Distillation Control & optimization An Ebook by Béla Lipták
proloGue
G
iants o the automation automation industry are ew ew.. Béla
plants. Although volumes have been written on dis-
Lipták is one. From his beginnings in the
tillation control, most columns are not optimized. In-
Hungarian resistance movement to his vol-
strument and process engineers will nd a practical,
umes o work on automation and control, he has ha s proven
readable treatment o the subject that can be applied
himsel to be one o the great thinkers thin kers and authorities authorities in
quickly and eectively. e ectively.
the industr industry. y. In this ebook, Distillation Control & Op-
Like other publications rom ControlGlobal.com, ControlGlobal.com, the
timization, he provides a concise,
Distillation Control & Optimiza-
yet complete analysis o distilla-
tion ebook is eective without
tion control, showing “the po-
being laborious. It provides oSM
tential or savings through better
cus on the important aspects o
control and optimization.” optimi zation.”
the subject or the time-pressed
In a practical, intuitive w ay, ay, Lipták emphasizes reli-
engineer who is responsible or distillation optimiza-
able stable control strategies aimed at optimization o
tion. We are proud proud and honored honored to assist in bringing bringing
process perormance. perormance. He claims claims that these optimiza-
this important work to the automation community.
tion techniques can improve “productivity and protability by 25% 25%.” .” I believe him. Distillation is arguably
martin Berutti
the most complex unit operation in most processing
MYNAH Technologies
www.controlglobal.com
––
proloGue
G
iants o the automation automation industry are ew ew.. Béla
plants. Although volumes have been written on dis-
Lipták is one. From his beginnings in the
tillation control, most columns are not optimized. In-
Hungarian resistance movement to his vol-
strument and process engineers will nd a practical,
umes o work on automation and control, he has ha s proven
readable treatment o the subject that can be applied
himsel to be one o the great thinkers thin kers and authorities authorities in
quickly and eectively. e ectively.
the industr industry. y. In this ebook, Distillation Control & Op-
Like other publications rom ControlGlobal.com, ControlGlobal.com, the
timization, he provides a concise,
Distillation Control & Optimiza-
yet complete analysis o distilla-
tion ebook is eective without
tion control, showing “the po-
being laborious. It provides oSM
tential or savings through better
cus on the important aspects o
control and optimization.” optimi zation.”
the subject or the time-pressed
In a practical, intuitive w ay, ay, Lipták emphasizes reli-
engineer who is responsible or distillation optimiza-
able stable control strategies aimed at optimization o
tion. We are proud proud and honored honored to assist in bringing bringing
process perormance. perormance. He claims claims that these optimiza-
this important work to the automation community.
tion techniques can improve “productivity and protability by 25% 25%.” .” I believe him. Distillation is arguably
martin Berutti
the most complex unit operation in most processing
MYNAH Technologies
www.controlglobal.com
––
Distillation Control & optimization
D C d o – p 1
FeeD Co Controls 22
Maximizing Feed Flow . . . . . . . . . . . . . . . . . . . . . . . . . 22
tHe proCess 4
The Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Feed eed Tem Tempe pera ratur turee and and Entha Enthalp lpyy Con Contr trol ol . . . . . . . . . . . 24 reFluX Co Controls 24
Col Column Vari ariables and The Their Pairi airing ng . . . . . . . . . . . . . . . 7 Assigning Assig ning Variables, Relative Gain Calculations . . . . . . 7 Composition Control 9
Indirect (Tempera (Temperature-Based) ture-Based) Composition Composition Control . . 9 Composition Me Measurement . . . . . . . . . . . . . . . . . . . . . . 10
D C d o – p 3 Controlli llinG tHe WHole Column 25 Constant separation 27
Analyzer Selection Select ion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Maximum Re R ecovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Analyzer Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Com Composi positi tioon Cont Contro roll Using sing Anal Analyz yzer erss . . . . . . . . . . . . . 14 Smith Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Com Compos positi ition Con Control trol o Two Prod Produ ucts cts . . . . . . . . . . . . . 27 Two Products With Interaction . . . . . . . . . . . . . . . . . . . 29
Triple Ca Cascade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Feed Com Compositio tion Com Compensation . . . . . . . . . . . . . . . . . 30
Selective Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
multiple proDuCt Di Distillat lation 31 31
Multivariable Control . . . . . . . . . . . . . . . . . . . . . . . . . . 31
D C d o – p 2
Model-Based Co C ontrol . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Articial Art icial Neural Networks (ANN) . . . . . . . . . . . . . . . . . 34
pressure Control 16
Cool Coolin ingg Wat Water er Cont Contro rol, l, Negl Neglig igib ible le Iner Inerts ts . . . . . . . . . . . 17
The To Total Mo Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Sub-Op b-Opti tim miza izatio tion o Un Unit Ope Operatio tions. . . . . . . . . . . . . . 36
Distillate With Inert s . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Operating Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Bypass Co Control Co Congurations . . . . . . . . . . . . . . . . . . . 18 Optimization: Mi Minimum Pressure . . . . . . . . . . . . . . . . 19 Vacuum Vacuum Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
total optimization 39
Product Price Considerations . . . . . . . . . . . . . . . . . . . . 39
Reboilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Operating Constraint s . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Vapor Recompression . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
www.controlglobal.com
––
Distillation Control & optimization
p 1
D
uring the global transition rom the present oil-based economy to one based on clean and inexhaustible energy, the importance o distillation distill ation will increase. Ethanol will need to be distilled, dist illed, and new oil reneries will need to be built. It is hoped that the designers o these new reneries will benet rom this di scussion, which describes the state o the art in the control and optimization o distillation processes.
Distillation is a common, energy-intensive energy-intensive method o separation in the petroleum, chemical, ood, pulp and paper and pharmaceutical pharmaceutical industries. Globally, Globally, more than 80 million barrels o crude oil are rened daily. daily. In the t he U.S., 146 reneries operate, employing over 65,000 people and producing a total value that exceeds $151 billion. But no new renery has been built in the U.S. since 1976, 1976, and the t he technology technology in use today is not much dierent rom that used on the rst distillation columns in the 19th century. The thermal energy requirements o distillation are enormous. The thermodynamic eciency o distill ation processes is less than 10%. The amount o energy used or distillation is approximately 8% o the total energy used in the industrial industria l sector o the t he U.S. Reneries Reneries spend 50% to 60% o their operating costs (excluding excluding capital costs and depreciation) on energy, energy, while the t he chemical industry spends only 30% to 40%. This T his dierence di erence shows shows the saving potential o implementing implementing better bet ter control and optimization o the distillation process. My main reason or writing this ebook is to show how these increases in efciency and savings can be accomplished accomplished by the applicaapplication o state-o-the-art process controls.
I will rst describe the distillation process, its components components and dynamics. Next I will discuss the traditional PID-based control system congurations, which are still used in the majority o the operating reneries. I will conclud concludee with a review o st ate-o-the-art distill ation optimization strategies, including including multivariable, model-based and articial neural network-based control systems, where the process variables are used only as constraints, and the production rate and eciency are continuously optimized. I hope to show that advanced controls can cut the operating costs and internal energy consumption o reneries and other distillation processes by 25%. tHe proCess
Distillation is the most requently used separation process. It separates the components o a mix ture on the basis o their boiling points and on the dierence di erence in the compositions o the liquids and their vapors. The product purity o a distillation process is maintained by the manipulation o the material and energy balances. Diculties in maintaining that purity arise because o dead times, t imes, nonlinearities and variable interactions. interactions. Distillation can be perormed either as a batch or a continuous operation. The main dierence between the two is that t hat in continuous distillation, the eed concentration is relatively constant, while in batch, the concen concentratration o the light li ght components components drops and that o t he heavy components components rises as distillation distillat ion progresses. Another basic dierence di erence between distillation operations is in i n the handling o the heat removed by the condenser at the top o the column. The more common approac approach h is to waste wa ste that heat by rejecting it into t he cooling water. In this case, “pay heat” must be used at the bottom o the column in the reboiler. reboiler. Figure 1 (p. 5) illustrates this conguration and identies its it s main components. Because a large part o the t he total operating cost is in providing the heat required at the reboiler in some distillation systems, the t he heat content content o the t he bottom product is used to preheat the eed to t he column. www.controlglobal.com
––
Distillation Control & optimization FiGure 1
V Condenser –Q Column
Accumulator
L Feed pump
D (y)
F (z)
Reflux pump
Preheater V
B
Reboiler +Q
B (x) Illustration o a tray-typ distillation towr, whr (without accumulation), th matrial balanc is F = D + B and D = V – L. Th mol r actions o t h light ky componn t in th bottoms, distillat an d d ar id ntid as x, y and z. For bin ary s paration, S = (y[1-x])/(x[1-y]).
The other option is to recycle the heat removed at the condenser by a heat pump (compressor). In this conguration, as the vapors rom the column (V) are condensed, the heat rom the condenser is used to vaporize a working fuid. These vapors are at the low pressure o the suction side o the compressor (heat pump). When the working fuid vapors are compressed, and these high-pressure (and temperature) vapors in the reboiler contact the bottoms liquid rom the column, they condense, and their heat o condensation serves to vaporize the liquid rom the column bottoms (Figure 13, let, p. 21). While vapor recompression is energy-e cient, it is not used very requently. th C
The main distillation equipment is the column, tower or ractionator . It has two purposes: First, it separates a eed into a vapor portion that ascends the column and a liquid portion that descends; second, it achieves intimate mixing between the two counter-current fowing phases. The purpose o the mixing is to get an eective transer o the more volatile components into the ascending vapor and a corresponding transer o the less volatile components into the descending liquid The separation o phases is accomplished by the dierences in vapor pressures, with the lighter vapor rising t o the top o the column and the heavier liquid fowing to the bottom. The portion o the column above the eed is called the rectiying section and below the eed, the stripping section. The intimate mixing is obtained by either lling the column with lumps o an inert material ( packing) or by the use a number o horizontal plates, or trays, which cause the ascending vapor to bubble through the descending liquid (Figure 2, let, p. 6). www.controlglobal.com
––
Distillation Control & optimization FiGure 2
X(t) Y 40
Y 10
Y 4
Y 2
Y 1
L
L
L Y 1 = 1 Lag Y 2 = 2 Lags Y 4 = 4 Lags Y 1 0 = 10 Lags Y 4 0 = 40 Lags
Sum of the time constants are equal Time
Lt: Th contact btwn liquid and vapor is mad intimat as th vapors ascnd through th liquids hld on ach tray as th liquid dscnds. Th dynamics o a multipl-tray column can b approximatd as a scond-ordr lag, plus dad tim. Right: Th rspons s o th distillat composit ion (y) o 1st-, 2nd-, 4th-, 10th- and 40th-ordr proc sss ar shown wh n a unit stp chang in bottoms composition (x) occurs. A 40-tray column is a 40th-ordr procss.
