EClassroom

USE OF DYNAMIC SIMULATION

TO CONVERGE
COMPLEX PROCESS FLOWSHEETS
WILLIAM L. LUYBEN

Lehigh University Bethlehem, PA 18015

ommercial process simulators are widely used for designing new pro cesses and for anal ysis of existin g
processes. Mo st senior design courses contain a signifi cant component of co mputer si mulatio n of pro ce ss
flow sheets using these tools. Th e most widely used comm ercia l process simulation software is that developed by Aspen
Technology-AspenPlu s for stea dy -s tate simulatio n and
AspenDynami cs for dynamic simulation-and these tools are
used in the exa mples in this paper. The standard As pen notation is used . For example, distillation co lumn stages are
counted fro m the top of the column: the condenser is Stage I
and the reboiler is the last stage.
The simulators contain models of most common unit operations, which can be conne cted into a process flow sheet. If
the units operate in series, with upstream units feedin g downstream units, the simul ation is usually reasonably straig htforw ard. If the flow sheet contains recy cle strea ms, however,
the simultaneous solution of the typically very large numb er
of simultaneous nonlinear algebrai c equations that make up
the steady-state model can be quit e challenging. There is no
guarantee that any algor ithm will find a solut ion. In additi on,
there are sometimes multiple solutions in these nonlinear systems. The convergence of these recycle loops (or "tear" streams)
is a major challenge in steady-state process simulation.
Energ y integrat ion can also produ ce complications because
of "e nergy recycle" between different units. These difficulties can sometimes be avoided by using the plant utility system to break the energy linkage.
Commerc'ial steady-s tate simulators contain a variety of
algorithms. For exa mple, the user of AspenPlu s can try such
methods as Wegstein , Bro yd en , and Ne wt on . Con ver gen ce
tolerances and the max imum number of iter ati on s ca n al so
be adjusted.
The experience of many users, particularly students, has
been that the conve rgence of recycle loops is the most diffi142

cult part of steady-state simulatio n. The normal procedu re is

to ass ume some conditions of a recycle stream (flow, tem perature, pressure, and comp osition ) and work down through
the flo wsheet until the calculated values of the rec ycle stream
are available. If the ass umed and calculated values are not
sufficiently close, new guesses must somehow be made. The
process is repeated until convergence between the ass umed
and calculated values has been attained. Often, however, co nvergence does not occ ur.
One wo uld think that this conve rgence should be fairl y
easily achieve d if the user has adjusted the design and operating parameters so the ass umed and calculated conditions of
the recycle strea m are fairly close, but this all too frequ entl y
does not occur. For exa mple, in one of the case s discussed
later, an assumed recycle strea m had a co mpositio n 98 mol%
methan ol and 2 mol% water, while the calculated stream is
only slightly different (0.4 mol% dimeth yl ether; 97.2 mol%
methanol; 2.4 mol% water). The assumed flowrate is 72 kmol/
hr, while the calculated flow rate is 72.9 kmol/hr. Th e temperatur e and pressures are ide nt ica l. Afte r co nnec ting the
recycl e strea m and defin ing it as a "tear" stre am in the
" Co n ve rg e nce " sec t io n o f t he " Dat a B row s e r" i n
Asp enPlus, the recycl e loop doe s no t co nverge when using an y of the algor ithms .
Thi s paper illustrates that recycle loops ca n be easily converged if the steady-state AspenPlus simulation (with the reWilliam L. Luyben earned degrees in chemicai
engineering from Penn State (8 S, 1955) and
Delaware (PhD, 1963). His industrial experience
includes four years with Exxon, four years with
DuPont, and three decades of consulting with
chemical and pe troleum compa nies. He has
taught at Lehigh University since 196 7 and has
participated in the development of several innovative undergraduate courses.

Copyrigh t ChE Divi sion oj ASEE 200.J

Chemical Engineering Education

cy cle loop s not co nnec ted) is "exported" into the dynamic

si mulator As pe nDynamics and the recycle co nnec tio ns are
made on the dy namic model. The steps in goi ng from a steadysta te simulation to a dynami c simulation are disc usse d in the
next sec tio n.
It should be noted that the issue of requiring goo d initi al
guesses of plant co nditio ns in orde r to converge is not a problem since all the unit s have been co nverged individu ally in
AspenPlus before go ing into dy na mics. The "guessed " and
the "calculated" values of the tear streams have
also been adju sted to be fairly close to eac h other.

