Sie sind auf Seite 1von 11

e Pergamon

Computers chem. Engng Vol. Ie}. No.3, pp. 321 ·331. IWS

0098-1354(94)00057-3
Copyright © !e}e}S Elsevier Science Ltd
Printed in Great Britain. All rights reserved
IK)'JR·13541e}5 $e}. SO + 0 (K)

PLANT-WIDE CONTROL OF THE TENNESSEE EASTMAN


PROBLEM
P, R, LYMAN and C. GEORGAKtst
ChemicalProcess Modelingand Control Research Center and Department of Chemical Engineering,
Lehigh University, Bethlehem, PA 18015-4732, U.S.A.

(Received 6 May 1993; final revision received 15 February 1994; received for publication 16 May 1994)

Abstract-This study focuses on the developmentand performanceof four plant-widecont~ol structures


for the Tennessee Eastman challenge problem. The control structures are developed to a tiered fashion
and without the use of a quantitative steady state or dynamicmodel of the process. The throughput or
production rate manipulator is selected first so. tha~ it is located on the major process path. The
inventory controls are arranged in an outward direction from this throughput maOlpulato~. The four
structures are described and commentsare givenon their effective handlingof the defineddisturbances
and setpoint changes, One structure provides effectivecontrol under all circumstances for 50 hours of
processtime, The effectivedynamicperformance of these structures supports the strength of the tiered
plant-wide control design methodology used.

l. INTRODUCTION AND BACKGROUND The synthesis of plant-wide control structures has


been the topic of recent research by Price and
The purpose of plant-wide control is to provide an Georgakis (1993), They have provided a systematic
overall structure for the coordinated control of method and a set of guidelines which enable the user
several important variables of a multi-unit process, to develop a plant-wide structure which provides
including the overall plant throughput and the pro- tight and effective control of the overall plant. This
duct quality control. The main incentive for this can be extremely valuable, especially in a system
approach is the coordination of the different local which includes large recycle streams which often
control tasks so that one can achieve substantial propagate and amplify process variation and the
reduction of costly intermediate storage facilities, effect of disturbances. The proposed method con-
The classical aproach to the plant-wide control prob- sists of a tiered framework in which control loops
lem has consisted in the past of the breaking down of are classified and ordered according to their import-
the over-all problem into a series of smaller control ance for the plant as a whole. The categories which
problems around the individual unit operations of compose the tiered framework are production and
the process, For this approach to work effectively, inventory control, product specification control,
the designer must provide sufficient material inven- equipment and operating constraints, and economic
tories between units which serve to isolate one from performance enhancement, respectively in the order
the dynamic characteristics of the others, These in which they are developed. One of the major
inventories might need to be large and are costly in recommendations of the cited references is to design
many ways, They contribute to a high capital cost, the structure so that the inventory control loops are
increase the overall holdup of valuable processing arranged in an outwardly fashion away from the
material and occupy valuable plant space. Large location of the production rate manipulator. Control
storage of possibly hazardous material also poses a structures which possess this characteristic are called
safety and an environmental hazard and the possible "self-consistent" .
degradation and contamination could significantly In this paper, attention is focused on the plant-
affect the product quality, Furthermore, the exis- wide control of the Tennessee Eastman (TE) chal-
tence of such large material hold-ups make required lenge problem, defined in Downs and Vogel (1993).
production shifts between different products or pro- They have also made available an open-loop dyna-
duct grades difficult and time consuming, The use of mic simulation of the already designed process but
an effective plant-wide control structure may lead to have refrained from providing even the simplest
a significant reduction in the size of these interme- SISO controllers. The primary solution to the chal-
diate inventories or possibly their elimination from lenge problem described in the present paper is a set
the plant design of SISO loops, labelled structure 2, Three additional
t To whom all correspondence should be addressed. control structures, labelled structures I.' and 4, will
321
322 P. R. LYMAN and C. GEORGAKIS