Generally trays work better in applications requiring high fow, such as those encountered in high pressure distillation columns—depropanizers, debutanizers, xylene purication columns and t he like. Packing works best at lower fow parameters because the low pressure drop o structured packing makes it very attractive or use in vacuum columns or ethylbenzene recycle columns o styrene plants. The infuence o plate eciency in the operation o the distillation tower becomes important in the control o the overhead composition. Because plate eciencies increase with increased vapor velocities, the infuence o the refux-to-eed ratio on overhead composition becomes a nonlinear relationship. Column dynamics are a unction o the number o trays, because the liquid on each tray must overfow its weir and work its way down the column; thereore, a change in composition will not be seen at the bottom o t he tower until some time has passed. These lags are cumulative as the liquid passes each tray on its way down the column. Thus, a 30-tray column could be approximated by 30 rst-order exponential lags in a series o approximately the same time constant. The eect o increasing the number o lags in series is to increase the apparent dead time and increase the response-curve slope. Thus, the liquid trac within the distillation process is oten approximated by a second-order lag, plus dead time (Figure 2, right). www.controlglobal.com
––
Distillation Control & optimization
C Vb d th pg
Controlled variables include product compositions, column temperatures and pressure, and tower and accumulator levels. Manipulated variables include refux, coolant, heating medium and product fows. Load and disturbance variables include eed-fow rate, eed composition, steam-header pressure, eed enthalpy, environmental conditions (e.g., rain, barometric pressure and ambient temperature) and coolant temperature. The general guidelines or pairing manipulated variables with controlled variables are as ollows: • Manipulate the stream that has the greatest infuence on the controlled variable. • Manipulate the smaller i two streams have the same eect on the controlled variable. • Manipulate the stream that is more nearly linear with the controlled variable. • Manipulate the stream that is least sensitive to ambient conditions. • Manipulate the stream least likely to cause interaction. In a binary distillation process, the number o independent variables is eleven, and the number o dening equations is two. Thereore, the number o degrees o reedom is nine. Consequently, the maximum theoretical number o automatic controllers that can be used on a binary distillation process is nine, but usually only ve are controlled. These variables are the compositions o the bottom and top products ( x and y), the levels in the column base and accumulator, and the column pressure. The manipulated variables that can be assigned to control these are the distillate ( D), bottoms (B) and refux (L) fows, the vapor boil-up ( V set by heat input QB), heat removal ( QT ) and the ratios o L/D or V/B. These ve single loops can theoretically be congured in 120 dierent combinations, and selecting the right one is a prerequisite to stability and eciency. Column pressure almost always is controlled by heat removal ( QT ). This loop closes the heat balance around the column, while the levels are controlled to close its material balance. Thereore, the key task is the assignment o the manipulated variables to the composition controllers. No matter how we make that selection, these two loops will interact. A change in one will upset the other because whenever the openings o their control valves change, the material and heat balance o the column will also change. Thereore, the most important decision in designing the distillation controls is to assign the least-interacting manipulated variables to the composition control loops. The tool used in making that selection is the relative gain (RG) calculation. agg Vb, rv G Cc
The RGs o the t wo composition loops are calculated as the ratio o t heir open-loop gains when the other loop is open (in manual,) divided by their open-loop gains when the other loop is closed (in automatic). The open-loop gain can be measured by manually changing t he valve opening by, say, 1% and reading the resulting percentage change in the controlled variable when the new steady st ate is reached. The higher the ratio o controlled-to-manipulated variable response, the higher will be the open-loop gain and, thereore, when the loop is closed (placed in automatic), the lower the controller gain (wider proportional band) has to be to obtain stable control. Naturally, the ideal RG is 1.0. RG = 1 indicates that the loop gain is unaected (the tuning o the loop does not need to be changed) when the other loop is switched rom manual to automatic or back. Consequently, our goal o selecting the combination o loops with the least interaction is to nd pairings with RG values near unity (not much above or below RG = 1). For the various equations to be used in making RG calculations, see Chapter 2.25, Vol. 2, The Instrument Engineer’s Handbook, 4th edition. I RG is zero, that indicates that the manipulated variable has no direct infuence, and the loop can only control through interaction with the other loop. Consequently, the process is controllable only when the www.controlglobal.com
––
Distillation Control & optimization
other loop is closed (in automatic). Similarly, a very high RG value (say, 100), indicates that the process is controllable only i the other loop is open (in manual). Negative RG values will reverse the control ler action and should never be considered. I the RG is bet ween 0 and 1, the eedback rom the interaction i s negative and when RG > 1, the eedback rom the interaction is positive. Loop congurations with RG values that are < 0.5 or > 10 should be rejected, while RG values in the range o 0.5 to 10 are controllable. Naturally, the urther the RG is rom 1.0, the worse the interaction. About t he same amount o interaction is indicated by an RG value o 0.5 and an RG value o 10. Figure 3 illustrates a possible result o calculating t he RG values. Here it was concluded that xing the production rate (heat input to the column QB) and controlling only the distillate composition ( y), while allowing the bottoms composition ( x) to foat, will give the most stable and ecient operation. FiGure 3
Qt
LC
l a
PC
FC
(D, y)
L
FC
TC
(F, z)
V
FC
l b
LC
QB
FC
(B, x)
In this xampl, th v manipulatd variabls ar so assignd to th v controlld variabls that th hat input at th r boilr (Q B) and th distillat composition (y) ar xd and, thr or, th bottoms fow (B) and c omposition (x) ar allowd to chang with th variations in d fow (F) or composition (z). www.controlglobal.com
––
Distillation Control & optimization
Composition Control Conceptually, product quality is determined by the heat balance o the column. The heat removal determines the internal refux fow rate, while the heat addition determines the internal vapor rate. These internal vapor and liquid fow rates determine the circulation rate, which in turn determines the degree o separation bet ween two key components. The rst task in conguring the control system or a distillation column is to congure the primary composition control loops. This conguration must consider the interaction between the proposed control loops, the column’s operating objectives and the most likely disturbance variables. The measurements o the composition control loops can either be direct or inerred. Figure 4 provides some guidance on how to select t he manipulated variables or controlling the compositions (and levels) o distillation columns. FiGure 4
Dyc r d svy l h pg D C Vb (I both compositions ar important, thy should not both b controlld by matrial balanc [B, D]) Manipulatd Variabls Controlld Variabls
Composition o Ovrhad Product (y)
Distillat Flow (D)
Vaporizatio n Rat (V) or Hat Input at Rboilr (Q)
Rfux Flow Rat (L)
Nots 1 and 2
Not 2
Nots 1 and 2
OK i trays ≤20
Not good with urnac. OK i V/B ≥ 3
OK i L/D ≥ 0.5
OK i L/D ≥ 6 Not 3
Composition o Bottoms Product (x) Accumula tor Lvl
Bottoms Product Flow (B)
Not 3 OK i L/D ≤ 6
Bottoms Lvl
OK i V/B ≤ 3
Not good i urnac is usd. OK i diamtr at bottom ≤ 20 t.
Nots: 1. Controls th concntration (x or y) which has th shortr rsidnc tim by throttling vapor fow (v). 2. Mor pur product should control sparation (nrgy). 3. Lss pur product should control matrial balanc. 4. Whn controlling both x and y, th only choics or possibl pairings ar a. Control y by D and x by V, b. Control y by D and x by L, c. Control y by L and x by V, d. Control y by B and x by L. O ths choics, d is not rcommndd bcaus a y/B combination is not rsponsiv dynamically.
idc (t-Bd) C C
I the eed composition and the column pressure are constant, temperature can be used as an indirect measure o composition. I the column pressure is not constant, the temperature measurement must be pressure-compensated. When the bottom product composition is controlled, the temperature sensor is located in the lower hal o the column, and when overhead composition is controlled, in the upper hal. The temperature sensor should be located on a tray t hat strongly refects changes in composition. This means a 1% change in the manipulated variable (refux or steam fows in Figure 5, p. 10) should result in a temperature change o at least 0.1 ºC to 0.5 °C, and that this change should be symmetrical, meaning that a 1% drop or rise in these fows should result in approximately the same size o drop or rise in the temperature on the control tray. When two compounds o relatively close vapor pressures are to be separated (or example normal butane rom isobutene), temperature measurement is not sensitive enough to measure composition. In such cases, two temperatures or a temperature dierence (say, between Trays 5 and 15) can be used instead o a single sensor. This conguration can also be used to eliminate the eects o column pressure variations. www.controlglobal.com
––
Distillation Control & optimization FiGure 5
V Water
Condenser
(-Q) PRC
PT TRC
Set TT
Accumulator
FRC
LT
LRC
Set
FT FRC
L
FT
D(y)
TRC
TT
Set FRC
Reboiler
FT
(+Q)
Steam LT
LIC
B(x)
Top: Distillat composition can b controll d by a c ascad tmpratur mastr on th uppr par t o th column, which ma nipulats th rfux fow L. Bottom: Similarly, th bottoms composition can b controlld by a cascad tmpratur mastr locatd on th lowr hal o th column, throttling th rboilr hat input.
C m
The direct measurement o composition is more expensive, but also more accurate and versatile than is indirect temperature measurement. Intermittent analyzers, such as chromatographs (with cycle times o a ew minutes), are oten provided with dead time compensation or closed-loop control. The analyzer update time must always be less than the response time o the process. For improved accuracy, one usually measures the impurity concentration in the controlled stream. This way the upsets caused by eed composition changes, tower pressure or eciency variations can be more accurately corrected. In the case o ractionator trains, t he bottom or distillate analyzer on one column can also be the eed analyzer on the next one. In such processes, one can control either the concentration o a single component or the ratio o two components (Figure 6, let, p. 11). As will be shown later, such decisions can be based on protability and market considerations. www.controlglobal.com
– 10 –
Distillation Control & optimization
In order or the composition to be close to the terminal composition, the point at which the measurement is made must also be near the column terminals. This will also minimize the transportation lag (dead time). In some applications though, the sampling point is moved closer to the column eed because that provides earlier recognition o eed composition changes or because the key control component is present at a higher concentration. Figure 6, right, describes some o the actors that should be considered in sample point selection. The samples should always be ltered, dried and cooled, and their dead t imes (transportation lags) must not exceed 30 seconds. Transportation lags can be reduced by the installation o high-fow bypasses. ay sc
Most o the 66 types o analyzers that are discussed in Chapter 8, Vol. 1, The Instrument Engineer’s Handbook, 4th edition, can also be used to control the dist illation process. On the LNG train (Figure 6, let), in the depropanizer (where isobutane is to be measured in the presence o ethane, propane and normal butane) and in the deisobutanizer (where isobutane is to be measured in the presence o normal butane and isopentane), an inrared analyzer is recommended. In the debutanizer, where the combined isopentane plus normal pentane concenFiGure 6
AX C3 Ethane product
NGL
r e z i n a h t e e D
AX C2/C3
Ok (fast but not repeatable) AX NC4
AX IC4
r e z i n a p o r p e D
AX C3/IC4
AX C5+
r e z i n a h t u b e D
NG
Isobutane product
Propane product
Ok (slow)
Distillate product NG (D)
r e z i n a h t u b o s i e D
Feed (F)
N-butane product
Ok (fast)
AX IC4
Straight run gasoline
Ok (slow & low pressure)
AX NC4
Ok (slow)
Bottoms product (B)
Lt: Rcommndd analyzr locations on ractionator trains. Right: Sampling point considrations on individual columns.
trations are measured in the presence o isobutane and normal butane to control the butane-pentane separation, gas chromatography is recommended. Some physical properties analyzers, such as boiling-point analyzers, are reliable enough to be used or online control (see Chapter 8.48, Vol. 1, The Instrument Engineer’s Handbook, 4th edition). Cut points between overhead products and side-cuts can be controlled on the basis o temperature, but doing so results in the downgrading o the more valuable product to the stream o lesser value. This downgrading can be minimized through the use o online boiling point analyzers. Justication o a boiling point analyzer depends upon the value o the products and the cost o analyzer maintenance. www.controlglobal.com
– 11 –
Distillation Control & optimization
Viscosity can also be measured continuously to give aster control in vacuum distillation. The use o a viscosity analyzer minimizes downgrading during major upsets and large eed composition changes. With such an arrangement, low-viscosity vacuum bottoms can be detected quickly and diverted to recoverable eed or protable reprocessing. Many other analytical instruments are being moved out o the laboratory and into the processing area. Mobile units containing several dierent kinds o analyzers can be used to learn the best place to locate the on-stream analyzers. In cases in which permanent analyzers cannot be justied, the mobile unit is connected to the process long enough to nd the best operating conditions. Then the mobile unit can be moved elsewhere. Recently articial neural networks (ANN) have also been used as indirect analyzers. In Fig ure 7, the product specications are based on the Reid vapor pressure in the bottoms product and on the 95% boiling point o the distillate. Using ANN sotware eliminates the problems o dead time, cost and availability limitations o direct analyzers. ANN uses a nonlinear neural network model, which iners the analytical readings on the basis o other measurements. I historical data exists, the model can be developed based on collected laboratory or analyzer results, as commonly recorded on log sheets. Figure 7 illustrates the case where the Reid vapor pressure o the bottoms product and the boiling point o the distillate are predicted by an ANN model. This particular model has nine measurements (input nodes), our hidden nodes and two output nodes with bias. FiGure 7
RVP IN
DISTILLATE
BOTTOMS
95% BP
OUTPUT LAYER (2 NODES)
HIDDEN LAYER (4 NODES)
INPUT LAYER
BIAS
(9 NODES)
BOTTOMS
FEED
BOTTOMS
FEED
FLOW
FLOW
TEMP DISTILLATE
STEAM
FLOW
FLOW
PRESSURE
TEMP TOP TEMP
REFLUX TEMP
ANN modl-bas d sot war c an provid indirct analy tical r ading pr dictions, ovrcoming d ad tim, av ailability and cost limitations o onlin analyzrs.
