Cop\'ri~ht © .-\ir Products a lld Chemicals.

111 (,
Il)R'i

DYNAMIC MODELLING AND CONTROL OF

CRYOGENIC AIR SEPARATION PLANTS
J. A. Mandler, D. R. Vinson and N. Chatterjee
Air I'mill/d' IIlId Chnlliml.,. 111( .. 72UI Hall/il/ulI Rh·d .. :1.//~II/Vll'lI.
1'.4 18 1 <) 5- I 5() I . L'S,4

Abstract. An air separation plant is an interesting and chal l eng i ng system from
a control standpoint. The process involves up to three h i gh l y interconnected,
mu lti component, high-purity distillation columns coupled to a series of complex,
efficient heat exchangers. The system can have over five liquid and gaseous pro-
duct streams . A high degree of energy integration is achieved. The plant oper a-
tin g conditions can change over a wide range depending on the mode of operation ,
on the variation s in demand of the on -lin e customer and on weather co nditi ons .
Plant opt im ization is sought in the face of these and o ther process c han ges .

Together with an introductory overview of the ai r sepa rati on process and it s

main control issues, this paper describes our current wo rk on dynami c modelling
and simulation of cryogenic air separation units. Rigorous mathematical models
were derived for the mult i component - multiphase p r ocesses, in clud ing distillation
column models with minimal a-priori assumptions a nd dist ributed /dynamic models
for the reboiler/condenser units. The paper h i gh li ghts speci al features of
cryogenic distillation systems that distinguish them fr om more convent i ona l
d i stil l at i on systems and that a l so give rise to special modelling needs. The
paper presents the basis of our dynamic modelling/nume ri cal solut i on approach
which fully addresses these special needs. The benefits of a different ial /alge-
braic model formulation and solution are discussed.

Keywords. Mode llin g; dynamics; air sepa rati on; distillation; heat exchangers;
d iffer e ntial /algebraic syst ems; computer cont r ol; o ptimization.

OV ERVIEW OF THE AIR SEPARATION and any heat leak from the surround ings , a nd in
PROCESS AND MAIN CONTROL ISSUES order to produce any liquid products .

Cryogenic air separation technology is used to

supply nearly all of the oxygen consumed by DISTILLATION

industry today. Air separation plants are also a

source for pure gaseous and liquid nitrogen, and
AIR
for valuable by - products such as argon. Improve - COMPRESSION
AIR
ments in plant operat i on and process control can PURIFICATION
resul t in significant savings. AND
HEAT
Fo r ove r a de cade the Process Systems Group within EXCHANGE

Air Pr od ucts has successfully in sta lled compute r

PRODUCT LIQUID
cont rol/on-line opt i mizati on systems on ove r 120 COMPRESSION PRODUCTS
operat ing plants. These systems insure efficient
and stable operatio n while matching production to
demand. The benefits of optimization have been
substant ial (D aws on and colleagues , 1989; King and Fig. 1 . Air separation unit oper ati ons.
colle ague s, 19 87; Vinson and Chatterjee, 1986). As
process inte grat i on increases, plants become more
comple x and specifications more exacting, the need The need for thermodynamic efficiency in air
for advan c e d, r obust control schemes designed on sep arati on plants result s in comp lex integrated
the basis of a detailed understanding of the distillation co lumn systems. Figu re 2 illustrates
process dynamics becomes more urgent. Good dynamic this comple xity for the l ow pressure cyc le which
models are als o needed to improve contro l produces nitrogen, oxygen and argon products. After
strategies in certain ' hard to con trol' systems for the air is cooled to near its liquefaction tempe-
which traditional control approaches are n o t rature, it is fed to the high pressure distillation
entirely adequate. co lumn . In the high pressure (HP) column the air is
separated into a pure nitrogen fraction and an
In the air separation process, Fig . 1, air is oxygen rich liquid fraction , The overhead from the
compressed, c leaned from impurities, cooled to top of the HP column is c ondensed by boiling liquid
cryogenic temperatures and distilled into its oxygen in the bottom of a l ow pressure (LP) column,
component parts . The produc t streams may be remove d physically located on top of the HP column. This is
as liquids, or warmed against incoming air to done through a reboiler/ condenser operating in a
ambient temperature and supplied in gaseous form. thermosiphon fashion. The pressures in the two
Refrigeration is provided to overcome the finite columns are such that the boiling point of pure
temperature differences in process heat exchangers nitrogen at the high pressure conditions is higher

