Process Control: Potential Benefits and Wasted Opportunities

M.L. Brisk

Department of Chemical Engineering

Monash University
Melbourne, Au.
e-mail: [email protected]

Abstract course, manufactured by the engineer’s employer. On the

other hand, end users of process control in industry, whilst
Seventeen years ago the University of Sydney Warren hopefully having an image of a well-equipped control room,
Centre’s study demonstrated that introducing advanced but more often than not having to think of a stand-alone
control systems could potentially provide bottom-line controller with a level of sophistication not beyond simple
benefits in the range of two to six percent of operating feedforward, should have quite a different immediate
costs. Since then the process industry world-wide has thought: process control’s beneficial impact on
reported many successful applications of a range of profitability. Industry’s natural focus should be on the
advanced control technologies which achieved, or obvious commercial advantages of the technology.
exceeded, those benefits. There are now a plethora of
multivariable predictive controls, inferential sensors, These advantages have been well demonstrated over many
dynamic modelling tools, fuzzy logic algorithms, and neural years. Over the last two decades, the process industry
networks, and we know how to use them. So does it follow world-wide has reported many successful applications of a
that industry is consistently achieving the best possible range of improved and advanced control technologies
process control performance and realizing all those which have achieved significant economic benefits, and a
potential benefits? The answer, unfortunately, is a few further examples, if indeed any are needed, are
depressing no! Recent studies have shown that only about a presented below. There is an overwhelming supply of
third of industrial controllers are achieving acceptable technology available, with a plethora of multivariable
levels of performance, and this is especially true at the most predictive controls, inferential sensors, dynamic modelling
basic levels of control, needed to underpin the advanced tools, fuzzy logic algorithms, and neural networks, and we
controls. Indeed, there are reported trends which suggest the know how to use them So it should follow that process
gap between desired and actual controller performance is industry is consistently achieving the best possible process
widening! This paper revisits the Warren Centre study and control performance and realizing all the potential benefits.
its benefits analysis methodology, and addresses the issues Sadly, there is a great deal of evidence that this is simply
of actually achieving those benefits. This is shown to be not the case. Much of this evidence identifies, not so much
more than just the introduction of advanced control an unwillingness to introduce today’s advanced
technology. There is a need to re-focus on the humble – but technologies, but a neglect of the performance of more
vital – PID controller, and the often overlooked issue of basic controls. The theme I wish to pursue is that there are
ensuring continued maintenance of the best performance of real benefits to be had, but industry needs to take a hard
existing controls, if industry is to stop wasting the look at how it is using all its process control technology,
opportunities to gain those benefits. and especially the humble – but vital – PID controller which
underpins advanced controls, as well as itself contributing
1 Introduction to that gain in economic performance which should be the
main goal.
A little over a decade ago, during a talk at the IFAC
Congress in Sydney in 1993 [1], I asked the assembled 2 Control delivers benefits
delegates what they first thought of when confronted with
the words “process control”. I suggested that an engineering Why do we use process control? We control processes to
academic researching in control might well focus on overcome the effects of disturbances, and the inherent
complex non-linear models, or advanced mathematical variability that occurs in all processing operations. We
algorithms, which can at times be quite formidable, as many know that by keeping processes close to a steady state, or
students will know. An engineer with an instrument and by ensuring they accurately follow desired trajectories, we
control system vendor company might visualize a large can achieve a number of key objectives:
control room with the latest DCS installation, preferably of
• Safer operation and reduced environmental impact. Garden Isle (UK) who applied improved PID with some
feedforward control to the process for making frozen
Keeping process operation steady will always help reduce French fries. They reported more than 50% reduction in
