Ind. Eng. Chem. Res.

1992,31, 1679-1694 1679

T,= sampling time Hoskuldsson, A. Partial least squares regression methods. J. Che-
U = matrix derived from the SVD of X mom. 1988,2,211-228.
V = matrix derived from the SVD of X Jorgensen, S.; Goldschmidt, L.; Clement, K. A sensor-location pro-
W = projection matrix of X cedure for chemical processes. Comput. Chem. Eng. 1984,8,
195-204.
W,+ = weights used for SSV analysis described in Figure 4b; Kumar, S.; Seinfeld, J. Optimal location of measurements in tubular
1=1,2 reactors. Chem. Eng. Sci. 1978,33,1507-1516.
Wd = disturbance weight Laughlin, D.; Jordan, K.; Morari, M. Internal model control and
W, = performance weight process uncertainty: Mapping uncertainty regions for SISO con-
X = input data, each row corresponds to one input sample; troller design. Znt. J. Control 1986,44, 1675-1698.
each input sample consists of 12 temperatures and 2 ma- Lee, J.; Morari, M. Robust control of nonminimum-phase systems
nipulated variables through the use of secondary measurements: Inferential and in-
X, = lower dimensional data set obtained from the projection ferential cascade control. Automatica 1992,submitted for pub-
ofXonW lication.
Lee, J.; Morari, M. Robust measurement selection. Automatica
y = measured concentration 1991,27(3),519-527.
yn = nominal concentration defined in (28) Luyben, W. L. Parallel cascade control. Znd. Eng. Chem. Fundam.
6 j = deviation for the variable j 1973,12 (41,463-467.
A" = uncertainty matrix Mandler, J. A. Robust Control System Design for a Fixed-Bed
Ai* = individual uncertainty elements Catalytic Reactor. Ph.D. Thesis, California Institute of Tech-
A = IMC filter constant nology, 1987.
p = structured singular value (SSV) Mejdell, T.; Skogestad, S. Estimate of process outputs from multiple
Z = matrix of the singular values of X secondary measurements. Proc. Am. Control Conf. 1989,
T~ = controller integral reset time in (46)
2112-2121.
Morari, M.; Zdiriou, E. Robust Process Control; Prentice Hall:
Literature Cited Englewood Cliffs, NJ, 1989.
Tham, M. T.; Montague, G. A.; Morris, A. J.; Lant, P. A. Soft-sensors
Brosilow, C.; Joseph, B. Inferential control of processes. AZChE J. for process estimation and inferential control. J. Process Control.
1978,24,485-509. 1991,l (3).
Doyle, J. Analysis of feedback systems with structured uncertainties. Van Herwijnen, T.; Van Doesburg, H.; De Jong, W. Kinetics of the
ZEEE Roc. Part D 1982,129,242-250. methanation of carbon monoxide and carbon dioxide on a nickel
Geladi, P.; Kowalski, B. Partial least squares regression: A tutorial. catalyst. J. Catal. 1972,28,391-402.
Anal. Chim. Acta 1986,185,1-17. Webb, C. Robust Control Strategies for a Fixed Bed Chemical Re-
Gulandoust, M. T.; Morris, A. J.; Tham, M. T. Adaptive estimation actor. Ph.D. Thesis, California Institute of Technology, 1990.
algorithm for inferential control. Znd. Eng. Chem. Res. 1988,27, Webb, C.; Budman, H.; Morari, M. Identifying frequency domain
1658. uncertainty bounds for robust controller design-theory with ap-
Harris, T.; MacGregor, J.; Wright, J. Optimal sensor location with plication to a fixed-bed reactor. Proc. Am. Control Conf. 1989,
an application to a packed bed tubular reactor. AZChE J. 1980, 1528-1533.
26. Wold, S.; Ruhe, A.; Wold, H.; Dunn, W. The collinearity problem in
Hill, C. Chemical Engineering Kinetics and Reactor Design; Wiley: linear regression: The partial least squares approach to general-
New York, 1977. ized inverses. SZAM J. Sci. Stat. Comput. 1984,5(3), 753-743.
Holt, B.; Morari, M. Design of Resilient processing planta-V The
effect of deadtime on dynamic resilience. Chem. Eng. Sci. 1985, Received for review March 24, 1992
40, 1229-1237. Accepted April 13,1992

Separation System Synthesis: A Knowledge-Based Approach. 2.

Gas/Vapor Mixtures
Scott D.Barnicki and James R.Fair*
Separations Research Program, Department of Chemical Engineering, The University of Texas at Austin,
Austin, Teras 78712-1062

A description is given for a prototype knowledgebased expert system, the separation synthesis advisor
(SSAD),for synthesis of separation sequences for gas/vapor mixtures. The core of the SSAD is
the separation synthesis hierarchy (SSH),a highly structured, taak-oriented framework for repre-
senting separation knowledge. The hierarchy, based on interviews and information from the literature,
emulates the approach that an expert process engineer follows. In ita current implementation, the
SSH is limited to the preliminary sequencing of multicomponent gas/vapor mixtures using the
following separation methods: (1)physical absorption; (2) chemical absorption; (3) cryogenic dis-
tillation; (4) membrane permeation; (5) molecular sieve adsorption; (6) equilibrium-limited absorption.
Several examples of practical industrial separation problems are included.

Introduction problem for gaslvapor mixtures. The SSAD is a prelim-

This paper is the second of a series on the development inary process design tool. Ita purpose is to formulate a
of a prototype expert system for the syntheais of separation limited number of feasible separation systems for a given
sequences for fluid mixtures; the system is called the multicomponent mixture. Final comparisons and opti-
separation synthesis advisor (SSAD).Part 1 concentrates mization must be carried out with the aid of a process
on separation system synthesis for liquid mixtures (Bar- simulator, as the SSAD currently does not have the ca-
nicki and Fair, 1990). Part 2 focuses on the parallel pability to perform a detailed economic analysis.
0888-5885/92/2631-1679$03.00/00 1992 American Chemical Society
1680 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992
THE SEPARATION
SYNTHESIS HIERARCHY
Split
!wuamQQ
mass Distiilatbn
RmiQDQu
Condensation
Figure 1. Separation synethesis hierarchy, showing each manager with its complement of selectors and designers.
The separation of gas and vapor mixtures is a significant
part of many key activities in the chemical process in-
dustries, ranging from the recovery of carbon dioxide in
enhanced oil recovery to environmental concerns over the
removal of solvents and acid gases from exhaust and
process streams. In spite of its obvious importance, the
synthesis of separation sequences for gas/vapor mixtures
has been completely neglected in the process design lit-
erature. In the 23 years since the first proposals of Rudd
and Masso advocating a systematic approach to separation
system synthesis (Rudd, 1968; Masso and Rudd, 1969),not
one article has appeared on any aspects of gas/vapor
separation system synthesis.
As with liquid mixture separation synthesis, the general
gas/vapor synthesis problem involves method selection and
sequencing subproblems. However, beyond these super-
ficial similarities the specifics of the synthesis problem for
gas/vapor mixtures are fundamentally different from the
corresponding problem for liquids. Whereas liquid method
selection is clearly biased toward simple distillation, no
such dominant method exists for gases. Several methods
can often compete favorably. Moreover, the appropri-
ateness of a given method depends to a large extent on
specific process requirements, such as the quantity and
extent of the desired separation. The situation contrasts
markedly with liquid mixtures in which the chemical
characteristicsof the components to be separated are often
the dominant factors (Barnicki and Fair, 1990).
This paper addresses the complexities of separation
method selection and sequencing for gas/vapor mixtures.
The design experts knowledge is organized as an auton-
omous problem-solving entity, called the gas split manager
(GSM). The purpose of the GSM is to provide control over
the overall synthesis activity. The GSM is further sub-
divided into three essentially independent subproblems,
referred to as separation method selectors. Each selector
pertains to a distinct aspect of the gas mixture separation
method problem.
The GSM and its complement of selectors are part of
a larger, highly structured framework for representing
separation knowledge, the separation synthesis hierarchy
(SSH)(Figure 1). The hierarchy emulates the approach
that an expert process engineer follows. The overall sep-
siwa9mm
Azeototropic Bulk, Sharp
Zeotropic Enrichment
Purification
l2wiamM
Simpb Distillation
Azeotrapic Disliilatlon Phyaicai Abrorptbn
Extractive Distillation Chemical Absorption
Liquid-Liquid Equiibuhn-Limited
Earaction Adsorption
Equilibrium-Limited Mokcular Sbvo
Adsorption Adsorption
Mokcular Sieve Cryogenk DlSlltrtbn
Adsorption Membrana Permeation
Melt Cystalliutbn Condensation
Siripping Catalytic Comembn
Membrane Penneation
aration problem for fluid mixtures can be divided into four
distinct synthesis phases composed of unique selection,
sequencing, and design subproblems, but all with parallel
problem-solving structure. The four managers which
comprise the SSH are the phase split manager (PSM), the
distillation split manager (DSM), the liquid split manager
(LSM), and the gas split manager (GSM). Each manager,
selector, and designer represents a clearly defined and
essentially independent subtask of the overall separation
synthesis activity. The first three of these (the PSM,
DSM,and LSM) have been described previously (Barnicki
and Fair, 1990). This paper concentrates on the manager
and selector subtasks for gas/vapor mixtures.
The paper is organized into three sections. The first
summarizes briefly the structure of the manager subtasks,
concentrating on the particulars of the gas split manager.
The basic task-oriented problem-solving philosophy of the
SSH and the SSAD has been described in part 1 of this
series of papers; it will be outlined only briefly here to
orient the reader. The second section presents the con-
cepts involved in separation method selection for gas/va-
por mixtures. The process attributes which can be used
to categorize gas/vapor separations are described, together
with a discussion of the conditions under which the in-
dividual separation methods are feasible. Emphasis is
placed on identifying the component properties and pro-
cess attributes which determine the utility of a particular
technique. The final section presents several industrially
significant gas/vapor separation examples which illustrate
the capabilities of the SSH.
T h e Gas Split Manager
Structure of the Gas Split Manager (GSM). A
schematic overview of the problem-solving strategy em-
ployed by the SSH is shown in Figure 2. An initial sep-
aration problem (Le., a fluid mixture) is formulated from
the problem specifications pertaining to the feed compo-
nents and desired products. This first mixture is placed
on an agenda, which is initially empty. The agenda is
expanded as separations of the initial and succeeding
submixtures are specified. Only those submixtures that
require further separation (Le., those that do not match
a desired product directly) are added to the agenda.

