You are on page 1of 9

Process Control Performance

for Efficient Manufacture -


Measures & Metrics
Performance Solutions Division

By W. L. Bialkowski, President, EnTech Control

ABSTRACT

In many unit operations the focus is on the advanced process control (APC) system, while the
performance of the base regulatory control layer is ignored, often resulting in performance
decay until eventually the APC is turned off. Many control practitioners ignore tuning methods
preferring to “tune-by-feel”, and fewer still consider control valve response as part of the
control system performance. Moreover, most loop tuning concepts ignore the multivariable
dynamics needed to optimize overall process performance. The paper reviews how control
loops shift process variability, and its impact on process performance. It reviews tuning
methods from a process variability standpoint, and presents performance concepts designed
to optimize efficient process manufacturing by coordinating the tuning of the base layer
regulatory loops. Performance measures are presented and performance metrics are
recommended.

INTRODUCTION

Process control should enable chemical process plants to produce uniform product safely and
efficiently. It has long been assumed that controller tuning should be “as tight as possible” and
that this would somehow translate into process performance. Findings of process variability
audits have shown this not the case. A new understanding of process variability in an
integrated plant environment has emerged which redefines the requirements for process
control performance. The paper attempts to put this in perspective. It reviews the concept of
controller induced variability shifts, discusses final control element performance, and process
variability pathways. Early tuning methods are compared to Lambda Tuning, which is put
forward as the tuning method to achieve uniform manufacturing. Finally, control loop
performance measures and metrics for uniform manufacturing are presented.

PLANT CONTROL STRUCTURE – FROM BASE LEVEL LOOPS TO APC

The control system pyramid base-layer consists of the final control elements and field devices.
Next come the base-level control loops, cascade loops and feedforward controllers, with the
advanced process control (APC) algorithms such as dynamic matrix control (DMC) [1] or other
versions of model predictive control (MPC) [2] residing at the top of the pyramid for plant
control optimization. The MPC dynamics play out at the speed of the slowest variable (typically
minutes or hours), with all variables driving towards the prediction horizon together. The lower
level loops on the other hand engage the fluid flows and have dynamics that play out in
seconds and minutes. MPC is multivariable in nature, and most people consider this as the

Presented ISA Houston, September 2001


Process Control for Efficient Manufacture - Measures & Metrics Page 2 of 9

only place where the multivariable nature of the process needs to be considered. The attitude
towards the base level control loops is often that they require little attention, especially if the
dominant dynamics of the process are slow (e.g. as in a distillation column). However, plant
variability audit and Lambda Tuning experience emerging from the pulp and paper industry [3,
4, 5] leads to the conclusion that significant advantages can be gained by careful selection of
final control elements, as well the design and tuning of the lower level loops. These steps can
drastically reduce the degree of process variability propagation through the process. Good
integrated engineering design modifications can eliminate sources of variability and variability
pathways. This should leave the MPC to handle a simplified plant with easier dynamics and
higher constraint values that can operate more economically with better long-term
sustainability.

THE CONTROL LOOPS AND VARIABILITY SHIFTING

Variability enters a process with the raw materials and travels through the process network.
Control loops attempt to reduce the variability of the process measurement (PV) by
manipulating the controller output (O/P). On manual the O/P is constant, while the PV drifts.
On automatic the O/P is active while the PV is held near setpoint (SP). Clearly, the controller
attempts to shift the variability from the PV to the O/P. How much variability is shifted is a
function of the controller tuning, the process deadtime, and the state of the final control
element. This concept is well developed in Downs and Doss [6]. The fact that the controller
shifts the variability elsewhere raises the key question: will this variability harm the process
operation there, or will the impact be benign? The controller action impacts the process by
causing a flow of mass or energy to change (e.g. grams/s, lbs/hr, gpm, BTU/hr, Watts,
Joules/s). For example, pH can be controlled by causing the flow of acid to change. As
discussed later, this flow coupling is the key element that causes variability transmission
through the process network to occur.