267
 .J. :'I.. \landl er. D. R. \ 'insu n a nd \:. C haltcrj ec

LPGAN4-----------r---~

-....,~~-I-=~;f:::=::~
UN ••
WASTE •• ----r
HP UN
-ARGON

GOx ••- - - -

HP GAN ~=====~:~~~
AIR -

Fig. 2. Air separation distillation column system .

than that of pure oxygen at the low pressure upset to plant ope ra t i on may be caused by very
condit i ons. Because the condensing duty in the small amoun ts of nitrogen ente rin g the s id earm
HP column provides the reboiler duty for the LP feed. Cont r ol of the feed compos i tion to the
co lumn. there is a strong heat integration between sidearm co lumn to maximize the contained arg on in
the two co lumns linking the column pressures . Thi s the feed without intr od ucing an unacceptable amount
is a major cause of intera ct ions between the of nitrogen is among the most diffi cu lt control
cont r ol loops associated with each column . problems in many air separati on plant s.

To achieve thermodynamically efficient separation Most plant ope rat ors are conse rvativ e in ope rating
the low pressure column illustrated in Fig. 2 has the sidearm co lumn because of the serious
four sections. Other cyc le designs may have LP conse quen ces of a major co lumn upset. I mprov ing the
columns with more sections to allow a c lo se resulting argon recovery c an often pay for the
matching of the operating line to the equilibrium entire cost of a computer control system. Because
line . A major function of the HP column is to of the imp o rtan ce of argon op timizati on and the
generate the reflux streams for the LP column. The nature of the ab ove problems. the dynami cs of the
condensed nitrogen from the top of the HP column is low pressure/argon co lumn system was c hosen as the
split into two reflux s treams . one for each column. initial focus of a comprehensive program on
A second liquid s tream is withdrawn from an inter- dynamics and cont r o l of air separation plants.
mediate tray in the HP column t o provide an impure
reflux for the LP column . Optimizing the se two
reflux streams. whose purities are c losely linked. ARGON SYSTEM STRATEGIES
is an imp or tant aspect in the overall optimization
of the p lant . A short s ummary is presented of the main directions
taken by the Pr ocess Syst em Group at Air Produ cts
Argon is a valuable by-product of oxygen pro- to addre ss the arg on system issues. The combined
duction. The re covery of argon. whic h comprises implementation of the following appr oa che s and the
slightly less than one percent of the feed air. can effort of many teams has led over the last several
have a significant effect on the plant e conomics. years to signific ant improvements in argon
Argon has a volatility between that of nitr oge n and production and in plant operability. This brief
oxygen. Its concentration reache s a peak near the discussion will lead us to the dynamic modelling
middle of the LP co lumn . A gaseous side stream is studies. the main focus of this paper.
withdrawn from an interme di ate tray in th e LP
co lumn and is fed t o a parallel column (commonly
called a crude argon sidearm co lumn ) where the a ) Pr ocess Design Changes
argon- oxygen separation is performed. Note the
tightness of the integration between th is column Improvements in the c rude argon sidearm column
and the other two. To maximize argon recovery it is design have resulted in co lumns that can tolerate
important to have a clear understanding of how the significantly higher levels of nitrogen in the
argon profile shifts in the LP co lumn as conditions sidearm feed before getting into an upset situa-
change. in order to minimize the t ot al amount of tion. The benefits are not only improved control-
argon lost through the GOX and WASTE streams. lability and operability. but also increased argon
recovery . This is a result of the in c reased ability
Affecting the design of any argon optimization to drive the nitrogen profiles down the LP column
strategy is the fa c t that any nitrogen entering the before affecting operation of the sidearm column.
argon column will concentrate in its overhead In this way the amount of argon lost in the WASTE
stream . At the very least. this nitrogen is an stream can be reduced (GOX is purity controlled)
undesired impurity in the crude argon product. At and the argon concentration peak can be located
worst. the nitrogen may blanket the surface of the c loser to the sidearm point.
condenser at the top of the argon column to a point
that it may cause los s of feed flow to the sidearm b) Improved Instrumentation
column and dumping of the liquid inventory on the
trays back into the LP column. This can in turn The levels of nitrogen at the sidearm feed that can
contaminate the oxygen product. Such a serious cause serious trouble to the argon column operation
 Cn()gellic .-\ir Sep;lrati()1l Plallts