Pmbkm spsclflution:
creme inltbl midun.. r w
- nda
I I
top 01 the agonda
NO
Manager rslectlon
t
Split gonemion
+ heuristics lo determln
SSH
Split sequencing
Seprratlon design
Figure 2. Overview of the problem-solving strategy of the separa-
tion synthesis hierarchy.
For each mixture accessed from the agenda, one must
select the appropriate manager. This choice is trivialized
here, as all example separations described here require only
the GSM. However, more complicated examples requiring
manager selection are given in Barnicki and Fair (1990)
and Barnicki (1991). A manager oversees five problem-
solving activities:
1. Split generation: Possible separation points (splits)
are identified. The possible splits of a given mixture de-
pend on product specifications and on the order of the
components in ranked property lists.
2. Selector Analysis: The appropriate separation
methods (if any) for each possible split are determined.
A potential split is defined as a split which may be ac-
complished by at least one separation method.
3. Split sequencing: The potential splits are compared
to determine which are the most appropriate to perform
next. This analysis is guided by well-known design heu-
ristics.
4. Separation design: Each potential split chosen in
step 3 is subjected to a shortcut separation design proce-
dure to find the distribution of components in the resulting
submixtures.
5. Submixture analysis: The submixtures generated
by each separation must be analyzed to determine if they
meet product specifications. Those that require further
separation are added to the agenda.
The procedure is repeated for each mixture on the
agenda and continues until the agenda is empty (i.e., all
product specifications are met).
Ind. Eng. Chem. Res., Vol. 31, No. 7,1992 1681
GAS MIXTURE
Generate ranked lists for
a11 pertinent propertks
IdeMlfy possible splits In ranked lists
appropriate#elector
1-(
(-)
Split selector Spilt selector
(-) Split selector
I 1 * * *
CONSULT
11 rpproprhte
Analyze all
submixtures
ANALYSIS COMPLETE. PROCEED
TO NEXT MIXTURE ON AGENDA
j
Figure 3. Logic diagram for the gas split manager.
The gas split manager (GSM) is the only split manager
of the SSH devoted exclusively to predominantly gaseous
mixtures. No clearly dominant separation method exists
for gas separations. However, the problem-solving ap-
proach of the GSM still follows the general strategy out-
lined above. The logic diagram for the GSM is presented
in Figure 3.
Split Generation. Every separation method is based
on a difference between one or more physical or chemical
properties of the components in a mixture (King, 1980,
Rudd et al., 1973). A given method can achieve a sepa-
ration between any components in which these charac-
teristic properties differ significantly. Since the term
differ significantly is difficult to quantify, one could
conceivably separate between any two components of a
mixture. We will call these two components the keys of
the separation. For a large multicomponent mixture the
number of possible pairs of keys clearly precludes a de-
tailed examination of all options. A workable approach
requires a compromise between thoroughness and the need
to eliminate infeasible splits with minimal effort.
The strategy adopted in the SSH utilizes ranked prop-
erty lists. A ranked property list orders the components
by the magnitude of the characteristic property of the
separation method (e.g., relative volatility for cryogenic
distillation). Thus, a ranked list determines, in a quali-
tative sense, the possible distribution of components for
a given separation method. The possible separation points
are further restricted by product specifications. Splits
between components appearing in the same product are
prohibited.
Not all separation methods (especially those requiring
mass separating agents, MSA) are well-characterized by
a single, easily-calculable property. For example, once the
mass separation agent is known, the distribution of com-
ponents can be determined readily, and from this the
possible splits. However, the selection of the MSA, which
1682 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992
Table I. Separation Methods and Their Corresponding
Characteristic Properties
seDaration method properties
cryogenic distillation relative volatility
physical absorption chemical family
chemical absorption chemical family
catalytic conversion chemical family
membrane permeation critical temperature,
van der Waals volume
molecular sieve adsorption kinetic diameter
equilibrium-limited adsorption equilibrium loading
condensation relative volatility
itself is nontrivial, is entirely dependent on the separation
method chosen. One cannot determine definitely the
proper separation method until the possible separation
points are known.
In these situations (i.e., for essentially all processes ex-
cept distillation), ranked property lists are still used to
generate the possible splits. Although these characteri-
zations are not perfect, they are generally accurate enough
in a qualitative sense for a preliminary analysis. The
selection of the appropriate property (or properties) which
can reliably predict the component distribution for a given
separation method is open to some debate. Table I
presents a list of the gas/vapor separation methods cur-
rently implemented in the SSH along with their corre-
sponding characteristic property or properties.
Split Sequencing. Separation sequencing is a critical
step in the generation of optimal to near-optimal separa-
tion system designs. Limited aspects of the topic have
received serious consideration in the literature. This effort
has resulted in some relatively simple, reliable heuristic
methods of generating near-optimal distillation sequences
(Kelly, 1987; Liu, 1987; Nishida et al., 1981). These
methods generally make use of a series of ranked heuristics,
which are applied sequentially. If a heuristic is not ap-
plicable, the next one on the list is considered.
For these techniques, sequencing is based primarily on
process characteristics, such as the relative magnitudes of
the desired products and the ratio of the expected distillate
to bottoms flow rates. Since simple distillation typically
haa been the only separation method dealt with in previous
studies, relative volatility is an adequate measure of how
easy it is to accomplish a particular separation method.
Thus, there is no need to consider separation method
characteristics as well as process characteristics; the se-
quencing and separation method selection problems be-
come decoupled.
When a variety of separation methods are available,
however, one must now compare the relative ease of sep-
aration of competitive techniques. For example, if mo-
lecular sieve adsorption and cryogenic distillation processes
yield similar product distributions (and would thus trigger
the same process characteristic heuristics), how does one
compare the relative volatility and the difference in kinetic
diameters to determine which separation method results
in a more efficient separation?
The sequencing of gas separations does not lend itself
as readily to the highly qualitative approach used for
distillation sequencing. However, several of the general
separation sequencing heuristics formalized in previous
studies are still applicable to gas separations, albeit in
rather weak forms. Table I1 shows the sequencing heu-
ristics as modified for gas separations.
The presence of corrosive or hazardous materials tends
to increase the expense of equipment. Therefore these
Components should be removed as early as practical. Many
gas-phase separation processes are affected adversely by
the presence of trace impurities. Small amounts of
Table 11. Sequencing Heuristics for the Gas Split Manager
1. Remove corrosive and hazardow materials first.
2. Remove troublesome trace impurities first.
3. Favor separations which match the desired products
directly. If a separation resulta in a substream which
requires no further separation, and is a desired product,
and if that product is the most plentiful in the mixture,
remove it next.
4. Favor separations which give equimolar splits. When ease
of separation and compositions are similar, perform the
separation which divides the feed as equally as possible.
Table 111. Typical Special Processing Conditions for the
Gas Split Manager
1. Favor condensation for the removal of high boilers from
noncondensable gases when cooling water can be used as
the condensing medium. Condensation is one of the
simplest and cheapest unit operations.
2. Favor catalytic conversion when the impurities can be
converted into 4 desired product. Further purification
and/or separation steps may be unnecessary.
3. Favor adsorption for small-scale desiccation operations.
Solid-phase desiccant systems are relatively simple to
design and operate. They are generally the lowest cost
alternative for processing small quantities of gas.
4. Favor adsorption for processes which require essentially
complete removal of water vapor. Adsorptive dehydration
is capable of achieving dew point depressions of 80 O F or
more.
5. Favor glycol absorption for large-scale desiccation operations
required dew point depressions of 50 O F or less. The
initial and operating costa of high-volume glycol absorbers
are typically much lower for small to medium dew point
depressions than the corresponding costa of solid-phase
desiccation.
freezable components (water, carbon dioxide) may foul
cryogenic units. High gas humidity or moisture content
reduces the effectiveness of many adsorption processes
(particularly the adsorption of low molecular weight com-
pounds). Other components, high boilers and polymeriz-
able compounds, may permanently foul an adsorbent.
Such components should be removed first, downstream of
any important, larger-scale separations.
It is obviously advantageous to perform a separation
which removes a component directly as a product. This
generally will improve overall recovery and purity. The
specification of equimolar splits tends to reduce the
downstream separation load more effectively. In terms of
energy usage, two smaller separation units are generally
more efficient than one very large unit.
Some processing conditions are especially favorable for
certain separation methods. These special circumstances
may override the more general sequencing and selection
heuristics. A list of some typical special processing con-
ditions is given in Table III. See also the sections Physical
Absorption, Adsorption, and Condensation.
The sequencing procedure presented above does not
guarantee that only one potential separation will be found
in all cases. Because of their highly qualitative nature, the
heuristics often cannot differentiate between several al-
ternatives. When such a situation arises, all alternatives
are considered to be equally feasible at this stage of the
process design; a detailed economic analysis is often nec-
essary to determine which method is actually preferred.
The qualitatively equivalent separations are propagated
through the remainder of the manager activity and lead
to unique submixtures. If such branching does occur, the
final output of the gas split manager will contain several
competing separation system designs. On the other hand,
if no equivalent separations are found at any stage of the