www.controlglobal.com
– 1 –
Distillation Control & optimization
ay sg
The sample system must condition the sample to remove traces o oreign materials t hrough ltering, maintain pressure and temperature, and maintain or change phase or introduction into the analyzer. The description o the wide variety o samplers, probes, lters and other sampling system components is beyond the scope o this work and is covered in detail in Chapter 8.2, Vol. 1, The Instrument Engineer’s Handbook, 4th edition. The system must transport the sample rom the sample point to the analyzer with a minimum o transport lag (preerably less than 30 seconds and denitely not greater than one minute). Transportation times are minimized by using high-fow-rate bypass streams taken rom the process sample point. With a circulation sample pump, care must be taken to prevent cavitation by locating the pump close to the sample take-o. Single-line transport is the most direct approach and is used when the sample line volume is small in relation to the analyzer sample consumption, so that the transport time lag is reasonably short. (See the tabulation at the bottom o Figure 8, p. 14). Ater selecting the appropriate sample transport method, a calculation o the sample time lag should be made, based on the ollowing: • Available dierential pressures, • Total length o the ast loop rom the sample take-o point to the analyzer location and back to the sample return point, • Line sizes, • Viscosity o the sample. Sample disposal is a critical consideration, both rom an economic and an ecological perspective. One goal is to prevent the emission o most hydrocarbons into the air. When there is an economic justication or saving the sample, as when dealing with liquids at the boiling point and viscometer analyzers, a sample collection and return system must be urnished to collect the sample at atmospheric pressures and pump it back at high pressure into the process. For gases with no sample return point, the sample can be pressurized back in to the process, or as is most requently done, vented in to the fare system. For chromatographs, liquid sample points are generally preerred (Figure 8) because o condensation at the sample probe and in the sample lines when hydrocarbons with high boiling points are present in the sample. When the sample lines are long, some separation between components can also occur. A satisactory point or measuring bottoms product composition is at the point o highest pressure immediately ater the product pump. However, i liquid holdup in the reboiler and kettle is large, a long lag is introduced, so an alternative sample point, such as a bottoms tray or seal pan may be used. A satisactory sample-point location or measuring the distillate is the outlet liquid o the overhead vapor condenser. Sampling the overhead accumulator liquid ater the refux or distillate pump should be avoided because o the tremendous process lag it introduces. Sampling the overhead vapor reduces the process lag o sampling ater the condenser i a repeatable, representative sample can be obtained. Figure 9, Part B, (p. 15) shows the same conguration as Figure 9, Part A, except that the analyzer controller is equipped with a rst-order Smith predictor (discussed on page 15) which provides dead time compensation. www.controlglobal.com
– 1 –
Distillation Control & optimization FiGure 8
Analyzer Sample Out
Pressure Temperature gauge gauge
Cooling Water in out
In
PI Control valve
TI
Flow indicator
FI
Self cleaning filter Sample cooler
Pressure regulator
Shut-off valves FI
Calibration sample
Shut-off valves
Coalescer
Lab sample take off
Pressure relief valve
Pressure gauge
Flow indicator FI with needle valve
Flow indicator
Flow indicator with needle valve
FI
PI Check valve
Check valve
Shut-off valve To drain
Top: Rnry chr omatograph liquid sampling systm a nd its componnt s. Bottom: Tubing or pip volums and dimn sions.
D d V tbg d p ud s sy Typ
316 stainlss stl tubing
Schdul 40 pip
Nominal Volum peR DIAmtr, in.
Innr Diamtr, in.
/8
Intrnal Ara in2
cc
.0787
.0048
.9571
¼
0.1850
0.0268
5.3035
/8
0.0253
0.0684
13.4417
½
0.4055
0.1290
25.2984
¼
0.3642
0.1040
20.4521
/8
0.4921
0.1891
37.4904
½
0.6220
0.3038
59.7408
¾
0.8268
0.5863
106.6800
C C ug ay
Analyzer controllers in a eedback conguration can be considered only when the dead time caused by analysis update is less than the response time o the process. Naturally, direct control o the composition o the product by an analyzer gives more accurate results than indirect control by temperature. The composition controller provides a eedback correction in response to eed composition changes, pressure variations or changes in tower eciencies. In Part A o Figure 9, the analyzer controller (ARC) uses the chromatographic measurement to manipulate the refux fow by adjusting the set point to the refux fow controller (FRC). A liquid sample rom the condenser run-down line is obtained by a sample probe, and a sample system is used to condition and vaporize the liquid sample to provide a representative vapor sample to the chromatograph. sh pdc
Oten the analy zer is slow and introduces a signi icant delay time that degrades the controllability o t he process. In that ca se, some type o dead time compensation is use d. (Section 2.30, Vol. 2, The Instrument Engineer’s Handbook, 4th edition.) A Smith-predictor compensator can serve to model the process to www.controlglobal.com
– 1 –
Distillation Control & optimization FiGure 9
SP
SP FT
FT
FRC
∑ ARC ARC
PRC
PT
TT
FT
Accumulator
PT
D
L
AT
X
PRC
SP FRC
AT
+ ∆ + AY AY + -
FRC
TT
AY
1
2 Lag
SP FRC
A
AY
FT
AY 3 Dead time Accumulator D
L
B SP FT
ARC
SP
PRC
r e p p i r t s r e b r o s b A
FRC
AT
TRC
PT
TT
D
SP
FY
<
TRC
SP FRC
FT
Accumulator
FRC
FT
L
D
Reboilier ARC
C
Lean oil to absorber
Part A: Ovrhad composition controls by cascading rfux fow as th slav controllr (FRC). Part B: Th Smith prdictor, which is th sam as Par t A, but with dad tim compnsation addd. Part C: Ovrhad composition control by tripl cascad o ARC to TRC to FRC. Par t D: Absorbr bottoms composition control (ARC) cascadd to rboilr hat input (FRC) with tmpratur ovrrid (TRC).
predict what the analyzer measurement should be between analysis updates. When the actual measurement is completed, the model’s prediction is compared to the actual measurement, and the input to the controller is biased by the dierence. Controllability o the process is degraded by the dead time between measurement updates. (Dead time compensation in detail is discussed in Chapter Section 2.30, Vol. 2, The Instrument Engineers Handbook, 4th edition). As shown in Part B o Figure 9, the Smith predictor compensator provides a process model in terms o its time constant and dead time, and thereby predicts what the analyzer measurement should be between analysis updates. When an actual analysis is completed, the model’s prediction is compared to the actual measurement, and the input to the controller is biased by the di erence. In Figure 9, Part B, the multiplier ( AY1), rst-order lag ( AY2) and dead time ( AY3) provide the required inputs to the calculation o the predicted analysis. This predicted response is subtracted rom the act ual measurement ( AY±) to give a di erential o the actual process rom its own model. This delta is added to the model ( AY∑) without dead time to provide a modied pseudo-measurement to the analyzer controller. Thus, the analyzer measurement, which has a signicant dead time due to sampling and cycle times, is provided with a trim to obtain the predicted measurement o the model. www.controlglobal.com
– 1 –
Distillation Control & optimization
t Ccd
The purpose o all cascade systems is to provide slaves that will correct or dist urbances beore they can upset the primary or master controller. Part C in Figure 9 illustrates a triple-cascade loop, where a temperature controller is the slave o an analyzer controller, while the refux fow is cascaded to temperature. This conguration is used when maintaining a stable temperature on a particular t ray, while the tower operates at a constant and controllable pressure is desired. Since temperature is an indicator o composition at constant pressure, the analyzer controller serves only to correct or variations in eed composition. Cascade loops will work only i the slave is aster than the master, which adjusts its set point. Thereore, in the case illustrated in Part C, the time constants o the fow controller (FRC) must be much smaller than those o the TRC and similarly, the TRC must be aster than the A RC. Another important consideration in all cascade systems is that t he integral mode in the master will cause the output to saturate when that output is blocked rom reaching and modulating the set point o the slave (when the slave is switched to local set point). This is called “reset windup,” and it is prevented by providing the integral mode o the master with an input (an “external reset”) that is never blocked. The external reset o the master is always the measurement o the slave (temperature or the ARC, fow or the TRC, in Part C). These signals are not shown in Part C scv C
Part D o Figure 9 illustrates a limit control conguration where the analyzer controller is overruled when the temperature reaches its high limit. T he temperature controller is a constraint controller preventing the temperature rom exceeding a limit at the bottoms o an absorber stripper. The reason or this limit is energy conservation, because no additional stripping o the light component can be accomplished once the boiling point o the impurity is exceeded. Thereore, even though an analyzer controller may call or more heat, this heat would only increase the bottoms temperature without removing the impurity, thereby wasting heat. Selective control congurations also require external eedback to protect them rom reset windup. In Part D, we have a combination o selective and cascade systems as the master o the FRC is selected to be either the TRC or the ARC. In such a conguration, the external reset (ER) signal (not shown in the gure) is taken rom the measurement o the slave controller (FRC).
p 2 pressure Control n controlling the pressure o a column, the key pieces o equipment are the condenser and the accumulator. First the overhead vapors enter the condenser (partial or total), and next the liquid condensate is collected in an accumulator vessel. Some o the accumulated condensate is returned to the column as refux, while the remainder is withdrawn as overhead product (distillate). I the condensation is incomplete, the condenser is called a “partial” condenser, and the overhead product is withdrawn in both vapor and liquid phases. Total condensers are usually designed or accumulator pressures up to 215 psia (1.48 MPa) at an operating temperature o 120 ˚F (49 ˚C). A partial condenser is usually used between 215 psia and 365 psia (1.48 to 2.52 MPa), and rerigerant coolant, such as propane, is used i the operating pressure is greater than 365 psia (2.52 MPa). The condenser pressure drop is usually about 5 psia (34.4 KPa). Most distillation columns are operated under constant pressure, because at constant pressure, temperature measurement is an indirect indication o composition, but foating the operating pressure can have advantages in many applications. When the column pressure is allowed to foat, t he composition must be measured by analyzers or by pressure-compensated thermometers. The primary advantage o foating-pressure control is t hat one can operate at minimum pressure, and this reduces the required heat input needed at the reboiler. Other advantages o operating at lower temperatures include increased reboiler capacity and reduced reboiler ouling.
