Professional Documents
Culture Documents
Modeling
To Optimize
Process Control
A control-room operator
typically works off a set of
throughput, yield and quality
targets established by the
production department
n the chemical process industries In the operations area, however, sig- and data servers, and reduces the need
CPI : . the drive to improve safety, nificant improvements in the prof- for interface equipment or gateways .
around
product quality, product yield, turn- itability of the process can still be real- The most common fieldbus standards
time and overall profitability ized by increasing asset utilization . or in rise today are Foundation Fieldbus
of ` tile " process is relentless . Industrial- "sweating the assets ." Process control and Profibus . Many of the benefits of
grade computers, with superfast cen- technology can be utilized to raise the information integration can also he ob-
tained with conventional 4-20mA
tral processing units and ever-expand- effectiveness of the total plant-control or
ing memory and bandwidths, can be infrastructure, from field devices to Hart systems.
had at relatively low cost . These hard- process optimizers. Among the new
ware advances have been matched by technologies being implemented are : Managing the control loop
significant software. developments in " Fieldbus and intelligent field de- In a typical plant control system,
information and production manage- vices to improve control-system in- shown conceptually in Figure 1, the
ment. systems, such as enterprise re- frastructure operator works off of a set. of through-
sources planning (ERP) and manufac- " Loop performance monitoring using put., yield and quality targets estab-
turing execution systems ( MES). statistical models to improve basic lished by the production department .
Typically, ERP is focused on financial regulatory control More closely related to the process op-
and transactional information . MES " Process modeling far advanced ap- eration, specific target values for
has evolved from its origins as the link plications to improve overall process variables, such as individual
between the process and the enterprise process performance pressures, temperatures, flows, and
to take on a larger role in optimization An integrated information system be- levels, am based on design and safety
and decision support for ebusiness . gins with intelligent field instruments, considerations and a history, of opera-
Equipped with these tools, companies positioners and actuators. Standard tional experience .
are further integrating plant informa- open architectures minimize network process variables are measured
tion to streamline production manage- and systern incompatibilities . This en- using instruments and analyzers . Mea-
ment and boost profitability . ables tight integration of field devices surement signals are transmitted to
0 Ideal on
U
C control
E
t Optimization
n potential
r
a
0 " Off control - - Poorly
6 toned on
Increased c control
production
Loop Loop or quality
variability , tuned Time
FIGURE 2. Profitability of the process can be improved by FIGURE 3. Poorly maintained control loops can compromise
reducing the variability in process control loops product yield, quality and throughput 1171
the control system, and am used by the itable operating conditions without vi- control loop performance, how does one
various automatic controllers to deter- olating any plant. constraints . Yet know if a loop is correctly tuned" The
mine lire required actions to keep the studies show that many" plants have a quickest method is to observe a trend
plant at target conditions, Signals rep- number of poorly performing loops - plot of the controlled variable typically
resenting these actions are -sent to and some plants have up to a third or flow, pressure, temperature ) versus de-
plant actuators such as valves and more of their loops on manual control' sired setpoint, If the controlled vari-
motor drives to control the process to Ideally, trained staff of the plant in- able is not. a t its target . or if oscillations
the desired conditions, With modern strurnentation and control (I&C) appear in the trend, then the loop tun-
digital control systems in most plants, group maintains the performance of a ing needs to he adjusted . Worse, if the
information from the process and the process control loop, This group should amplitude of the loop increases, then
control system is visible to the operator check control loop performance on a the loop should temporarily be put in
on a human system interface ( I HSI I dis- regular basis and have analysis tools manual control, as the tuning is caus-
play WE . November 199h, pp . s2 941, on hand to facilitate work, The follow- ing an unstable condition .