also be briefly described and compared to the prim- which uses valve 11 as the production rate manipu-
ary solution. All these control structures have been lator, shown in Fig. 1, will be called control struc-
described in great detail by Lyman (1992). We focus ture 2.
here on structure 2 because this has been able to As valve 11is being closed, the temperature of the
successfully handle all of the process disturbances reactor product stream at the exit of the condenser
described in the posed problem. The other struc- rises and a lesser amount of the G and H products
tures are also of interest because they provide alter- are condensed and separated. More of G and Hare
native structures that handle the great majority of sent back to the reactor where they cause a rise in
the defined disturbances. The paper is organized as the reactor level. The reactor level is controlled with
follows. The development procedure which was the flow rate of stream 4. This stream has a substan-
used to synthesize the control structure is described tial effect on the reactor level because it is the
in Section 2. The three additional structures, result- largest feed stream and contains both A and C which
ing from alternate locations of the variable that take part in both product producing reactions. The
manipulates the production rate, are described in separator and stripper levels are controlled by the
Section 3. The response performance of the main liquid efflux stream flow rates from each vessel
structure to the applied disturbances and setpoint following the rules defined previously. For example,
changes and its comparison with the other structures as the condenser cooling water flow rate is reduced
is described in Section 4. Finally the conclusions of in response to a reduction in the product flow rate
this work are provided in Section 5. setpoint, the separator and stripper efflux valves
close to maintain their corresponding vessel levels
2. SYNTHESIS OF A PLANT-WIDE SISO CONTROL
resulting in a reduction in the final product flow rate.
STRUCTURE
This sequence illustrates the action of a self consis-
In the following discussion we will describe how tent structure.
the tiered framework method proposed by Price and The one inventory which has not been addressed
Georgakis (1993), leads to a plant-wide SISO is the gas content in the reactor/recycle system,
control structure for the TE process. For complete- indicated by any of the different but interrelated
ness, the Appendix details the tuning approach used pressure values. In all of the structures described in
for the different control loops. The description of this paper, the reactor pressure is left uncontrolled
the control structures starts with the selection of the and is allowed to float. Although the reactor pres-
variable that will manipulate the production rate of sure can cause large variations in reaction rates, its
the plant and with the design of the inventory effective control would require the simultaneous
control loops. The last ones are to be located along and coordinated manipulation of the many factors
the primary process path and oriented in an outward which can influence the reactor pressure. This list
direction from the location of the manipulation includes the raw material flow rates, the recycle flow
variable for the production rate. This arrangement rate and the reactor temperature and level. Because
of inventory controllers follows the recommenda- there are substantial interactions between the differ-
tions made by Price and Georgakis (1993) who ent variables, one should consider the design of a
called it a self-consistent structure through the prim- multivariable controller for this reactor. Such a
ary process path. Consequently, it is important to controller was not considered here for two primary
first identify the primary process path. reasons. First we wished to consider the simplest
The primary process path for the TE process control structure possible and we chose to restrict
starts at the raw material feeds to the reactor, goes the problem solution to SISO structures with PID
through the reactor, condenser, and separator, goes type controllers. The second reason that restrained
on to the stripper, and through the liquid stream of us from considering a multivariable controller is its
the stripper to the liquid outflow from the stripper, requirement for a dynamic process model. We con-
the product stream. The next step is to locate the sidered it of interest to investigate what type of
production rate manipulator. A second recommen- control performance one could obtain without the
dation from the same references is to locate the use of a process model. This completes the descrip-
production rate manipulator near the center of the tion of the inventory control tier of the development
process flow path. With such a choice, the produc- of control structure 2.
tion rate changes will propagate in both directions Additional tiers in the structure development
simultaneously and thus faster. Although it might must provide for reactor temperature control,
not be obvious on first sight, the condenser cooling inventory control of individual species in the recycle
water valve (valve 11) can be considered to be on loop and final product purity. The reactor tempera-
the major process flow path. The control structure ture is controlled through a cascade controller by
~ ~
:. i
~ Purge >!
~ !
XQ ~
! @
I I ~ • \

9
71 -® :l

[
ISEPARATOR I 8.,
:r
(b

ews ;;i
:l
:l
(b

!l!J '"

(b
(b

rn
!!
! ... , ""
'"
3
·····i··@ , 12 : ""
:l

® @ "c:
.® @ <ID ~
sr

Al"" '"
Ir-R-EA-cr-oR-1 :3
~ ® '
;.. 4 .0- 1

L1 ...

I
L .

