Proceedings of the American Control Conference

Arlington, VA June 25-27, 2001

Control Design for an Industrial Boiler

Wen Tan, H. J. Marquez, Tongwen Chen R. K. Gooden

Dept. of Electrical and Computer Eng. Syncrude Canada Ltd.
University of Alberta RO. Bag 4009 MD 5300
Edmonton, AB T6G 2G7 Fort McMurray, AB T9H 3L1
Canada Canada
{wtan,marquez,tchen} @ee.ualberta.ca gooden.kent @syncrude.com

Abstract characteristics, utility boilers are used to regulate the steam

pressure, while CO boilers and OTSG boilers are used to
In the control of an industrial boiler system, multi-loop (de- track steam load and regulate the steam temperature, respec-
centralized) PI control is used because of its simplicity and tively. At present, utility boilers are controlled via a multi-
the existence of simple on-line tuning rules. We show that loop (decentralized) proportional plus integral (PI) type con-
such control schemes sacrifices robustness and performance troller. This configuration, however, ignores the fact that
of the overall system. In particular, under normal boiler oper- there exist interactions among the variables to be regulated.
ating conditions, we design a robust multivariable controller The result is that under normal operation, the 900# header
using the Hoo loop-shaping technique. For reason of imple- pressure exhibits oscillatory modes that the controller is un-
mentation, we then reduce this controller to a multivariable able to damped out as quickly as desired. This motivates the
PI structure. Both the Hoo controller and its PI approxima- redesign of the utility boiler controllers to improve overall
tion are tested extensively in the frequency domain as well system performance.
as in the time domain, using a complex nonlinear simulation fr -,~
software; the results show that the designed controllers are
superior in robustness and performance compared to the ex- Utility ~'~ 9oo# s t e ~
Boilers v"
isting multi-loop controller.
_~ .ower __r~
Generation
~ Electricity
CO ~ 900#
Boilers ~ Header
1 Introduction

Industrial co-generation systems invariably present a chal-

lenge to control system designers. The Syncrude Canada Ltd 5O#steam
(SCL) integrated energy facility located in Mildred Lake, Al-
berta, utilizes a complex header system for steam distribu-
tion, which includes headers at four different pressure levels Figure 1: A simple diagram for the Syncrude utility plant
(900, 600, 150 and 50 psi). The 900# header receives steam
from three utility-type boilers burning refinery gas, three CO-
type boilers burning coker off gas and refinery gas, and two In this paper we investigate the use of modern multivariable
once-through steam generators (OTSG). The steam is then control techniques as applied to co-generation systems such
distributed for three different usuages: i) to extract bitumen as the one described here. More explicitly, we will investi-
from oil sand; ii) to numerous turbines to generate electricity; gate the following issues:
iii) to three other headers to generate steam at different pres-
sure. The overall plant, like many similar available world-
wide, is thus a rather complex, nonlinear, interconnected sys- • Multivariable control design: We will proceed to de-
tem. A simple diagram of the utility plant is shown in Fig- sign a multivariable controller for the utility boiler and
ure 1. compare our new design to the existing one. The new
controller should be robust, i.e., it should maintain sta-
The normal plant operation requires tracking the steam de- bility as well as performance despite the existence of
mand while maintaining the steam pressure and the steam modeling errors. We emphasize that in the existing
temperature of the 900# header at their respective setpoints, controllers, on line tuning makes it virtually impossi-
despite variations of the steam load. Due to their physical ble to incorporate robustness constraints.

0-7803-6495-3/01/$10.00 © 2001 AACC 2537

 • Hoo optimization and controller reduction: Our design In the fuel-air-flue system fuel and air are thoroughly mixed
will be cast as an Hoo optimization problem. Undoubt- and ignited in a furnace. The resulting combustion converts
edly Hoo optimization has been the leading design ap- the chemical energy of the fuel to thermal or heat energy. The
proach in robust control over the last two decades. It gases resulting from the combustion, known as the flue gases,
is known however, that the order of the resulting con- pass through the superheaters, the risers, and the downcom-
troller using this approach is no less than that of the ers, and leave the boiler. A schematic diagram of this type of
original plant to be controlled. There is some reluc- boiler is shown in Figure 2 (where the arrow points out the
tancy in industry towards the use of high order con- direction of the steam-water flow).
trollers, due to complexity of implementation and dif-
y3 -- Steam u3 -- Attemperator
Temperature I Spray Flow
ficulty associated with possible re-tuning. Thus, we
will approximate the optimal controller by a multivari- s.....o
900# header
~
~ a r 7 - Pdmary ~
.......
~ ' ~ u l -- FeedwaterFlow
[.-_
able PI type controller. It will be shown that the new Super Heater Super Heater / \