uimlm
PHYSCAL ABSORPTION
No
EfinUMW
hlnow fouling
conponrmr err(
Ad8O~l.fOuing
c#rpornn(r p v n l 7
Rwnow f#lltng
conpomnla err(
*d.orb.nl.lorling
cOnlpommtmpvn17
EfifdMW!
CONDEHSAllON
Figure 4. Logic diagram for the selection of separation methods for gas-phase enrichment separations, enrichment split selector.
synthesis process, only one final separation system design
will result.
Separation Method Selection for Gas Mixtures
SeparationSpes. Gas-phase separations are classified
into three categories based on the purity, recovery, and
magnitude of the pertinent Separation. Each category is
the basis for a dietinct separation selector in the SSH (1)
enrichment separation; (2) sharp separation; (3)purifica-
tion separation.
The classification system allows for a certain amount of
synergy, as several separation methods may be combined
in order to achieve the desired result. Each separation
category is organized as a distinct selector, with its own
favored separation methods. The applicable separation
techniques for gas mixtures are shown in Table I.
A. Enrichment Separations. An enrichment is de-
fined as a separation process that results in the increase
Ind. Eng. Chem. Res., Vol. 31, No. 7,1992 1683
YEUBRANE PERLlEITlON
z
Elifdm
MOLECULAR SIEVE
bN (h C O n l p W W I l 8
of oimilr aim/- 7
Efimim
EOUIUBRIUU-UYTED
ADSORPTION
EWIUBMUY
--C
rrcuvlry Mol 7
in concentration of one or more species in one of the
product streams and the depletion of the same species in
the other product stream. Neither high purity nor high
recovery of any components is achieved.
Because of a lack of stringent purity and recovery
specifications, enrichments are the most general gas sep-
aration type. They can be accomplished with a wide va-
riety of separation methods: physical absorption, molec-
ular sieve adsorption, equilibrium adsorption, cryogenic
distillation, condeneation, and membrane permeation. The
logic diagram for the enrichment split selector (ESS) is
shown in Figure 4.
B. Sharp Separations. A sharp separation results in
two high-purity, high-recovery product streams. No re-
strictions are placed on the mole fraction(s) of the com-
ponent(s) to be separated. A separation is considered to
be sharp in the present work when the key component
1684 Ind. Eng. Chem. Res., Vol. 31,No. 7, 1992
mimi-
PMVSlCAL ABSORPTION
\ ADSORPTION
splita are greater than 9.0-9.5 or less than 0.111-0.105. A
key component split is defined as:
Figure 5. Logic diagram for the selection of separation methods for gas-phase sharp separations, sharp split selector.
Skey = c1/c2 1 9.0-9.5 for c1 > c2
Skey cl/c2 5 0.105-0.111 for c2 > c1
where Shy= split of light or heavy key, c1 = flow rate of
key component in product 1, and c2 = flow rate of key
component in product 2.
The separation methods that can potentially obtain a
sharp separation in a single step are physical absorption,
molecular sieve adsorption, equilibrium adsorption (for
componenta which comprise less than 10% of feed mix-
ture), and cryogenic distillation. The sharp split selector
(SSS)is illustrated in Figure 5.
Chemical absorption is often used to achieve sharp
separations, but is generally limited to situations in which
the componenta to be removed are present in low con-
centrations. These special cases of low mole fraction,
high-recovery, high-purity separations are treated as a
distinct separation type, purification separation.
C. Purification Separations. A purification separa-
tion involves the removal of one or more low concentration
impurities from a feed stream. A low concentration im-
purity is arbitrarily defined here as a component or group
CaJbr PHVS ABS.
+ CHEYCAL ABS
n
4 VES
EuIdma
EQUIUBRUYUYTED
ADSORPTION
i
of componenta which comprise less than 2 mol 9% of the
parent mixture. A purification separation typically resulta
in a product of very high purity (e.g., >99% impurity
(1) removal, depending on the separation method used). It
(2) may or may not be desirable to recover the impurities.
The separation methods considered here as applicable
to purifications are limited to equilibrium adsorption,
molecular sieve adsorption, chemical absorption, and
catalytic conversion. Physical absorption is not included
in this list. With low inlet concentrations of the impurity
(characteristic of a purification separation), physical ab-
sorption processes are typically not able to achieve ex-
tremely high purities (Tennyson and Schaaf, 1977). One
notable exception to this rule is the absorption of water
vapor by glycol solutions. Glycol dehydration processes
are able to achieve dew point depreasiom of 200 OF or more
(Valerius, 1974). However, adsorptive desiccation is gen-
erally more economical when dew point depreasions of 80
OF or more are neceesary (Kohland Riesenfield, 1986)
(refer also to sections Physical Absorption and Adsorp-
tion).
In some cases an enrichment may be coupled with a
purification step in order to achieve the desired separation
sharpness (e.g., physical absorption followed by chemical
absorption, condensation followed by a purification op-