In CPI facilities, such as petroleum ing act ributes are key to the success of There are standard manual and
refinerles, chemical plants, pulp and the I&C group [1]: computer-based timing techniques
paper mills, mining and metals facil Training: Qualified members of the available that. can be used to determine
ities, and food and pharmaceutical I&C group should have froth formal how well a control loop is tuned. Man y
plants, there may he anywhere from training and field experience. The ex- companies now use computer-based
tens to hundreds or thousands of tent of training should be based on job tools, because then are usually more
basic control loops, Inopportunely, requirements. Some universities and precise than traditional tuning-mile
while the operation of the process automation suppliers provide formal methods, such as Zirgler-Nichols A
control loop can be described simply training, Typictolly, automation suppli- good computer-based control tuning
enough, there are a number of areas ers offer courses :r. lair! of their train- package should be capable of simulat-
where less than optimum perfor- ing curriculum for prices
., control ing the Imp and providing
mance can occur, resulting in lower Regular performance checks : The recom-mended tuning parameters based on
plant profitability, For example, tuning for ;ill (-one rol loops should be the control algorithm and the tuning
process measurements may be sub- checked nn it regular oasis Follow- criteria The ability" to
ject to noise, calibration errors, or in- ing any process changes to piping, specified. simu-late the Impwil help confirm that the
strument drift, Control valves may instrument or actuator location, new tuning parameters will provide ac-
stick, display hysteresis, he subject to trained personnel should review con- ceptable loop performance,
wear, or respond with poor accuracy" trol loop performance, Changes in Clearly there are benefits from tun-
due to improper sizing or lack of a po- process operating conditions, such as ing up and maintaining the basic
sitioner. Control tuning may he solo- a significant change in feedstock process control loops in a plant, In ad-
optimal due to process nonlinearities quality or quantity or a change in dition .
or to the process moving outside nor product grade, will also require a re-
mat operating ranges, view of affected control loops. If no studies[2] have shown Ihat ad-vanced control can lava processindus-tries 2- 6% 111 annual operating cost-:,
Properly tuned loops enable plants process changes have occurred, then
to run smoother and more safely, A a rule of thumb is to check basic Using statistical tools
fully functional and tuned regulatory loops such as flow, pressure and tem- Micro n control I+lop is first commis-sioned.itsperformance is usually
control system is the foundation for perature everv six months, Loops
an y" advanced control application. Un- that include special measurement very good . Over time, however, per-
less the basic regulatory control sys- equipment, such as stack gas analyz- formance tends to degrade for rea-
tem is operating at its best, the bene- ers. pH meters or gas chro- sons already noted. Figure 3 shows
fits of advanced control are likely to be matographs, should be checked more the typical progression of control
compromised, Bv reducing the vari- frequently, Both loop-tuning para- loop performance.
ability in the regulatory process con- meters and analvzer calibrations 'there's no doubt that loop timing
trol loops, as shown in Figure 2. ad- should be checked k,, according to man- improves operations . Yet., how often is
vanced controls can then be used to I it n done? As previously noted, loop per-
push the process toward more prof- Proper
ufacturers' loop tuning : 's1 When
recommendations checking formance ideally" would he managed
CHEMICAL ENGINEERING WWW.CHE .COM AUGUST 2001 67
On control
Off control
Y, (t) +~ GUltl G
u
r * controller -- k process
5etpoint
z " ai .rl
FIGURE 4. Standard deviation provides a benchmark for FIGURE 5. In a control loop, the controller tracks the setpoint
controller performance for each variable
by the trained staff of the I&C group, trail, related to each device, and as- models for the controller, plant and dis-
In some cases, a member of the plant. sist in documenting information rele- turbances leg, measurement noise and
I&C staff actually has the time to reg- vant to individual devices such :i . cal- or either other disturbances, respec-
ularly consider control loop perlor ibration data, tively . In a well-performing controller,
mance . Occasionally, a control special- Such tools allow the stove from cor- the controlled variable should track the
ist is assigned from elsewhere within rective maintenance Unexpected re- setpoint and the deviation between the
the company, or one is brought in from pairs V through preventive maintenance two should be minimal - ideally con-
a vendor to manage loop performance, ;elapsed-lime-scheduled repairs, to sisting of only " small random noise,
Too often, however, loop mainte- predictive maintenance based upon re- However, process dynamics are charac-
nance goes undone, and operators altime device assessments, Smart de- terized by some amount of dead time,
must compensate far any control loop vices and predictive maintenance poli- -Noting that controller action can take