I
www.controlglobal.com
– 1 –
Distillation Control & optimization
In the ollowing paragraphs, foating and constant pressure control strategies will be described or operations when (1) liquid distillate is withdrawn in the presence o non-condensables; (2) vapor distillate is withdrawn in the presence o non-condensables; and (3) liquid distillate is withdrawn when the amount o non-condensables is negligible. Cg W C, nggb i
In distillation processes where the distillate is in the liquid phase and the quantity o inerts is negligible, the column pressure is usually controlled by modulating the rate o condensation in the condenser. In the control system shown or Column A in Figure 10 (p. 18), the column pressure is controlled by throttling the cooling water fow through the condenser. This control scheme is recommended only when the cooling water is treated, because condenser tube ouling due to high temperature rise across the tubes can occur. The advantage o this conguration is simplicity and low maintenance cost, because the control valve is on t he water side and gives acceptable control perormance, provided the condenser is the bundle type, and t he cooling water is fowing through the tubes at a rate over 4.5 eet per second (1.35 m/s), or the water has residence time o less than 45 seconds. With such short residence time, the pressure controller requires only a nar row proportional mode. As the residence time increases, the controller will require wider and wider throttling ranges and will also need the addition o an integral mode to compensate or the load changes. Once the proportional band is wide, the control quality will no longer be satisactory or precision distillation applications, because the recovery time rom upsets will be long, and because proper tuning o the integral mode is prevented, due to the dead time varying with load. Thereore, one should not use this control system when the process time lag is large, such as in a case where a condenser box with submerged tube sections is used, because o the large volume o water in the box. I the amount o inerts is nearly zero, and one would like to have a more responsive control system, one can relocate the control valve rom the water to the condensate liquid line on Column A in Figure 10. In such a conguration, when the column pressure is dropping, the pressure recorder controller (PRC) reduces the opening o this valve, which causes the condensate level to build up, and the tube surace exposed to the condensing vapors is reduced due to the fooding o the heat transer tubes. This reduces the rate o condensation and increases the column pressure. I it is expected that over time the inerts will accumulate and blanket the condenser, a vent valve should be added to the condenser. D wh i
The problem o pressure control can be complicated by the presence o large quantities o inert gases, which must be removed, because otherwise they will blanket the condensing surace and cause the loss o pressure control. In renery applications, the non-condensables are oten sent to the uel gas system or to fare. In other cases, a xed fow o inerts and vapors is purged to a lower pressure absorption tower or to a vent condenser, where the condensable vapors are recovered. When the amount o inerts is variable, the purge stream has to be modulated. In the control system shown or Column B in Figure 10, as the inerts build up in the condenser, the pressure controller will rst open up the control valve (PCV-1), which lowers the fooding level in the condenser and, by increasing the heat transer surace area, also increases the rate o condensation. When PCV-1 is nearly ully open (say 95%), the valve position controller (VPC-2) is activated. Its set point is 95%, so when PCV-1 has opened to 95%, it starts purging the inerts by opening control valve (VPCV-2). The control system shown or Column C in Figure 10 is used when the distillate is in the vapor phase and inerts are present. As the overhead product is removed under pressure control, the system pressure will quickly respond to changes in the distillate vapor fow. Here, the level o the overhead receiver regulates the condensate generated by throttling the cooling water supply. This way it will condense only enough material to provide the column with refux. This control system will give acceptable perormance only i the condenser residence time on the coolant side is short. I this is not the case, the cooling water fow should be held constant, and the control system shown or Column www.controlglobal.com
– 1 –
Distillation Control & optimization
o re-vaporizing the condensate, part o the overhead vapor stream is never condensed, but is sent to a vapor bypass valve around the condenser. This allows the condensate fow to stay constant, while varying the rate o condensation by changing the degree o fooding o the tubes. Column F in Figure 10 illustrates the controls when, in the case o liquid distillates with neglig ible inerts, the condenser needs to be mounted below the accumulator or ease o maintenance or to save on the steel supports. In this conguration, when the bypass valve is open, the condenser is fooded. When the column pressure drops, the PRC increases the opening o the bypass valve, which increases the pressure in t he accumulator and pushes some o the condensate back into the condenser to reduce its rate o condensation. I the column pressure rises, the opposite occurs, and the rate o condensation increases. It is usual practice to elevate the bottom o the accumulator 10 t to 15 t (3 m to 4.5 m) above the suction o the pump in order to provide the required positive suction head. Under normal operating conditions, the subcooling that the condensate received in the condenser is su cient to reduce the vapor pressure in the receiver. The dierence in pressure permits the condensate to fow up the 10 t to 15 t o pipe between the condenser and the accumulator. When the condenser is mounted above the accumulator and no inerts are present, two bypass valves can be used, as shown or Column G in Figure 10. The column pressure is maintained by PRC-1, which is throttling the fow o vapor through the condenser, while PRC-2 is maintaining the accumulator pressure by throttling t he bypass. This conguration provides aster pressure regulation or the column. In most applications, some inerts are also present, and in such cases the controls shown or Column H in Figure 10 are applicable. Here, the column pressure controller is throttling the hot vapor bypass, and the accumulator pressure controller is throttling the vent valve o the inerts. The accumulator PRC is set at about 5 psig below the tower pressure.
o: m p
Operating the column at the minimum possible pressure minimizes the energy cost o separation within the constraints o the system. Lowering this pressure increases the relative volatility o distillation components, and thereby increases the capacity o the reboiler by reducing the operating temperature, which also results in reduced ouling. Reducing pressure also aects other parameters, such as tray eiciencies and latent heats o vaporization. Yet composition control must take precedence over pressure optimization, and thereore the pressure optimization loop response must be much slower than that o the composition control loop. The eects o pressure changes also must be considered on the upstream and downstream units. For example, i the pressure o an upstream tower provides the driving orce to move product to a downstream tower, pressure minimization may not be practical. Fractionators using vapor recompression, such as a propylene splitter (Figure 13, p. 21) with a heat pump, may actually benet rom increasing the pressure rather than reducing it. Minimum pressure operation can be achieved by adjusting the set point o the pressure controller so as to keep the condenser ully loaded at all times. In order to prevent upsets caused by rapid set point changes, the valve position control (VPC) scheme shown or Column A in Figure 11 (p. 20) can be used to minimize the set point o the pressure controller, and thereby keep the condenser control valve in nearly the ully open position. The VPC should be an integral-only controller, with an integral time setting approximately 10 times that o the overhead composition controller. In addition, bumpless transer rom VPC to direct PIC control should be guaranteed by providing external reset, and the set point adjustment capability o the VPC should be limited i required by the process. The controls or Column B in Figure 11 serve to minimize the operating pressure o a partial condensing system with a vapor distillate product. Here the level controller (on the rerigerant side) serves to ully load the condenser in the long term, while the pressure controller corrects the short-term upsets. www.controlglobal.com
– 1 –
Distillation Control & optimization FiGur e 11 Refrigerant vapor Condenser valve
Water
Condenser valve
LC Keeps condenser flooded in the long run
Refrigerant V
PI
PT
PIC
SP
I
VPC
Set at 90%
External feedback
A
Remote set PT
B
I VPC=10 ( IPIC )
PIC
D
L
V
PIC
PT
Remote set
F(x)
Steam
Air
Water PRC
Bias Vapor
PIC
PT
(D)
C
Condensate temperature sets PIC to control vapor composition
LT
TT
LIC
D D
D
Optimization and vacuum control stratgis. A: Minimizing (foating) prssur by maximizing coolant valv opning; B: Floating prssur control o part ial condnsr by maximizing condnsr lvl; C: Floating prssur control whn th distillat is both vapor and liquid; D: Vacuum controlld by air bld-in.
An additional control loop is required to control the composition o the vapor product (Column C, Figure 11) i the column pressure is foated and both liquid and vapor distillates are produced. In this control system, the column pressure is controlled by throttling the fow o the vapor distillate, while the set point o this PIC is adjusted i changes occur in t he accumulator (condenser outlet) temperature, which is characterized to represent the desired composition. I changes are to be made in the product composition, the bias is adjusted. Vc sy
To separate some liquid mixtures, t he temperature required to vaporize the eed would need to be so high that decomposition would result. These columns are operated under vacuum and steam jet ejectors can be used, singly or in stages, to generate the required vacuum. Ejectors have no moving parts and require litt le maintenance, but are designed to operate at a xed capacity and steam condition. Increasing the steam pressure above the design level will not increase and sometimes will decrease the capacity o the ejector because o choking. Reducing the steam pressure causes unst able operation. For the above reasons, the steam pressure is controlled at a constant value (PIC) in the conguration or Column D, Figure 11. In order to reduce the time lag, instead o the column pressure, it is the accumulator pressure that is controlled by the PRC, which modulates the amount o air or inert gas that is bled into the system. At low loads this results in a substantial waste o steam. Thereore, operating costs can be lowered by installing t wo ejectors and by automatically switching between the larger one and a smaller one to match the load. rb
As shown in Figure 12 (p. 21), reboilers can be internal or external, and operated by either natural or orced circulation. The kettle reboiler is the most common design or external orced-circulation applications. Thermosyphon reboilers (vertical or horizontal) operate by natural circulation, which is induced by the hydrostatic pressure imbalance between the liquid inside the tower and the two-phase mi xture in the reboiler tubes. www.controlglobal.com
– 0 –
Distillation Control & optimization
A newer development is the use o sel-cleaning, shell-and-tube heat exchangers or applications where heat exchange suraces are prone to ouling. Common heat sources include hot oil, steam, uel gas (red reboilers) and the condensation o compressed vapors in vapor recompression systems. FiGure 12
V
V
Q Q
V
Q Q
L B
B
B Internal
B Horizontal thermosyphon
Vertical thermosyphon
External kettle
Rboilr dsign variations ar distinguishd by th typ o circulation—natural and orcd (pumpd)—and by th point rom which thy tak t hir liquid. Most r boilrs tak thir inlt r om th column bottoms, whil horizontal thrmosyphon rboilrs tak it rom th bottom trays.
V rc
Vapor recompression is another means o improving energy eciency. As shown in Figure 13, the overhead vapor rom the distillation column is compressed to a pressure at which the condensation temperature is greater than the boiling point o the process liquid at the tower bottoms. T his way the heat o condensation o the column overhead is reused as the heat or reboiling the bottoms. This scheme is known as vapor recompression. It is oten used when the boiling points o the top and bottom products are similar. Examples o such processes are the cryogenic demethanization processes, where the column pressure is controlled by throttling the speed o the recompression compressors or the propylene ractionation process. As shown on the right o Figure 13, the propylene column pressure is controlled by throttling the bypass valve around the vapor recompression heat pump. FiGure 13
L
PT PR C
Heat pump
D M F
Compr.
Work
r e w o t
F
CW
e n e l y p o r P
TR C
SP Accumulator
FR C FT
Reboiler B
Recovered heat
Propylene – (C3 )
D
Propane (C3)
Lt: Th vapor rcomprssion systm uss rcovrd hat. Right: Th prssur o such a distillation procss can b controlld by modulating th spd o th comprssor or by throttling th bypass around it.
www.controlglobal.com
– 1 –
Distillation Control & optimization
FCCU main ractionators and crude towers also make use o compressors to “draw” vapors rom their essentially atmospheric towers and maintain the column pressure by manipulating the steam fow to the turbine, which sets the speed o the compressor.