incidents which may create hazardous conditions, or standard deviation of product quality (assessed by moisture
undesirable emissions to the environment. One does not and fat content, colour, texture, and size distribution).
seek to put a monetary value on ensuring personnel safety,
although it unquestionably exists. The same is true of • Agility gains
preventing environmental incidents, although the fines
From the turn of the century, and into the immediate future,
imposed by protection agencies for violations are very real!
new factors governing processing profitability include
• More sustainable manufacturing. manufacturing flexibility, customer responsiveness, and the
related need to reduce working capital by processing to
Better control can achieve more efficient raw material and order, not to stock. Indeed, some visions for the process
energy usage per unit of product. Apart from the financial industry [8] predict that in less than thirty years a chemical
benefits this brings, there is waste reduction, and the better plant will have to be designed to produce product quantity
conserving of non-renewable resources. Increasingly, equal to the smallest order size, even if that is one drum of
modern sustainable plant designs with their high degrees of chemicals! This requires an agile processing capability,
material and energy recycle, are so complex that high with responsive plant exploiting the full potential that well
performance control systems are mandatory in order to performing control can provide.
operate them at all.
2.1 The Warren Centre study
• Improved bottom line returns. Seventeen years ago, in 1987, the Warren Centre for
In business terms, the key benefits from the good use of Advanced Engineering at the University of Sydney
control are economic. They arise from three main sources: undertook a major study which set out to show that the
efficiency, quality and agility gains. introduction of advanced process control could lead to
significant bottom-line benefits [2][3]. The study, carried
• Efficiency gains. out by a team of engineers from both industry and
These have been achieved from the very earliest days of academia, was led by Dr Tom Marlin, then from Stone and
process control application, and certainly for advanced Webster (now Professor of Chemical Engineering at
control, from the seventies onwards. They are still being McMaster University, Canada). The study developed and
achieved today. In the area of improved raw material yields, refined methodologies to identify opportunities for
Suncor Extraction, processing Canadian oil sands, in 2002 improved control, and to predict the economic benefits that
reported a 7% increase naphtha recovery by introducing should follow. It then employed the tools in seven
improved basic regulatory control, with a three month Australian plant case studies to demonstrate the premise
payback! [4] that improved control delivers real economic benefits. The
studies, covering conventional oil refining and
In the area of energy savings, in 1998 Hickson & Welch petrochemical plants, a coal-fired power station, a sugar
PLC, a UK detergent manufacturer used standard PID refinery, alumina production, and a municipal waste water
algorithms as building blocks, with some limited treatment plant identified potential annual economic
“advanced” features, mainly feedforward, to achieve 30% benefits of from 2 to 6% of operating costs.
energy reduction by eliminating the need for a re-
distillation step [5]. The project also saved some ǧ300,000 Although the project focused on the benefits from
pa in raw material costs, and reduced volatile organics introducing advanced control, I have used the term
emissions by more than 90%! improved control. Advanced control is usually understood
to mean that process knowledge, generally encapsulated in
Efficiency gains also apply if one can increase throughput a process model embedded in the control strategy, is used,
with existing equipment, avoiding new capital investment. together with appropriate hardware and software platforms,
In 2000, the introduction of model predictive control (MPC) to implement on-line control. Whilst new advanced controls
together with upgrading the performance of basic PID often lead to dramatic savings, and were the bases for most
controls, increased production at a Noranda zinc roaster in of the benefits identified in the case studies, a proportion of
Canada. The company reported gains worth $900,000 a those benefits arose from simply improving existing more
year, excluding the savings from not having to install new basic controls. All of them were only achievable if all the
plant [6]. basic controls could be assumed to perform according to
• Quality gains their design intent.