TRANSITION FROM STEADY STATE

TO DYNAMIC SIMULATION
Th ere are seve ral items that must be taken care
of to co nve rt a steady -state simulatio n into a
dynamic simulation: all equipment must be sized
a nd a co ntro l struc ture mu st be de vel op ed .
Lu yben' " prese nts many details of these necessa ry steps, which are summarized below. Not
all of the units that are available in steady -s tate
As penPlus are suppor ted in As penDy namics, so
th is lim itation mu st be kept in mind . For ex ample, neither the "separa tor" (a ficti tious co mponent splitte r) nor a liqu id-l iqui d extractor is
supported in the curre nt version (Version 11.1)
of As penDynamic s.

Th e Tray Sizing sec tion of a distillation column block in

As penPlus ca n be eas ily used to pro vide the
column diamet er. Th e default weir height of
0.05 m can be used. Th e volumetric flowrates
The
of liquid into the reflu x drum (Stage I, the
convergence
"condenser" in Aspen terminology) and the
liquid into the base of the col umn (the last
of steady-state
stage, or "s ump" in Aspen terminology) can
simulations of
be used to size the two vessels by using the
flowsbeets
heuri stic of a 10-minute holdup time . These
volumetric flowrates are given in the Hydrau with recycle
lics
pag e tab of the Profiles section of the colstreams is
umn block . To have these result s made availfrequently
able, yo u must go to the Report section of the
very difficult.
co lum n bl ock , se lec t th e Property Opti ons
page tab and c lic k the Include Hydraulic PaAn alternative
rame te rs box be fore running th e pr ogram.

is suggested in
this paper and
an example
illustrates the
proposed
method.

When the steady-s tate simulation in Aspe nPlus

is ex po rted into Aspe nDynamics. a "p ress uredri ven" dynami c simulation should be used. Thi s
requires that all the " plumbing" mu st be specified in the flowsheet. Pumps and co mpressors
mu st be insert ed where needed to provide the
required pressure drop for materia l tlow. Co ntrol valves mu st
be installed where needed , and their pressure drops selected.
Th is is one of the more imp ort ant educa tional aspects of
the procedure since mo st students have a poor grasp of plumbing. Common errors include inserting two va lves into a liquid-filled line, inse rti ng a va lve in the suctio n of pumps. or
inse rting a valve at th e di sch ar ge of co mpressors (co mp ressor speed or its eq uivale nt co m pre ssor work sho uld
be manipul at ed ).

Equipmellt Sizillg - For steady-state simulation, the size

of the equipment is not needed , except for reactors. For dynam ic simulation, the inve ntories of mater ial co ntained in all
the pieces of equipment affec t the dynam ic res po nse , so the
ph ysical dim en sion s of all unit s must be known .
In distillation co lumns. the diameter of the co lumn, the weir
height , and the sizes of the reflux drum and the co lumn base
mu st be spec ified. Of co urse . before these ca n be ca lculated,
the number of stages and the feed stage location must be set
by some heuri stic or rigorou s optimization meth od . Perh ap s
Sprin g 2004

the eas iest heuri stic approach is to fix the distillate and bottoms spec ifications (using the Design Spec and \.ill)' tool s in
As pe nPlus ) and keep increa sin g the number of stages unt il
the requ ired reflux ratio stops decreasin g-this gives the mini mum reflux rati o. Th en the ac tua l reflu x ratio is set at 1.2
tim es this minimum. Finally, the optimum feed stage can be
de ter mi ned by vary ing the fee d stag e until the min imum
reboi ler energy co nsumptio n is found.

For example, the liqu id holdup in the reflu x

dru m of a col umn with a total co nde nser is
ca lculated from the volumetri c flowrate of liquid leavin g the drum (dis tillate plu s reflux),
[Drum volume (mJ ) ]

= [Liquid volumetric flowrate (mJj

min)][10 minutes]

If an aspe ct ratio (len gth to diameter, LID) of

2 is used, the diameter of the drum is
0= [2(Volume)/rcjl"

Th e same procedure can be used for flash tanks and vaporizers. Fla sh tank vapor velocity should also be checked .!"
Heat exc hange r tube -and- shell vo lumes can be calculated
fro m the heat-tran sfer area, which is known from the steadystate design, if a tube diameter D (typica lly 0.0245 m) is selected.
Area = [number of tubes][rcD][tube length L] = N1ube(rcDL)
[Volume of tubes] =[rcD1j4][L][N1ube]
= [rcDW ][L][Area)/[rcDL] = D[Area]/4
Shell vo lume is approxim ately equa l to tube volume in most
tub e-in-shell heat exc hange rs. If the proc ess streams in the
heat exc hange r are gases, the dynam ics are very fas t and can
usually be ignore d (spec ify Instantan eous in the Dynamic
sec tio n of the heat- exch anger block .