are very low and are also very difficult to measure have an adequate representation of the strong
on-line. It is easier to measure the nitrogen at dynamic interactions existing between the three
the point of highest concentration, the crude argon columns. This is a necessary step in the design of
product stream. However, control strategies based improved control/optimization strategies for the
on the measured nitrogen at the outlet rather than overall system. In the context of argon opti-
at the inlet to the argon column are prone to mization we seek to understand the movement of the
failure due to the nature of the dynamic relation- concentration profiles inside the low pressure
ships and the nonlinearity of the nitrogen input/ column to determine ways to better prevent the
output response in the argon column. Air Products' dynamic buildup of nitrogen around the sidearm
proprietary nitrogen analyzer has been recently point while still operating at close to maximum
perfected to a point that it can provide reliable recoveries.
on-line measurements of the nitrogen at the sidearm
feed conditions, this being a major development in The remainder of the paper focuses on the
our overall argon optimization efforts. mathematical models and the numerical solution
considerations for the work of phase a). Note that
the main objective of the paper is to describe the
c) On-Line Optimization modelling needs and our resulting modelling
approach, and not to present a detailed analysis
Argon optimization modules have been implemented and solution of the control issues just described .
within our current hierarchy of regulatory control
and on-line optimization (Dawson and colleagues,
1989). The on-line optimization is based mainly on SIDEARM COLUMN DYNAMIC MODEL
steady-state models whose parameters are constantly
updated as conditions change. In the case of argon Efficiency in cryogenic distillation cycles is
optimization this approach has been successful in a c hieved through use of small driving forces for
certain environments, while in others a clear need flow and for heat and mass transfer. Note for
appears for more detailed dynamic models.In example on Fig. 2 that there are no pumps nor
addition, the current trend toward larger and more restrictions in the argon reflux flow line into the
rapid adjustment of production rates to precisely sidearm column, the main driving force being just
match customer product demand and the implemen- the liquid head. Similarly there are no restric-
tation of unattended operation impose more severe tions in the sidearm feed line into the column. The
conditions on the argon system . This has spawned sidearm flow is therefore a direct function of the
our dynamic modelling work. condensation rate in the reboiler/condenser at the
top of the column and is effectively controlled by
changing the pressure in the boiling side of the
DYNAMIC MODELLING OF THE LOW reboiler/condenser. This highlights the high degree
PRESSURE/CRUDE ARGON COLUMN SYSTEM of interaction existing among different sections of
the system as well as the extreme need for accuracy
The main thrust of the dynamic modelling and not only in the description of temperature and
simulation work is the design of improved control concentration profiles but especially on all
and optimization strategies. Once reliable dynamic flow/pressure relationships. For these reasons, the
models are obtained and linearized approximations simultaneous solution of all differential and
derived, any of a host of advanced model-based algebraic equations is the only adequate approach
techniques can be employed for controller design to the dynamic simulation of this system (and the
purposes. The nonlinear dynamic simulator can serve preferred one for cryogenic systems in general).
as the initial test site for evaluation of the
control schemes prior to on-site implementation. Correct representation of the flow/pressure respon-
Although in many cases this approach is not needed ses in the argon column is attained by incorpora-
as loops are simple enough that they can be tuned ting full and accurate information on both the
very well directly at site, certain systems such as liquid and the vapor hydraulics. This is done
the argon column have been traditionally more employing carefully studied techniques and
difficult to address and have required many hours correlations. A high degree of preCision is also
of on-site tuning work without resulting in needed in the evaluation of all thermodynamic
strategies reliable enough to be fully accepted by properties. A proprietary, equation-of-state based,
the operators. The modelling and simulation work local thermodynamics package is employed for
are also needed for a clearer understanding of why accurate and fast evaluation of the properties of
certain phenomena occur, which can later be the nitrogen-argon-oxygen system.
translated into better process and/or controller
designs . The dynamic model of the distillation column is
formulated as a set of differential/algebraic
Along the above lines, the following phases were equations (DAE's). The basic equations describing a
identified for the dynamic modelling work in the general tray without sidestreams are presented
context of argon optimization: below as Eqs (1)-(7), written for tray k and
component j.
a) Study of the argon sidearm column by itself, to
obtain a clear understanding of its dynamic
behavior, and in particular of the responses to Liquid Hydraulics
nitrogen disturbances at the feed. The objective is
to understand the exact nature of the phenomena Pb ' p.
that may lead to a dumping situation once a bit of - (---- ----) + HEADS. ,
L
nitrogen has entered the column, i.e. the nature of g p.
the flow/pressure responses inside the argon column
and its reboiler/condenser following concentration with HEADS. = ~,(L.,geometry, ... )
changes at the boundary. To reflect dumping
behavior the model has to incorporate a larger Component Balances
amount of physical detail than sometimes needed in
distillation column modelling. Through these dw~ .•
investigations we may be lead to column designs or o
control strategies that result in improved dt
tolerance to nitrogen incursions. + WEEP. _ '~2 (x J. • - I ,X J. • - 2) + ENI'R .. I X J .•• I
- vJ .• - (L. + ENI'R.) XJ .•
b) Study of the interconnected column system, to - WEEP. ~2 (xJ.. ,xJ.. - I)
~7() .J. :'I.. \\andler. D. R. \'insoll and :-.:. Chattcrjec