Maintaining consistent product quality is a key factor in 2.2 Other benefits studies and estimates
ensuring and potentially growing a company’s market Shortly after the Warren Centre project, the DuPont
share. From the early nineties onwards, in an increasingly Company commissioned a consultant (A T Kearney Inc) to
competitive and often global marketplace, control focusing undertake a benchmarking study of the way leading
on product quality became important. An interesting companies were exploiting advanced process control
example from the food industry was reported in 2002 [7] by technology [9]. The results, published in 1991, found the
so-called “best of the best” companies were achieving 15% Figure 1A shows an example of a reduction in variability
savings on operating costs, with an average figure in achieved by implementing improved control of a single
DuPont itself of 7%. The company concluded that it had an variable. In some cases this reduction could account for the
8% “incentive” to chase. In its first year of a renewed thrust entire economic benefit. Examples where cost or quality
in process control it achieved US$6M in benefits. penalties for moving either way from the desired operating
level occur include effluent pH control, boiler flue gas
A year later, in 1992 the ICI group of companies undertook
excess air control, and paper thickness control. There may
a global benchmarking study within its divisions and
also be less tangible benefits from steadier operation, such
subsidiaries [10][1]. Three different approaches to
as reduced equipment wear. However, in general variability
estimating the potential benefits of advanced control:
reduction alone does not achieve the major benefit.
extrapolation of the (limited) existing ICI applications; the
Warren Centre methodology applied to a range of company Figure 1B presents the more usual case where variability
plants; and a scaling of the DuPont data, all led to very reduction allows operators to move the mean operating
similar conclusions. The ICI Group world-wide had the level closer to a constraint. In this case there is some
potential for at least 7% savings on operating costs (ǧ400M penalty associated with violating the constraint, but also a
pa at the time) from improved control. ICI achieved ǧ4M in cost incurred for operating away from the constraint.
the first year (1993) in the UK alone, and identified a Tighter control reduces this cost. A simple example would
further potential of ǧ16M in that year. be avoiding product giveaway by not over-purifying a
stream, whilst still adhering to a product specification. Once
A discerning reader could well note that all these studies
the extent of the movement of the mean is known (the
and benefit estimates were more than a decade ago, and
“improvement” shown in Figure 1B) it is straightforward to
perhaps are now less relevant. A strong counter to that view
calculate the benefit in financial terms. The issue is how to
was presented by a senior spokesperson for the Shell Group
estimate the amount of movement a priori.
at an automation forum in Paris in September 2002.
Reflecting on 15 years experience in the implementation of 2.4 Predicting the benefits
advanced process control (APC) in the petrochemical The Warren Centre project presented a number of possible
industry, Rouby [11] stated that APC can increase the plant approaches to estimate benefits from improved control
margin by 10 to 20%, and its application is generating introduced to existing plants. For the common case shown
savings of US$300M pa for the Shell Group, with potential in Figure 1B, the approach is illustrated in Figure 2.
to rise to US$500M pa. Good process control is still
delivering benefits and will continue to do so! Predicted with improved control
“Normal”
2.3 Delivering the benefits distributions M2 - M1 = 3*(σ1 - σ2)
How does control deliver these economic benefits? It does

