f~~~~~~~~~~ J ~I~~~~~~;~~.~~~.~~~~~~~ …whitehouse-consulting.com/Improved training of... · Control Engineers Distributed control systems (DeS) have been in use for over 30 years.

f~~~~~~~~~~.~~I~~~~~~;~~.~~~.~~~~~~~~~c ~~~~~~~~~c~~ JThemCl: Industriele Clutomatisering

Improved Training Of ProcessControl Engineers

Distributed control systems (DeS) have been in use for over 30 years. Yet

the majority of the process industry does not properly understand many of

their features. On almost every process there exists the opportunity to

improve basic control and thus to increase process profitability. The key to

this is to provide better training and design tools for the engineers that

implement and support such systems. The problem is that most of the avai-

lable training does not properly address practical issues and most of the

design tools fail to deliver effective controllers.

Mr. M. King is Director of

whitehouse Consulting,

Redway House, East Lane,

Merstone, Isle of Wight,

P030 3D),United Kingdom.

E-mail:

[email protected]

tel.: +44 (0)1983529931

Control engineers are of course aware that good basiccontrollers will improve process operation. The pro-cess will recover quickly from disturbances - maintai-ning product quality constant and enabling plantconstraints to be approached more closely.Withoutgood control it may be necessary to operate wellwithin product specifications to ensure that disturban-ces do not cause off-grade production. This 'giveaway'incurs increased operating costs and reduced productyield. Similarly it may be necessary to operate wellaway from other limitations to ensure that criticalequipment constraints are never violated, thus under-utilising plant capacity. It is common for improvedcontrol to increase process revenue by several percent. While the process industry is generally good atensuring that the hardware of the instrumentation issound, problems arise in configuration both fromusing the wrong control algorithm and from imple-menting the wrong tuning.

Fig. 1 Process flow

diagram Choice of ControL ALgorithmThere is not always an awareness that the DCSoffers awide range of control algorithms. Or, if the engineerhas read the manual thoroughly, he or she may notunderstand why there are so many choices. Not fullyappreciating the benefit of each algorithm the engi-neer will select the default or the one that most close-ly matches his or her understanding of proportional-integral-derivativ:e(PID) control. In almost every cir-cumstance this will lead to the wrong selection. The

engineer will have missed the opportunity to significantlyimprove the response of the process to disturbances -needlessly extending (typically by a factor of three!) thetime that the unit takes to recover.

Consider subjecting the process in Figure 1 to a reductionin feed rate caused by the operator reducing the set pointof FCl. The two heater outlet temperature trends are inresponse to the same disturbance to feed rate. Each algo-rithm was tuned to give virtually the same response totemperature set-point changes. Algorithm A, chosen bymost, is the conventional version of PIDcontrol. Careneed be taken in understanding how the DCSvendor hasconverted the algorithm to the discrete form but, in conti-nuous form, it can be represented as:

[If dPV]M=Kc E+- E.dt+Td--t; dt

•

t-e---

npt procestechnoiogie CD februari 2005

and algorithm Bas:

[If dPVJM = s; PV + T; E.dt+Td Tt

Controller TuningThe next challenge is ensuring the controllers are pro-perly tuned. Many are not. Probably the most commonexample is 'averaging' level control. Rather than thecontroller being tuned for a fast return to set point,this permits the vessel level to vary (within alarmlimits), thus minimising changes to the downstreamflow. There are many situations where level control-lers can exploit surge capacity within the process. Thisreduces downstream disturbances - often givingremarkable improvements to process stability. Manycontrol engineers appreciate this, but few properlycalculate and implement the correct tuning.

Figure 2 shows the effect of changing from tight toaveraging control. Tuning constants for tight controlcan be derived from:

K = 0.8VC F.ts

T=_V_, l2.5f

and for averaging control from:

K = 80fC F.d

T=~t l2.5f

Care is needed in ensuring that the engineering unitsused for each parameter are consistent with the wayin which the PIDcontroller is coded. In the examplecited, the corrective action taken by the averaginglevel controller is almost 3000 times slower than thattaken by the tight level controller.