Fig. 1. TE process- control structure 2. ...,


...,
N
324 P. R. LYMAN and C. GEORGAKIS

manipulating in the master loop the set point of the Structure 3 controls the production rate by mani-
outlet temperature of the cooling water in the reac- pulating the outflow from the separator tank. As this
tor. In the inner or slave loop the outlet cooling valve is closed the flow into the stripper is reduced
water temperature is controlled by manipulating the and the level within the separator will rise. The
cooling water flow rate. This minimizes disturbances outflow from the stripper is then reduced in order to
to the reactor temperature from the cooling water maintain the stripper bottoms level as in structures 1
inlet temperature. A number of additional loops are and 2. The condenser cooling water flow is used to
used to control the overall material balance of some control the separator level in this structure. As the
individual components from the reactor/recycle separator level rises the condenser cooling water
system. Stream 1 is used to maintain a constant valve closes in order to maintain this level. This
composition of A in the reactor feed stream. This condenses less reactor product and sends it back to
controller is required to compensate for a variable the reactor where it causes a temporary increase in
amount of A which might enter the system from reactor level. The reactor level is maintained by
stream 4. Stream 3, which is nearly pure E, is used in manipulating the flow rate of stream 4 as in
order to maintain the proper product ratio which is structure 2.
measured at the product stream. The purge flowrate Structure 4 controls the production rate by mani-
is manipulated in order to maintain a constant com- pulating the stripper underflow valve. A closure of
position of the inert component B in the recycle this valve will cause the stripper level to rise. The
stream. The purity of the product at the bottom of stripper level is controlled with the separator out-
the stripper is related to the composition of E, the flow, and the separator level is controlled with the
heaviest of the impurities. It is controlled through condenser cooling water valve. The level in the
the outer loop in a cascade structure by manipula- reactor is controlled as described for structures 2
tion of the set point of an inner temperature control and 3. The above description illustrates the differ-
loop. The temperature controller manipulates the ences between the four control structures studied.
steam flow rate. Several controllers are identical in all control
The performance characteristics of this control structures. These include control of the reactor tem-
structure will be examined in the section after the perature by the use of the cooling water flow and the
next and are also detailed by Lyman (1992). control of the composition of the inert component B
with the flow rate of the purge stream. The product
3. ADDITIONAL PLANT·WIDE CONTROL STRUCTURES
quality control loop described above is also common
to all control structures examined and the compo-
This section briefly describes three additional sition of A in the reactor feed is controlled using the
structures which result from the use of the tiered feed rate of A to avoid accumulation of this compo-
approach mentioned above. These structures differ nent. The composition of the components D and E
in the location of the manipulated variable for the are controlled in structures 2, 3 and 4 in a similar
production rate. They are all self-consistent and manner to prevent their accumulation. Lastly and
have several of the same control loops as structure 2. very importantly, the recycle flowrate is maintained
Figure 2 shows the location of the production rate at a constant value in structures 2, 3 and 4 using the
manipulated variable for the structure already des- compressor recycle valve. This valve is kept at a
cribed and for the three additional structures. constant position in structure 1 to avoid interaction
Detailed information about these additional control with the reactor level control loop. None of the four
structures is presented elsewhere (Price et al., 1994). structures include control of the reactor pressure for
Structure 1 sets the production rate by manipulating reasons explained earlier.
the feed flow rate of D into the reactor. In order to
4. CLOSED LOOP PERFORMANCE OF PLANT·WIDE
reduce the production rate the flow rate of D is
CONTROL STRUCTURES
reduced. This lowers the production rate of G and
causes a drop in the level within the reactor. In this In this section we will concentrate on the perfor-
structure, the condenser cooling water valve is used mance of structure 2 and we will make some com-
to maintain the reactor liquid level. It closes in order parisons between structure 2 and the other ones with
to condense less product vapor and return it to the respect to their response performance to the
reactor as part of the recycle gas. The reduction in provided disturbances and the recommended set-
condensate flow to the separator is carried down- point changes. The discussion of the control struc-
stream by the separator and stripper level con- ture's performance is divided into three segments:
trollers resulting in a reduction in the product flow operating costs, some disturbance response results
rate. and the responses to the defined setpoint changes.
~
__
A ~
Structure 1

i ,
!!...........