Plalltwide Control Wh en the file containing the flow sheet

is opened in AspenDynamic, a default control sc heme is al143

ready installed on some loops. For example, level and press ure con trollers are inserted on all distillation columns and reactors in the
flowsheet. This default control must be modified and supple mented
with other control loops to incorporate a stable basic regulatory
control structure.

Plantwide Control Structure A simple heuristic metho d for development of an effec tive plantwide control structure is presented
by Luyben, et al.13] Ge nera l principles and many exa mples are give n
in great detail. The pro posed nine-step procedure has been successfully app lied to a large num ber of rea listically complex industrial processes. Some of the key co ncepts are: ( I) the con trol structure should guaran tee that all chem ical components fed into the
sys tem are either reacted or ca n leave in some ex iting strea m; (2) a
flow controller should be install ed so me where in all liqui d recycle loop s; and (3) all liq uid le vel s mu st be co ntro lled and
pressures in gas systems , whic h ca n be multipl e units connected
togeth er, mus t be co ntro lle d .

eters . Th e che mistry is the exo thermic revers ible deco mposition of meth anol to for m DME and water in an adia batic , tubular, gas -phase reactor.
2 MeOH <=> DME + Water

The reac tion is exot hermic . and the adiabatic tem perature
A
~_--:-

I
i
I
I

I
J
_
U
I

i
I

,~--!---,--i----i---I,--,

.--,1,

l ength (meters)

Controller Tuning Most of the controllers are eas ily tuned by

simply using heuris tics. All liquid levels should use prop ortionalonly controllers with a gain of 2. All flow con trollers sho uld use a
gai n of O.S and an integra l time of 0.3 minutes (also enab le filtering
with a filter time of 0.1 minutes).

B
150

'"
0

:;;
C

The defa ult values in AspenDynamics for most press ure contro llers seem to work reasonably well. Temperature controllers ofte n
need some adj ustments. The defau lt tra nsmitter ranges are usually
too large, and spans should be set at abo ut 10% of the absol ute
te mperat ure leve l (ty pically a spa n of 100 K for mod erate- tem pe rature processes).
Distillation colum ns are typ ically con trolled by manipulating
reboi ler heat input to contro l the temperature on some selected tray.
Th e heuristic procedure of finding a tray where the temperature
changes from tray to tray are large is easy to use and provides effective composition con trol in most cases . Direct com position measurements can be used if temperature changes are too sma ll. If very
large temperature cha nges occur in the col umn (over 100 K), an
average tem per atur e can be used (measuring the te mperatures at three or fo ur trays,
DMEProcess
calculating the ave rage, and using this for
co ntro l).

100

50

450

Tin (I<)

600

750

g
'5
~

700
650

600
550

500
450
450

Figure 1.
(A] Adiabatic reactor temperature profile (10 m eters
leng th].
(B] Effect of inlet temperature for different reactor
lengths.

It should be kep t in mind that the objective at this point is not to come up with the
"best" co ntrol structure or the opti mum
co ntro ller tuning . We only need a stru cture and tun ings that dr ive the simulation
to a steady sta te .

EXAMPLE ("OME" PROCESS)

To illustrate the use of a dy nam ic sim ulator to converge a flows heet, we se lect the
di methy l ether (DME) process discusse d by
Turton, et alY ]Th is design text gives a flowsheet and some pre liminary desig n param144

Frnb

Melb a nol
Feed

Water ProductV21

Figure 2. DME flowsheet.