Figure 6. Logic diagram for the selection of separation methods for gas-phase purification separations, purification split selector.
eration). A logic diagram for the purification split selector
(PSS) appears in Figure 6.
Separation Methods. A. Membrane Permeation.
The ease of separation of two gaseous components by
membrane permeation is characterized by the ratio of their
permeabilities in the membrane material. This permse-
lectivity is often represented in the literature as consisting
of solubility and diffusivity contributions:
The ability of a polymer membrane to act as a selective
separatingagent for a particular mixture of gaseous species
is a function of the physical properties of the polymer as
well as those of the components to be separated. The
magnitude of the diffusivity ratio is dependent on the size
and shape of the molecules, while the solubility ratio is an
expression of their relative condensability (Koros and
Hellums, 1989).
In general, an aij* 2 15 is required for a membrane
permeation process to be commercially feasible (Hogsett
and Mazur, 1983). Moreover, permeate purity is rela-
tively unaffected by an aij* > 20 (Stookey et al., 1986).
Permeability data are readily available for many com-
mon gaseous systems, such as C02/CH4and 02/N2(Koros
et al., 1988; Walker and Koros, 1991; Teplyakov and
Meares, 1990). However, when experimental data are
nonexistent or unavailable, a preliminary screening of
potential membrane processes is possible by examining the
combination of component properties which would result
in a permselectivity of 15 or greater. Barnicki (1991)
presents a generalized method for predicting whether a
favorable permselectivity (e.g., aij* 2 15) can be obtained
for a given gaseous separation using any of 51 glassy or
rubbery polymers. One needs only a knowledge of a dif-
fusion-related property (effective kinetic diameter or van
Ind. Eng. Chem. Res., Vol. 31, NO. 7, 1992 1685
der Waals volume) and a solubility-related parameter
(effective Lennard-Jones well depth or critical tempera-
ture) of the components in question.
B. Catalytic Conversion. Catalytic conversion is not
a separation method in the conventional sense. Impurities
or other objectionable Components are not removed, but
rather chemically transformed on the surface of a solid
catalyst into less objectionable species. The new compo-
nent(s) may then require further separation.
Because of its destructive nature, catalytic conversion
can be eliminated from further consideration i f the im-
purities are a desired product.
Catalytic conversion is especially favorable for separa-
tions in which extremely high purity is needed. In fact,
almost complete removal of the objectionable components
is possible (down to 1-10 ppm). Conversion is best for a
stream with a low concentration of impurities (less than
about 5000 ppm), for high temperature, and for low
pressure, e.g., flue gases and purges (Kohl and Riesenfeld,
1985; McInnea et al., 1990). The removal of small amounta
of the leas volatile components (i.e., liquid-type compounds
is also a favored situation. The utility of catalytic con-
version hinges on a difference in reactivity of impurities
and bulk stream components. Industrially significant
purification conversions can be classified as either com-
bustion or hydrogenation reactions. Other conversion
methods specific to particular compounds, (e.g., the cata-
lytic reduction of nitrogen oxides with ammonia) are not
considered in this work.
Combustion (an oxidation-reduction reaction) entails
the addition of oxygen and heat, often over a precious
metal catalyst, to yield water, carbon dioxide, and some-
times sulfur dioxide, depending on the composition of the
impurities:
-
impurities + O2 + heat H 2 0 + N2 + COP+ SO2
Typical industrial applications of this technique involve
1686 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992
Table IV. Hydrogenation Reactivity
reacting chemical family" hydrogenation products
alkynes olefins
olefins paraffins
acyl fluorides aldehydes, hydrogen fluoride
acyl chlorides aldehydes, hydrogen chloride
nitriles-aromatic nitros-oximes primary, secondary amines
aliphatic nitros primary, secondary amines
aldehydes alcohols
ketones alcohols
aromatic rings cycloalkanes
heterooxygen aromatics cyclic ethers, n-alkanes
heteronitrogen aromatics nitrogen cyclics
anhydrides polyesters
esters alcohols, acids, ethers
organic acids alcohols
amides primary amines
thiole-sulfides hydrogen sulfide, hydrocarbons
heterosulfur aromatics hydrogen sulfide, n-alkanes
a Ease of hydrogenation decreases from top to bottom of table.
the removal of species that are environmentally objec-
tionable or detrimental to downstream processes, (e.g.,
hydrogen, elemental sulfur, hydrogen sulfide, and organ-
ics). Combustion is not recommended for halogenated
organics, as the reaction products are generally as objec-
tionable as the original compounds.
For safety reasons the concentration of the impurities
should be no more than 10% of their lower explosive limit.
The selectivity of the combustion reaction is generally
poor; almost all organic compounds will catalytically ox-
idize. Close control of the reaction temperature generally
improves selectivity, but the most favorable temperature
for oxidation is difficult to predict. Reaction temperatures
are usually between 250 and 700 "C, assuming an impurity
composition of approximately 10% of the lower explosive
limit (Suter, 1955).
From the standpoint of preliminary process analysis
catalytic combustion is feasible for purification processes
only when the impurities are at concentration levels below
10% of the lower flammability limit and when the bulk
stream already consists of oxidation products, e.g., air
stream, off-gases,and other inerts. In addition, catalytic
oxidation should not be used when the process stream
contains halogenated organics.
The hydrogenation reaction involves the addition of
hydrogen to specific functional groups. Conversions of
impurities H, -
9599% are typical for the reaction
+ addition products
Hydrogenation requires much milder conditions than
combustion, typically at temperatures lower than 100 "C.
High selectivity is possible by a controlled addition of
hydrogen, depending on the functional groups present.
The order of reactivity of various functional groups as well
as their hydrogenation products is listed in Table IV
(Rylander, 1985; Streitwieser and Heathcock, 1981).
Groups higher in the table hydrogenate more easily than
those lower down.
Catalytic hydrogenation is a feasible purification op-
eration only when the impurities contain functional
groups listed in Table IV. Moreover, the reactivity of the
functional groups in the impurities must be higher than
that of the bulk stream species, i f the bulk stream is to
be unaltered by hydrogenation. Hydrogenation is espe-
cially favorable for processes in which the impurities can
be converted into desired products.
The conversion of acetylene to ethylene during the
production of ethylene is an excellent industrial example
of the use of hydrogenation for product purification
(Reitmeier and Fleming, 1958).
C. Physical Absorption. Physical absorption is
characterizsd by specific nonchemical interactions between
the absorbent liquid and the solute gas. These interactions
are t y p i d y a linear function of the solute partial pressure
in the gas phase and the solute concentration in the liquid
phase. Consequently, a physical solvent maintains its
absorptive properties even when the partial pressure of the
solute in the feed is high (England, 1986). This contrasts
markedly to a chemical solvent which typically loses its
effectiveness as the solubility limit of the solute is ap-
proached. However, unless the solute-solvent solubility
is extremely large, the product stream concentration gen-
erally cannot be reduced much below 100 ppm with a
physical solvent (Tennyson and Schaaf, 1977). Thus the
best applications of physical absorption involve sharp and
enrichment separations.
One exception to this rule is the widespread use of glycol
absorption for the dehydration of natural gas and other
process streams. For large-scale operations with dew point
depression requirements of 50 O F or less, glycol absorption
is generally the most economical alternative. When dew
point depressions of 50-80 O F are necessary, glycol ab-
sorption and adsorption are competitivetechnologies (Kohl
and Riesenfeld, 1985).
Selective physical absorption is based on a difference
in solubility resulting from the intermolecular forces be-
tween the gaseous solutes and the absorptive liquid.
Fundamental intermolecular force calculations, involving
the species' dipole moments and polarizabilities (Kaliszan,
1987), are not accurate enough to be even qualitative in-
dicators of the feasibility of physical absorption. There-
fore, one is forced to turn to bulk thermodynamic mea-
surements of the solubility selectivity.
The selectivity exhibited by a particular absorbent can
be expressed in terms of the ratio of the liquid-phase mole
fractions of two gaseous solutes in the liquid solvent.
(4)
For purely physical absorption at low to moderate
pressures, gas-phase solute-solute interactions are gener-
ally small and tend to cancel (Le., = 4).
Moreover, as
a first approximation the activity coefficient ratio can be
replaced with infinite dilution values. Upon substitution
eq 4 becomes
(5)
The standard-state liquid fugacity of a component can
be determined by any of the commonly used methods
listed below:
1. Extrapolation of the vapor pressure curve to a hy-
pothetical liquid state. This is the simplest approach, but
it can be extremely unreliable for temperatures much
above the critical temperature (Prausnitz et al., 1986).
2. Use of semiempirical fugacity correlations. Praus-
nitz and Shair (1961)and Yen and McKetta (1962) present
correlations for nonpolar and polar solvent systems. Ac-
curacy is varied.
3. Use of the 'ideal" solubility concept (Gjalbaek, 1952,
England, 1986). The expression for ideal solubility is
derived form the Clausius-Clapeyron equation and
Raoult's law. Results are best for simple non-polar gases
well above their critical temperatures.
4. Use of Henry's constants. Sander et al. (1983) de-
scribe the use of a modified version of the UNIFAC group
contribution method for calculating Henry's constants.