FeeD Controls One o the best means o stabilizing the operation o a distillation column is to hold both the eed fow and temperature constant. Feed composition is seldom subject to adjustment. Having constant eed conditions simplies the operation o the control system. For example, i the eed fow to a column is controlled by a liquid level controller, that controller can be tuned with a low gain, (wide proportional band) so that the level will be allowed to swing over a wide range without causing much change in fow. Nevertheless, i the eed comes rom another column, the eed to the second column will eventually change i the load on the rst column changes. As shown in the control loop or Column A in Figure 14 (p.23), the temporary fow fuctuations rom the previous column and/or the pressure fuctuations caused by the distillate/refux pump can be minimized by cascading the eed-fow controller to a nonlinear level master (LRC) on a surge tank. Such level controllers can be so congured that as long as the level is between 25% and 75%, the set point to the FRC remains constant. Naturally, i the level drops below 25% or rises above 75%, the FRC set point is changed to protect the surge tank rom being drained or fooded. Feedback controls are capable o compensating or upsets only ater the upsets have occurred and been detected. With eedback controls, the operators have to manually adjust the set points o the loops when plant conditions change. Feedback control is usually sucient to keep the distillation column in operation, but it is not sucient to provide optimal perormance. Usually, eed-orward-based product composition control o distill ation can provide 5% to 15% energy savings. The goal o a eed-orward control application can be to maintain the material balance o the column as eed rate varies. Feed-orward controls represented the rst steps towards model-based process control. They were rst applied in well-understood processes, such as heat transer and distillation, where the rm knowledge o material and heat balance equations made it possible to predict and anticipate the consequences o changes in the inputs beore they had time to evolve, and to take corrective action beore the tower is signicantly a ected. This correction is accomplished by considering process dynamics (dead times and time lags), the nonlinearities between separation eciency and column loading, loop interactions and process measurements. I eed-fow variations are unavoidable, the impact o these disturbances can be reduced by eed-orward correction o the material balance (Column B in Figure 14). In this conguration, i the distillate fow is changed in the right proportion (m) and at the right time, the upsets resulting rom eed-fow variations can be minimized. This dynamic compensation (FY ) is not only a ratio adjustment, but must also be tuned to refect the time constant and dead time o the process. However, i the eed composition changes, the value o m will have to be readjusted. mxg h Fd Fw
In cases where both the demand or the product and the availability o eedstock is unlimited, increasing the throughput o the column maximizes protability. In such installations, a valve-position controller can be cascaded to the eed-fow controller in order to increase the eed rate until an equipment constraint is reached. In many cases, it is not known which constraint will be encountered rst, as eed fow is maximized because the limiting constraint may vary over time. In such cases, a multiple constraint network is implemented. In such a conguration, the outputs o the valve-position controllers (VPCs) detecting the loadings o the critical equipment are sent to a low-signal selector, and the eed fow is set by that. I the constraint on maximum production is the cooling capacity o t he condenser, the opening o the columnpressure control valve can be measured (VPC on Column C in Figure 14), and the eed rate can be manipulated by the valve-position controller to keep t he back-pressure control valve always nearly open (V PC set at 90%). I a hot-vapor bypass around the condenser exists, as is the case on Column C, the opening o the bypass valve can also be used to indicate the condenser load. www.controlglobal.com
– –
Distillation Control & optimization
I the const raint on maximum production is the heating capacity o the reboiler, a similar cascade control coniguration can be used to keep the reboiler ully loaded (Column D, Figure 14). This strategy is particularly e ective when the cost o reboiler heat is negligible, such as when a waste steam is used t hat otherwise would be vented. The VPCs used on Columns C and D are usually selected as integral-only controllers and are set at around 90% o valve stem lit, which on an equal-percentage valve corresponds to about 70% o maximum fow. The integral time setting o the VPCs is slow, about 10 times the integral setting o the FRC, which the VPC adjusts in a cascade arrangement. In order to eliminate reset windup, the VPC is provided with external eedback (FB) rom the slave transmitter (FT, the eed-fow transmitter). FiGure 14
Nonlinear Distillate
LRC
LT
Set FRC
A
B
FT
FT
X
SP m
Feed
Set @ 90% INT.
Back-pressure valve Maximum coolant flow rate
VPC
FB
PRC
F
LRC
FRC
FT
PRC
PT
FB
L LRC
Set (F) FRC
FT
FC
FY
C
FRC
FRC
Set (Q)
INT
FRC
D
VP C
Set (F)
Heat (Q) FB
Set (D)
B FRC
D
F B
TRC
To column
PID FRC
TT
Economizer
E
Set
T F
F
F
TT
FT TRC
Steam VP C
SP
FB Reboiler
Feed INT. set @ 10% bypass
Fd control systms. A: Surg t ank with non-linar lvl controllr as cascad mastr; B: Fd-orward adjustmnt o distillat product fow; C: Maximizing throughput by ully loading th condnsr; D: Maximizing throughput against a rboilr constraint; e: Fd pr-hatr controls; F: Maximum rcovry o th hat contnt o bot tom product by conomizr.
www.controlglobal.com
– –
Distillation Control & optimization
Fd t d ehy C
The thermal condition o the eed determines the required heat input at the reboiler, but constant temperature alone does not necessarily mean constant eed quality. For best separation eciency, the eed should be preheated to its bubble point (the temperature at which bubbles rst appear when a liquid mixture is heated—when the lightest component in the eed starts to boil). I the composition o the eed varies, its bubble point will also vary. As the eed becomes lighter, some o it will vaporize, but this variation can be handled by subsequent controls. The controls shown or Column E in Figure 14 show the cascade controls or a steam-heated preheater. In this conguration, the temperature controller usually is a three-mode (PID) controller. The role o the derivative mode (D) during start-up is to provide a large correction initially, which helps to quickly get the unit up to stable operation. An economizer can be used to preheat the eed by sending it through an economizer (Column F, Figure 14), which is heated by the bottoms product. The economizer operation is optimum when the amount o heat recovered rom the bottoms product is maximized. To achieve this goal, a valve position controller (VPC) is used as the cascade master o the eed temperature controller, which controls the bypass around the economizer. The goal o optimization is to keep the bypassed fow at a minimum. Thereore the VPC is usually set at about 10% o valve opening. Control o eed enthalpy instead o temperature can be achieved i the right measurements are available. For example, i the cold eed (at temperature T F ) rst passes through the bottoms-to-eed economizer and then through the steam preheater, one can calculate the steam fow required to maintain t he eed enthalpy constant by the calculation: S = [F( ∆HF ) – CPFT F) – B(CPB ∆T )]/ ∆Hstm
Where: ∆HF = F = S = CPF = T F = B = CPB = ∆T = ∆Hstm =
eed enthalpy as it enters the tower, BTU/lb (kcal/kg) eed fow to preheater, lb/hr (kg/hr) steam fow, lb/hr (kg/hr) eed heat capacity, BTU/lb °F (kcal/kg °C) eed temperature beore preheaters, °F (°C) bottoms fow to economizer, lb/hr (kg/hr) bottoms heat capacity, BTU/lb (kcal/kg) bottoms temperature dierence through the economizer, °F (°C) steam heat o vaporization minus condensate heat, BTU/lb °F (kcal/kg °C)
The primary eect o increasing eed enthalpy is to decrease vapor-liquid circulation below the eed tray relative to that above the eed tray. When preheating the eed is less expensive than reboiler heat, or when the reboiler is the limiting constraint on production rate, the process is optimized by maximizing eed preheat. When condenser capacity is limiting or fooding is encountered above the eed tray, preheat should not be used.
reFluX Controls Stable column operation is guaranteed by keeping the internal refux o the distillation tower constant. Consequently, internal refux controls are designed to compensate or changes in the temperature o the external refux caused by ambient conditions. Column A, Figure 15 (p. 25) is controlled by a typical internal refux control system. The equations or calculating the required external refux rate are shown at the bottom. This control sy stem corrects or either an increase in overhead vapor temperature or a decrease in external refux liquid temperature and, obviously, the required control actions are in the opposite direction. www.controlglobal.com
– –
Distillation Control & optimization
The need or internal refux control can be eliminated in some cases when the fow o disti llate product is controlled under the cascade control o the accumulator level. I the speed o response o this control system is not sucient, it can be speeded up by reducing o the accumulator volume. To overcome the accumulator lag, the refux rate, L, is manipulated in direct proportion to a change in distil late rate, D, rather than by waiting or the response o a level controller. In the control system or Column B, Figure 15, when K = 0, the refux is adjusted by the level controller only. In other cases, the refux fow is immediately altered by some percentage or a change in distillate, and the level controller orces the balance o the change. The response is a rs t-order lag. I K = 0.5, the refux fow is changed to the exact new steady-state value, because K equals the ratio o Dmax/Lmax, and thereore the computation is exact. The lead equals the lag and the net eect is the elimination o dynamic contribution. I K = 1.0, the initial response is a rst-order lead-lag unction. In this case, the refux is greater than required or the new steady state, and the level controller eventually corrects the fow. The value o K aects the transient response only, but does not change the steady-state fow; thereore, it can be used to adjust the dynamics o the loop. The greater the value o K the aster is the response. In order to maintain stability, K should not be set greater than 0.75. FiGure 15
T
V
O
TT X FRC
UY
∆
B
TY
A FT
FY
TT
L ∑ FIC
SET
L=m–KD
–K D
+m LT
FY
T
FY
LIC D
F
FY
X FY
K
FIC
FT
150-250ºF=0-100% T TT O
FT
D
125-225ºF=0-100% C X L =L(1+ p ∆T) ∆ T –T T i F o F ∆H TT TY UY [Li’=0.754L’(0.885 +0.115∆T’)] 0-50ºF=0-100% 0-10,000 GPM FT
L2
FY
L
Lt: Th controls rquird to kp th intrnal rfux fow o th column constant; Right: Th controls ndd to liminat th ct o accumulator lag in controlling th xtrnal rfux to a column.
p 3 ControllinG tHe WHole Column reviously, I have discussed the individual loops used to control the dierent segments o the distillation process. Now I will talk about various control strategies that can be used to control and optimize the complete column. Here, I will treat distillation as a single unit operation. In unit operations’ control, the individual column variables are treated only as constraints and so long as the values o these constraints are within acceptable limits, the column is controlled (optimized) to maximize production rate, protability, etc. In order to optimize the unit operation o distillation, one not only wants to keep the top and bottom product compositions within specication, but might also want to maximize the throughput, minimize energy use, enhance column stability or operate against equipment constraints. In addition, one might shit less protable components into more protable products. To optimize the protability o the operation, one must know the acceptable variations in product specications, the relative economic values o the product streams and cost o energy used in the separation.
P
www.controlglobal.com
– –
Distillation Control & optimization FiGure 16
V, y
F L 2
Dynamics
2
F
ARC
AT
LC
FY
m=
D F 2
2
LT
FRC
L
D 2
SP 2 L
F
FFY
X
FT
FT
Q 2 F
Distillate (D)
Q 2
Feed FIC
(F,z)
FT FY LC
X
Q
F L 2 F L - Lagged feed flow rate B, x
Lead/lag
Lag - lead/lag
Step change in load i nput
Impulse
Multiple lag
Lag
Load input
Time
Top: Fd-orward co ntrol sys tm that provids c onstant spara tion by manipula ting th dis tillat f ow. Bottom: A varit y o dynamic compnsators that can b usd to matc h th “dynamic prsonality” o th procss. www.controlglobal.com
– –
Distillation Control & optimization
Economics o individual ractionators may continually change throughout the lie o the plant, because energy savings can be important at one particular time, while product recovery can be more important at other times. Other goals include the desire to minimize the disturbances propagated to downstream units, minimize rework or recycle o o-spec products and maximize the consistency o product quality. Thus, a given column’s operating economics, and, thereore, its optimization objectives may change with time.