Chemica! Enginee ring Educat ion

rise is abo ut 120 K. The temperatu re profi le is show n in Figure IA for a reactor that has a diameter of O.n m and is 10 m
long. Fig ure IB shows how reactor inlet temperat ure Tin affects the produ ct ion of DME and the reacto r exit tem perature
T UUI for three different reactor lengths. The inlet tem perature
req uired to achieve maxi mum conversio n decreases as the
reac tor is made bigger. S ince the react io n is exotherm ic .
the maxim um conversio n decrease s slo w ly as te mperatures inc rea se .
Th e equilibri um constant is abo ut 6 at a reactor tempera ture of abo ut 600 K. so the per-pass convers ion is abo ut 80%.
T his requires a recycle of methan ol back to the react or from
the separatio n section. Figur e 2 gives the As penPlus flowsheet. and Figure 3 gives strea m co nditions. T he NRTL physica l prop er ty package is used in the simulation.
Fresh meth anol and recycl e met hanol are vaporized and
heated to 555 K. T he vapor-phase reacti on occ urs at ab out IS
atm . Reactor effl ue nt is cooled and fed into a two-column
separation section. The low-boilin g DM E is the distillate product in the first co lumn C I. which ope rates at a pressure of 10
atrn so that coo ling water ca n be use d in the co nde nser (reflu x-drum temperature is 318 K with 99.9 mo l% DME purity). Th e column has 32 stage s. is fed on Stage 15. and is
opera ted with a reflu x ratio of 0.5.
Th e bott om s of Column C I is fed to the seco nd col um n C2
in whic h wa ter is removed from the bott om and recycled
meth anol is remove d from the top. Thi s co lumn operates at
1. 1 atm and has a refl ux-drum temperatu re of 340 K. It has
22 stages and a reflux ratio of 1.5.

Reaction Kinetics' One of the most difficu lt parts of developing a tlowsheet is ge tting the reactio n kinetic param eters co rrec tly spec ified. Any errors in unit co nve rsions are

r-'; .

if

a lAiN

Si nce the reac tio n is reversible. both the fo rward and the
reverse reac tio n rates mu st be specified. Thu rton , et al.P provide a kinetic ex pression for the forward reac tion rate
R F ( kmol l hr 1 m 3 ) = 1.21x 10 6 e - SO"+SO/ RT P~ I EOH

0]

B IREC(CLE

piMET-lll

liJO.O

.26672E-9

MEOH

262.0000

73.02976

335.0264

0.0

.5078546

.5078378

i ~ 0.0

.1619Ell

.5693E-ll

.3900267

.9930939

ROUT

RF(kmoll scc/m -') = 1.2 1x 106 e - SO.-lSOIRT r pkl' a 113600

= 1.21 x 10 6 e - 80.-lS0/RT [ p Pa ]/3600

1000
= 0.336 1e - 80.-l801RT P Pa

(2)

Turto n. et (/ I..' ~ I also give informa tion abo ut the equilibrium

con stant. but state that the pub lished data does not seem to
match the calculations using free energies . Th is was confirmed
by running an R-Gibbs reactor in Aspe nPlus. which gave a
much lower reactor conversio n tha n wo uld be pred icted by
the published data.
T he heat of reaction is A = - 11.770 kJ/kmol. so we ca n
esti ma te the ac tiva tion energy of the reve rse react ion fro m
that of the forwa rd reaction .
E F - E R = A~ E R = 80.480 - (- I I. 770 ) = 92.250 kl I krnol

(3)
...:.. IDI

B 181

B 82

Fl3Y

.J::.., 1

.WATER "'-

ME~

WATE,:\
rolalFlowklnor/hr

-.

1-,qooooo

I,

0.0

262.0000

5.26672E-9

.8486E-23

52.34693

73.29178

.1309983

73.16076

.1309980

29.31671

130~~_~73 ;<-,
73.29176 .-

131.3752

6.7731E-13

131.3751

130.8672

52.55007

131.3751

.9990000

.5750E-ll

.9379E-25

.3900267

.3900268
21 ~4 33 1

'"

.9 ~8 48 65

.2184331

.00000E3

_3 5 7 6 9 1 ~

.99998E'"

.2184331

~.90605E-3

.51352E-3

.3915402

5.1703E-15

.6423084

.9990000

.3915402

.391 5~1

73.53762

335.5343

335.5343

130.9983

204.5358

130.9982

134.2137

335.5342
10744.12

folalFlowkg/hr

8395.046

2349.180

10744.12

10744.12

6033.130

4710.988

2361.807

4297.648

folalFlowI/min

175.3164

52.67718

17092.40

21437.16

162.4282

108.3552

43.21094

8574.864

243.3503

fampa'CI!Il ,a K

298.1500

342.2560

555.0000

672.7188

318.1082

424.6567

379.1812

672.7186

325.0000

'rassu,e IlIm

17.00000

17.00000

14.90000

14.40000

10.00000

10.21094

1.242897

14.40000

12.00000

0.0
nnnnnn

0.0
nnnnnn

1.000000

1.000000

00

0.0

0.0

1.000000

lopo,F,oc
' ;n ,,;" ",""

nn

nnnnnn

nnnnnn

nnnnnn

nn

'"

'"

.=

......