~
[
€p ....® ISEPARATOR] g,
L->-" - "2 s
(1)
cws
;3
e
~
'"
'"
(1)
(1)

[STRIPPER! [I1
~
3"
@ '"
::l
"0
a
IREACTOR I CJ

"3
4

Fig. 2. TE process-locations of the production rate manipulated variable for four SISO control structures.
...,
N
Ul
326 P. R. LYMAN and C. GEORGAKIS

Table I. Total operating cost


another disturbance or setpoint change. In order to
Control structure satisfy this recommendation, a 15% setpoint reduc-
Disturbance 2 3 4
tion in the production rate, was selected since it
None 8459 8470 8471 8472 represents a plant-wide shift and therefore provides
1 8225 8226 8230 8228 a rigorous test for these plant-wide control struc-
2 14,144 14,042 14,038 14,037
3 8460 8471 8472 8472 tures.
4 8459 8470 8471 8472
8475
The results shown in Table 1 indicate that all four
5 8459 8464 8647
6 SD @7.5h 14,899 14,943 14,881 structures are capable of rejecting the majority of
7 8493 8470 8475 8473
8696
the disturbances and Structure 2 is capable of con-
8 8548 8678 8698
9 8459 8469 8470 8470 trolling the TE process simulation for 50 hand
10 8458 8470 8470 8471
8472
responds efectively to all disturbances and set-point
11 8459 8470 8473
12 8469 8489 SD @28h SD@ 19h changes postulated, A total of six process shutdowns
13 8371 8346 8647 SD @50h
6957* 6758* 6765 6775
occur in the other three structures all of them due to
14
15 6956* 6763* 6890* 6791* high pressure. Control structure 1 has a slightly
16 6957* 6758* 6765* 6775*
6965* 6778* 6854* 6824*
lower operating cost at steady state but will shut-
17
18 6992* 6812* SD@7.5h* SD @8.0h* down with disturbance 6. Control structure 3 has
19 6957* 6759* 6772* 6780*
6981* 6772* 6776* 6790*
two shutdowns: disturbances 12 and 18. Control
20
21 6957* 6758* 6765* 6775* structure 4 has three shutdowns: disturbances 12, 13
22 8563 8392 8544 8404
8461 8461 8460
and 18. Detailed descriptions of the results from
23 8347
24 7527 7559 7581 7562 each of the four structures with each disturbance is
* These runs include a 15% reduction in the production rate and
provided by Lyman (1992). For clarity and brevity,
can be expected to have lower operating costs. SD denotes process only selected plots will be presented here.
shutdown at the simulation time given.
Disturbance 12 and the setpoint changes are des-
cribed in more detail below.
4,1. Operating cost
The total process operating costs calculated from 4.2. Disturbance 12
the simulation runs that lasted 50 hours and repre- Disturbance 12 provides some interesting results
senting each control structure's response to the since it occurs in the center of the main process path.
different disturbance upsets and setpoint changes The control structures described above can be
are presented in Table 1. The operating cost is a sum expected to respond very differently to this distur-
of costs associated with the purge stream, the pro- bance since the production rate manipulated vari-
duct stream, the recycle compressor and steam able is in a different location for each structure.
usage and has been defined in the problem defining Disturbance 12, random variation in the condenser
paper (Downs and Vogel, 1993). At the base case cooling water temperature, has the effects on pro-
provided, about two-thirds of the total operating duction rate and reactor pressure shown in Figs 3
cost can be attributed to the purge stream. Since and 4. Control structures 1 and 2 transmit the
each structure incorporates identical purge stream variation out through the primary process flow
controllers the operating cost does not show large stream resulting in variation of the product flow
differences between structures. Disturbance 2 which rate. This is contrasted to control structures 3 and 4
is an increase of component B in stream 4 results in which maintain a nearly constant product flow rate
very high operating costs for all structures because by rejecting the disturbance's effect and propagating
of the high purge rate needed to maintain a low it back to the feed flow rates, through the recycle
concentration of B in the system. Disturbance 6, loss stream, resulting in increased reactor pressure varia-
of stream 1, also gives high operating costs since one tion, eventually causing the process to shutdown. In
mechanism utilized to remove excess C from the this case, it is preferable to transmit the disturbance
system is a higher purge flow rate. This is imple- out of the process as quickly as possible in order to
mented with the use of an override controller on the minimize its effect upon the reactor system. Control
purge stream flow rate. The override responds to an structures 3 and 4 could be made more robust by
excess of C in the sytem by purging more material to either reactor pressure control such that the pres-
control the composition of C instead of B in the sure variation could be removed from the process or
recycle stream. Entries denoted with a (*) symbol by relaxing the production rate control in appropri-
are runs in which the production rate is decreased ate situations to allow specific disturbances to be
and they should be compared only with other entries transmitted out of the process. To summarize, the
of the same type. The problem statement suggests location of the production rate manipulated variable
using disturbances 14 through 20 in conjunction with relative to a specific disturbance to a large extent
Control of the Tennessee Eastman problem 327