PI multivariable controller retains the features of the

f ~~ /~ - - Dr m Level
optimal control with little performance deterioration.
Steam Drum [
• Simulation and time domain performance comparison: I

As mentioned, the true system is nonlinear and rather

Heat
complex. So the final result should be tested under
more rigorous conditions. Namely, it should be tested
with a more accurate description which should incor-
porate the nonlinearities encountered in the true plant.
Mud Drum

y4--Excess Oxygen

In the present case, Syncrude Canada Inc has available a E

u2 - - Fuel Flow
simulation package, known as SYNSIM [1]. The SYNSIM
model was developed with the purpose of studying the causes n uce ra an

of certain upset conditions that have been sporadically de-

tected, as well as a general tool for stability analysis. The q u4 -- Air Flow

model has been extensively tested and correlation between

measurements from the true plant outputs and predictions by
SYNSIM are excellent. The final controller will be simulated
in SYNSIM and compared to the existing design under fairly Figure 2: Utility boiler
realistic conditions.
As shown in Figure 2, the principal input and output variables
are
2 Utility Boiler
Ul: feedwater flow rate (kg/s),
u2: fuel flow rate (kg/s),
The utility boilers in the plant are watertube drum boilers.
u3: attemperator spray flow rate (kg/s),
This type of boiler usually comprises two separate subsys-
u4: air flow rate (kg/s);
tems: the steam water system (also called water side), and
Y]: drum level (m),
the fuel-air-flue gas system (also called the fire side).
Y2: drum pressure (KPa),
In the steam-water system, preheated water is fed into the Y3: steam temperature (Celsius),
steam drum. Liquid water exits the steam drum and flows Y4: excess oxygen level (percent).
through the downcomers into the mud drum. The mud drum
distributes the water to the risers, where the water is heated The following basic requirements must be satisfied for proper
to saturation conditions. The saturated steam-water mixture functioning of the boiler system:
then re-enters the steam drum in which the steam is separated
from the water and exits the steam drum into the primary and
1. Steam pressure of the 900# header must be maintained
secondary superheaters. In the two superheaters, the steam
despite variations in the amount of steam demanded by
is further heated, and then is fed into the 900# header. In
users.
between the two superheaters is an attemperator which regu-
lates the temperature of the steam exiting the secondary su- 2. The amount of water in the steam drum must be main-
perheater by mixing water at a lower temperature with the tained at the desired level to prevent overheating of
steam from the primary superheater. drum components or flooding of steam lines.

2538
 3. The steam temperature must be maintained at the de- • It can incorporate both performance and robust stabil-
sired level to prevent overheating of the superheaters ity requirements.
and to prevent wet steam entering turbines.
• It is conceptually and computationally simple and re-
4. The mixture of fuel and air in the combustion cham- tains many of the features encountered in the design of
ber must meet standards for safety, efficiency, and en- classical PI loops using frequency domain techniques.
vironment protection; this is usually accomplished by This is important since experienced engineers work-
maintaining a desired level of excess oxygen. ing in industry can incorporate their knowledge of the
plant and classical control design into modem, power-
ful techniques.
To simplify the control design problem, we assume the ratio
of fuel-to-air flow rate is kept constant. With this assump-
tion the last requirement can be omitted, and moreover, the Given a plant model G, the design approach consists of three
resulting model has three inputs (Ul, u2, and u3), and three steps:
outputs (Yl, y2, and Y3). The nominal operating conditions
are specified as follows: 1. Loop Shaping: Use pre- and/or post-compensators W2
and W1, to shape the singular values of the original
1. Two utility boilers, three CO boilers and two OTSG's plant G. This step contains all of the ingredients of the
are on-line. classical techniques. The shaping functions W1 and W2
are controlled by the designer and the properties of the
2. The total steam load for the 900# header is 315.08 kg/s. resulting controller depend upon these functions in an
essential manner. Guidelines for choosing W1 and W2
3. The total steam load for the 50# header is 155.69 kg/s.
may be found in [7].
4. The total steam turbine electrical output is 145.53 MW.
2. Robust Stabilization: A feedback controller which ro-
bustly stabilizes the 'shaped' plant ( ( ~ - W2GW1 ) is
At this operating point, for each of the two utility boilers, we found. More explicitly, we solve the following Hoo op-