Table V. Estimated H2S/C02 Selectivities
method Selectivity
vapor pressure extrapolation 6.56
Shair correlation 4.15
ideal solubility 2.59
Henrys constants 4.87
Only the activity coefficients in eq 5 are solvent-de-
pendent; the standard-state liquid fugacities depend only
on the properties of the individual solutes. For regular
solutions, the infinite dilution activity coefficient is given
by the Scatchard-Hildebrand-Flory-Huggins equation as
(Walas, 1985)
Vi vi vi
In T~~= -(a,
RT
+
- (si)2 In - - - + 1
v, v, (6)
Note that eq 6 is a function only of the properties of
component i and the solvent, not those of component j .
This is consistent with the assumption of negligible so-
lute-solute interactions. Substitution of eq 6 into eq 5
yields
v: (V:- V:)
I
(7)
At the preliminary process design stage, the choice of
the best solvent cannot be known a priori (Barnicki,
1991). In general, a desirable solvent for selective physical
absorption will form an ideal solution with some of the
solutes and not the others. AB a rough approximation, one
can assume that this yet unknown solvent will have
properties very similar to one or more of the solutes (and
therefore form an ideal solution with these species). For
the binary case of components i and j in solvent s, if the
solvent is similar to component i , then 6, = (si and V , = Vi.
Component i will be preferentially absorbed:
f.01.* Vj vj - -
vj
In = In -
I + -((si - + In -
vi vi + 1 (8)
PI .* RT
f.01
1
One can now estimate the selectivity of a potential ab-
sorption separaton solely from the properties of the two
competing solutes. The selectivity calculated from eq 8
tends to be within 20-40% of experimental values, de-
pending on the method chosen for calculating the stand-
ard-state liquid fugacity and on how different the chosen
solvent is from the solutes. The selectivity achievable with
a physical solvent that is not an exact analog of one of the
solutes will rarely exceed a value of 10 and is generally in
the range of 3-8 (Astarita et al., 1983).
Taking into consideration the inaccuracies of the above
analysis one can formulate a general heuristic on the fea-
sibility of physical absorption for a given separation:
If the selectivity calculated from eq 8 is 3 or greater for
an enrichment process or 4 or greater for a sharp sepa-
ration, then physical absorption should be considered as
a feasible separation method.
The utility of eq 8 is illustrated by predicting the se-
lectivity for a mixture of hydrogen sulfide and carbon
dioxide, common industrial gas components. The selec-
tivities obtained by each of the four methods of correlating
the standard-state liquid fugacity are shown in Table V.
The solvent is assumed to be similar to H2S,and the partial
pressures of the gases are assumed to be equal. The
magnitudes of the estimated selectivities indicate that
physical absorption is a feasible separation option for these
two gases. The ideal solubility method, however, does not
Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992 1687
Table VI. Acid-Base Functional Groups and Molecules
basic ~TOUDS acid groups
ammonia carbon dioxide
amines sulfur dioxide
water hydrogen sulfide
alcohols thiols
aromatic amines hydrogen bond donors
heteronitrogen aromatics (see Kaliszan, 1987)
thiols
indicate a favorable selectivity. This method is probably
inappropriate, as the system temperature is close to the
critical temperatures of both components. Astarita et al.
(1983) report an experimental H2S/C02 selectivity in
methanol of 5.50 for the commercial Rectisol process.
D. Chemical Absorption. Chemical absorption is
characterized by nonlinear interactions that are particu-
larly strong at low concentrations or partial pressures.
These interactions tend to weaken considerably as one
approaches the solubility limit of the solute; the solvent
loses ita absorptive properties. In general, chemical ab-
sorption is favored when the partial pressure in the feed
of the components to be removed is low and when the
desired removal is high (purities at the p p m level are not
uncommon) (Astarita et al., 1983; Tennyson and Schaaf,
1977).
Although results have been published for selected sys-
tems (Astarita et al., 1983; Kohl and Riesenfeld, 1985), a
generalized predictive method for chemical absorption
equilibrium is not currently available. Without selectivity
information, determining the feasibility of chemical ab-
sorption is difficult but not hopeless. Chemical absorption
often involves the complexing of the acid-base functional
groups of the solvent and solute. Table VI lists common
acid-base functional groups (Ho, 1977). Note that only
a limited number of functional groups exhibit acid-base
behavior. Thus i f the species to be separated contain
different acid-base functional groups (or i f one contains
neither),then chemical absorption (based on an acid-base
reaction) may be a feasible alternative.
The above rule is a crude indication of potential utility
only; it does not categorically ensure that an appropriate
chemical solvent can be found.
E. Cryogenic Distillation. The feasibility of a cryo-
genic distillation can be determined from the relative
volatility, a,of the key components in much the same way
as high-temperature distillation. The relative volatilities
of condensed gaseous systems tend to be larger than those
of liquid systems because of the wide boiling point ranges
of the gases normally encountered. For typical industrial
applications 2.0 I CY I 5.0 (Timmerhaus and Flynn, 1989),
and in general cryogenic distillation can be considered
as a feasible bulk separation alternative when CY 12.0.
Although comparatively high relative volatilities are
common for cryogenic distillation separations, one cannot
categorically state that such a process will be the clearly
favored separation method as is the case for high-tem-
perature distillation (see Barnicki and Fair (1990)). The
economics of a cryogenic separation are dominated by the
scale of the process as well as the thermodynamics.
Cryogenic distillation is rarely cost-efficient for small-scale
separations or purification operations which produce less
than 10-20 tons/day of product gas. For example, energy
consumption for air separations drops from approximately
500 kW-h/ton of gas to less than 300 kW-h/ton of gas as
the process scale increases from 10 tons/day to 100
tons/day (Springmann, 1985).
Cryogenic distillation is feasible only for bulk, sharp,
or enrichment separations involving high throughput.
1688 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992
Moreover, when cryogenic distillation is considered as an
alternative, one must ensure that components with high
melting points are removed before the distillation is
carried out (i.e,,species that may freeze at processing
conditions).
Any solids formed may foul reboilers, condensers, and
other piping. The nitrogen-oxygen distillation of air is a
good example of a separation in which freezable com-
pounds cause problems. The inlet air typically contains
carbon dioxide and water that freeze at the temperature
and pressure at which nitrogen and oxygen liquify. Isalski
(1989) lists other freezable impurities that are commonly
present in cryogenic plant feed gases.
F. Adsorption. F.1. Adsorbent Fouling and
Chemical Damage. The ultimate lifetime and capacity
of an adsorption bed depends to a large extent on the types
of components that are processed. High-boiling organics
(those with normal boiling points above 150-180 C) tend
to be preferentially adsorbed and are extremely difficult
to remove during the regeneration cycle. Under favorable
conditions, low molecular weight organics may polymerize
on the surface of the adsorbent. Dialkenes, 1-alkenes,
alkynes, and epoxides are especially susceptible to this
behavior.
Highly acidic or alkaline moieties may also cause per-
manent chemical alterations in the adsorbent. Aluminas
are sensitive to acid solutions, while silica gels are strongly
attacked by alkalies and hydrogen fluoride. Zeolites are
generally resistant to chemical attack when the pH is kept
in the range of 5-12 (Ullmanns, 1988).
When possible, adsorbent-fouling and adsorbent-dam-
aging components should be removed upstream of the
adsorber inlet.
F.2. Molecular sieve Adsorption. The effect of dif-
ferences in adsorbate molecular structure and size on se-
lectivity can be eapecially dramatic when using zeolites and
carbon molecular sieves. Certain sizes and shapes of
molecules may be excluded completely from the micro-
pores of the adsorbent due to the extremely narrow dis-
tribution of pore sizes. A number of industrially important
vapor-phase (and liquid-phase) adsorptive separations are
based on this molecular sieving effect, notably Union
Carbides IsoSiv processes (Cusher, 1986) and certain
Sorbex processes of UOP (Mowry, 1986; Johnson and
Kabza, 1990).
The molecular dimension of importance in sieve-baaed
adsorption processes is the minimum kinetic diameter. It
is a combined measure of the cross-sectional area and
shape characteristics of a molecule (see Barnicki (1991)for
methods of estimating kinetic diameter). Commercially
available molecular sieves fall into five distinct categories
according to their nominal aperture sizes (Le., pore size
distribution). Thus gaseous species can only be separated
by molecular sieving effects when their kinetic diameters
fall into different zeolite aperture size categories.
Table VI1 presents the nominal aperture size and cor-
responding zeolite types for each category. This classifi-
cation system was developed by Barrer (1959) and is re-
peated in modified form in many other references (e.g.,
Collins, 1968; Yang, 1987; Kovach, 1988). There is con-
siderable disagreement in the literature on the subject of
kinetic diameters of gas molecules. Breck (1974) presents
one set of values, whereas several other authors report
considerably different figures (Barrer and Brook, 1959;
Collins,1968,Ullmanns, 1988). Barnicki (1991)describes
methods of estimating kinetic diameters for limited classes
of compounds when no experimental data are available.
These estimates are consistent with the results of Barrer
Table VII. Aperture Size Categories for Major
Commercial Zeolites
category nomind aperture size (A) zeolite type
5 3 3A Linde
3A Davison
4 4 4A Linde
4A Davison
3 5 5A Linde
5A Daviaon
2 a 1OX Linde
1 10 13X Linde
13X Daviaon
and Brook (1959). In spite of the inconsistencies in the
reported values of kinetic diameters of individual mole-
cules, there is general agreement on which molecules are
excluded from the pores of a given zeolite type.
Recent advances in the understanding of zeolite mor-
phology have enabled the fabrication of molecular sieves
with aperture sizes tailor-made for a specific separation
application (Vaughan, 1988; Ruthven, 1988). However, the
use of custom-made sieves adds considerably to the cost
of the adsorption process and is not considered as an op-
tion here.
Molecular sieves are extremely effective desiccants be-
cause of their highly polar surface environment. Because
of this high affinity for water, molecular sieve drying
processes can achieve essentially complete dehumidifica-
tion of gas streams. Dew point depressions of 80 O F or
more are readily obtainable (Kohl and Risenfeld, 1985).
If water vapor is present in a gas stream, it typically will
be the most strongly adsorbed species. Thus i f the ob-
jective is to recover adsorbed components which are free
of water vapor, then the inlet gas stream should be dried
before the molecular sieve adsorption process occurs.
F.3. Equilibrium-LimitedAdsorption. As stated in
the section Separation Types, the primary uses of equi-
librium-limited adsorption are restricted to purifications
and the separation of dilute components from bulk streams
(i-e., for components consisting of less than 10% of the
feed). In order to limit the necessary size of the adsorbent
bed and to facilitate the subsequent regeneration steps,
it follows that equilibrium-limited adsorption will be a
favorable alternative only when the adsorbent affinity is
greater for the impurities or dilute components than for
the bulk stream. The mutual affinity of a given adsor-
bate-adsorbent pair is typically reported in terms of
equilibrium loading on the adsorbent. The equilibrium
loading is expressed as a function of adsorbate partial
pressure at a single temperature (i.e., an isotherm ex-
pression (Yang, 1987; Ruthven, 1984)). Once the isotherm
expression is known, the design of an adsorber is a rela-
tively simple task (Fair, 1969; Kovach, 1988; Wankat,
1990).
The ultimate utility and cost of an adsorption process
is closely related to the interrelation between the amount
of time that the product gaa(es) can be collected (i.e., the
cycle time) and the size of the required adsorption unit.
As the cycle time increases, the adsorber length (and
separation cost) increases correspondingly. For a large-
scale industrial process a cycle time of 2 h is typical.
Depending on the magnitude of the equilibrium loading
of the preferentially adsorbed components, the length of
the adsorber needed to achieve such a cycle time may
result in an uneconomical process. Thus, the required
adsorber length is a criterion of the feasibility of an ad-
sorption separation.
In general, for a standard cycle time of 2 h, i f the de-
sired separation or purification requires an adsorber that
is longer than 20 f t , then equilibrium-limited adsorption