130.8673

130.8673

~ol e Fr~

DIMET-lll

I ~?:
~.26566E-9

c~~

~.

.::..

..

violaFlowklnol/hr

(I)

where P~I EOfI has units of kPa and the act iva tion energy has
units of kJ/k mol. As pen insists on exp ressi ng reacti on rates
in krnol/sec/m' , so we must divide by 3600. More serio usly,
As pe n also insis ts that partial press ure be in Pa, not in kPa . A
co mmon erro r is to mult iply by 1000 to co nve rt the pressure
in Eq. ( I) fro m kPa to Pa. T his is incorrect. T he pressure in
Eq. ( I) is in kPa . and Aspen uses Pa. so the pre-exponential
fac tor must be d ivided by 1000.

--,

IMEOH

'ubslrel>ltl:)\,UXED .'

amplified by the exponentia l ex pres sio ns.

.~

~O , ~.,

nnnnnn

Figure 3. DME stream dat a.

Spring 2004

/ 45

To find a reasona ble pre-exponenti al factor for

the reverse reaction, the conversion data (reactor size; inlet flowrate, temp eratur e, and
composition; and exi t co mposi tion) give n
Turton, et aI., were used to back-c alculat e this
parameter by trial and error. The resultin g kinetic expression is
R R (k mol l

VAPour
13

secl m 3 )

= 2.7 x IO--{; e - 92.250/ RT (PDM E )(PWater)

Energy Tear Steam

HXl

(4)

where PDME and PWater have units of Pa.

Thermal Recycle Loop A feed-efflu ent

heat exchan ger (HX I) is used to preheat reactor feed, using a portion (20 I kmol/hr) of the
hot reactor effluent to heat the vapor
from the vaporizer from 427 K to 555
K. Some of the reactor effluent (134
kmol/hr ) is bypassed through valve V2
to control reactor inlet temperature.

VOUM

Figure 4. Flowsheet with energy tear stream.

,--------1"""'1-------.

Thu s there is a thermal recycle loop

in the process. Unless the reactor effluent stream is known , the HX I equations cannot be solved. The conventional procedure to get started is to use
a tear stream (see Figure 4). We specify
a stream "HIN" and provide estimates
of its flow ( 193 kmol/hr), temperature
(672 K), pressure (14.4 atm), and composition (40 mol % DME, 20 mol%
methanol , and 40 mol% water).
When this flowsheet is converged,
the source of "HIN" is changed to the
splitte rT2 (deleting stream "5" in Figure 4). In the Convergence section of
the Data Browser, we select Tear and
specify HIN to be a tear stream. Then
we rerun the program to get the converged flowsheet around the reactor
and heat exchanger HX I.

Figure 5. Flowsheet with m aterial tear stream RECYCLE.

r-------IHOUTl---------

Material Recycle Loop Figure 5

shows the flowsheet with a tear stream
RECYCLE, whose stream conditions
have been estimated to be 72 kmol/hr,
345 K, 17 atm, 99 mol% methanol ,
and I mol% water. The calculated conditions for stream RCALC are only
slightly different from those assumed
for RECYCLE (73.2 kmol/hr , 342 K,
17 atm, 99.3 mol% methanol, and 0.1
mol% water).
The default con vergenc e meth od
146

Figure 6. Defaul t con trol scheme; DME process.

Chemical Enginee ring Education

Wegstein is used (with a limit of 30 iteratio ns), the source of the

RECYCLE stream is specified to be the discharge of pump P22
and it is specified as a Tear under the Conve rgence Sectio n of the
Data Browser Th e program is run .
The convergence loop fails to converge.