Further, few understand the adaptive nature of the'error squared' algorithm, i.e.

zeo •..... , ....

'"r<:

V~m I

/ ~\zse r \ i

\~, i-:/~ ~'=--".

OlV.~~~~~i~hl':!I(;:~~rOl)Im ..

" s to rs zo " " "rime~min"l") •••

•

[If dPVJM=KcIEI rr »= E.dt+Td-T; dt

Used normally only for averaging control, tuning canbe similarly calculated. But the formulae can varygreatly - depending on how the vendor has coded thealgorithm. '

Another example is the use (or not!) of derivativeaction. It is seen as additional complexity, making thecontroller more difficult to tune. Many believe thatderivative action should only be applied to temperatu-re controllers. Derivative control is highly beneficial ifthe process dead time (8) is large compared to the pro-cess lag (t). Its anticipatory nature helps the controllerrespond much more quickly to disturbances. Manytemperature controllers have such dynamics, but notall. The use of derivative action here would bring littlebenefit and could cause stability problems. Similarlythere will be many other controllers where 8/L ismuch greater than unity; ignoring the benefit of deri-vative action will greatly extend the time taken by theprocess takes to recover from disturbances. Figure 3 isbased on the use of the conventional PIDcontroller ona process that has a lag of one minute. Tuning con-stants were determined to minimise ITAE(integralover time of absolute error) while avoiding excessivecontroller output overshoot. It illustrates the impactthat dead time (8) has on the choice ofh

Multivariable control (MVC)packages are becomingincreasing common. These adjust the set points of thebasic controllers to permit further increases in processprofitability by more closely approaching operatingconstraints. As part ofMVC design, a major 'step-testing' exercise must be undertaken to obtain theprocess dynamics. On complex processes this can takeseveral weeks, often working shifts to cover round-the-clock testing. It is not something that the plant

• Fig. 2 Response to

changes in feed rate

,................................

----- /"»> ,

»>//

-:-:

/'

.>01

Fig. 3 Effect of process

deadtime,::; 0.5

0.0

0.0

t----------,--t-------;---~-~~Tuuu--;--J

Definition of Symbolsd maximum deviation from set pointE error between PVand set pointF instrument range of manipulated flow

controllerf normally expected flow disturbanceKc controller gainM controller outputPV process value (measurement)t timeTd derivative action timeTi integral action timets controller scan intervalV vessel volume between 0 and 100% oflevel

indication

Themo: Industrtele outomotisering

owner would willingly undertake more often thanabsolutely necessary. Changing a basic control algo-rithm, or re-tuning one, changes the process dyna-mics. Thus, once step testing has started, the engineerhas effectively committed the site to retaining poorbasic controllers at least until the next major processmodification, when step testing would have to berepeated in any case.

ConclusionsThe first priority in addressing these problems is trai-ning. However care should be taken in selecting thebest provider. DCSvendors are generally poor atexplaining why their systems support so many diffe-rent algorithms. MVCvendors provide good product-specific courses covering their technology but usuallytry not to become too involved with the basic controls.Academic institutions provide a broad range of cour-ses but usually only address theoretical issues usingmathematical techniques that have littlerelevance to the process industry.

The control engineer will need aneffective tuning aid. There is a bewil-dering array of methods. Almostevery month a journal publishes anew one or a new product is announ-ced. With only a few exceptions thesemethods are usually flawed. Rarely dothey account for the variety of algo-rithm types and usually they applyincorrect tuning criteria. In selectinga method, care should be taken inensuring that it handles both self-regulating and non-self-regulating(integrating) processes. The version ofPIDalgorithm it assumes should beidentical to that configured by theDCSvendor. The method should bebased on discrete, rather than conti-nuous, control. And it should allowthe user to specify the tuning criteria.

Finally the control engineer should begiven time to address the basic con-trol problems. Effort should be focus-sed on those controllers where a noti-ceable improvement is possible andadequate time should be allowedbefore beginning any installation ofMVC.