Table VIII. Favorable Components and Chemical Families
for Desiccation by Equilibrium-Limited Adsorption
gases chemical families
argon aliphatics
helium hydrocarbon aromatics
hydrogen chlorides
chlorine fluorides
hydrogen chloride oxygenated compounds
sulfur dioxide
ammonia
air
can be eliminated as a potential separation method.
Two distinct applications for equilibrium-limited ad-
sorption are possible. The first entails the removal of
adsorbable components from an inert carrier gas (e.g.,
removal of organics from air, oxygen, nitrogen, helium, etc).
In thia case, the equilibrium loading of the inert ~ 8 on
9 the
adsorbent is negligible and can be ignored.
The second application of equilibrium-limited adsorp-
tion involves the separation between adsorbable compo-
nents. The objective here is to collect the less adsorbed
component in pure form for a period of time (typically
about 2 h) until the adsorber bed is exhausted. At that
point the more adsorbed component will break through
and will begin to contaminate the product. Such a process
will be feasible only under the following conditions:
1. The more adsorbed component must be in the mi-
nority in the feed (less than 10 mol %). If the majority
feed component(s) were to be adsorbed, the adsorber bed
would fill rapidly or would be impractically long.
2. For a cycle time of 2 h, the adsorber length required
to achieve breakthrough of the more ahorbed component
should be less than 20 f t . The length of an adsorber can
be found by several methods such as those given by Fair
(1969)or by Wankat (1990).
3. The ratio of the equilibrium loadings of the two
components should be at least 2, and preferably higher
(Chu, 1991). A high loading ratio ensures that simulta-
neous adsorption will be minimal. Because of its high
concentration in the feed, the less adsorbed component
may displace the more adsorbed component if the loading
ratio is too low.
A limited number of bulk enrichment separations (i.e.,
adsorbed components consist of 10 mol % of more of the
process stream) are now routinely performed with pressure
swing adsorption cycles. Examples include hydrogen re-
covery, methane enrichment from biogases, oxygen en-
richment, carbon dioxide recovery, and natural gas re-
covery. Further details are available in Richter (1987).
These cases currently are not covered by the SSAD.
The use of equilibrium-limited adsorption for desiccation
operations has been notably successful. Silica gels, zeolite
molecular sieves, and activated aluminas have high affin-
ities for water. The following heuristic reflects current
industrial applications (Keller et al., 1987;Yang, 1987):
I f the process stream to be dried contains less than 3
wt % water and is composed of gases or organic species
which are members of the chemical families listed in
Table VIII, then equilibrium-limited adsorption will be
a feasible (and probably the best) alternative. The ap-
propriate adsorbent (some type of silica gel, zeolite mo-
lecular sieve, or activated alumina) for the particular
application in question cannot be determined at this
stage.
As is the case with other separation techniques requiring
mass separating agents, the appropriate adsorbent for a
given separation is not known in the early stages of process
development. With hundreds of commercial adsorbents
available, the examination of each potential adsorbate-
Ind. Eng. Chem. Res., Vol. 31,No. 7,1992 1689
Table IX. Mixed Solvent Recovery Specifications
mol mol boiling point component
component % wt (K) type
nitrogen 70.645 28.0 77.4 gas
oxygen 28.855 32.0 90.2 gas
ethyl acetate 0.256 88.1 350.3 liquid
toluene 0.244 92.1 383.6 liquid
adsorbent pair would be prohibitively time-consuming.
Moreover, even if an exhaustive search could be done
quickly, the available isotherm data are relatively limited
(Valenzuela and Myers, 1989).
When experimental isotherm data are unavailable, ad-
sorption affinity can be estimated for activated carbon
adsorbents from a generalized Dubinin-Polanyi charac-
teristic curve developed by Barnicki (1991). The method
described by Barnicki requires only molar volume and
fugacity data.
G. Condensation. Condensation is a basic separation
technique in which a gas stream is brought to its saturation
(dew) point where the low volatility components begin to
liquefy. As these Components condense out, the dew point
rises and the temperature must be lowered further to
continue the process. A condenser is typically equivalent
to only one or two theoretical equilibrium separation
stages. Consequently, condensation processes exhibit poor
selectivity unless the relative volatility or boiling point
temperature difference of the components is extremely
large.
Condensation should be explored as a potential sepa-
ration method for enrichment operations when the rela-
tive volatility between key components is greater than
approximately 7 or the boiling point difference is greater
than 40 "C.
Condensation is most favorable for the separation of
high-boiling organic vapors from noncondensable gases,
especially when cooling water can be used as the con-
densing medium. In such situations, extreme purity (e.g.,
ppm levels) cannot be achieved, but generally greater than
95% removal is possible.
Example Separations
Mixed Solvent Recovery from a n Air Stream. In
order to comply with strict environmental regulations on
the extent of toxic emissions, many chemical synthesis
processes include one or more steps involving the removal
of trace amounts of organics from process off-streams. Fair
(1967)presented a detailed study of such a process for the
removal of toluene and ethyl acetate from an air stream.
Table IX gives the input specifications for the solvent
recovery problem. The objective is to recover 99% of the
ethyl acetate and essentially all of the toluene. Note that
the organics are to be separated and recovered, rather than
removed and possibly destroyed. Air is considered to be
71 mol % nitrogen and 29 mol % oxygen for this problem.
The separation analysis starts with the phase split
manager (PSM) as described in part 1 of this series
(Barnicki and Fair, 1990). The input stream includes both
gas and liquid compounds. However, due to the extremely
low concentration of liquids (0.5 mol %), no phase sepa-
ration is required. The analysis proceeds directly to the
purification split selector (PSS) of the gas split manager
(GSM).
For purification operations, the possible separation
methods are chemical absorption, catalytic conversion,
molecular sieve adsorption, and equilibrium-limited ad-
sorption. The ranked component property lists are given
in Table X. Examining the ranked lists, the possible key
component pairs are nitrogen-ethyl acetate using chemical
1690 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992

Table X. Ranked Property Lists for Mixed Solvent Recovery producing 2 X lo6 standard ft3/day of gas.
ranked Property The key to the economic viability of such a reclamation
method DroDertv components values process centers on the effective separation and purification
chemical absorption chemical family oxygen inorganic gas of the methane and carbon dioxide components. A rep-
catalytic oxidation nitrogen inorganic gas resentative gas composition consisting of seven components
ethyl acetate is given in Table XI1 (simplified slightly from the com-
acetate positions given by Magnani (1984) and Schumacher
toluene alkylbenzene
molecular sieve kinetic diameter oxygen <3 A (1983)). Blakely (1985) reports typical specifications for
adsorption nitrogen <3 A salable carbon dioxide. Stockmann and Zollner (1987)
ethyl <8 A indicate typical chemical synthesis methane gas specifi-
acetate cations. These are repeated in Table XIII. For merchant
toluene <a A gas applications, the carbon dioxide product must be es-
equilibriumlimited equilibrium loading toluene 0.00202
adsorption (mol/g of ads) ethyl 0.00132 sentially free of all impurities, including methane.
acetate Methane synthesis gas also must be essentially free of
oxygen =o.o impurities (ppm levels permissible).
nitrogen =o.o Separation synthesis for the landfillgas proceeds directly
absorption, catalytic oxidation, or molecular sieve ad- to the split manager rather than the phase split
sorption and ethyl acetateoxygen using equilibrium-lim- manager. The low concentration of liquid-phase compo-
ited adsorption. One must now refer to Figure 6, the nents ( ~ 0 . 8 1mol ?% combined aromatics and halohydro-
purification split selector (PSS),to determine which of carbons) precludes the need for both gas and liquid sep-
these separations are feasible. aration systems. The GSM analysis begins with the gen-
Looking at the first branch of the PSS, one can eliminate eration of ranked property lists and the identification of
immediately chemical absorption as a possible separation possible separation points. Because of the presence of the
method because neither nitrogen nor ethyl acetate contains trace impurities benzene, chloroethane, and hydrogen
acid-base functionalgroups (see Table VI). Although ethyl sulfide, the initial separation step will involve the removal
acetate and toluene are oxidizable impurities, catalytic of these components (see sequencing heuristics in Table
oxidation cannot be used because these two components 11). The separation of trace impurities is a purification
are desired products. They would be destroyed in the process (refer to the section Separation Types and Figure
oxidation process. 6). Therefore, the potential separation techniques are
The kinetic diameters of nitrogen and oxygen are clearly limited to chemical absorption, catalytic conversion, and
much smaller than those of the organics; the Components adsorption. Ranked property lists for these separation
can be separated by molecular sieves. However, in this methods are shown in Table XIV.
case, the oxygen and nitrogen would be adsorbed and the Looking first at chemical absorption, one finds that the
ethyl acetate and toluene would be excluded from the process stream contains acid-base functional groups,
molecular sieve pores. Adsorption of 99.5 mol ?% of the namely hydrogen sulfide and carbon dioxide (see Figure
feed stream is obviously impractical. Therefore, molecular 6). Moreover, it is known that amine solvents are available
sieve adsorption is also eliminated. which can selectively remove hydrogen sulfide (Kohl and
The final potential separation method is gas-phase Riesenfeld, 1985). Since chemical absorption is capable
equilibrium-limited adsorption. From the generalized of removing one of the impurities, it is kept for the moment
characteristic adsorbent curve for activated carbons as a feasible separation alternative.
(Barnicki, 1991),the equilibrium loadings of ethyl acetate Again referring to Figure 6, catalytic conversion is ex-
and toluene are found to be 0.00132 and 0.00202mol/g of amined next. Catalytic oxidation is eliminated from fur-
adsorbent, respectively. The adsorption of oxygen and ther consideration because the methane oxidizes more
nitrogen is negligible. In addition, no adsorbent-fouling readily than the impurities themselves. Hydrogenation
components are present, and the amount adsorbed is small. is also inappropriate. Only benzene is amenable to hy-
Thus, equilibrium adsorption is the favored separation drogenation and the saturated hydrocarbon product from
method. These results are summarized in Table XI. this reaction must still be separated from the mixture.
Separation and Purification of Landfill Gases. One Thus, all forms of catalytic conversion can be eliminated.
ton of municipal waste contains approximately 200 kg of Adsorption based on molecular sieving effects (using
organic material. In the anaerobic environment of a 1OX sieves) theoretically could be employed to remove the
landfill, the organic refuse is decomposed by microorgan- chloroethane and benzene from the rest of the feed stream,
isms into a gaseous product consisting of 40-60 mol 5% However, as chloroethane and benzene are the larger
methane, 40-50 mol ?% carbon dioxide, and 5000 ppm or molecules (and are thus excluded from the zeolite pores),
more of impurities (various hydrocarbons, halogenated it is clearly infeasible to adsorb 99.7% of the feed stream.
compounds, and sulfur compounds). Large municipal Molecular sieving adsorption also can be eliminated from
landfii are capable of producing (0.2-8.0) X 106 standard further consideration.
ft3/day of gas for up to 20 years. Thus, landfill gas rep- As indicated by the ranked list of equilibrium adsorbent
resents a significant potential source of carbon dioxide and loadings (see Table XIV), equilibrium-limited adsorption
methane (Ruf and Egli, 1988; Malik et al., 1987). For this is favorable for the removal of all three impurities si-
exercise, the landfill is assumed to be of moderate size, multaneously. This alternative is clearly superior to the
Table XI. Summary of Mixed Solvent Recovery Separations
separation method logic
Rejected Methods
nitrogen/ethyl acetate chemical absorption no acid/base functional groups
nitrogen/ethyl acetate catalytic oxidation oxidizable Components are desired product
nitrogen/ethyl acetate mol sieve adsorption adsorbed components (02,Nz) are majority of feed
Selected Method
oxygen/ethyl acetate equilibrium-limitedadsorption selectivity for toluene, ethyl acetate adsorbates high
 Ind. Eng. Chem. Res., Vol. 31, No. 7, 1992 1691

Oxygen Separation by
f'' methane MOL SIEVE ADSORPTlON
benzene
Separation by
-chloroethane
----
\-
EOUILIBRIUM-LIMITED
carbon dioxide izmfmmu
mhane ADSORPnON ntLpea
Methane
nitrogen methane Product
oxygen
methane

methane
nitrogen Sanaratinn hv
oxygen
carbon dioxide G171EM''
---- Separationby
methane CRYOGENIC DlSTlLLATlON
Carbon Dioxide \
Product

Figure 7. Summary of landfill gas separation process alternatives.