FINAL CONTROL ELEMENT SMALL SIGNAL NONLINEAR DYNAMICS

How the flow actually changes as the


55
Flow valve moves in response to the O/P
change determines the control
53 performance to a great extent. The
controller expects the flow to track the
controller output faithfully in a repeatable
51
%

Controller Output and timely manner. If this is not the case


the control will likely cycle and increase
49 variability. Recent attention has been
paid to the small signal nonlinear
performance of control valves and other
47 final elements [7, 8]. It is now well known
0 500 1000
that nonlinearities such as backlash,
Time seconds deadband, hysteresis and stiction will
Figure 1 – Typical Limit-Cycle caused by a result in limit cycles as illustrated in
control Figure 1 and cause the control loop to
valve with backlash / stiction or deadband increase variability.
Process Control for Efficient Manufacture - Measures & Metrics Page 3 of 9

INTEGRATED PROCESS DESIGN / CONTROL CONCEPT

The notion of the process as a network of vessels and piping is developed in Downs and Doss
[6], and leads to the development of “Integrated Process Design and Control” concept. The
goals are to eliminate variability sources where possible, to effectively attenuate variability
through process design, and to shift the remaining variability through process control to places
in the process network where it can dissipate benignly without impacting final product
uniformity or operating constraints. The topic includes: variability sources, variability sinks and
variability transmission pathways through the process.

VARIABILITY SOURCES, SINKS AND PATHWAYS

Process variability sources include raw materials, chemical additives, cycling control loops,
pressure pulsations, multi-phase flow disturbances, and discontinuous or batch operations as
part of a continuous process. The best course of action is to eliminate variability sources if
possible [3, 4]. Variability sinks occur naturally because process dynamics tend to “average-
out” fast variability. The amount of attenuation at a given frequency depends on the process
time constant. Typical
variability sinks include process
mixing, agitation and recycle
paths, which tend to smooth out
variability in concentration or
composition, while thermal
mass smoothes out variability in
heat transfer and temperature.
Control loops are not variability
sinks. Instead they selectively
attenuate slower PV variability
by shifting it via the O/P to
Figure 2 – Variability shifts due a closed vessel
elsewhere. The exact amount
of attenuation at a given
frequency depends on the
speed of response of the
control loop. Control loops with
process deadtime or other
difficult dynamics (e.g. wrong-
way response) will resonate at
a frequency above the loop cut-
off frequency and will increase
variability in this region. So
even with a perfect actuator,
control in the presence of
Figure 3 – Buffer storage variability transfer deadtime will increase
variability at frequencies above
the loop cut-off frequency [9].
Variability pathways determine
where process variability goes in the process network. They include controller induced flow
changes in or out of closed vessels with multiple flow streams such as headers. Here mass
accumulation and fan or pump head characteristics results in pressure / flow relationships that
cause the other flows to change.
Process Control for Efficient Manufacture - Measures & Metrics Page 4 of 9

Let us consider four common chemical process examples. As a controller modulates one flow,
other flows change elsewhere because the fluids share a common vessel volume or a fluid
pressure / flow relationship. Figure 2 shows a closed vessel with multiple flows such as a fluid
header or hydraulic column. The primary mechanism for the transfer of variability is the
dynamic interaction between the fluid flows and the vessel pressure. Figure 3 shows a buffer
storage tank designed to take out variability through mixing. In many cases the level controller
has tight tuning and its control action transfers the variability present in the demand stream to
the feed stream where it damages upstream reactions and impacts product uniformity. This is
a very common problem and a major source of variability transfer in many plants. Figure 4
shows a reactor feed system with
multiple reagents, where it is
critical to keep the feed
composition and reactor
stoichiometry as constant as
possible in order to maintain
product purity and conversion
efficiency. The last example
concerns stream composition and
is shown in Figure 5. Maintaining
a constant stream composition is
Figure 4 – Variability transfer around a reactor one of the most important
considerations in operating any
chemical process (e.g. chemical
concentration, pH, slurry density,
pulp consistency). The
composition is usually controlled
by adjusting a dilution or
sweetener flow stream in response
to a composition measurement.
The potential exists for variability
transfer between the demand flow
controller and the composition
controller, which will have serious
consequences when the
Figure 5 – Flow / composition variability transfer composition is disturbed.