Phase Equilibrium

r---~
~------,-I -- ----T- -t- -----
Vapor Hydraulics

vk

....... ·.·.·.·.·.·.·.~·.·."""·~ ... ·.· ... i....................

• •••• .•••.•• •• •.. • • . • • • • • . • • . . • • • • . • • . ! •••••..••.•••....••.•

Liquid Holdup

o = - --------- + VOL~ ,
L
Pk
with VOL~ = ~.(zk,Lk,Vk,geom., ... )

Energy Balance
dE k
o = - ----- + Vk+ l Hk+l + Lk - 1 hk - 1 (6)
dt Fig . 3. Di st illati on tray variables.
+ WEEP k - 1 ~5(hk-l,hk -2) + ENTR k.l h k • 1

- Vk Hk - (L k + ENTR k ) hk
- WEEP k ~5(h k_ l,hk) both "differential " and "algebraic " variables as
the integration proceeds. Standard ODE methods, on
Energy Holdup the other hand, require a manipulation o f the dif-
ferential equations to c ast them in the standard
o = - Ek + u~ WT~ (7) form (Gallun and Holland, 1982) and/or require the
use of separate nonlinear algebraic equation sol-
vers to solve f o r any variable whose time deriva-
The model cons ists of the fundamental componen t and tive does not appear explicitly on the left-hand
energy balances as well as algebraic equations side. This may lead to unnecessarily complex and
des c ribing the phase equilibrium relationships even obscure pieces of co de and to a lack of
(Eq, (3)), vapor hydraulics (Eq, (4)), liquid effic iency in the numerical solution. The flexibi-
hydraulics (Eq. ( 1 )), and algebrai c constraints on lity a c hieved through the DAE formulation is
the liquid h o ldup (E q . (5)) and total energy holdup particularly important when there is a need to link
(Eq. (7)). In this mod el the tray vapor holdups are to a physi c al properties code, the performance of
assumed negligible compare d to the liquid holdups. whi c h may depend on the set of variables that are
This assumptio n can be easily overriden, if specified as inputs. The DAE approach has been
necessary . The liquid and vapor leaving each tray adopted for dynami c simulation in packages such as
are assumed to be in thermal equilibrium but not in SPEEDUP (Pantelides, 1986), DIVA (Kroner and
phase equilibrium. The component balances employ colleagues, 1989) and CONSYD (Holt and others,
uniform liquid and vapor con c entrations for ea c h 1987; Mandler and Morari, 1986).
tray. This is necessary to reduce the complexity of
the problem since detailed models of ( improper) The set of solution variables for the DAE system
mixing on trays involve partial differential (2)-(7) for tray k consists of the component liquid
equations. The departure from phase equilibrium as holdups Wj,k , the component vapor flowrates Vj,k ,
well as the effects of liquid maldistribution on the pressure Pk , the total liquid flow through the
the trays or impr o per mixing are appropriately downcomer Lk, the total energy holdup in the liquid
inc o rpo rated in our calculation of the tray Ek, and the temperature Tk ' The numerical solution
efficiencies. scheme is given below.