Constraint/target
so by reducing the variability of the controlled variables. Original If σ2 = α * σ1
plant α < 1

A value for α allows

σ1
3*σ benefits estimate by
3*σ
3* σ2 converting M2 - M1
M1 $ M2 to $

Figure 2: Predicting improvement

Data collected for a process variable during the “base case”

study are shown distributed about the original mean M1. A
Figure 1A: Variability reduction normal distribution is shown, but the method can be adapted
to other distributions. It is assumed that M1 will be
maintained three standard deviations (3σ1) from a
constraint by the operators to meet quality constraints
99.7% of the time. The value of a mean under better control
(σ2 < σ1) can then be predicted, and the movement in the
mean used to calculate the expected benefit. For other
distributions, integration under the curves can be applied.

How does one predict σ2 ? The most widely adopted

approach is to assume α in Figure 2 is 0.5. For the
introduction of advanced control to an existing plant, the
Figure 1B: Moving the mean (constraint pushing) assumption of 50% reduction in variability is, if anything,
often conservative, and justifiable. For other improved
control applications one may have to seek relevant, variables at the setpoints provided by the MPC, as shown
comparable experience, and exercise some judgment. If a schematically in Figure 4.
plant model is available, and there is some knowledge of
the nature of process disturbances and measurement noise, a
lower bound on α can be estimated from minimum variance A dvanced
theory [12] and then increased based on realistic judgment.
C o n tro l
The Warren Centre project benefits analysis methodology M PC
also presented other approaches to estimating the
improvement. One can identify historically achieved
consistent best performance, and seek to achieve this as the S e tp o in ts ↓
new mean. If a process model is available simulation, ↑ M e a s u re m e n ts
preferably dynamic, although even steady state may be
useful for simpler processes and controls, is a powerful
approach. Finally, one can monitor the performance of S IS O S IS O
existing controls and using a benchmarking approach to DDC DDC
assess the improvement potential. This is discussed further lo o p s lo o p s
below.
2.5 Achieving the benefits PR O C ESS PLANT
It is clear that there is ample evidence that there are real
benefits to be had from improved process control. We have
well-developed methodology to estimate these benefits in Figure 4: Implementing Advanced Control
order to justify the investment in better control technology,
and there is no question about the availability of a wide These regulatory controllers almost invariably use a version
range of very applicable technology. So it should follow of the PID algorithm. If the basic controllers do not
that industry is consistently achieving the best possible perform, then the advanced control cannot deliver its
process control performance and realizing all those benefits. The issue is wider, however, than basic control just
potential benefits. Unfortunately, there is a great deal of being a potentially weak link between advanced control and
evidence that this simply is not the case. We can – and benefits. A 1995 study by Monsanto, in collaboration with
should – do very much better. Perhaps surprisingly, it is at several other chemical companies [14], sought to predict the
the basic regulatory level, rather than at the advanced savings achievable by adopting “best practice” in each of a
control level where there appear to be the biggest problems. number of categories of process control from basic through
to advanced. Table 1 presents selected major categories,
Figure 3 shows the lowest three levels of the commonly with savings expressed as a percentage of production costs.
presented pyramid representing the hierarchy of process
control. The missing upper levels deal with integration of Table 1 Savings in production costs from
data and information across a multi-plant site or an categories of process control
enterprise, and are outside the scope of this discussion. % cost
Category
Performance savings
Final control device performance and
ADVANCED CONTROL Regulation 1.5
basic loop tuning
BASIC PROCESS CONTROL Measurement Unit operations control 0.8
Advanced control 1.4
PLANT LEVEL DATA COLLECTION On-line optimisation 0.5