Constant separation A distillation column operating under constant separation conditions has one ewer degree o reedom because its energy-to-eed ratio is constant. This means that or each concentration o the key component in the distillate, a corresponding concentration exists in the bottoms. Thereore, i the eed composition is constant, and i the concentration o a component in one product stream is held constant, that xes the concentration o that component in the other. Figure 16 (p. 26) shows a eed-orward control system or constant separation in which distillate is the manipulated variable. The control system is so adjusted that the output (m) o the trim analyzer controller ARC is 50% when the design or normal distillate-to-eed ratio is required. I the gain o the multiplier ( FY ) is set at 2, the output tracks the load when this normal distillate-to-eed ratio occurs. The block labeled “dynamics” in Figure 16 is a special module that serves dynamic correction by modiying the transient response. This matches the time response o the distillate to a eed-rate change. The dynamic block most oten is a combination o a dead-time and a lead-lag module in series, which in the steady state makes no correction at all; its output equals its input. Because the “dynamic personalities” o the various dist illation processes are dierent, a variety o dynamic compensators are available to match them, as shown at the bottom o Figure 16. mx rcvy
I one product, such as the distillate ( D), is worth much more than the other, the control system can be designed to maximize the more valuable stream. I we assume that energy is ree, the material balance or distil late in such a column can be expressed as D = m(KF + K 2F2 ) where: D = distillate rate F = eed rate K = adjustable coecient K2 = 1 – K m = eedback trim Figure 17 (p. 28) shows the controls or this maximum recovery system. In this conguration, the boil-up (heat input rate) is constant, and consequently the distillate product fow is not linear with the eed rate. To increase the response o the system (minimize accumulator lag), the refux fow set point is adjusted by the distillate fow measurement. A summing block (FY-1) is provided to compute ( KF + K 2F2). The values o m can be calculated on the basis o eed composition. A typical range or m is 0.35 to 0.65. This is the output signal range o the eedback controller (ARC-2). While the coecients can be calculated with reasonable accuracy, online adjustment is also easy. (These coecients are accessible in most DCS and PLC systems.) I energy is not assumed to be ree, and the composition o only one product needs to be controlled, then the distillate fow can be determined by the ollowing linear relationship: D = m1(K3F). C C tw pdc
I the composition specications or both products are tight, the controls described in Figure 17 (or the constant separation strategy) are not good enough. This is usually the case whenever the eed composition is unpredictable. In such cases, one must directly control the compositions o both products. The main benet o dual composition control is minimized energy consumption. The main limitation is caused by the severe interactions between the two composition loops. Naturally, the eed composition and the tower design determine the compositions that can be achieved. The let side o Figure 18 (p. 29) gives an example o a eed-orward dual composition control system. In this conwww.controlglobal.com
– –
Distillation Control & optimization FiGure 17
KF + K 2 F 2
D = m (KF + K 2 F 2 )
D
L IC X FY
m
∆ F Y
ARC 2 LT ∑ FY 1
AT
√ F IC
F 2
F
FY
F Y
V
√
F Y
FT
√
SP
FR C
L
FT
FT D F
Q
LC
B
Control systm that guarants th ma ximum rcovry o th mor valuabl product, in this cas, th distillat (D).
guration, the distillate fow is manipulated to control distillate composition by maintaining the relationship D = F [(z-x) / (y-x)]. However, in order to also enorce composition control o the bottom product, an additional manipulated variable is needed. This cannot be another product stream, because that would confict with the material balance and change the accumulation in the column. Consequently, the bottoms composition ( x) has to be controlled by manipulating the energy balance o t he column. The relationship between separation ( S) and the ratio o boil-up to eed ( V/F ) over a reasonable operating range is V/F = a + bS, where a and b are unctions o the relative volatility, the number o trays, the eed composition and the minimum V/F. The control system thereore computes V based on the equation V = F(a + b[V/F]), where [V/F ] equals the desired ratio o boil-up to eed. www.controlglobal.com
– –
Distillation Control & optimization FiGure 18
L = K 1Q – K 2F
z–x D = F(m) = F ( ) y–x FY
F
FY
X AR C
AT
F (t) FY
F FY
F
+ K 1Q Q
– X
L
FY
FR C
AR C
FY
LC AT
FT
LC
V,y FY
2
V = F(m 2 ) = FT V = F(a+b ) F,z F SP V
F
√
FY
FI C
D
X FY
SP
F
FI C
(V/F) min + (V/F)
LC
B,x
FT
K
K 1 D,y
FI C
L
Dynamics FY
√ FT
FT
√
√ FY
Dynamics K 2
√
X
FY
Q = kF ([V/F] min + [V/F])
Dynamics
F m 2
∑
LT
AR C
Q
√ FY
FT
Q
AT
Q AR C
LI C
AT
B,x
B Lt: Conguration or controlling th composition o both products o a distillation column without much intraction; Right: With intraction.
The column on the let side o Figure 18 illustrates how the two product composition controllers (ARC) are congured to throttle the material balance (D) and energy balance ( V ) to maintain the compositions o both products. The loops include multipliers (FY-1 and FY-2) and the FY block labeled “dynamics.” This last block serves the unction o dynamic compensation, the role o which has already been explained in connection with Figure 16. tw pdc wh ic
Interaction is unavoidable between the material and energy balances in a distillation column. The severity o this interaction is a unction o eed composition, product specication and the pairing o t he selected manipulated and controlled variables. Severe interaction is likely to occur when the composition controllers o both products are congured to manipulate the energy balance o the column. An example o such a case is when refux fow and steam fow are manipulated by the two product composition controllers (Figure 18). In such cases, the heat input is Q = kF([V/F]min + [V/F]) while the refux rate is L = K1Q – K2F. Thereore, both product fow rates are dependent on energy balance terms. This conguration, without decoupling, results in severe interaction, because i the composition controller on the bottom product is increasing heat input to the reboiler, this action will orce the overhead composition controller to increase refux fow in order to increase heat withdrawal. Thereore the two composition controllers “ght” each other. In the decoupling equations just dened, the values o K1, K2 and k are determined by using the actual process values o [L/F], [V/F]min and [V/F]. The decoupling scheme shown on the right side o Figure 18 serves to minimize the tendency o a change at one end o the column upsetting the controller at the other end by implementing the decoupling equation L = K1Q – K2F. The system is now hal-decoupled: A change in heat input at the bottom will not upset the top composition, because the decoupling loop adjusts the refux independently. However, the heat input is still coupled to refux, because a change in refux will still cause the bottom temperature controller to adjust steam fow. This type o hal-decoupling is enough to reduce the interaction approximately twenty-old. The limitations o decoupling include the act that overrides can drive the loops to saturation when conwww.controlglobal.com
– –
Distillation Control & optimization
straints are encountered. Also, in some distillation columns, small measurement errors can transorm a system that provides complete decoupling into one that provides no control at all. Since the proper decoupler gains (K1 , K 2 and k) depend on the process gains, and because process gains inevitably change with variations in eed rate, product specications and column characteristics, these systems require constant attention and adjustment. Such adjustments are beyond the capability o the average plant operating personnel and require sophisticated column models. The diculties associated with the application o decoupling systems have prompted a re-examination o interaction itsel. The problem may be postulated in two ways: 1) For a given column, is the interaction equally strong in each o the possible control congurations? 2) For a given control conguration, will the interaction be the same or every column to which it is applied? Some o the answers to these questions appear in the earlier discussion o relative gain calculations. One general rule worth remembering is that the composition controller or the component with the shorter residence time should adjust vapor fow, and the composition controller or the component with the longer residence time should adjust the liquid/vapor ratio. Fd C C
I the changes in eed composition occur too ast to be handled by eedback control, eed-orward compensation is required. I z is the concentration o the key component in the eed, the material balance, solved or disti llate FiGure 19
D = zF/m
x & ÷ z
FY
m
F ARC FY
Dynamics HIC
AT
1
LT
FY
SP L
AT
LIC
FY
FRC
FT FT
F, z SP FY
RIC
Ratio controller
D
LT
FT
LIC
Q B
Fd composition basd d-orward control o distillation. www.controlglobal.com
– 0 –
Distillation Control & optimization
fow is D = F [(z-x) / (y-x)], and thereore, when z is measured, the equation or distillate can be simplied to D = zF/m, where m is the output o the overhead analyzer eedback trim controller (ARC-1 in Figure 19, p. 30). The auto manual station (HIC) is used in the event o analyzer ailure. Dynamic compensation is included ( FY ) in the measurement o the eed fow ( F). The control o the bottoms fow (LIC) is indirectly obtained by the eedorward control o the reboiler heat input, which is also based on the dynamically compensated eed fow rate.
multiple proDuCt Distillation Most separations involve multiple components and produce two or more liquid or vapor products. Sometimes only one product is withdrawn at a time. When there is a side-stream product in addition to the overhead and bottom products, an additional degree o reedom is available or the control system, because the overall material balance becomes F = D + C + B, where C is the side-stream fow rate. Thereore, two product streams can be manipulated to control variables, while the material balance can stil l be closed by the third product fow. The presence o this added degree o reedom makes the careul analysis o the process even more essential to avoid mismatching o the manipulated and controlled variables. I the eed rate and column pressure are constant, ve degrees o reedom exist: three composition specications and two levels. These ve controlled variables can manipulate the material and heat balance by throttling three product fows and the loading o two heat exchangers (V and L). As was detailed in Part 1, interactions among the loops can be minimized by determining the relative gains o the possible combinations o the manipulated and control variables. Ater such evaluation, one might arrive at the controls shown on the let side o Figure 20 (p. 32). Here the ratio o heat input to eed and, thereore, boil-up to eed, is held constant, and separate dynamic elements are used or the distillate loop, the heat input and the side-stream loops. Multi-product ractionators—crude towers, vacuum towers and FCCU main ractionators—are common in the rening industry. Product quality is oten based on true boiling point (TBP) cut points, which approximate the composition o a hydrocarbon mixture and are numerically similar to the American Society or Testing and Materials’ 95 percentage points. As was shown in Figure 7 (p. 12), an articial neural network can calculate—on the basis o local temperature, pressure, steam fow and refux data—the 95% boiling point or TBP cut point o the products. I there is a boiling point analyzer, the ANN-based approximation can be used as the measurement or a ast inner slave loop o a cascade conguration in which the analyzer provides the measurement or the slower master, which trims the set point o the slave. Because o the volume o liquid/vapor loads within most multi-product ractionators, the greatest control sensitivity and the quickest response are usually obtained by thrott ling the product fows. Heat balance is oten controlled by throttling the pump around refux fows, as shown on the right side o Figure 20. The goal is to maximize the amount o heat that is transerred to the eed. Super-ractionators are used to separate st reams with close relative volatilities o the light and heavy key components. These include deisobutanizers, which separate isobutane rom normal butane, propylene splitters, which separate propane rom propylene, ethylbenzene towers, which separate ethylbenzene rom xylene, xylene splitters, which separate para and ortho-xylene rom meta-xylene. Super-ractionators require many trays; thereore, the column heights oten necessitate dividing them into two or even three sections. Super-ractionators require high internal vapor and liquid rates and high fow ratios o both refux-to-distillate and vapor-to-bottoms. These result in large tower pressure drops, long time constants and long dead times. Thereore, the response o compositions to the manipulation o eed and refux fows is slow. As a consequence, dynamically compensated material balance control is recommended to control the distillate compositions. mvb C
Traditional PID controls usually implement SISO (single input, single output) algorithms, while advanced controls work with multiple inputs and outputs (MIMO). The modeling o a process can be o the “white box” or “black box” type. White-box modeling is used or well-understood processes, such as distillation, where the knowledge o mass, energy and momentum balances allow the development o acwww.controlglobal.com
– 1 –
Distillation Control & optimization FiGure 20 SP FT
FRC
D PRC SP
LI C
LT
PT
X
SP
F Y
FI C
SP
Dynamics
V
F Y
Dynamics F,z1 ,z 2
FT
F Y
FT
FI C
FT
L
FT
AR C
AT SP RI C Ratio controller
D,y 1 ,y 2
e c n a c l i a g b o t l a e H
Accumulator
FRC
FT
SP
D
L Q
FT
FRC
SP FT
C SP
C,c1 ,c 2
LT
FT FI C
FRC
SP FT
Q
t n n i o i o t p a l g u n c i l l i a o c B
FRC
FT
FRC
SP
C
LI C
X F Y
F AR C
AT
B B,x 1 ,x 2
Lt: Multi-product ractionator controls whr, atr dynamic corrction, th boil-up (V), sid-draw (C) and distillat (D) fows ar ratiod to th d fow (F ); Right: Th tru boiling points ar controlld by throttling th product fows, whil hat balanc is controlld by manipulating th rfux fows.