Figure is a schematic diagram of a typi c al At time t DASSL provides to the function evaluation
distillation tray with some important variables routine the current (new) value for the solution
highlighted. The componen t and energy balanc es variables and for the time derivatives of Wj,k and
in clu de terms to a ccount explicitly for liquid Ek' Note that Eq. (1) is solved explicitly for the
entrainment and for weeping/dumping from the trays. liquid level zk in the downcomer, and therefore it
The latter term is particularly important for our is not included in the equation set for solution by
applicati on as it accounts for one of the main DASSL. The addition of the algebraic vapor hydrau-
effe c ts to be studied. The entrained flow as well li cs Eq . (4) and the total energy holdup constraint
as the weep rate are calculated as explicit Eq. (7) allow the pressure Pk and the temperature
functions of the integration variables and of the Tk to be inc luded in the set of solution variables
geometry. Note that the properties such as concen- for DASSL. This is an advantage, as suggested
trations in the weep or entrained flows may be previously, since it allows direct access to the
taken as being the same or different from those of thermodynamics package through a convenient
the main stream of liquid flowing through the down- pressure-temperature specifi c ation (p-T flash) to
comers, depending on the assumed mixing characte- obtain accurate thermodynamic properties for both
ristics on the tray . The effect of weeping and/or the liqUid and vapor phases with a minimum of
entrainment on the tray efficiencies is accounted iterations . The thermodynamic properties calculated
automatically through the explicit inclusion of at time t from the p-T flash include densities,
WEEP and ENTR in the material balances; in order to internal energies, enthalpies and the fugacity
avoid double-counting it is not incorporated into coefficients used to calculate the K values. The
the calculation of the tray efficiencies. thermodynamic and transport properties and the
geometric parameters are used to explicitly
The formulation of the problem directly as a system calculate all the remaining factors in the right-
of differential/algebraic equations (DAE) and its hand side of Eqs (1)-(7), including the different
numerical solution through efficient DAE codes such holdup, head and pressure drop terms, the tray
as DASSL (Petzold, 1982) has significantly efficiencies and the entrainment and weeping flows.
facilitated the modelling and simulation work. The Once calculated, the right-hand sides of Eqs (2)-
DAE formulation allows extreme flexibility in the (7) (the residuals of the DAE system) are passed to
specification of the set of solution variables and and used by DASSL to converge the current iteration
provides the ability to solve simultaneously for and/or advance the solution to the next time step.
 Cryogenic :\ir Separation Plants :!71

YAPOR FLOWRA TE PROFILES ARGON PROFILES

92 1.

91 0

90 0
M
0
89 0
1
88 0
F
o 87 r O.
1
, 86

85 0
t. = O.
0

82O+o----5~--~10----1+5----2r
O---2~5----3+0--~3r5---4~0~--4+5--~50 o 0~O----5~1--~16----1+15--~2~6---2~15~~3+6--~
3r5---4~0----4+~--~50

TRAY TRAY

(a) (b)

Fig. 4. Argon column simulation results - Step increase in feed flow

(below tray 50) at t=O+.