Thi s is a typical result in man y cases . It should be noted that

Design Sp ecs and Vmy features are used on both distillation column s. In Column C I , the distill ate DME produ ct purit y is specified to be 99 .9 mol % DME, and the di stillate flowrate is varied.
In Co lumn C2, the bottoms water produ ct purit y is specified to
be 99.9 mol % water, and the bottom s tlowrate is varied . The reflux ratio is fixed in both co lumns.
Incre asing the ma ximum iterat ion s does not achieve con verge nce. Swit ching to the Broyden algo rithm is equally unsuccessful.
Thi s failure to co nverge cert ainl y does not occur in all cases
for all flowsheets, but it does occ ur in many cases. Senior students
spend many frustrating hours trying to get recycle loops to converge.
Co nverting to Dynamic Simulatiol/ Th e dia meters of the two
columns are calc ulated in the Tray Sizing section: 0.6 1 m for C I
and 0.94 m for C2. Th e liquid tlowrates into the retlux drum s of
C I and C2 are 0.242 and 0.131 mvmin, respecti vely, as found in
the Hydraulics page tab. Reflux drum sizes (D X L) in the two
columns are set at 1.2 x 2.4 m and I x 2 m, respe ctiv ely, using an
aspect ratio of 2. Th e liquid flowrat es into the base (to the sump
from the next-to-Iast stage in the co lumn) of C I and C2 are 0.213
and 0.085 3 m' min, respecti vely. Reboiler sizes in the two co lumn s are set at 1. 1 x 2.2 m and 0.82 x 1.64 rn, respective ly.

Th e heat exc hangers are assumed to be "instant aneou s" since

they are gas phase. Reactor size is already specified for the steadystate simulation. The onl y other dynamic unit in the tlowsheet is
the va po rizer. Its liquid feed is 10,700 kg/hr with a den sit y of
Error Mess aae

- " .RJ

r- '"

Ge ne~ IFlUdOi 8flg IJrtlIieI O1lUlg 1

S"".

y~ ._

... . ..

_dby" , ,,,,,,,,,,..