[1o] E4o81 [ ] E ]
have timization problem:

Ylo 1.0
U20 -- 2.102 , Y2o -- 6450.34 . Ema x - - inf /((I + ~/~)-1 . (1)
K co
u3o 0 Y3o 466.7
where N/14-1 is a normalized left coprime factoriza-
tion of (~. The importance of this minimization is that
In addition, the following limit constraints exist for the three the controller obtained in (1) will guarantee stability of
control variables: any plant G* which belongs to the family of plants Gzx
0 ~ Ul ~ 120, defined as follows [6]"
0 ~ U2 ~ 7,
GA- {NAM-£1"II[NA-N,MA-~4]IIoo< Emax}.
0 ~ U3 ~ 10,
--0.017 _< /i 2 ~ 0.017. This also guarantees that the loop shape we selected in
the previous step can be well approximated with good
robust stability if emax is sufficiently large. The value
We note that the input at this operating point is small com-
Emax is used as a design indicator; usually it should be
pared with the magnitude limit, so these limits do not impose
between 0.3 and 0.5.
hard constraints for design. The rate limit for fuel flow rate
(u2), however, has a significant impact on the system perfor- 3. The final feedback controller is obtained as K -
mance. w1Rw2.

We now apply this method to the utility boiler model. We

3 Controller Design start by scaling the model so as to improve the condition
number of the plant, which is not good because the coeffi-
Various technic ues have been applied to boiler or boiler- cients of the first row and the first column of the model are
turbine controller design, e.g., inverse Nyquist array [2], too small compared with other columns and rows. In fact,
LQG [3], LQG/LTR [4] and mixed-sensitivity Hoo approach by scaling the drum level by a factor of 100 and the feed-
[5]. Here we will adopt the loop-shaping Hoo approach intro- water flow rate by 10, the condition number of the system
duced by McFarlane and Glover [6]. The essential features frequency response matrix is reduced significantly as shown
of this approach can be summarized as follows: in Fig.3.

2539
 . . . . . . . . i . . . . . . . . i . . . . . . . . | . . . . . . . . l . . . . . . . . i . . . . . . .
Since W1 is incorporated in the final controller, Ak always
has eigenvalues at the origin. We can decompose the zero
80

eignevalues from the rest nonzero eigenvalues by eigenvalue

70
decomposition, i.e, we find a similarity transformation T
such that
~" 60
TAkT-I--[ 0 0 1
0 A2 '
:~ 50
Clearly A2 is nonsingular. With this T, the new state-space
8 40 realization is given by

{ .~ - Ak£ + Bky,
'-.. .. u -- GX + D~y,

with Ak -- TAkT- 1, L)k -- Dk, and

lO , , ...... i . . . . . . . . l . . . . . . . . i . . . . . . . . i . . . . . . . . i . . . . . . . .
1 0 --4 10 -3 10 -2 1 0 -1 lO 0 101 lO a
I B1
Frequency(tad/s)
Ck -- Ck r - 1 -- [ C1 c2], ~ - - r B k - B2
Figure 3: Condition number of the frequency response matrices A PID approximation of the form
(solid: before scaling; dotted: after scaling)
xpm(s) - Xp + Xi/s + tCd~
For the scaled model, we will choose W2 = I and set W1 = can now be obtained by truncating the MacLaurin expansion
WaWi, where Wa is a static decoupler and Wi is a diagonal PI of the controller with respect to the variable s. Since
compensator that determines the desired open-loop shapes.
B1 1
I 1
The pre-compensator was selected as: sI 0
g(s)- [ C1 C2 ] ] --I-Ok
0 sI-A2 B2 .1