curate dynamic models. These internal model control (IMC) systems are useul in optimizing the process and in anticipating uture events. “Black box” or model-ree controls (MFC) include the ANN, uzzy logic and statistical process control strategies. These algorithms are trained on the data obtained rom the past operation o the controlled process. Their limitations include their required relatively long learning periods and the act that their knowledge is based on the past. Thereore, they are not well-suited to anticipating uture events, and i conditions change, they require retraining. md-Bd C
Once a process model has been established, it is possible to build the inverse o that model, which can be used as a controller. In that sense, the PID controller is a li near inverse model o a single process loop. A simple IMC is shown at the bottom o Figure 21 (p. 33). It has the same structure as a Smith predictor (Figure 9, Part B, p. 15), which is a rst-order system with dead time combined with a PI controller. Multivariable control (MVC) is particularly well-suited or controlling highly interactive ractionators, where several control loops need to be simultaneously decoupled. Figure 21 illustrates an MVC conguration, which simultaneously considers all the process lags and applies saety constraints and economic optimization actors in determining the required manipulations to the process. In Figure 21, two products and an impurity stream are produced by two towers. The objective is to control the composition o both products; thereore, when an adjustment is made to control one composition, that manipulation causes a disturbance in the operation o t he other composition loop. In this example, the MVC controls the two product compositions, while keeping the operation within the constraints o the process equipment capacities and considering the dead time introduced by the stripper. In this process, the eed-fow rate is the main dist urbance variable. The steam to the rst column and the temperature at the top o that column are the manipulated variables. A constraint vari able is the internal refux fow, which is calculated rom tower temperatures and fows. www.controlglobal.com
– –
Distillation Control & optimization
The technique o multivariable control requires the development o dynamic models based upon ractionator testing and data collection. Multivariable control applies the dynamic models and historical inormation to predict uture ractionator characteristics. Predicted ractionator responses result in planned controller actions on the manipulated variables to minimize error or the dependent controlled variable. For towers that are subject to many constraints, towers that have severe interactions, and towers with complex congurations, multivariable control can be a valuable tool. FiGure 21
PI C
PI C LI C
SP
FI C
D
L
TI C
FT
F
Calculated internal flow
SP
1 r e w o T
L i
F
r e p p i r t S
F (%HK D )
Multivariable controller
(%LK B ) SP
FI C
FI C
LI C
Q
LI C
Q AT
AT
B
B
Influence of disturbances
Set point
IMC controller
Process
Output
Model
Multivariabl intrnal modl controls (IMC) or controlling two product compositions (B), whil kping th opration within th constraints o th proc ss quipmnt and whil taking into account intractions and th dad tim introducd by th s trippr.
www.controlglobal.com
– –
Distillation Control & optimization
Dynamic Matrix Control (DMC) is also an MVC technique, but it uses a set o linear dierential equations to describe the process. The DMC method obtains its data rom process-step responses and calculates the required manipulations using an inverse model. Coecients or the linear equations describing the process dynamics are determined by process testing. During these tests, manipulated and load variables are perturbed, and the dynamic responses o all controlled variables are observed. This identication procedure is time-consuming and requires substantial local expertise. afc n nwk
Figure 22 (p. 35) shows a three-layer, back-propagation ANN, which predicts the manipulated steam and refux fows o a column. The process model is stored in the AN N by the way its processing elements (nodes) are connected and by the importance that is assigned to each node (weight). The ANN is “trained” by example, and thereore, it contains the adaptive mechanism or learning rom examples (somewhat similarly to the learning process o a child). During the “training” o these networks, the weights are adjusted until the output o the ANN matches that o the real process. Naturally, when process conditions change, the network requires retraining. The hidden layers help the network to generalize and even to memorize. In the SISO conguration, the ANN network builds an internal nonlinear model relating the controlled and manipulated variables. It builds this model by learning or “training” rom a data set o known measurements and process responses. This makes the neural controller more useul and more robust than the standard PID. Because the neural network paradigm can accommodate multiple inputs and outputs, an entire ractionator model can be built into a single controller. The neural controller can be thought o in t he same terms as model-based control algorithms, whereby the neural network is used to obtain the inverse o the process model. As shown on the top right o Figure 22, the back-propagation network can be trained to behave as an inverse model o the process, with load and controlled variables being input vectors and manipulated variables output vectors. To build such a model, all inputs and outputs must rst be normalized based upon expected minimum and maximum values, and be presented to the network or the t raining By using the same historical data, the network can be trained, and a nonlinear internal model can be created. The network’s ability to do the prediction o the dynamics o the ractionator improves as more data become available or training. Thus, the neural controller can be considered as a specic type o nonlinear, multivariable, model-based control algorithm. Instead o creating the nonlinear process model with explicit equations, the neural controller builds its own process model rom actual tower operation. Since the neural controller is an empirical, not a theoretical model, it is susceptible to errors i operated outside the conditions o the training set. Data or the training set need to be continually gathered and the network retrained whenever novel conditions occur in order to increase the robustness o the neural controller throughout its operating lie. th t md
It is possible to design a control system which will compensate or changes in any o the load variables: eed rate, composition, enthalpy, refux and bottoms enthalpy. The goal o these systems is to eliminate interactions and to protect the column rom the consequences o changes in ambient conditions. To provide a model t hat describes both the material and energy balance o the column, one has to develop both the steady-st ate and the dynamic equations or the ollowing: • Feed enthalpy balance, • Bottoms enthalpy balance, • Internal refux computation, • Reboiler heat balance, • Overall material balance. www.controlglobal.com
– –
Distillation Control & optimization FiGure 22
PT
PIC
AT LI C
Set point (L) TT
FI C TT
w
D
–
FT L
+
Q
X
LI C
FI C
Q
–
ANN model
e m
%HK D
Neural controller
Set point (Q)
y
Process
ef
F T O P T l y
r e w o T
F
u
ANN controller
+
%LK B
filter
T b
TT AT
B
Manipulated variables Stream flow (Q)
Reflux flow (L) Output layer Output node #2
Output node #1 W 1,4 W 1,3
W 1,0
W 1,2 W 1,1
Hidden layer Hidden node #1
Hidden node #2
Hidden node #3
Hidden node #4
Input layer
Bias Input node #2
Feed flow
Top temp Bottoms temp T Reflux temp o T Feed temp b T T
F
f
Input node #3
Input node #4
Input node #5
Input node #1
Input node #6
Bottoms composition
l
Disturbance variables
X
Input node #7
Distillate composition
Input node #8
Tower pressure
Y
Controlled variables
Th congur ation o a back-pr opagation nur al ntwor k (AN N) and its us a s an intrnal mod l controllr. www.controlglobal.com
– –
P
Distillation Control & optimization
sb-o u o
Every distillation column is unique and thereore, the control strategies described in Figure 23 (p. 37) are or illustrative purposes only and should not be considered to be the sub-optimal or optimal control solutions or every column. The goal o optimization o a single column is to saely operate at maximum prot, but this can only be done i the market value o each product is known. This is not t he case when the products o a column are not nal products, but the eed fows to other unit processes. When the product prices are unknown, it is still possible to perorm optimization, but the optimization goal changes. The criteria in t hat case becomes the generation o the required products at minimum operating cost. This can be called an optimum with respect to the column involved, but only a “sub-optimum” with respect to the system o which the column is a part. When the market values o the products are known, the column can be ully optimized, but additional variables, including the type o the market that exists or the products, must still be considered. I the market is limited, the goal is to generate the products at optimum separation and minimum operating cost. This cost varies as the eed fows and their compositions vary. When the market is unlimited and sucient eedstock is available, the optimization task is more dicult, because one must determine both the optimum separation and the value o the eed streams. In this case, the goal o optimization is either maximum loading or maximum energy eciency. In this case, one o three constraints can be the limiting one: 1. Throughput can be limited by t he maximum cooling capacity o the overhead vapor condenser, 2. The maximum heat input capacity o the reboiler, 3. The maximum separation rate o the column itsel. In some cases, this constraint will change rom time to time, depending upon product prices and other independent variables. Thereore, the design o an optimal control system or a single column should ollow three logical steps: 1. Designing the basic controls to regulate the operating variables, such as pressures, temperatures, levels and fows; 2. Conguring t he controls to regulate the reboiler heat input, the internal refux fow rate, eed enthalpy and the sources o heat to the reboiler and preheater(s); 3. Determining the controls required to maintain the specied separation. I the above control loops are provided or a single column, that column can be said to have been “sub-optimized.” This sub-optimum will generate products at close to the specied separation, whether or not that separation is ideal with regard to the total system. I product purities are higher than specied, the operation is not considered sub-optimum. og eq
When a single column is automated through the sub-optimum operation stage, it will still exhibit up to ve degrees o reedom. The let side o Figure 23 describes the controls o a column that has been “sub-optimized” by controlling bottom product fow to obtain the specied separation. Figure 24 (p. 38) shows some typical operating control equations used in operating the predictive controls or maintaining the energy balance (refux fow rate) and the material balance (bottom product fow rate) to give the specied separation (right o Figure 23). Both control systems shown in Figure 23 will achieve their goal o operating at sub-optimum; the dierence is mainly in their basic controls. The theoretical operating equations are normally developed by tray-to-tray runs o calculations, which are usually perormed by an o-line digital computer. Next, a statistically designed set o runs is made, and the inormation thus obtained is curve-tted to an assumed equation orm. Once the steady-state theoretical equation is developed and placed in service, the ex perimental part is determined by online tests. These tests involve operating the column at dierent loads to determine the correction required to (Li/F) in order to obtain the specied separation. Average overall eciency E is set to make ( Li/F) equal to the actual Li/F which exists. The loading tests are carried out under this condition. www.controlglobal.com
– –
Distillation Control & optimization FiGure 23
Computer input ( T o - T ) r
Computer input ( T o - T ) r
TDT
TDT
Specify set ( PC )
Hot vapor by pass
PRC
T o FRC
Specify set ( PC )
PRC
T o
PRC
Set
FRC
T r
PRC
T r
LRC
Set
Hot vapor by pass
LRC
Steam
Set (D)
Top product
Reflux
FRC
Steam
D
Feed (F) Reflux
Feed (F)
FT
AT
Set (L)
D
AT
FRC
Computer inputs
LRC
Set
LLK F LK F HK F F
Steam
Set (B)
HK B FT ∆HF
Steam
FRC
Q
Internal reboiler
LK B HK B FT ∆HF P E
set point outputs (L1 T f B)
Bottom product operating equation and dynamics
LLK F LK F HK F F m
Measurements Computer
Measurements
E
Specified inputs
Computer
Computer
Reflux operating equation and dynamics
P
Set (D)
B
Internal reboiler
Bottom product
LK B
Set (L) Set (Q)
LRC
Bottom product
FRC
FRC
FRC
Set
FRC
Computer inputs
Specified inputs
Set (Tf)
LLK F LK F HK F F
Set ( T f )
FRC
TRC
Top product
TRC
B
FT
T O - T r
Feed enthalpy regulation
Top product operating equation and dynamics Reflux operating equation and dynamics Feed enthalpy regulation
m= Measurements required to compute feed temperature for specified value of feed enthalpy
m LLK F LK F HK F F T O - T r
Computer set point outputs
T f
L
D
L = external reflux flow rate K 1 = ratio of specific heat to heat of vaporization of the external reflux
where
K = ratio of heat vaporization of external reflux to heat of vaporization of internal reflux 2
%LLK F = lighter than light key in the feed (mol%)
T O = overhead vapor temperature
%LK F = light key in the feed (mol%) = z %LLK D = lighter than light key in the distillate product (mol%)
T = external reflux temperature r E = specified constant average efficiency
%LK D = light key in the distillate product (mol%) = y
FT = specified value of feed tray location
%HK D = heavy key in the distillate product (mol%)
∆HF = specified value of feed enthalpy
%LK B = light key in the bottoms product (mol%) = x
P = specified value of column pressure
Suboptimization distillation controls aimd at producing th spciid sparation by controlling bottom product low (lt) or r lux low (right).