""' ~

1iJ;~

,oJ

e~OJ

Ff1!8

Suggested Action Message

----

_o x

S"
s<~,. I ~ -"''''_''''''''''''''
~'''''~'''''''M"Uo.~. D~.''
rlIr-----~rbutl;l\lorrno'lIl1llormellOtl
Yo..,,,.

~~~

l----"
~

e~

...... I'I

--~

N_S

f~ f re.

F-..t

F...

Ve!;le [)Ol':
170
)-6 0

-e
2

-g ,

~~ l't!

--,,1 "'*. I

00 ..

Figure 7. Error m essage and s uggeste d action m essage.

Sp ring 200-/

~--~~

"'~~~
.. --"~ ._-"'
Vanrblll

Th e fil e is pr e s sure c he c ke d and ex po rte d into

AspenD ynamics. The default control scheme is shown in
Figure 6. Note that eac h column has pressure and level controllers, some of which are fully connected and others without the co ntro ller output signal (OP) connected to a valve.
Thi s default co ntrol scheme must be modi fied to provide
an effec tive regulat ory contro l scheme.
Before doin g anything, an Initiali zat ion run and a short
Dyn amic run should be made to confirm that all the plumbing is okay and the process is co rrectly co nfigured.
Closing the R ec ycle Loop The proc edure for changing
the proc ess struc ture in AspenDynamics is the same as in
AspenPlus. Th e stream RECYCLE is deleted. The destination of the stream RCALC is made the inlet to valve V22 .
Another set of Initializati on and Dynamic run s should be
made. Now the default co ntro l structure ca n be modified. It
is a goo d idea to perform Initi alizat ion and Dynamic run s
afte r eac h new change in the co ntrol structure so that any
erro r in contro ller installa tion ca n be detect ed indi vidu all y.
Th e most co mmo n erro r is to have the wro ng action in the
co ntroller (for example, specifying reverse action when it
should be direct action).

An obvious alternativ e to deleting RECYCLE and reat tachin g RCALC is to delete R CALC and reattach RECYCLE.
If this is atte mpted, the little green box at the bottom of the
AspenDynamics screen turns red, indicating that something
is wro ng. Doubl e-clickin g the red box opens the wind ow
shown at the top of Figure 7, which states that the probl em
is overspec ified. Click ing the Analyze butt on opens the window show n at the bottom of Figure 7, wh ich says that the
temp erature and pressur e of the RECYCLE stream must be
changed from fixed to variable. Clicking theA CCEPT button turn s th e box gre en aga in . Then Initiation and Dy nomic are again run to make sure all is okay. Now the
default control sc he me can be modified.
Plantwide COl/trol Structure Usin g the method proposed in Luyben , et al.. [J] a plant wide cont rol scheme is developed that feature s the follow ing loops:

78 5 kg/rn ' , w hic h give s a d iam et er of 1.1 m and a len gth

of 2. 2 m to pro vid e ten minutes of holdup.

it

The fee d flow to the fi rst column is flow controlled. This

puts a fl ow controller in the liquid recycle loop. /t also has
the advantage of keeping a fi xed steady flow 10 the column.
Since the fi nal DME product is produced ill this column,
product qual ity variability is minimized by 1I0t permittin g
disturbances to enter this column.
The vaporizer level is CO li trolled by manipulating the f resh
methanol fee d. This guarantees that only the amount of
methanol that is being conslimed ill the reactor will be fe d
into the process. If more DME p roduction is required, the
setpo int of the fl ail' controller on Column C/ f eed can be
increased. Using reactor inlet temperature to change
produ ction is not effective because of the reversibility of the
reaction. For example. sometimes increasing reactor inlet
147

temp erature produ ces a decrease in the p roduction of

DME.
Vaporizer pressure is controlled by manipulating steam
(heat input Q) to the vapo rizer.
Reactor inlet temperature is controlled by the HX1 heat
exchange r bypa ss fl ow (valve V2).
Condenser HX2 exit temp erature is controlled by man ipulating cooling water (hea t removal Q).
The pressure in each column is controlled by condenser
heat removal.
The reflu x-drum level in each column is controlled by
man ipulatin g distil/ate fl ow.
The base level in each column is controlled by manipuluting bOIlOIllS product fl ow.
A temp erat ure in each
column is controlled by
manipulating rebo iler
heat input. The steep pa rt
of the temp erature profil e
in C 1 is at Stage 17 and
in C2 at Stage 19, so
these tray locations are
selected. The temp eratu re
setpoint in C1 is 380 K
and in C2 is 370 K.
The reflu x ratio in eac h
..
co lumn is controlle d.
V22
The distillat e fl o wra te is
m easu red , multiplied by
the desired reflux ra tio
(0.5 in CI and 1.5 in
C2) and th is signal sets
th e flo wrate of reflu x.

AspenDynamics flow sheet with this control structure installed. All

level controllers are proportional with K, = 2. Table I gives controller tuning parameters and transm itter spans . The dynamic simulation is run until it reache s a steady -state condi tion . Thi s may take
several "process hours " (several minutes of computer time ), depending on the complexity of the flowsheet. When running the dynamic simulation out to a steady state, valve V22 in the recycle line
went wide open and the reflux-drum level in Column C2 began to
rise. The power to the pump P22 was increased to solve this valve
