
and tuning using generalised minimum variance control’. monitoring and assessment’. Hydrocarbon processing, April.
Automatica, 38, pp. 2111-2119. Control Arts website, www.controlartsinc.com.
[28] Huang, B. and Shah, S.L. (1999). ‘Performance assessment [54] Control Arts Inc. (2000), Performance assessment tool-kit,
of control loops: Theory and applications’. Springer-Verlag; www.controlarstinc.com.
ISBN: 1-85233-639-0. [55] Honeywell (2000), Loop Scout, www.IAC.Honeywell.com.
[29] Desborough, L. and Harris, T.J, (1993). ‘Performance [56] Matrikon, (2000) Process Doc, www.Matrikon.com.
assessment measures for univariate feedforward/feedback [57] Aspen watch, (2000). www.aspentech.com.
control’. Canadian Journal of Chemical Engineering, 71, pp. [58] ACTC, Advanced Control Technology Club, www.isc-
605-616. ltd.com/actclub.
[59] ExperTune. www.expertune.com