In other cases, plant tests can be perormed to determine Li/F without consideration o the theoretical term. As to the internal refux fow rate, one can approximate it by making a heat balance around the top tray. I the operating equations that have been obtained or the steady state are used without alteration, the column will be upset when changes, such as sudden eed fow rate increase or decrease occur. Feed composition changes cause less severe upsets and, thereore, providing dynamic compensation is less o a concern. The purpose o dynamic compensation is to compensate or changes (such as eed-fow rate changes) in such a way that the column’s terminal fows respond to them at the proper time, in the correct direction and without overshoot. The simplest dynamic element, which can be used is the sum o a dead time plus a second-order exponential lag response. As shown in Figures 18 (p. 29) and 23, the eed fow rate signal is passed through this dynamic element beore being used to obtain the set points or the bottom product fow rate (B) and the refux fow rate ( L) set points. The bottoms and overhead dynamic constants, t, T 1 and T 2 in the operating equations listed in Figure 24, are not necessarily the same values. With the use o the operating equations in Figure 24 (p. 38), the system still has ve degrees o reedom. Thereore, eed tray location (FT ), eed enthalpy ( ∆HF), column pressure (P), concentration o heavy key component in the top product (%HKD), and concentration o light key component in the bottom product (%LKB ) can all be controlled according to specications. Although tray eciency is also included in the operating equations, it is a xed value. www.controlglobal.com
– –
Distillation Control & optimization
FiGure 24
Column loading limited by condenser Set (P)
Maximum coolant flow
PRC
D
Set (T F )
(PD), (%LLK D ) ( %HK D )
L
TRC
(%LK D ) ≥ ( %LK D ) SS FRC
F Set
( L)
(FT) O
FRC
Set ( F)
Set
(B )
FRC
B (PB), (%HHK B ) ( %LK B ) F
P
T f
F T o
L
(%HK B ) ≥ ( %HK B ) SS
B
Setpoints (B)
(FT)O
( L)
( T 1 )
( P) O
Suboptimization operating equations: ( B) = ( F)
Ke – ts
(%HK D )
( %LK B )
For liquid phase feed 100 –
( T 1 s + 1) ( T s + 1) 2
(%HK D ) 100 –
–
(%LLK F )
(%HK D )
–
–
(%LK F )
(%LK B )
where ∆H F = Specified value for feed enthalpy ( CPF ) = Specific heat of feed
– ts
( L) = ( F)
( L 1 /F ) t – (L 1 /F ) e
Ke
( T 1 s + 1) ( T s + 1) 2
K 2 [1
+
( T f ) = Temperature of feed
K 1 ( T O – T t ) ]
( T ) = Base temperature of transmitter b
( ∆H F ) + ( CPF ) ( T ) b
( T 1 ) =
( CPF )
( ∆H F ) = ( ∆H F ) o
( P)O
(FT)O
for ( L 1 /F ) t
for (L 1 / F) t
Optimization equations (empirically developed) (FT) O = f 3 [ (%LLK F ) , (%LK F ) , (%HK F ) , ( F) , (%HK D ) , (%LK B ) ] ( ∆H F ) o = f 4 [ (%LLK F ), (%LK F ), (%HK F ) , ( F) , (%HK D ) ,(%LK B ) ] ( P)O = f 5 [ (%LLK F ) , ( %LK F ), (%HK F ) , ( F) , (%HK D ), (%LK B ) ]
( F)
(%HK D ) O
(%LK B ) O
( P) O
1. Determine optimum separation (%HK D ) O (%LK B ) O 2. Determine optimum column pressure, ( P) O 3. Determine feed flow rate set point, ( F) that will load overhead vapor condenser.
Values determined to give maximum profit
Top: Th cont rol congurati on; Bottom: Oprating quations to minimiz oprating cost s whn th markt is unlimitd and product prics ( PD & PB ) ar known. www.controlglobal.com
– –
Distillation Control & optimization
total optimization Optimization implies maximum prot rate. An objective unction is selected, and manipulated variables are chosen that will maximize or minimize that unction. This is similar to the PID equation that is designed to minimize the error between the set point and measurement, but it views the process at a higher level. Optimization can be applied in several layers. Local optimization is the optimization o a single column. Normally, the goal o optimization o a single tower is to obtain minimum energy consumption or maximum throughput. For a detailed treatment o distillation optimization and o the development o the optimization equations, reer to Chapter 8.21, Vol. 2, The Instrument Engineer’s Handbook, 4th edition. Unit optimization addresses several columns in series or parallel. It is concerned with the e ective allocation o eedstocks and energy among the members o that system. Plant-wide optimization involves coordinating the control o distillation units, urnaces, compressors, etc. to maximize prot rom the entire operation. All lowerlevel control unctions respond to set points received rom higher-level optimizers. pdc-pc Cd
I the prices o the distillation products are not known, it is impossible to maximize the prot rate or that column without taking into account all other aspects o the overall plant. Optimization o a single column whose product prices are unknown involves determining the location o the eed t ray ( FT ), the eed enthalpy ( ∆HF), and the column pressure (P), which will result in minimum operating costs. Assuming that the appropriate operating equations are available, it is a relatively simple matter to establish optimum values or these t hree variables. Because these variables must stay within specic limits, a st atistical design study can be made ofine that allows correlation o the variables with each o t he three optimizing variables. In a majority o cases, the optimum column pressure ( P) can be ound by lowering the operating pressure until the condenser capacity is reached, or until liquid entrainment occurs in the vapor on the trays. The operating equations in Figure 24 are applicable or the case when product prices are unknown. When terminal product prices or a single column are known, the column is optimized to obtain the specied separation or the least operating cost. In this case, the main task is to nd the separation that will maximize prot rate. Any o the ollowing conditions can be true or a particular column. 1. The optimum separation can be determined independently o eed cost. 2. Optimum separation can be obtained by producing the product with the highest unit price at minimum specied purity. 3. The optimum separation is a unction o t he price dierence between products. 4. The optimum separation is a unction o the price dierence between the heavy key components in the top and bottom (PHKD – PHKB) and o the price dierence between the light key component in the top and in the bottom products (PLKD – PLK B). The above our policies are derived by the evaluation o the partial dierential equations that describe the prot rate or a single column with respect to t he specied separation, noted as ( LK B), (HKD). og C
When product prices are known, complete economic optimization almost always requires that the column be operated against a constraint or at a point where the specied separation, ( LK B), is such that an incremental gain in production is equal to a corresponding incremental gain in operating cost. Column loading is increased when either the separation is improved or when the eed rate is increased at a constant separation. The operating constraints are a unction o the capacities o the condenser, the reboiler and the column. As eed rate or separation is increased, one o these three constraints will be approached. Ambient conditions, steam pressure, eed composition, etc. can also infuence the choice o the constraint rst reached. www.controlglobal.com
– –
Distillation Control & optimization
Condenser Constraint: The maximum cooling capacity o a given condenser at maximum coolant fow rate is
a unction o the dierence between the temperatures o the overhead vapor and o the coolant. When operating against the condenser constraint, tests should be conducted to determine the maximum vapor fow rate as a unction o this temperature dierence. Such a correlation can be obtained by column testing. Column pressure and condenser ouling are also major variables that will aect overhead vapor temperature. Reboiler Constraint: The maximum heating capacity o the reboiler at maximum heating media fow rate is a unction o the temperature dierence between the heating media and the liquid being reboiled. As with the condenser, tests can be conducted to correlate maximum vapor fow rate with temperature dierence across the reboiler tubes. Column pressure and reboiler ouling are major variables that will aect this correlation. Generally, when using waste streams such as low-pressure steam that would alternatively be vented, it is optimal to operate against a reboiler constraint. Column Constraint: Column capacity is a unction o liquid and vapor fow rates and o column pressure. Oten, capacity is limited by liquid entrainment o the vapor. Operation at low internal liquid fow rates allows higher vapor fow rates. Also, column capacity increases as the operating pressure rises. I capacity is limited by entrainment, then loading can be increased by raising the operating pressure. However, i column capacity is limited by the tray downcomers, internal liquid fow rate can be increased by lowering pressures. Thereore, to optimize the operation, the capacity-limiting parameter must be known. An equation can be developed rom test data to cover a limited range o liquid and vapor fow rates and pressure. This relationship can be linear and have a orm o V max = a1 + a 2 (L) + a3 (P). This equation is useul in predicting the values o eed fow rate and o t he separation that will cause maximum vapor fow rate to exist. For a detailed description o optimization strategies or a variet y o product prices and market conditions, reer to Chapter 8.21, Vol. 2, The Instrument Engineer’s Handbook, 4th edition. Cc
The advanced process control strategies that are most applicable to the optimization o the distillation process are usually based on white-box modeling, where the theoretical dynamic models are derived on the basis o t he mass, energy and momentum balances o this well-understood process. Fuzzy-logic and black-box models are less oten used, as they are more applicable when it is acceptable to use a complete mechanistic empirical model constructed solely rom a priori knowledge. The amount o energy used or distillation is approximately 8% o the total energy used in the industrial sector o the United States. Reneries spend 50% to 60% o their operating costs (i.e., excluding capital costs and depreciation) on energy, almost twice as much as does the better controlled and optimized chemical industry (30% to 40%). This dierence shows the potential or saving s through better control and optimization. While the optimization techniques described in this book can improve renery productivity and protability by 25% compared to present operation, this goal will only be achieved i we stop using individual PID loops and treat distillation as a single and integrated unit operation. In multivariable unit operation control, the variables, such as fows, levels, pressures, etc. become only constraints, while the controlled and optimized variable is productivity and protability. The advances in process control have replaced or supplemented the single loop PID, but it is still true that we can only control a process i we ully understand it. As was explained, each distillation column has its own “personality” and one cannot control it without ully understanding it. Thereore, the importance o the role o process control engineers, who ully understand the controlled process did not diminish, but in act increased in C importance as our tools o control became more sophisticated. www.controlglobal.com
– 0 –
Distillation Control & optimization
Acronym references ANN – ar ticial nur al ntwor ks ARC – analy zr rcor ding controll r B – bottoms product CW – cooling watr D – distillat DMC – dynamic matrix control e – cincy FCCU – fuidizd catalytic cracking unit FRC – fow rcording controllr FT – fow transmittr or d tray HK – havy ky HHK havir than havy ky IMC – intrnal modl control L – rfux Li – intrnal rfux LK – light ky LLK – lightr than light ky LNG – liqud natural gas LRC – lvl rcording controllr MFC – modl-r controls MIMO – multipl input, multipl output MVC – multivariabl control PB – pric o bottom product PCV – prssur control valv PD – pric o distillat PI – prssur indicator PIC – prssur indicator controllr PID – proportional-intgral-drivativ PRC – prssur r cording controllr RFC – rfux fow controllr RIC – r atio indicating controllr RG – rlativ gain SISO – singl input, singl output TBP – tru boiling point TRC – tmpratur rcording controllr VPC – valv posit ion controll r VPCV – valv pos ition contr ol valv ∆H – nthalpy chang x – bottom pr oduct compos ition y – distillat composition z – d composition
www.controlglobal.com
– 1 –
According to Lipták, developing eective control strategies and optimization o distillation columns can improve “productivity and proitability by 25%.” Being able to test these control strategies and optimization techniques oline can provide additional signiicant savings in time and money. By using the MiMiC Distillation Modeling Package, the process or instrument engineer can: • Test control strategies and optimizations on an accurate, dynamic plant model • Optimize the operation o the distillation process
• Train operators ofine in an environment identical to plant Human Machine Interace
The MiMiC Distillation Modeling Package is a powerul tool or control strategy testing, initial and ongoing tuning o control loops, and dynamic optimization o distillation systems. It provides dynamic and accurate models or distillation, providing time and cost savings to maximize new capital projects and operational excellence initiatives. It provides: • Accurate dynamics or distillation simulation • Easy-to-use conguration wizard tool • Powerul data visualization tools • Quick and easy integration with process automation systems through MiMiC simulation tags
For more inormation about the MiMiC Distillation Modeling Package, please contact: MYNAH Tchnologs 504 Trad Cntr Blvd. Chstrld, MO 63005 USA +1.888.506.9624 (North Amrica) +1.636.681.1555 (Intrnational) +1.636.681.1660 (Fax) http://www.mynah.com
About mynAH tecHnologies
MYNAH Technologies (www.mynah.com) is the leading provider o Process Simulation Sotware or Automation system sotware acceptance testing and operator training. Process companies worldwide save time and money on their process auto-
mation projects by using MYNAH’s award-winning MiMiC Process Simulation Sotware. MYNAH also produces high-perormance products or IO and device connectivit y. MYNAH Technologies is a wholly-owned subsidiary o Experitec, Inc., located in Chestereld, Missouri, USA.