Variability pathways can be altered by control strategy changes, and can be eliminated or
created by process piping changes [6]. Tseng, Cluett and Bialkowski [10] discuss a pulp and
paper example in which the frequency response of variability pathways through a process
network is used as the methodology for the integrated process and control design.

TUNING METHODS

The first tuning rule was developed in 1942, nearly sixty years ago, by Ziegler and Nichols [11]
and is known as the Quarter-Amplitude-Damping method (QAD). Yet many practitioners still
prefer to “tune-by-feel”. In part this is because control engineering mathematics has been
taught as an abstract subject and few people have gained any practical knowledge from the
experience. Default tuning values abound and are written in “black-books” for future
reference. A gain of 1.0 and a reset time of 1.0 minute are typical “good numbers”. One
Process Control for Efficient Manufacture - Measures & Metrics Page 5 of 9

reason for the popularity of advanced control algorithms such as Dynamic Matrix Control [1] is
that the “tuning problem” goes away as it is handled by the APC application. Even amongst
the APC practitioners “classical control” and controller tuning skills are rare assets. The
objective of QAD and other early tuning methods [4] was to make the loop as fast as possible,
while still remaining stable. From a process variability propagation point of view this results in
each loop being tuned individually with no consideration for the integrated process dynamics.

LAMBDA TUNING

“Lambda Tuning” refers to all tuning methods where the loop speed of response is a
selectable tuning parameter. The closed loop time constant is often referred to as “Lambda” –
hence the name. Selecting a fast Lambda value compared to the open loop dynamics makes
the loop less robust, while slowing down Lambda increases robustness. This is in contrast
with earlier tuning methods which attempted to make the loop “as fast as possible”, thus
creating an arbitrary speed of response with very low robustness. Lambda Tuning originated
with Dahlin [12] in 1968. It is based on the same IMC theory as MPC [13, 14], is model-based
and uses a model inverse and pole-zero cancellation to achieve the desired closed loop
performance. Lambda Tuning is used widely in the pulp and paper industry [3, 4, 5, 15],
where it was realized early-on that a strong connection exists between paper uniformity and
manufacturing efficiency on the one hand, and control loop interactions with upstream
hydraulics on the other. Paper being a solid product, captures all this upstream variability.
Lambda Tuning offered a new way of coordinating the tuning of the paper mill loops to gain
improved process stability. By contrast the Lambda Tuning technique is not well known
outside of the pulp and paper industry at this time.

The technical aspects of Lambda Tuning are described elsewhere [3, 4, 5]. In summary the
method consists of the following steps. First the open loop dynamics of the process are
tested. Usually this is done by carrying out a series of open loop bump tests, and fitting simple
dynamics models to the data. The first order plus deadtime and integrator plus deadtime are
the two most common models. The bump tests usually reveal control valve deficiencies and a
host of other problems, which should be corrected to allow the process to work properly. The
user selects a closed loop time constant for the loop based on process requirements, and
applies the Lambda Tuning mathematics to calculate the PID controller gains. For a non-
integrating process the result will be a non-oscillatory closed loop response reaching steady
state in about four Lambda values. For integrating variables, such as tank levels, the
response will reach steady state in about six Lambda values, while it will “arrest” a level
disturbance (the level stops rising or falling) in one Lambda value. This is also known as
“averaging level control” in the literature.