0.00149 0.0133 0.00362 ] --- C!B1 ~- (Dk - C2A21B2) - C2A22B2s + " "
W1 - 0.00016 0.0075 0.0020 ] • (3)
0.0011 0.0126 -0.0394 we get
lO+4/s 0 0 -1
0 2.5 + 0.05/s 0 J . Kp -- D k - C2A21B2, Ki -- C1B1, Kd -- -C2A22B2 •
0 0 1+ 0.05/s It is then clear that based on this reduction procedure, the
(2) resulting PID controller achieves good approximation of the
The final design indicator for this design is Emax -- 0.318,
controller K at low frequencies.
so the designed loop shapes do not change much from the
desired. In the present case, our goal is to get a PID type controller to
approximate the 18-th order H= controller designed earlier.
Since the elements of the derivative term in KpID is too large
4 Controller Reduction compared with those of the first two terms, so we keep only
the first two terms in the expansion to get a PI approximation
A minimal state-space realization of the designed H~ con- 0.2069 -0.00919 -0.0290 ]
troller has order 18. Practical implementation issues dictate
the need to investigate the performance of a reduced order
controller. Moreover, all controllers presently used in the
plant are of the PI type. This structure is thus familiar to the
xpI(s)
I 0.0250
0.1208
0.0055
0.0099
0.0022 0.00022
0.00027 0.0002
0.0004
-0.0642
+

-0.00021 I 1
0 s
(4)

operators and it is easy to implement. Thus, in this section 0.00136 0.0003 -0.0009
we investigate the performance of a reduced order multivari-
able controller of the PI type. A secondary reason for the Figure 4 compares the singular values of the open-loop fre-
preference of PI type controllers is that anti-windup imple- quency response matrices [G(jo~)K(jco)] for both the H~
mentation is relatively easy, considering practical constraints controller K and its PI approximation KpI. Also shown is
such as rate limits and saturations of the control inputs. the desired open-loop shape we select. It is clear that both
controllers have similar characteristics within the plant band-
Consider now a controller K(s), given by a state-space real- width.
ization of the form
Note the controllers above are designed for the scaled model.
2--Akx+Bky The final PI controller for the unscaled model can be obtained
u -- Ckx + Dky. by scaling back.

2540
 100 . . . . . . . . , . . . . . . . . , . . . . . . . . , . . . . . . . . , . . . . . . . . , . . . . . . . .
with either the Hoo controller or the PI controller. Moreover,
80 no significant loss of performance was detected between the
Hoo controller and the PI approximation of section IV. Thus,
6O
from now on we will continue with the analysis of KpI only,
40
which as mentioned earlier is our preferred choice due to ease
m 2o in implementation and anti-windup configuration due to the
fuel flow rate limit.
o
-20 This frequency analysis, however, is based on the linear time-
-40
invariant approximation obtained in section II, and therefore
contains no information regarding the performance of any
-60
of these controllers connected with the true nonlinear plant.
-80 Clearly, this frequency domain analysis is not suitable for
-100
this purpose. To this end and to test our controller under
1 0 -4 10 -3 10 -2 10 -I 10 ° 101 10 z
Frequency (rad/sec)
virtually true plant conditions, we performed a series of test
using SYNSIM, but replacing the existing controller with the
Figure 4: Singular values for the open-loop frequency response final PI controller.
matrices (sold: Hoo controller; dashed: PI controller;
dotted: desired open-loop shape) Figure 6 compares the step responses (for the drum pressure
and the drum level. Response for the steam temperature is
omitted for brevity) for the designed multivariable PI con-
5 Simulation Results troller and the existing multi-loop controller. We see that
the multi-loop controller has a slow rising time and a large
In this section we evaluate the results obtained in the previ- overshoot for drum pressure change. However, it has a fast
ous sections. We start with a frequency domain comparison rising time and less overshoot for drum level change. This
among the three different controllers, namely: (i) the multi- is due to the fact that the original feed-water controller is a
loop PI controllers presently used in the plant, (ii) the Hoo cascade one using the steam flow as feedforward, which can
controller of section III, and (iii) the PI controller approx- improve the response of the drum level. We also observe that
imation of section IV. Note we must use the scaled model the drum pressure drop due to the change of the drum level
and the scaled controllers to compute the frequency domain for the multi-loop controller is much more than that for the
properties, since for unscaled model these properties are not designed PI controller.
in the same magnitude for comparison.
Durm level increases 10% Drum pressure increases 5%

1.15l
Figure 5 shows the maximum singular values, Os(jm) and
Or (jc0), of the sensitivity function S and the complementary
sensitivity function T for the three controllers in considera-
tion. Clearly a significant improvement has been achieved ~ 1.05

2o / ........ , ......... . ..y,' ".' '" ........ , ........ , .......