Product Flowrate, Disturbance 12


30r---....,...----,r__--r--~--..,_-....,..--r_-....,...--r__-__,

29
- Structure 1
28 - - Structure 2
.. Structure 3
~27 . - . - Structure 4
E
1126
~
~25
~
u::: 24
1)
~

~23
a.

22

21

2 4 6 8 10 12 14 16 18 20
time (hrs)

Fig. 3. Closed loop response of product flowrute to disturbance 12 for each control
structure.

determines the ability of the structure to reject that 4.3. Setpoint changes
disturbance. In the example provided above, it is
better to pass the disturbance through the process The following setpoint changes are defined In the
quickly using structures 1 and 2 rather than hold the postulation of the problem to allow comparisons
disturbance in the recycle loop as with structures 3 between structures and test their ability to move the
and 4. process to alternate operating points.

Reactor Pressure, Disturbance 12


3000r---,----r---r---,----r---r---,----r------,

2900

2800

(if
~2700

l!!
:::>
'~" 2600
0..

2500
- Structure 1
- - Structure 2
2400 Structure 3
. - - Structure 4

2300 '--- ....J...- - -'-- ---''--- ....J...- - -'-- ---'L-- ......L-- - -'-- --J
0 2 4 6 8 10 12 14 16 18
time (hrs)
Fig. 4. Closed loop response of reactor pressure to disturbance 12 for each control
structure.
328 P . R. LYMAN and C. G EOR GAKIS

includes the dynamic lag of the stripper column


while structure 4 could be retuned to give a faster
- s....",•• response. Figure 5(B) shows the response of the
- - Stuc\Jr.2
... .. Stue",.3 flow of D , the most expensive raw material , with the
. _ .• S...."'.4
same setpoint change . Control structure 1 turns
down the flow rate of raw materials the fastest as
expected while the other three structures are about
the same. In short, control structure 1 wastes the
fewest raw materials in a production rate reduction .
20
The impact on the reactor can be seen in Fig. 5(C).
Control structure 2 affects the reactor pressure first.
19.5
Although control structure 1 turns down the flow of
D and the other raw materials faster this has a much
20
more gradual impact on the reactor pressure since
Flow •• 01D. Solpon(11 the flows are small compared to the recycle flow.
3.00r-----.-...,.....-~____,-........-~___.~....,...-....... __,

The reactor pressure also equilibrates at the new


3700
-SANd 1 reaction rate sooner for structure 2. A production
- - Sttuch 2
3800 ... .. Swell."..3 rate reduction is easily accomplished using any of
. _ . ~ SlruaUl'••
the four structures. Those structures in which the
production rate is controlled towards the end of the
primary process path give a faster turn down of the
production rate as could be expected.
3200 4.3.2. Setpoint change 2. Setpoint change 2 is a
3100 shift in the product ratio of G/H from 1.0 to 0.67 in
mass units. The ratio of products is plotted for
3000
control structure 2 in Fig. 6. The setpoint change is
10 12 14 18 18 20 made directl y in the reactor feed composition loops
limol'n,
as a step change. The change in operating conditions
_ " '....... ~11
2'5O'r-~-""""-~___''''-'''''''-'''''''''-~____'-~---, is complete after 15 h. All four control structures are
2'00
able to move the process to produce this new pro-
duct rat io. The shift in operating point s takes about
/.[1':/'Vl'f't t \!':,:';./' 15 h using structure 2.
4.3.3. Setpoint change 3. The reactor pressure is
,' /.'0.1
r: stepped down in setpoint change 3. Since all four
structures are designed to let the pressure float, the
\\,,\:W//
, ~ :
pressure can only be adjusted by changing another
If' variable which is controlled. The reactor tempera-
2500
~' : : -s.ruc...... , tur e was found to be the best variable to use for this
- - S'rucU92
2450
.. .. $lruck.n3 purpose . A temperature shift of + l OC gives approxi-
_ . • Slructur.4
mat ely a 50 kPa reduction in reactor pressure . This
10 12 14 18 18 20
Imt~)
shift in reactor temperature was used with control
structure 2 giving the results shown in Fig. 7. The
Fig. 5. Closed loop response of product fiowrate, flowrate
of 0 and reactor pressure to setpoint change I for each pressure reduction per °C change in the reactor
control structure . temp er ature setpoint is greater for structures 2. 3
and 4 than for structure I (not shown). This effect is
4.3. J. Setpoint change I. The first setpoint a result of recycle flow rate control differences
change , which is a 15% reduction in production between the control schemes . For instance. in struc-
rate, is described by Fig. 5(A-C). Figure 5(A) ture I the recycle flow rate is uncontrolled; the
shows the response of the stripper underflow to a higher temperature gives a higher molecular weight
production rate change at 1 h. Control structure I recycle mixture however control structures 2. 3 and
gives the slowest respon se, as expected , since the 4 hold the molar flow rate const ant despite pressure
production rate is controlled at the start of the changes and the density shift. In the case of struc-
process . Control structure 3 is actually slightly faster tur e I. the molecular weight shift and the new
at turning down the production rate than control operating pressure cause the recycle flow rate to
structure 4. This is surpri sing because structure 3 drop . This reduced flow rate lessens the impact of
Control of the Tennessee Eastman problem 329