LAMBDA TUNING MULTIPLE LOOPS FOR UNIFORM MANUFACTURING

Consider the reactor feed problem shown in Figure 4 where the need is to maintain the
reagent ratios and reaction stoichiometry constant. In the following example the reagent flows
‘A’ and ‘B’ have target ratios of 68% for ‘A’ and 32% for ‘B’. A level controller cascades down
to the setpoints of the two-flow control loops via ratio stations. The example illustrates the
consequences of using old tuning methods such as QAD, and Figure 6 a) (heavy lines) shows
both loops responding to setpoint change at the same time. Both loops oscillate with an initial
overshoot of 50% as required by QAD. Clearly, the oscillations and unequal speeds of the two
loops have resulted in a significant, yet unnecessary disturbances being created in the flow
Process Control for Efficient Manufacture - Measures & Metrics Page 6 of 9

ratios of Reagents ‘A’ and ‘B’ as shown in Figure 6 b) (heavy lines). Oscillatory tuning offers no
benefit for uniform manufacturing. In fact such tuning is unacceptable for solid product
manufacture (e.g. plastic, rubber, metal, paper) in which the product captures all of the
upstream variability. This is less of a problem in the manufacture of liquid or gas products,
which equilibrate with time and which also allow post-manufacture blending. Nevertheless,
operating a reactor with such tuning leads to lower reaction conversion efficiency, lower
product purity, lower operating constraints and increased cost. Clearly, the requirement for the
two flows is to respond in a non-oscillatory manner, with exactly the same speed of response.
This requires Lambda Tuning with both Lambda values set exactly equal as shown in Figure 6
a) and b) (light lines). For illustration a Lambda value of 20 seconds was selected for both
loops, however any closed loop time constant would work as long as both loops had the same
speed of response. In Figure 6 the transient lasts about 80 seconds, while the flow ratios are
absolutely constant. This is only one example of how the selection of a specific speed of
response for multiple control loops impacts uniform manufacturing.

QAD & Lambda Tuning QAD & Lambda Flow Ratios

80
65
Lambda Reagent A
70
Reagent A
55
60
Flow Ratios %
Flows%

45 Flow Lambda Tuning 50


SP's Time to Steady State = QAD Lambda - Ratios Absolutely Constant
35 4 x Lambda = 80 sec
40

25 30
Reagent B
QAD Reagent B
15 20
0 20 40 60 80 100 0 20 40 60 80 100

Time seconds Time seconds

a) Flow Response b) Flow Ratios


Figure 6 – Comparison of Reactor Feed Flows Using QAD and Lambda Tuning

The example illustrates that the fastest possible speed of response for each loop does not
guarantee uniform manufacturing of a product in a continuous plant - in fact the opposite is
true. The example also illustrates how selecting specific closed loop time constants can
enhance uniform manufacturing.

MEASURES AND METRICS FOR UNIFORM MANUFACTURING

A general consideration is that process variability is “bad”, and should be measured and
minimized. Loops should be Lambda tuned and the closed loop time constant should be
selected as part of a coordinated plant tuning strategy based on the process operating
objectives. Final control element performance should conform to performance standards [7, 8]
to prevent limit cycling. If the loop operates in steady state the PV can be used to calculate
statistics, otherwise the controller error (SP - PV) is a better measure. A good general metric
is to calculate the mean and standard deviation on-line. Two-standard deviations expressed
as a percent of mean (2-sigma%) is a slightly better metric.
Process Control for Efficient Manufacture - Measures & Metrics Page 7 of 9

REACTOR FEED MEASURES AND METRICS

For the reactor feed example of Figure 4 the closed loop time constants of all of the reagent
flows should be the same, hence their relative values are a measure of appropriate tuning.
The metric to assess performance involves calculating each reagent ratio on-line as well as
standard deviation statistics. Alarm limits should be set for each ratio and monitored so that
corrective action can be taken.

PROCESS INTERACTION

Figure 2 shows a closed vessel with multiple flows streams entering and leaving. Control
loops interact via the pressure / flow coupling in the vessel and can potentially cycle if tuned
aggressively. A good rule of thumb is to select the loop that is most important operationally
and tune it as fast as possible, while maintaining robustness and avoiding resonance. The
next most important interacting loop should then be tuned slower (typically five-to-ten times)
than the loop already tuned, and so on down the chain. Lambda Tuning allows specific values
to be associated with these tunings.

If in Figure 2 the vessel pressure was selected as the fastest loop, the remaining flows will
“see” a constant pressure in their range of control frequencies. They will be totally decoupled
from each other, as the pressure controller will be fast enough to solve the mass balance
problem each time any flow changes. This is an effective means of decoupling multiple flows
entering or leaving a closed vessel.