The system consists of (2*ncp+4 ) equations per tray REBOILER/CONDENSER DYNAMIC MODEL
(ncp is the number of components) and is effi-
ciently solved through the banded Jacobian option Reflux for the argon/oxygen separation in the
in DASSL. Figure 4a presents results of a test sidearm column is obtained by condensing the crude
simulation of the argon column showing the response argon stream against a boiling mixture of nitrogen
to a step increase in the feed flow and the ability and oxygen in the reboiler/condenser located at the
of the model to adequately describe both the fast top of the column. The reboiler/ c ondenser is shown
initial flow transient as well as the slower in more detail in Fig. Sa together with an
subsequent flow moves that accompany the tempe- equivalent representation for modelling purposes
rature and concentration changes. Figure 4b shows (Fig. 5b). Typical designs are based on brazed
the movement of the argon concentration profiles. aluminum plate-fin heat exchangers, with the
condensing and boiling streams flowing counter-
In this particular example a fixed top-tray currently and the boiling side recirculating
pressure, constant liquid reflux and 50 trays were through a thermosiphon effect. To achieve high
assumed. About 500 DAE's were needed to describe thermodynamic effi c iency the heat exchanger is
the co lumn. The widths of the lower and upper designed to operate under small temperature
Jacobian bands were 13 and 12. The ratio of cpu differences between the condensing and boiling
time to real time was 0.05 on a VAX 8700 before streams. Accurate modelling is therefore needed to
code optimization, for DASSL error tolerances of adequately describe the exchanger's performance.
10- 5 . Since the thermodynamics package includes a
few iterative calculations, the convergence The mathematical model for the boiling side of the
criteria within the package were adjusted to be reboiler/condenser follows the representation of
tighter than those of the integration routine to Fig. 5b and is split into the exchanger section and
prevent, as with any case of "iteration-within- a reboiler sump section. The model assumes that all
iteration ", numerical noise to be generated from accumulation effects on the boiling side occur in
inside the function evaluation routine that would the sump. The recirculation rate is calcu lated as a
prevent convergence of the DAE solver and/or would function of the static head and the vapor quality
cause an incorrect numerical evaluation of the using two-phase pressure drop relationships. The
Jacobian matrix. high recirculation rate justifies the assumption of

l
;[1
, ', --
Fl F,
.. . ..,..

.
I
, ,

; "-
, .
..;. .....
r J\ le }- R
-i. '-'
--£-----
.~
L.

(a) (b)

Fig. 5. (a) Schematic diagram of the argon column reboiler/condenser.

(b) Equivalent representation for modelling purposes.
 .J . ..'I•. ~Ialldlcr. D. R. \ ' illsoll alld :\. C:hat!erjl'l'