Figure 3: Process control hierarchy The total savings, including advanced advisory systems,
process data access and production management control
The pyramid shows clearly that advanced control, which totalled 7.4%, consistent with the estimates mentioned
focuses on improving performance, usually in economic earlier. Note that 20% of this total comes from the most
terms, requires a sound foundation of basic regulatory basic control system performance issues.
control. Everyone knows this, but somehow it seems to be
overlooked, despite the fact that advanced control is The process industry has a huge installed base of PID
increasingly being used. In 1997 there were more than 3000 regulatory controllers. In 1999 there were approximately
installations of some form of MPC worldwide [13], and no three million regulatory controllers in the continuous
doubt there are double that number today. But the most process industries in the USA, and fewer than 3% of these
common form of MPC implementation relies on existing were not PID [15]. So inadequate PID controller
SISO regulatory controllers to maintain the process performance affects all the industry’s attempts to secure
benefits from process control.
Sadly, inadequate performance abounds! Studies in Canada various industrial control systems, ranging from sub-
and the USA in 1993 by Bialowski [16] and Ender [17] of standard to world best practice. The metrics were selected
several thousand control loops showed that only about one for ease of measurement by non-specialist personnel, and
third delivered an acceptable level of performance. Factors sought to minimize subjective viewpoints, whilst being
ranged from poor tuning (even leaving the delivered applicable to both basic and advanced controls in a range of
default parameters of Kc = 1, 1/Ti = 0.1 repeats/min set in different processing environments. Some examples, with
DCS loops!), and loops left permanently in manual, to the then estimates for best practice are:
incorrect valve sizing, valve sticking, and sensor problems.
Ender [17] maintained that 30% of loops in automatic • Control infrastructure
actually showed increased variability over manual control o loops/operator >130
due to some of these factors. o loops/control engineer <150
• Level of control technology
By coincidence, the author, who was unaware at the time of o % intermediate
the studies reported above, carried out a benefits analysis
(ratio, cascade, feedforward) >50
investigation at a large North American plant in 1993. The
study report, which identified 80% of installed loops at the o % advanced >25
plant were either in manual, or oscillating badly, was • Performance
initially dismissed by management as lacking credibility, o % in manual <1
although it was subsequently verified. o % poor quality 0
o % operator intervention time <2
One would like to feel the industry has improved over the
past decade, especially since academic control research has The full set of benchmarks are given by Brisk and Blackhall
developed tools for measuring performance on line [12]. [10].
Yet Miller’s study of 26,000 PID loops in 2000 gave Of course the fact that a plant may rank poorly does not
virtually identical results, with only 32% rated as automatically imply that there are economic benefits to be
“acceptable” or better, and 36% operating open-loop had by improving the control. Plant managers were also
because of problems [15]. required to undertake a rapid “back of the envelope”
estimate of possible opportunities and their value, using a
The process industry may well be embracing advanced short checklist. Items included issues such as “could energy
controls such as MPC with enthusiasm, and rightly so costs be reduced?”; “is off-specification product
because of the perceived benefits, but it is wasting those produced?”, “could warehouse inventory be reduced?”, and
opportunities, as well as the potential of control in general “is the frequency effluent fails to meet regulations
by neglecting to ensure the performance of basic PID unacceptable?”. ICI used this tool to increase management
controls. awareness of the beneficial scope for improving control
systems.
3 Ensuring basic control performance In 1997 the Process Control Society of The Institution of
Engineers Australia applied the technique in a survey [18]
There are two major activities which management should of a range of Australian plants, ranking the results on the
undertake to ensure they are achieving the potential of all ICI scale of 0 (poor, “neglecting the potential of well-
their controls, as well as guaranteeing the benefits of the proven commercially available technology”) to 100
advanced controls. These are firstly, assessing (excellent, “leading control technology development
appropriateness and performance of the current level of proactively”). Whilst the focus was on the use of advanced
control application in their plants – both basic and control, the full spectrum of control was included. The
advanced. In other words, benchmarking their existing use results are summarised in Table 2.
of control. And secondly, continuously monitoring the
performance of installed controls to maintain it at its peak.
3.1 Benchmarking Table 2 1992 Australian benchmark survey
The concept of benchmarking, in this case auditing one’s Potential
own use of control technology and comparing it with world Number Overall
Industry benefits $M
best practice with the aim of identifying worthwhile of plants Score
pa
improvements, is well understood. The application,
however, is usually seen as too demanding of time and Chemical 12 55 17
resources, so it is neglected. Wastewater 12 51 14
ICI in the early nineties developed a rapid, simple approach
Minerals
to permit a low cost best practice self assessment [10]. Plant 7 50 70
processing
managers were asked to rate their control facilities against
a set of metrics developed from comparative reviews of
The scores are depressing! The industries all rated just and speed of recovery after a disturbance, measured relative
average (“applying control technology developments to recorded “good” historical performance. Modern
reactively” [10]), yet they identified significant potential performance monitoring software is able to look at
benefits. controller performance trends compared to theoretical
minimum variance control. This not only identifies a