Figure 5 shows an important example of interaction between stream composition and demand
flow. The composition is likely to be operationally the more important of the two loops.
Because the composition transmitter is located downstream there will be deadtime present;
hence the composition loop is likely to be naturally slower than the flow. The mass balance
dictates the coupling between the flow and composition loop, as increases in demand flow
cause an immediate need for more dilution / sweetener flow. To solve the dynamic coupling,
the flow loop should be tuned slower than the composition loop.

PROCESS INTERACTION MEASURES AND METRICS

The issue with interacting loops is that an operationally important loop will experience
increased variability due to control actions of less important loops. A measure of the variability
shift from a less important loop to the more important loop can be obtained by observing the
drop in variability in the important loop when the less important loop is placed in manual.

A metric for the variability transfer from a less important loop (Loop 2) to a more important loop
(Loop 1) can be calculated by observing the variability drop in Loop 1 (standard deviation, or
2-sgma, or 2-sgma % of mean) when Loop 2 is placed in manual. Calculate the percentage
decrease in Loop 1 variability when Loop 2 is in manual as opposed to automatic. This
requires Loop 2 to be turned off temporarily to make this measurement. Ideally the duration
of the Manual trial should be at least sixty times the closed loop time constant (ten cycles at
the cut-off frequency), and this should be compared to variability on Automatic over the same
period of time.
Process Control for Efficient Manufacture - Measures & Metrics Page 8 of 9

CASCADE LOOPS

Cascade loops interact dynamically because the output of one loop becoming the setpoint of
another as shown in Figures 3 and 4. For the cascade pair to be stable the inner loop must be
faster than the outer loop, and a rule of thumb is to start with five-ten-times faster initially.
When a level controller is the outer loop, the rule should be more stringently set at ten times
faster, due to the higher sensitivity of level controllers to slow inner loop dynamics. The upper
cascade controller transfers its variability to the setpoint of the inner cascade. How the two
loops are tuned should depend on their relative importance. In some cases the level controller
variability may be critically important, as in boiler drum level control. Large excursions of drum
level will cause a boiler trip. The drum level controller transfers variability to the boiler
feedwater flow, where the variability should be benign. However, the variability in the
feedwater flow will be transferred via the header pressure to the other boilers on the same
feedwater supply. Clearly, this is a case for stabilizing the feedwater header pressure through
integrated design, as none of the boilers should be allowed to trip on drum level excursions.

In case of the buffer storage tank level of Figure 3 the tank level is probably not critically
important because the buffer storage is designed to absorb disturbances. On the other hand
the feed flows to the buffer storage may be critically important as these are often closely
coupled to the reactors that supply these streams. Hence the amount of variability transferred
from the cascade level may be critical, and may need to be minimized.

LAMBDA TUNING STRATEGY FOR BUFFER STORAGE LEVEL CONTROL

The case of buffer inventory storage tanks is classic case of variability transfer gone mad. The
plant designer provided the buffer storage tank to blend-out incoming variability before the
next process downstream. In many plants however the level controllers have been tuned very
tightly because the operators do not like to see the level vary. As a result the level controller
transfers variability to the feed flows thus destabilizing the reactors upstream, and often plant
performance suffers severely as a result.

To minimize the upstream variability transfer, buffer storage level controllers should be tuned
as slowly as possible, while still preventing the level from overflowing the tank, or from
dropping too low when the demand flow changes. The selection of Lambda for the tank needs
to take these issues into account. A closed loop time constant roughly equal to the residence
time in the tank is probably a good choice. More exact numbers can be worked out by taking
into account the level setpoint relative to the top of the tank, how much level excursion is
permitted, and how much the demand flow is likely to change.