very fast dynamics in the boiling side of the f o r ea c h of the different reg i mes and swit c hes to
exchanger and the use of algebraic (steady-state) determine the ap p r o priate s et o f equations to be
balances for this side. The modelling of the used . Spe c ial diffic ultie s u s ed t o arise to
condensing side, on the other hand, requires a ade q uately handle tran s iti o n s fr om two -phase to
complete dynamic treatment sin c e among our studies all-liquid flow.
we need to investigate the possible accumulation of
nitrogen in the condenser and its effects on the In our approach, in a fun c ti on evaluation, the
operation of the column system. The rest of this pressure p, the internal energy u and the mole
section discusses the basis of our approach to fra c tions are obtained from the DASSL solution set.
modelling and simulation of distributed, dynamic, These are used as inputs to the thermodynamics
multiphase heat exchangers. This same approach has package accessed through a p-u (pressure-internal
been successfully applied for the dynami c energy) specification . The p-u flash calculates the
simulation of complex wound-coil heat exchangers split , if any, between the liquid and vapor phases,
used in liquefied natural gas plants. and all the thermodynami c properties, which are
then used to c al c ulate all term s o n the right-hand
The basic dynami c model for soluti o n by the side of the DAE system . The p-u spe c ifi c ation is
integration routine consists o f the comp onent, adequate for b o th compressible and in compressible
momentum and energy balanc es, plus an algebrai c fl ow. Th is is not, h owever, the c a s e for a density-
c onstraint for the total material holdu p . The internal energy spe c ifi c ati o n. A d ensity-internal
equations below apply to comp onent j at the kth e ne rgy spe c ifi c ati o n wo uld b e t he natural one t o
discretization point, with the index k increasing arise in a standard ODE f o rmulati o n of the problem .
in the direction of flow . If density and internal energy are the soluti o n
variables, small perturbati ons on the density by
Component Balances the integration pa c kage c an result, for the all-
dWJ . , liquid case, in very large c hanges in the pressure
o = - -----+ f J .' -1 - f J ., (8) c al c ulated from thermodynami cs . This is due to the
dt steepness of the pressure vs density relationship,
Holdup Constraint and it c an result in seri o us numeri c al diffi c ulties
o r even drive the s o lution to an unphysical region.
o = - WT , + HOLDUP , , These problems are avoided thr o ugh a p-u based
so lution approa c h .
with HOLDUP, = p, VOL k for homogeneous flow
Figure 6 shows simu lation results for a heat
Momentum Balance exc hanger for whi c h part of the heat load is
s uddenly cut thr o ugh a step reduc tion in the inlet
(10) flow of the condensing stream . The results show how
the vapor qualities c hange at different points
Energy Balance al ong the exchanger (as time pro c eeds the fully
condensed front moves ba c k towards the condensing
o = - WTk --- + fT . k - l(hk_I-Uk) s tream entrance). Sev e ral transitions from
dt two-phase to all-liquid fl ow are shown.
- fT . k(hk-uk) - Qk
CONDENS ING SIDE QUALITIES
For simplicity in the treatment here, the momentum
a cc umulation and momentum transport terms in t h e 1.
momentum balance have been taken as negligi b le
c ompared to the fri c tional and gravitati o nal
pressure drop terms. A similar treatment applies 0 .8
when these terms are inc luded . Equations ( 8 ) - ( 11) V
A
are derived from a discretization of the PDE's P
along length using finite differen c es, and use an 0
R 0 5

~SiOg,S
assumption of homogene o us flow (slip ratio = 1 ) .
Other discretization techniques and other assump- Q
tions for the slip ratio may be employed as well . U
A 0.4
Thermodynamic equilibrium is assumed to exist L
between the liquid and vapor phases. I
T
Y
The system (8)-(11 ) is formulated and solved as a 0 .2
DAE system. By including the algebrai c holdup
c onstraint and the momentum balance in the equati on
set for DASSL we can solve f o r the flow-pressure o.
relationships in the exchanger simultane o usly with 0 200 250 300
the holdup and internal energy dynamic changes . TIME ( 0 )
This is very important when it comes to computer
implementation of the solution algorithm . The Fig. 6. Dynami c resp o nse of a co untercurrent heat
solution variables are the component holdups Wj , exc hanger - Vap o r qualities along c ondensing side.
the total molar flow of liquid plus vapor fT, the
pressure p and the average specific internal energy
u at each discretization point. CONCLUDING REMARKS

Note that in the system (8)-(11) there is no basic The complexity, energy intensity and operational
distinction on whether the flow is all vapor, requirements of cryogeni c air separation systems
two-phase or all liquid except of course in the make the study of their dynamic behavior interes-
different pressure drop and heat transfer corre- ting, necessary and challenging, requiring state-
lations. One unique system of equations thus of-the-art modelling and computational techniques.
applies for all regimes, and transitions between This paper has presented the basis of our approach
the different regimes at any point in the exchanger to this modelling and dynamic simulation, which is
are handled straightforwardly through the same successfully meeting these challenges. The paper
formulation. In the past the fact that the flow- has also given at least a flavor for some of the
pressure relationships could not be solved simulta- special control issues that are relevant to these
neously with the dynamiC balances resulted in systems, the study of which has spawned our dynamic
intricate pieces of code with special conditions simulation work.
 Cnogellic .-\ir Separatioll Plallts