Ratio 01 GIH in Product. Setpoint(2)


1.05r---"T""""--r--~----,r---"T""""--r--~----,---.----..,

- Structure1
- - Structure2
0.95 Structure3
. - .- Structure4
~ 0.9
-.
C}
0>0.85
e
J:
<3 0.8
"0
.2
~0.75

0.7

0.65

5 10 15 20 25 30 35 40 45 50
time (hrs)

Fig. 6. Closed loop response of product ratio to setpoint change 2 for control
structure 2.

the recycle stream composition hence the pressure 4.3.4. Setpoint change 4. The composition of the
within the reactor is only marginally reduced. In purge stream is changed in setpoint change 4. The
summary, the reactor pressure can be changed indir- mechanism used to achieve the new composition is
ectly by manipulating the reactor temperature set- the same in each of the control structures since loops
point for control structures 2, 3 and 4. In order to 19 and 6 are identical in all four structures. Control
control the reactor pressure for structure 1 the structure 1 causes a step change in reactor pressure
recycle flow rate must be fixed. as shown in Fig. 8. Structure I rises to a much higher

ReactorPressure. Setpoint(3)
2720...---r----,r---....----r--.......--,---,.----r---~-..,

2710

2700

'i"2690
~
~2680
i
~2670
ia: 2660
2650

2640

5 10 15 20 25 30 35 40 45 50
time (hrs)
Fig. 7. Closed loop response of reactor pressure to setpoint change 3 for control
structure 2.
330 P. R. LYMAN and C. GEORGAKIS

ReactorPressure, Setpolnt(4)
2900

2850

<il2800
a..

!i',/
, NA ,t\,i~~i';': , fl ~;~~W ,I, ~i!t
~
~
.,
:>

~ 2750 II
;pr:~ : :t~};·¥/(!'V~~'A!v~N~~.tr\f:J:\Hti~Y:: <lN~r~IUW~:
l .·
f! ;"0: (;11;;iW
o,

~
'"
CD
a:: 2700
\n'j~F > ~ j',!;,
I~ . _ Structure1 1 : ' •
il - - Structure2
2650
..... Structure3
.- . - Structure4

2600
0 5 10 15 20 25 30 35 40 45 50
time (hrs)

Fig. 8. Closed loop response of reactor pressure to setpoint change 4 for each
control structure.