deficiencies, In the meantime, produc- cies have proven cost saving-. . place no sooner than the process dead
tion suffers due to reduced quality, Automated technology for time, a minimum variance controller is
yield or throughput, An automated ap- monitor-ing control lo p performance is being defined simply as that which restores
proach to this problem is warranted . used successfully in industrial envi- control as soon as possible once the
With the availability of historical ronments . The theoretical basis for dead time has elapsed . That is, it mini-
and realtime plant data, statistical and many of the benclunarks used for loop mizes the variance of the output,
other computational modeling tech- monitoring comes from statistical Using the benchmark that a mini-
niques are being used increasingly to models. Relatively simple statistics mum variance controller is best, then
track performance of a plant:: asset-c . such as on-slream time and the stan- the ratio of the output variance or.-, of
For example, spectral analysis has dard deviation, ~t, of the controlled a minimum variance controller to the
been used for a number of years to de- variable about its setpoint provide in- output variance it -, of the actual con-
tect . vibration in rotating equipment, dicators of control loop performance . troller can be used to define a loop
The use of statistical process control It. may seem too obvious to state, quality index (LQI) 131, Defining a as
tools is well known as a key component but a control loop must be on ,auto- the mean deviation of i from v, to ac-
for maintaining production quality lev- matic operation) to have any positive count for steady offsets from zero then :
els in a number of process operations . effect on process performance, Sur
This combination of plant data and prisingly, during a plant ,survey of d r
LQI- 1-d'-r,
:'~
- .1
analysis tools has evolved into asset control loops, toe finds that many
optimization or asset management mops have been turned off (replaced
programs . The objective of these pro- by manual operation, for one reason This index, defined by Harris and
grams is to optimize overall equipment or another, and have never been Desborough, is used by" a number of
effectiveness by focusing on turned back on, For loops that are on, industrial and commercial control-
maximiz-ingasset availability while minimiz- the standard deviation provides an in- loop per] formance monitors 1.51 An im-
ing operating and maintenance costs dication of performance, Often, two portant aspect of the index is that it
over the life of the asset, sigma is the metric for loop perfor- provides numerical bounds for con-
How can this approach be used to mance. The smaller the standard de- troller performance benchmarking,
improve the performance of the hun- viation the better the control, a .. illus- The lower bound for the mdex is zero,
dreds of control loop "assets" in a trated in the histogram in Figure 4 . When LQI = 0, the controller is indi-
plant" Asset optimization exploits the In recent Years, there has been con cating good performance_ If the LQI is
information available from intelligent . siderable development beyond these near zern, yet the output variance is
devices, Profibus and Foundation and other simple loop-performance unacceptable, changes such as process
Fieldbus digital protocols. a s well as metrics, Automated assessment of con- modifications or new control design
analog protocols such as Hart, sup- trol loop performance uses the concept will be necessary to improve perfor-mance, The upperbound is LQI = l.
port digital status signals, called of minimum variant-(- control 13, 41 .
maintenance triggers . These Consider, for example, Figure 5, the When LQI is not zero, there may be
numeri-cal or computedvariablesair inter- block diagram representation of the opportunities to reduce output vari-
preted to trigger a maintenance re- control loop from Figure l, Here, v, rep- ance using the existing controller do-
quest ant] initiate a work order in a resents the setpoint . F repre ents the sign by improving the loop tuning or
Computer Managed Maintenance loop output or controlled variable, u is the control valve performance,
System (CMMS) Software can also the control signal, and u, is white noise, LQI is relatively" simple to imple-
manage work orders, provide audit (: G;, and C ;, represent. mathematical ment and to use, since only normal op-
Model
Model Current
Measurements tuning settings
Process
Estimator
model
. -
Measurements
MPC
Process MPC 1 eng. tool
FIGURE 7. Model predictive control (MPC) allows a controller FIGURES. With a processmodel, MPC is mom efficientfornonlinear
to be modeled and refined before it applied to the actual process process control becausethe controller canbe adaptedin realtime
skill: available, and it has been ap- Return on asset investments their instrumentation and control sys-
plied successfully in the CPI 1101 . Process operations are asset intensive, tems . Using the latest model-based
Some of the more challenging issues and profit margins are often slim . technologies, these processors can im-
surrounding NIPC include configura- Small deviations from optimal process plement advanced control schemes,
tion, tuning and nonlinear applica- and maintenance operations are costly operate closer to constraints safely
tion .For configuration, a standard as the "economies of
scale" peak at a turn around faster and perform "what.