Our model formulation and numerical solution fully Subscripts/superscripts

exploit the capabilities offered by differential/
algebraic solvers. Through our approach we have j of component
been able to recast and solve in a direct manner k at kth tray; at kth discretization segment
certain simulation problems, such as dynamic phase L liquid
transitions in distributed heat exchangers, that in T total (all components)
the past required intricate pieces of code. average of liquid and vapor
Although the simultaneous solution of differential
and algebraic equations is not a new topic, our
experience indicates to us that a full awareness of REFERENCES
its benefits has not been yet realized within the
simulation community. Dawson, B.K., J.A. Mandler, A.P . Odorski, and S.F.
Salvo (1989). On-line optimizing control of an
In our recent efforts we have concentrated on the air separation plant. Proceedings of Control
development of a general dynamic simulation infra- Expo '89, Chicago, IL. May 24 1989 .
structure to allow us flexibility in setting and
accuracy in solving any flowsheet. This infrastruc- Gallun, S.E. and C.D . Holland (1982). Gear's proce-
ture employs an equation-based solution approach dure for the simultaneous solution of differen-
while preserving a complete modularity for model tial and algebraic equations with application
implementations . The end result will permit to unsteady state distillation problems.
improved computer control and optimization of Comput. and Chem. Engng, £, 231-244.
complex air separation plants as well as other
processes as operating conditions and product Holt, B.R. and others (1987). CONSYD - Integrated
specifications become more stringent. software for computer aided control system
design and analysis. Comput. and Chem. Engng,
11, 187-203.

King, D.C., R.L. Hutchison, K. J. Murphy, and A.P .

NOMENCLATURE Odorski (1987). The benefits of optimizing air
separation plant performance. Proceedings of
DAE differential/algebraic equations the Metallurgical Society of the Canadian Ins-
DPFRIC frictional pressure drop (kPa) titute of Mining and Metallurgy, Vol. 11, 1987.
DPGRAV gravitational pressure drop (kPa) Pergamon Press, pp 199-208.
DPTOT total pressure drop across a tray (m)
E energy holdup (kJ) Kroner, A., P. Holl, W. Marquardt, and E. Gilles
ENTR entrained liquid flow rate (kmol/s) (1989). DIVA - An open architecture for dynamic
f total flow of liquid + vapor in heat simulation . Proceedings of CACHI '89. European
exchanger's condensing side (kmol/s) Symposium on Computer Applications in the Che-
g acceleration of gravity (9.81 m/s2) mical Industry. Erlangen, FRG, April 1989.
GAN gaseous nitrogen
GOX gaseous oxygen Mandler, J.A. and M. Morari (1986). SNTEG: A dyna-
H specific enthalpy of vapor (kJ/kmol) mic simulation program for closed-loop systems.
h specific enthalpy of liquid (kJ/kmol) 1986 New Orleans National Meeting, A.I.Ch.E.,
HEADS sum of all liquid heads in downcomer New Orleans, LA.
backup calculation (m)
HP high pressure Pantelides, C.C. (1986). SpeedUp - Recent advances
K K-value in process simulation. Paper 4e, 1986 Annual
L total liquid flowrate through Meeting, A.I . Ch.E., Miami, FL.
downcomer (kmol/s)
LIN liquid nitrogen Petzold, L.R. (1982). A description of DASSL: A
LP low pressure differential/algebraic system solver. Sandia
MW molecular weight (kg/kmol) Tech. Rep., 82-8637.
ODE ordinary differential equations
pressure (kPa) Vinson, D.R. and N. Chatterjee (1986). Microcom-
P
PDE partial differential equations puters in process control. Proceedings of the
heat transfer rate (kJ/s) 1986 Industrial Energy Technology Conference
Q
R recirculation flowrate, reboiler (kmol/s) and Exhibit, Houston, TX, June 1986 .
t time (s)
u specific internal energy (kJ/kmol)
V total vapor flowrate from tray (kmol/s)
v component vapor flowrate (kmol/s)
VOL volume; in distillation column model,
volume of liquid on tray + downcomer; in
heat exchanger model, volume of discreti-
zation section (m 3 )
w component holdup (kmol)
WEEP liquid flow rate down through sieve holes
(weeping) (kmol/ s)
WT total molar holdup (in distillation column
model, liquid phase only) (kmol)
x component mole fraction in the liquid
y component mole fraction in the vapor
z liquid level in downcomer (m)

Greek letters

t.z length of heat exchanger discretization

segment (m)
~ ... ) explicit function of ...
11 tray efficiency
P density (kg/m 3 )