saturation problem . The detail s of how this is done are given in
Luyben.!''
The same valve saturation problem occurred in V 12 when the

TCalnd

'--+tISJt-.,!~l

LC12

r vw _ _ ~ j

,.~

dl:'
'''

':---- l: :"t: ~-1~~b~~

Figure

gives

,.@ _.:

~::~

Jf f

C1

P21

the

Figure 8. Plantwid e control struc ture: DME process.

TABLE 1
DME Controller Parameters

....

!'IJ .'ftroi . ......... !loft'.

Controller

Transmitter
Range
mill/max

Valve
Rallge
mill/max

(Action)

(units)

(Units}

0/663 (kmollhr)

0/100 (%)

_~FC

(Rev)

TCin (Dir)

0.3

0/100 (%)

2
2

300/400 ( K)

-2030/0 (kcal.sec)

20

10/20 (atm)

0/1910 (kcallsec)

",. 12

TCI (Rev)

300/400 (K)

0/690 (kcal/sec)

20

LCII (Dir)

012.5 (rn)

0/100 (%)

'''''2

LCI 2 (Dir)

0/ 3.4 (m)

0/100 (%)

(Rev)

-"'-

. .
~

,g~

~
"'r
j

!
~

~!.

,l:

PCI (Dir)

5/15 (atm)

-442/0 (lCcallsec)

5'" '" 12

TC2 (Rev)

300/400 ( K)

0/688 (kcal/sec)

1. ;; i, ~

LC21 (Dir)

0/2 (m)

0/100 (%)

PC2 (Dir)

C"

0/2 .52 (m)

0/100 (%)

0/2 .2 (atm)

-86010 (kcal/sec)

20

l
l

.,:;,

'j

20

i',11:-1~ i:I
,
, , 'j !

LC22 (Dir)

148

>!! 0.5

(mill)

01100 (%)

J!> PCvap

.,.

K,

500/600 ( K)

TCcond (Rev)

7[

' 0/3.26 (m)

~p (Rev)

'O'. IiII_ Clp : _lS I '" ,." ~ ...,,ro.,..,-

Integral
Time

s !

'.\ I
f

..

'

~.

......

"

+ .;. . ; . . .

"!::U :

~ ~
,

, ....;..... ,

: ~I

T T

o'-, ~..,,--...,..--. ,.

.; :.

Figure 9. Dyn amic responses to 20 % in crease in feed to

colum n ci .
Chemical Enginee ring Education

setpoint of the C I co lumn feed flow controller was increase d 20%

(fro m 236 to 402 kmol/hr). Th e power to pump P 12 was inc reased .
Not ice that the steady-s tate signals to valves V22 and V 12 are 20 %
and 29 %, respec tivel y, instead of the normal 50 %, beca use of the
higher valve press ure drops.
Figure 9 shows the dynamic respon se of the sys tem to a 20 %
increase in col um n C I feed. Th e initia l condition is the steady-state
con dition. The fresh feed of metha nol increases from 262 to 308
kmo l/hr and DME produ ct inc reased from 13 1 to 154 kmol/hr
(17.5 % incr ease in product ion rate ). The pu rities of both the
DM E and the wa ter products are maint a ined for thi s large di sturb ance. T he system tak es abo ut two process hours to come to
the new steady state.
The one-recy cle process illustrates how easi ly the flowsheet ca n
be conve rged by usin g dynami c simulation. Figures IO and II give
an exa mple of a process with two recy cles . Th e meth yl acetate and
butanol reactants not consumed in the reactor are separated in a threecolumn separation section and recycled in two different streams. More
details of this example are available from the author.

CONCLUSION

recycle streams is frequently very diffic ult. An alternative is suggested in this paper and an example illustra tes
the propo sed met hod . The steady-state simulation is con verted into a dy namic simulation, and the recycle loop s
are converged by letting the dynami c simulation run to
steady-state conditi ons.
The method depends on the development of a baselevel regu latory plantwide control structure, which can
be obtained by following a simple heuri stic design procedure. Simple controller tun ing rules ca n be applied to
eliminate detailed and length y contro ller tuning effo rts.
Th ere is, of course, an additiona l benefi t for this approach. Th e dynamic simulation ca n also be used to look
at the dynamic effec ts of alternative design conditions
(flowsheet structure, operating co nditions, equ ipment
sizes, etc.). Th is approac h, which is called "simultaneous
desig n," is a design philosop hy in which both the steady
state and the dynam ic performan ces of a process are
co nsidered at all stages of the development of a pro ce ss. Th e book by Sei der, et al., [5J disc usses this approach in mo re de tai l.

The converge nce of steady-state simulations of flow shee ts with

NOMENCLATURE
D
DME
E
FF
Kc
L
Pj
Te n
Tin
Tout

diameter
dimethyl ether
activa tion energy
fres h feed
contro ller gain
length
parti al pre ssur e of component j
tray temperature co ntroller in co lumn n
reactor inlet temperatu re
reactor exit temp erat ure
1:I integral time constant in PI co ntro ller
(minutes)
A heat of reactio n

Butyl Acetate Process

." Butyl Acetate

REFERENCES

Figure 10, Flowsbeet for two-recycle bu tyl acetate process .

Figur e

I. Luyben, W.L., Plantwide Dynam ic Simulators fo r

Chem ical Proce ssin g and Cont rol .
Marcel Dekker (2002)
,..-- - - - - - - - - - -iRClol.C1
2. O'Bri en. N.G.. and R.G. Franks. "Development and Application of a General
. - - + - - - - - - - - -----i ACALC2 1 - - - - - - - - - - - - - - - - ,
Purp ose Analog Computer Circuit to
Steady-State Mult icomponent Distillation," Chem. Eng.. 55. 2 1 (1958)
3. Luyben, WL. , B.D. Tyreus, and M.L.
Luyben , Plantwide Process Control,
McGr aw-Hili ( 1999)
4. Turton. R.. R.C. Bailie. WB . Whiting, and J.A. Shaeiwitz, Analysis. Synthesis and Design of Chemical Processes. 2nd ed.. Prent ice Hall. pg. 939
(2003)
5. Seider. WD .. J.D. Seader. and D.R.
Lew in. Produ ct lind Process Design
P31
Principles. John Wiley and Sons, New
11 , Plantwide con trol structure for two-recycle process .
York. NY(2004 ) 0

Spring 2004

/ 49