I would like to offer a challenge to my readers: have you
decline in performance, but can be used to estimate the cost
benchmarked your plant, and are you confident you are
of that decline through the benefits analysis approach noted
better than “average” in your beneficial application of
earlier. The industry process control end user of my
control technology?
introduction will certainly show real interest in this KPI !
3.2 Monitoring and maintenance
Even if you are confident of using control technology
4 The “take-home” messages
proactively to achieve real benefits, the question remains:
how do you know your control systems are performing We now have ample indisputable evidence, collected over
well? The surveys mentioned above strongly suggest most very many years in all parts of the world that improving
systems are not performing as they should. To achieve process control yields real, predictable economic benefits. It
consistently high levels of performance from any equipment is also well-established that advanced control can deliver
implies an adequate level of maintenance, and that in turn major economic gains. However, the latter is only true if the
implies a knowledge of when maintenance is required, or at underlying basic regulatory control is maintained at peak
least some form of regular scheduled maintenance. performance. Further, there are significant gains to be had
Consider the simple example of basic PID loops. How often just from improving, and maintaining the performance of
are they checked and tuned, and why? Problems can arise regulatory control alone.
even in well-designed control loops for a variety of reasons, So if safe environmentally responsible operation, and
ranging from a need for re-tuning due to process changes, to shareholder value mean something to you, I urge you to
difficulties with the sensor, actuator or valve operation consider benchmarking your existing control applications
which can occur in an unpredictable fashion. The chances and performance, and to implement proactive monitoring of
are nothing happens until the operators complain very this performance to keep it at its peak. Gain those benefits,
loudly. Even then, not much may happen because of limited and do not waste your opportunities.
personnel resources. Then the operators will switch to
manual control, and in many cases the economic benefits References
drop to zero. Certainly, the value of advanced controls
which rely on well performing regulators will fall. Ideally, [1] M L Brisk, Process control: Theories and profits, 12th
one would want to look at the PID loops as soon as there is IFAC Triennial World Congress, Sydney, Australia,
evidence that their performance is declining, but in Vol. 7, pp.~241 – 250, 1993
maintenance terms, control loops are almost always “run to [2] T E Marlin, G W Barton, M L Brisk & J D Perkins,
failure”, and treated by “breakdown maintenance”, if then! Advanced Process Control: Project Report and
It may be too idealistic to hope that control loops could be Technical Papers, Warren Centre for Advanced
on a scheduled maintenance list, and perhaps this is not the Engineering, Sydney, 1987
best way to proceed in any case because of the random [3] T E Marlin, G W Barton, M L Brisk & J D Perkins,
nature of faults. But there is no question that controller Benefits from process control: results of a joint
tuning and control loop support need maintenance that is industry-university study, J Process Control, Vol 1,
proactively data-driven, not reactively complaint-driven, as pp.~68 – 73, June 1991
does any other important plant item. It is commonplace to
monitor bearing temperatures, heat exchanger coefficients [4] Suncor Energy, Annual Report, 2002
and compressor vibration signatures to provide early [5] UK Department of Trade & Industry, Advanced
warning of possible problems. Why should we not monitor Control Technology Transfer Program Case Studies,
performance indicators to check the health of control loops? http://www.dti.gov.uk/actt/casestudies/casestu1.htm
Why indeed, given that modern intelligent digital field [6] M Grenier & L Lariviere, A quest for a perfect roast,
instrumentation, including transmitters and digital valve http://www.controlmagazine.com/Web_First/CT.nsf/A
controllers, are capable of monitoring their own “health”, as rticleID/DFUO-5CYNTD , Oct 2002
well as that of the process. But even with more traditional
instrumentation, a number of key performance indicators [7] G A Montague et al., Advanced control methods for
(KPI) for control loops exist, and are readily accessible with French fry quality improvement, Comp & Control Eng
today’s control technology. These include simple items J., Feb 2002
such as service factors ( % time spent in automatic); [8] R S Benson, From single loop to cohesive process
frequency of operator changes (auto to manual; cascade control: the future, First African Control Conference,
switched off). They also include more sophisticated, but Cape Town, SA, Dec 2003
more informative metrics such as the extent of oscillation
[9] J F Swallow, Best of the Best: Leaders in Advanced
Process Control, CPC IV, Texas, Feb 1991
[10] M L Brisk & N W Blackhall, Benchmarking for best
practice in process control, CHEMECA 95
Conference, pp.~74-79, Adelaide, Sep 1995
[11] Jean-Luc Rouby, 15 years of APC implementation
experience in the petrochemical industry, Advantic
Advanced Automation Forum, Paris, Sep 2002
[12] B Huang & S L Shah, Performance assessment of
control loops: Theory and Applications, Springer
Verlag, London, 1999
[13] S J Qin & T A Badgwell, An overview of industrial
model predictive control technology, Chemical
Process Control V, pp.~232-256, Tahoe, AIChe, 1997
[14] Emerson Process Management, The process
variability challenge in 2002, The EnTech Report, Vol
14(2), pp.~1-5, 2002
[15] L Desborough & R Miller, Increasing customer value
of industrial control performance monitoring –
Honeywell’s experience, Chemical Process Control
VI, pp.~169-189, Tucson , Jan 2001
[16] W L Bialowski, Dream versus reality: A view from
both sides of the gap, J Pulp & Paper Canada, Vol
94(11), pp.~19-27, 1993
[17] D B Ender, Process control performance: Not as good
as you think, Control Eng., Vol 40, pp.~180-190, 1993
[18] I Tanner, P Signal & R Newell, Advanced process
control trends survey, Process Control Society,
IEAust, Australia, 1997