controller requires process informa- certain point and turn around to work if simulations. These process ad-
tion that is generated typically by against the profitability goal . vances contribute directly to the oper-
making small perturbations to process Market leaders on every sector of ating company's bottom line,
variables. Sometimes, process infor- the CPI can do more than manage Edited in Deborah Hairston
mation from existing variations is
used to minimize the need to Perturb References Authors
1 Adler I u ,[l1q . //que stion 1 paperloop .com 1 rty Richard Lu l :, .
the proces . Product quality is under- ;"t' , I d. l I I :,- :3 ub-
"
turned when perturbations use is un- 2 Marlin, 'T E., Perkins, .1 . D. . Barton, G. \4 . ,tinns tl :U:[nn'rrtior.
and Brink, \I . 7--. Benefits room process con- 1t.lef,'j t`rodudr Iii":. oh
avoidable- This provides a transfer trol R ult et a court indu stry university L lac e Lt a \1 'Ti, ., C:,
function model as described above . A Control, Vol 1. pp .
57 -83. 1991 1.3 L173 -7140:
rigorous model call eliminate the need 716- '13, Emaf ;
to do this by providing a basis for de- 3. Astrom h Into duct ion to Stochastic Control [ rich r Cu . nblrconl~.
Theory Academie Press l, New York, 1470 .
termining a controller model offline, L "I~~,ru1 . mark,"-ine :md re
t Morris . T I \' m n[ -t Omitted loop per :r. wt .I,c:eitic:rtion Ii,r
and then, through simulation, allow formance t Journal of C. " mical
d ian
1,I , :I puntizat, Iriltn :nioi,ic
for controller refinement before it is Erg 6., pp. 856-861. cr .1 I It :, -n Ill nul .1111 n u" . I ,I I n :Il r:cn :,-
c: o msn :,I;c'rsnl lo,a,nrd: Kt, r...,-d in :tppli
applied to the. real process. 5. Harris T. ante
. Desborough h, L . Perf . rmance
c .u n I l:l t , I c, t l,a sl. par: in
assessment measures t r univariate . feed-back h. . t, . Ill: A1311 H, h,l I' a h,l) ir6
Applying a suitable process model
Eng., 70, pl, . 11,6-1197. .1 ,u1.t1 rnl;tIauinq Ill III till I'nivcrsity of
to the process improves MPC and r tL. .1 .R,I-: . .1v L:,1:u,ll
takes it a step further. With this ap- 6. Harris, T, .l nndSrppr. :,, .(' T. . hcc."n[devel- Rn,hnrt7 E. (Rick. sdunotwr
cpnv,nl it controller performance monitor ti alni,al spcci:dtat in the
proach, one can configure, tune, and
ing :n ( assessment t . I . techniques Chemical ' .lobo 1 tl ,. :Ir.1_v Pr-l
:study prospective control strategies Process Control \°I . 'III I" _*Q, , r" Ihv .,f .4RR Inc. II'hene-
\riz uI , : .tn t ts' .4ass5-
without perturbing the process. Fur- . F." n:uf. n l, .vrd ahmelu"r
thermore, reduced implementation 7. 1 Hugo A Pr,ws,. : o: :, I 11 p, aar - :ut,e :rs- a~t.,r, r', Ire ,em ,Lv
ses" nr nl . Heir,l :,r 7 " r April : . .1 = :: .1 ii-Ipa :rnt f,,r thr,
time and, as a result, improved control 2001, .,p s5-90 - .v .19 onNn'nn:aI of ABB'a
make this approach very attractive, as d -ins,"" d aatonotion sn-
8. Qin, 4. .1 ., Control performance monitoring latd.ns. Pt:er to - non¢ ARP,
shown in Figure 7. a revise' and assessment Computers and ~:Iun cc':,er ,a .r7:od R.r RP,
Chemical Engineering 2:i .:21 . pp . 173-156, ,'here rte held n numb- of IuWlt`w.n8 in
process,
The model approach has particular 1998.
. ., . . .sing :uid de,sloptnont III also
power in the case of nonlinear ea.rk for 91,e71 Ilevelopman[ Co . . to honest) .,
Ell Kozub 11 J \I t:u oty and diavm , of ss r It t "It!, 6 1 lu. .
processes. Classically, the simplest eh.. . . . . ... . 1 116 s .,,1, '601 automated pr c".sv s nu I pr :mot , I? I1, ;, Ph 11 111
control lb local t4u Control l, 1996, R,,",
method to invoke nonlinear NIPC is to I . : l ester L'ntrer-
1