Table XII. ReDremntative Landfill Gas Com~osition~ use of chemical absorption to remove hydrogen sulfide
component mol % followed by a second separation step to remove the chlo-
roethane and the benzene. The second process would be
methane 47.50 necessity (see analysis above), involve equilibrium-limited
carbon dioxide 47.00
nitrogen 3.70 adsorption. Therefore, the best initial separation for the
oxygen 0.99 feed mixture is equilibrium-limited adsorption to remove
hydrogen sulfide 0.01 the chlorobenzene, hydrogen sulfide, and benzene in one
aromatics (benzene) 0.30 step.
halohydrocarbons (chloroethane) 0.50 For the preliminary process analysis it is assumed that
a Magnani (1984);Schumacher (1983). the chlorobenzene, hydrogen sulfide, and benzene are
completely removed, leaving only oxygen, nitrogen,
Table XIII. Methane and Carbon Dioxide Product methane, and carbon dioxide (3.7, 1.0,47.9,and 47.4 mol
Swcifications % respectively). Because of the product specifications,the
merchant carbon dioxide' synthesis methane gasb next separation is required to be sharp. The potential
separation methods are limited to physical absorption,
cryogenic distillation, and adsorption (see the section
carbon dioxide 99.985mol % methane 99.98 mol % Separation Types and Figure 5 ) . Ranked property lists
total sulfur 0.3 ppm max chlorides 0.25 g/100 SCF'
compounds and split points for these separation methods are shown
total hydrocarbons 5 ppm max sulfur 1.25 g/100 SCFc in Table XV. One must now refer to Figure 5 to deter-
compounds mine the feasibility of the indicated splits.
OBlakely (1983). bStockmannand Zollner (1987). 'SCF p standard The relative volatility between methane and oxygen is
ft.3 favorable for cryogenic distillation (a= 2.7). Moreover,
Table XIV. Ranked Property Lists for Purification Separations of Landfill Gas
mol sieve adsorption equilib adsorption
chem absorption nominal kinet equilib loading
component chem family component diam (A) component (mol/g of ads)
carbon dioxide acid gas oxygen <3 oxygen =o.o
hydrogen sulfide acid gas nitrogen <3 nitrogen so.0
nitrogen inorg gas hydrogen sulfiide <4 methane 0.0005
oxygen inorg gas carbon dioxide <4 carbon dioxide 0.0035
chloroethane chloride methane <4 benzene 0.0046
benzene alkylbenzene chloroethane <5 hydrogen sulfide 0.0069
methane n-alkane benzene <8 chloroethane 0.0070

Table XV. Ranked Property Lists for Sharp Separations of Landfill Gas
cryogenic distillation equilib adsorption mol sieve adsorption
physical absorption re1 equilib loading nominal kinet
componenta
oyxgen
chem family
inorg gas
component
nitrogen
volatility
1.13 nitrogen
-
component , (mol/a of ads)
so.0
Component
oxygen
diam (A)
<3
nitrogen inorg gas oxygen 2.73 oxygen =o.o nitrogen <3
carbon dioxide acid gas methane methane 0.0005 carbon dioxide <4
methane n-alkane carbon dioxide freezes carbon dioxide 0.0035 methane <4
1692 Ind. Eng. Chem. Res., Vol. 31, No. 7,1992

the landfill gas separation is a large-scale process, pro- synthesis of separation sequences for gas/vapor mixtures
ducing approximately 21 tons/day of methane. However, has been presented. The architecture of the SSAD is based
the presence of large amounts of carbon dioxide precludes on a combination of rule analysis and task-oriented
its use; the carbon dioxide will freeze and foul condenser methods. The cornerstone of the task-oriented problem-
surfaces (see Cryogenic Distillation). Oxygen and nitrogen solving methods used in the SSAD is the separation syn-
can be separated from methane and carbon dioxide by 3A thesis hierarchy (SSH). The separation synthesishierarchy
molecular sieves, with the oxygen and nitrogen as the ad- (SSH) is the first comprehensive, systematic analysis of
sorbed components (see Table VII). Equilibrium loadings separation synthesis domain knowledge to appear in the
on activated carbon are favorable for the preferential ad- chemical engineering literature. In ita current imple-
sorption of carbon dioxide over methane. However, as the mentation, the SSH includes all of the major separation
problem is stated, almost 50% of the stream would be methods commonly encountered in industrial practice.
adsorbed (carbon dioxide as well as some of the methane). Two industrially significant separation problems have been
This is not a reasonable alternative. presented to illustrate the capabilities of the SSAD. The
The final separation method to examine is physical resultant separation sequences compare favorably with
absorption. The selectivity calculated from eq 8 between actual industrial processes.
carbon dioxide and methane is 4.6 at 298 K using the Shair
correlation. Thus, physical absorption is a feasible alter-
native (a common solvent, Selexol, gives a selectivity of
approximately 6.5 (Kohl and Riesenfeld, 1985)). Since high Acknowledgment
purity is required, the physical absorption process should
be followed by a chemical absorption step (see Figures 5 We gratefully acknowledge the partial support of this
and 6). work by a grant from the Exxon Foundation.
Two splits, the molecular sieve adsorption of nitrogen
and oxygen as well as the physical/chemical absorption
of carbon dioxide, have been found by the selector analysis Nomenclature
to be feasible. Comparing these two separations, one sees
that the physical absorption of carbon dioxide is the fa- Symbols
vored separation. Heuristic 3 of Table I1 indicates that ci = flow rate of key component in product i
the separation which matches a desired product directly Di = diffusivity
should be done next. Assuming essentially complete re- f j o l = standard-state liquid fugacity
moval of carbon dioxide, the remaining mixture consists Pi = permeability
of 91 mol ?% methane, 1.9 mol ?% oxygen, and 7.1 mol % Pisat= vapor pressure
nitrogen. pi* = partial pressure
The analysis of the separation of methane from oxygen R = ideal gas constant
and nitrogen is quite similar to the previous exposition for Si = solubility
carbon dioxide. Cryogenic distillation is feasible this time Ske= split of light or heavy key
because the carbon dioxide has been removed. In addition, Si# = physical absorbent selectivity
oxygen and nitrogen can be separated from methane and T = system temperature
carbon dioxide by 3A molecular sieves, with the oxygen Tb,i = normal boiling point
and nitrogen as the adsorbed components (see Table VII). T, = critical temperature
Methane is preferentially adsorbed on activated carbon. V , = van der Waals volume
However, again, this would require the adsorption of the Vi = molar volume
majority of the feed. It is worth noting that this separation x i = liquid-phase mole fraction
may be accomplished with incomplete recovery of methane yi = vapor-phase mole fraction
(with recycle), but the SSAD currently does not handle a = relative volatility
such a case. Physical absorption is also infeasible. aij* = membrane permselectivity
Since both the distillation and molecular sieve adsorp- 6i = solubility parameter
tion proceases result in the same product distributions, one yi = activity coefficient
cannot determine the best alternative without a detailed yijm= infinite dilution activity coefficient of component i in
economic analysis. Both separations are assumed to be component j
feasible at this point. A summary of two alternative sep- +i = mixture fugacity coefficient
aration sequences is given in Figure 7.
It should be pointed out that a proceas stream containing Superscripts
carbon dioxide and methane can be treated successfully 1 = liquid phase
using membrane permeation. This is a fairly common O = standard state
process in the natural gas industry. However, as the
problem is stated here, both pure methane and pure car- Subscripts
bon dioxide are desired products. Membrane permeation a = adsorbate
is an enrichment process only; it is not feasible to obtain i = component i
two products of high purity and high recovery. If the j = component j
problem had been stated so that enriched carbon dioxide s = solvent
and methane streams were the desired products, then the
selector analysis would have followed the enrichment split Abbreviations
selector (Figure 4) rather than the sharp split selector DSM = distillation split manager
(Figure 5 ) . ESS = enrichment split selector
GSM = gas split manager
Conclusions LSM = liquid split manager
A discussion of an extension of the prototype expert MSA = mass separating agent
system, the separation synthesis advisor (SSAD), for the PSM = phase split manager