pressure than the other structures. Since the recycle in control structure 1 if the recycle flow rate were
stream is now lighter (B is a noncondensible and low controlled.
molecular weight component) the recycle flow rate
is higher. The additional flow of a lighter stream S. CONCLUSIONS
gives a much higher reactor pressure. In the case of
the other control structures, the recycle flow rate is The TE process was found to be particularly
held constant by loop 5 despite the shift in the difficult to control because the recycle stream and
recycle stream molecular weight. This results in a the reaction characteristics introduce the potential
lower pressure rise compared to control structure 1. for excessive interaction between reactor tempera-
Control structure 3 gives significantlymore variation ture, reactor pressure, reaction or production rate
in reactor pressure than any of the other structures. and reactor level. Despite these difficulties, four
The same behavior was observed with structure 3 plant-wide SISO control structures were developed.
without any disturbances present and is attributed to The development of these structures is consistent
interaction between the separator level control loop with the tiered structure development framework.
and the recycle flow rate control loop. In structure 3, In this method, the production rate manipulator is
the separator level is controlled using the condenser chosen first and located on the primary process path.
cooling water flow rate and recycle flow rate loop is The inventory control loops function in an outward
the same as shown in Fig. 1. In light of this evalu- direction from the production rate manipulator.
ation of control structure performance, control This ensures that production rate and inventory
structure 2 is selected as the best of the four studied. controls are self consistent as described in Price and
This control structure has reasonable operating costs Georgakis (1993) and Price (1993). Control struc-
and more importantly is able to withstand all of the ture 2 offers the best control system of those studied
disturbances and setpoint changes without exper- for the TE process. This structure weathers all
iencing a process shutdown. Control structure 1 is disturbances and setpoint changes without exper-
the second structure of choice since it only shuts iencing a process shutdown. It has the further bene-
down with disturbance 6, the most severe distur- fit of being composed of a simple set of SISO loops
bance. Also, structure 1 has a lower operating cost and a few cascade systems. This would enable it to
with no disturbances. The reactor pressure can be be easily implemented and understood by operating
manipulated by changing the reactor temperature in personnel. If additional complexity could be toler-
control structures 2, 3 and 4. This could also be used ated a supervisory level of control may allow for
Control of the Tennessee Eastman problem 331

indirect control of the reactor pressure. The feasibi- APPENDIX


lity of this algorithm was shown by the use of the Controllers and tuning decisions are based on the fol-
reactor temperature setpoint to change the reactor lowing criterion. Level controllers for the separator tank
operating pressure in setpoint change 3. The deve- and the stripper column bottoms use proportional action
only with a dimensionless gain of one. The range of the
lopment of the supervisory algorithm is left for valve is set such that a maximum allowable level corre-
future work. sponds to a fully opened or closed valve depending on if
the valve is located downstream or upstream of the vessel.
Acknowledgements-The authors would like to acknow- respectively _ Series cascade arrangements arc used for
ledge William L. Luyben for helpful discussions during the reactor level or composition control. In these cases. the
course of this study. slave portion is a very simple flow controller. Thc slave
controller uses proportional action only while the master
loop uses proportional-integral action. The slave controller
is tuned such that the closed loop time constant is one half
REFERENCES the open loop time constant. The ultimate gain and period
for the master loop are calculated from an integrator-
Downs J. and E. Vogel, A plant-wide industrial process deadtime model of the open loop response of the
control problem. Computers chern. Engng 17, 245-255 controlled variable to a step change in the manipulated
(1993). variable. In this arrangement the manipulated variable 01
Lyman P., Plant-wide control structures for the Tennessee the master loop is the set point of the slave loop. During the
Eastman process. Master's thesis, Lehigh University identification procedure. the step change is implemented in
(1992). the manipulated variable of the slave loop and not the
Price R., Design of plant-wide regulatory control systems. manipulated variable of the master loop. This difference
Ph.D. dissertation, Lehigh University (1993). may be neglected if one assumes that the dynamics of the
Price R. and C. Georgakis, Plantwide regulatory control slave loop are sufficiently fast as to be negligible. The usc
design procedure using a tiered framework. Ind. Engng of proportional only control for the slave loop helps to
Chern. Res. 32,2693-2705 (1993). ensure that this is the case. Once the ultimate properties
Price R., P. Lyman and C. Georgakis, Throughput mani- are obtained. the controller tuning is calculated using
pulation in plantwide control structures. Ind. Engng settings described in Tyreus and Luyben (1992). These
Chern. Res. 33,1197-1207 (1994). settings are referred to as TL settings and are less oscilla -
Tyreus B. and W. Luyben, Tuning PI controllers for tory than Ziegler-Nichol settings. The remaining loops usc
intcgrator/deadtime processes. Ind. Engng Chern. Res. proportional-integral action and arc tuned in the same
31,2625-2628 (1992). manner described above for the master controllers.

CACE 19-3-F

Das könnte Ihnen auch gefallen