°915 2'0
0 10 20

(b)
6480 7000 j . •

o. . . . . . . iiiiii ;o,oo~ I

~'6440Ii i oot
o0oooF/_
642°Iiil
6 5 0 0 V " ""

6400 ''" " '

0 10 20 30 0 10 20 30
Time(min) Time(rain)

Figure 6: Boiler time response (solid: multivariable PI controller;

dotted: original multi-loop controller)

-50 , , ..... , ........ I ........ | ........ , ........ i .......

10-4 10 -3 10 -2 10 -1
F r e q u n c e y (rad/s)
100 101 102 Now suppose that the unmodeled 900# steam load increases
by 20%, Figure 7 shows the responses of the 900# header
Figure 5" Maximum singular values of S and T (dotted: original pressure under the original controller (dotted) and the de-
controller; dashed: Hoo controller; solid: PI controller) signed multivariable PI controller (solid). We observe that

2541
the designed PI controller damps out the oscillation modes generation system with Syncrude Canada. Comparing with
much faster. the existing multi-loop PI controller, we have conducted a se-
ries of frequency domain validations, based on an LTI model
under normal operating conditions, and time domain tests,
using a complex nonlinear simulation package (SYNSIM).
We conclude that the designed PI controller outperforms the
--..
~.632o[1 " existing one in both robustness and performance.

~v6310
Acknowledgment: We are indebted to Mr. David White of
. Syncrude Canada and to Dr. Ray Rink of the University of
6300 ". - -.. ..
. . ... Alberta. David White spent countless hour discussing sev-
eral aspects of this project during the course of this research.
~ 6290 '. .."

Ray Rink, Kent Gooden and David White are the 'brains'
• behind what today is called SYNSIM, an outstanding simu-
lation package that took years to develop. This project would
have never been completed without the technical expertise of
..

62600 5
I
1; 1;
I
20
I
25 30
Ray Rink who instructed us on the details of this program.
~me(min)

The project was supported by Syncrude Canada Ltd. and

Figure 7: Responses for unmodeled steam load change (dotted: the Natural Sciences and Engineering Research Council of
original controller; solid: multivariable PI controller) Canada (NSERC).

Finally, an extreme case is simulated for the designed con-

troller. Suppose one of the CO boilers in the plant is shut References
down at t = 0 and the unmodeled 900# steam load increases [1] Rink, R., D. White, A. Chiu and R. Leung, SYN-
by 20% at t = 30 (minutes), thus changing the operating SIM: A computer simulation model for the Mildred Lake
point. The response is shown in Figure 8. In this case the Steam/Electrical System of Syncrude Canada Ltd., Techni-
original controller cannot even stabilize the system and hence cal Report, University of Alberta, May, 1996.
its response is not shown. But the designed controller is quite
[2] Johansson, L., and H.N. Koivo, Inverse Nyquist array
robust and still functions properly.
technique in the design of a multivariable controller for a
solid-fuel boiler, Int. J. Control, vol. 40, 1077-1088, 1984.
[3] Cori, R., and C. Maffezzoni, Practical optimal control
of a drum boiler power plant, Automatica, vol. 20, 163-173,
1984.
6350
n [4] Kwon, W.H., S.W. Kim, and EG. Park, On the multi-
"~6300
g
variable robust control of a boiler-turbine system, IFAC Sym-
~-6250 posium on Power Systems and Power Plant Control, Seoul,
Korea, 219-223, 1989.
6200

[5] Pellegrinetti, G., and J. Bentsman, Hoo controller de-

6150
sign for boilers, Int. J. Robust and Nonlinear Control, vol. 4,
465-671, 1994.
6050 [6] McFarlane, D.C., and K. Glover, Robust Controller
6000 0
J 1; 20 30 4; 50 60
Design Using Normalized Coprime Factor Plant Descrip-
"rime(min) tions, New York, Springer-Verlag, 1990.
[7] Skogestad, S., and I. Postlethwaite, Multivariable
Figure 8" Response for one CO boiler shutdown and unmodeled
Feedback Control: Analysis and Design, John Wiley & Sons
steam load change.
Ltd., England, 1996.

6 Conclusions

Using Hoo loop-shaping techniques, we have designed a mul-

tivariable PI controller for a utility boiler in an industrial co-

2542