CASCADE LOOPS MEASURES AND METRICS

Cascade loops can be considered like other interacting loops. In some cases the upper
cascade is the more important loop (e.g. boiler drum level), and in other cases the lower
cascade is the more important loop (e.g. buffer storage tank). The measures and metrics are
the same as those for process interaction.
Process Control for Efficient Manufacture - Measures & Metrics Page 9 of 9

LAMBDA TUNING FOR UNIFORM MANUFACTURING – SUMMARY

The Lambda Tuning strategy for uniform manufacturing can be summarized as follows:
1) Identify the process dynamics and actuator nonlinearities through a series of bump tests.
2) Loops should never be tuned to cycle -- control should be 'smooth' – use Lambda Tuning.
3) Set closed loop time constants based on plant goals and a coordinated tuning strategy.
4) Tuning should be robust for all operating conditions.
5) All reactor feed flow controls should have equal speeds of response to maintain feed ratio
targets.
6) Key loops should be faster (five-to-ten times typically) than other process coupled loops.
7) Inner cascade loops should be faster (five-to-ten times typically) than outer loops.
8) Tune buffer storage tank level controls as slow as possible while satisfying production
needs.

CONCLUSIONS

The paper has presented the concepts of process variability transfer caused by control action
in a typical chemical process plant and has linked control loop performance to the efficient
manufacture of uniform product. Integrated process and control design focuses on ensuring
that variability does not harm manufacturing efficiency or product uniformity. The Lambda
Tuning concept has been presented as a means of coordinating the tuning of all loops in a
process to minimize process variability transmission. Control loop measures and metrics have
been provided to achieve such performance.

REFERENCES

1. Cutler, C. R. and Raemaker, B. L., Dynamic Matrix Control – A Computer Control Algorithm, Proc.
of the JACC, San Francisco, CA, 1980.
2. William S. Levine, W.S (Editor), CRC Control Handbook, CRC Press and IEEE Press, 1996.
3. William S. Levine, W.S (Editor), CRC Control Handbook, Chapter 72, "Control of the Pulp and
Paper Making Process". CRC Press and IEEE Press, 1996. Chapter 72 – Bialkowski W
4. Gregory K. McMillan (Editor in Chief), Process / Industrial Controls and Instrumentation Handbook,
5th Edition. McGraw-Hill, 1999. Isbn 0-07-012582-1. Section 10.17
5. Sell N, editor, Bialkowski W.L. and Thomasson F.Y. contributors, Process Control Fundamentals for
the Pulp & Paper Industry, TAPPI Textbook, TAPPI Press, 1995
6. Downs J.J and Doss J.E., Present Status and Future Needs - a view from N. Amer Industry, Fourth
Int. Conf on Chem Proc Control, Padre Island, TX, Feb. 17-22, 1991 (AIChE)
TM
7. EnTech - Control Valve Dynamic Specification (Version 2.1, 3/94, Version 3.0, 11/98)
8. ANSI/ISA S75.25 Control Valve Response Measurement from Step Responses Inputs
9. Bialkowski W. L., A Review Of Deadtime Compensator And PI Controller Regulation Performance,
Pulp and Paper Canada, 98:10, (1997) pp.1386-90
10. Tseng J., Cluett W.R. and Bialkowski W. L., Variability Propagation Through A Stock Preparation
System, Pulp and Paper Canada, 89:9 (1997) pp. T322-T325
11. Ziegler J.G. and Nichols N.B., Optimum settings for automatic controllers, Trans. ASME, pp. 759-
768, 1942
12. Dahlin E.B., Designing and Tuning Digital Controllers, Instr and Cont Syst, 41 (6), 77, 1968.
13. Morari M. and Zafiriou E., Robust Process Control, Prentice Hall, 1989.
14. Chien I-Lung and Fruehauf P.S., Consider IMC Tuning to Improve Controller Performance,
Hydrocarbon Processing, Oct. 1990.
15. Bialkowski W.L., Haggman B.C. and Millette S.K., Pulp and Paper Process Control Training Since
1984, Pulp and Paper Canada 95: 4 (1994)
Emerson is a trademark of Emerson Electric Company; EnTech is a trademark of Emerson Electric Canada Ltd.

© Copyright 2001 Instrument Society of America. All rights reserved. EnTech Form wlb-010910e1

You might also like