PSS = purification split selector
SSAD = separation synthesis advisor
SSH = separation synthesis hierarchy
SSS = bulk, sharp split seleotor
Literature Cited
Astarita, G.; Savage, D. W.; Bisio, A. Gas Treating with Chemical
Solvents; Wiley: New York, 1983.
Barnicki, S. D. Separation System Synthesis: A Knowledge-Based
Approach. PbD. Dissertation, Department of Chemical Engi-
neering, The University of Texas a t Austin, 1991.
Barnicki, S. D.; Fair, J. R. Separation System Synthesis: A
Knowledge-Based Approach. 1. Liquid Mixture Separations.
Ind. Eng. Chem. Res. 1990,29,421.
Barrer, R. M. New Selective Sorbents: Porous Crystals as Molecular
Filters. Br. Chem. Eng. 1959,May, 267.
Barrer, R. M.; Brook, D. W. Sorption and Reaction of Small Organic
Molecules in Chabazite. J. Am. Chem. SOC.1959,81,940.
Blakely, P. Carbon Dioxide Recovery and Sale. Proceedings From
the 8th International Landfill Gas Symposium; Government
Refuse and Disposal Assn.: Silver Spring, MD, 1985;p 6.
Breck, D. W.Zeolites and Clay Minerals as Sorbents and Molecular
Sieves; Academic Press: London, 1974.
Chu, S. Personal communication, Shell Oil Company, Houston, TX,
1991.
C o h , J. J. Where to Use Molecular Sieves. Chem. Eng. Prog. 1968,
64, 66.
Cusher, N. A. UCC IsoSiv Process. In Handbook of Petroleum
Refining Processes; Meyers, R. A., Ed.; McGraw-Hik New York,
1986;Chapter 8-2,pp 8-11to 8-17.
England, C. Gas Solubilities in Physical Solvents. Chem. Eng. 1986,
93,63.
Fair, J. R. Case Study 7: Mixed Solvent Recovery and Purification;
Department of Chemical Engineering, Washington University St.
Louis, MO, 1967.
Fair, J. R. Sorption Processes For Gas Separation. Chem. Eng. 1969,
76 (July 14), 90.
Gjalbaek, J. C. The Solubility of Hydrogen, Oxygen, and Carbon
Monoxide in Some Non-Polar Solvents. Acta Chem. Scand. 1952,
6 , 623.
Ho, T. L. Hard and Soft Acids and Bases Principle in Organic
Chemistry; Academic Press: New York, 1977.
Hogsett, J. E.; Mazur, W. H. Estimate Membrane System Area.
Hydrocarbon Process. 1983,52.
Isalski, W. H. Separation of Gases; Clarendon Press: Oxford, 1989.
Johnson, J. A.; Kabza, R. G. SORBEX Industrial-Scale Adsorptive
Separation. In Advances in Separation Processes; Institution of
Chemical Engineers Symposium Series 188; Hemisphere: New
York, 1990.
Kaliszan, R. Quantitative Structure-Chromatographic Retention
Relationships; Wiley: New York, 1987.
Keller, G. E.; Anderson, R. A.; Yon, C. M. Adsorption. In Handbook
of Separation Process Technology;Rousseau, R. W., Ed.; Wiley-
Interscience: New York, 1987;Chapter 12.
Kelly, R. M. General Processing Considerations. In Handbook of
Separation Process Technology;Rousseau, R. W., Ed.; Wiley-In-
terscience: New York, 1987;Chapter 4.
King, C. J. Separation Processes, 2nd ed.; McGraw-Hill: New York,
1980.
Kohl, A.; Riesenfeld, F. Gas Purification, 4th ed.; Gulf Publishing:
Houston, TX, 1985.
Koros, W. J.; Hellums, M. W. Transport Properties of Polymers-
Part I. Rubbers, Part 11. Glasses. In Encyclopedia of Polymer
Science and Engineering; Wiley Interscience: New York, NY,
1989;p 724.
Koros, W. J.; Fleming, G. K.; Jordan, S. M.; Kim, T. H.; Hoehm, H.
H. Polymeric Membrane Materials for Solution-Diffusion Based
Permeation Separations. Prog. Polym. Sci. 1988,13,339.
Kovach, J. L. Gas Adsorption. In Handbook of Separation Tech-
niques for Chemical Engineers, 2nd ed.; Schweitzer, P. A,, Ed.;
McGraw-Hill: New York, 1988;Chapter 3.1.
Ind. Eng. Chem. Res., Vol. 31,No. 7, 1992 1693
Liu, Y. A. Process Synthesis: Some Simple and Practical Develop-
menta. In Recent Developments in Chemical Process and Plant
Design; Liu, Y. A., McGee, H. A., Jr., Epperly, W. R., Eds.;Wiley
New York, 1987;Chapter 6.
Magnani, 0. M. Landfill Gas, From the Pail to the Customer. Pro-
ceedings from the GRCDA 7th International Gas Symposium;
Government Refuse and Disposal Assn.: Silver Spring, MI), 19%
p 118.
Malik, V. A.; Lerner, S. L.; MacLean, D. L. Electricity, Methane and
Liquid Carbon Dioxide Production From Landfill Gas. Gas Sep.
Purif. 1987,1 (41,77.
Masso, A. H.; Rudd, D. The Synthesis of System Designs: 11.
Heuristic Structuring. AIChE J. 1969,15 (l),10.
McInnes, R.; Jelinek, S.; Putache, V. Cutting Toxic Organics. Chem.
Eng. 1990,Sept, 108.
Mowry, J. R. UOP Sorbex Separation Technology. In Handbook of
Petroleum Refining Processes; Meyers, R. A., Ed.; McGraw-Hik
New York, 1986.
Nishida, N.; Stephanopoulos, G.; Westerberg, A. W. A Review of
Process Synthesis. AIChE J. 1981,27 (31,321.
Prausnitz, J. M.; Shair, F. H. A Thermodynamic Correlation of Gas
Solubilities. AIChE J. 1961,7 (12),682.
Prausnitz, J. M.; Lichtenhaler, R. N.; Gomes de Azevedo, E. Mo-
lecular Thermodynamics of Fluid-Phase Equilibria, 2nd ed.;
Prentice-Hall: Englewood Cliffs, NJ, 1986.
Reitmeier, R. E.; Fleming, H. W. Acetylene Removal From Poly-
ethylene Grade Ethylene. Chem. Eng. Prog. 1958,54 (12),48.
Richter, E. Industrial Processes for Gas Separation by Pressure
Swing Adsorption. Part 2: Processes. Brenmt.-Chem. 1987,40
(lo),432.
Rudd, D. The Synthesis of System Designs: I. Elementary Decom-
position Theory. AIChE J. 1968,14 (2),343.
Rudd, D.; Powers, G. J.; Siirola, J. J. Process Synthesis; Prentice-
Hall: New York, 1973.
Ruf,A.; Egli, S. Application of Gas Separation Shown Through the
Purification of Landfill Gas. Gas. Sep. Purif. 1988,2 (2),90.
Ruthven, D. M. Principles of Adsorption and Adsorption Processes;
Wiley-Interscience: New York, 1984.
Ruthven, D. M. Zeolites as Selective Adsorbents. Chem. Eng. Prog.
1988,84 (2), 42.
Rylander, P. N. Hydrogenation Methods; Academic Press: New
York, 1985.
Sander, B.; Skjold-Jorgensen, S.; Rasmussen, P. Gas Solubility
Calculations. I. UNIFAC. Fluid Phase Equilib. 1983,11, 105.
Schumacher, M. M. Landfill Membrane Recovery; Noyes Data
Corportion: Park Ridge, NJ, 1983; p 17.
Springmann, H. Cryogenics: Principles and Applications. Chem.
Eng. 1985,92(May 13),59.
Stockmann,R.; Zollner, R. Production of Pure Methane. Ger. Chem.
Eng. 1987,Feb, 72.
Stookey, D. J.; Patton, C. J.; Malcolm, G. L. Membranes Separate
Gases Selectively. Chem. Eng. Proc. 1986,36.
Streitwieser, A.; Heathcock, C. H. Introduction to Organic Chem-
istry, 2nd ed.; Macmillan: New York, 1981.
Suter, H. R. Range of Applicability of Catalytic Fume Burners. J.
Air Pollut. Control Assoc. 1955,5 (3),173.
Tennyson, R. N.; Schaaf, R. P. Guidelines Can Help Choose Proper
Process for Gas-Treating Plants. Oil Gas J. 1977,Jan 10, 78.
Teplyakov, V. V.; Meares, P. Correlation Aspects of the Selective Gas
Permeabilities of Polymeric Materials and Membranes. Gas Sep.
Purif. 1990,4 (6),66.
Timmerhaus, K.D.; Flynn, T. M. Cryogenic Process Engineering;
Plenum Press: New York, 1989.
Ullmanns Encyclopedia of Industrial Chemistry, Unit Operations;
Adsorption, Part 9-6,Gerhartz, W., Ed.; VCH: Weinheim, F R G
Vol. B3, 1988.
Valenzuela, D. P.; Myers, A. L. Adsorption Equilibrium Data
Handbook; Prentice-Hall: Englewood Cliffs, NJ, 1989.
Valerius, M. Dehydration of SNG. Presented at the 167th National
Meeting of the American Chemical Society, Los Angeles, CA,
April 1, 1974.
Vaughan, D. E. W. The Synthesis and Manufacture of Zeolites.
Chem. Eng. Prog. 1988,84 (2),25.
Walas, S. M. Phase Equilibria in Chemical Engineering; Butters-
worth: Stoneham, MA, 1985.
1694 I n d . Eng. C h e m . Res. 1992,31, 1694-1704
Walker, D.; Koros, W. J. Gas Separation Membrane Materials Se- Yen, L.; McKetta, J. J. A Thermodynamic Correlation of Nonpolar
lection Criteria: Weakly and Strongly Interacting Feed Compo- Gas Solubilities in Polar, Nonassociated Liquids. NChE J. 1962,
nent Situations. Polym. J. 1991,23,481. 8 (51,501.
Wankat, P. C. Rate-Controlled Separations; Elsevier Applied Sci-
ence: New York, 1990. Receiued for review October 25, 1991
Yang, R. T.Gas Separation by Adsorption Processes; Butterworth Revised manuscript received March 17, 1992
Boston, MA, 1987. Accepted April 5, 1992

Probabilistic Approach to Robust Process Control

Charles D. Schaper,* Dale E. Seborg, and Duncan A. Mellichamp
Department of Chemical and Nuclear Engineering, University of California, Santa Barbara, California 93106

A probabilistic approach to robust process control is developed. First, a statistical measure of a

controllers ability to reject disturbances is introduced. Next, a new robust control framework of
characterizing model uncertainty descriptions by probability distributions is developed. The statistical
measure of disturbance rejection is then incorporated within the framework. In the proposed
probabilistic approach, process knowledge can be incorporated in the design procedure and controller
performance can be analyzed by probability measures. Several simulation examples demonstrate
the advantages of the new approach.

Introduction could occur when a fundamental physical model of the

An important objective in designing a process control
system is robustness to modelling error. Previous ap-
proaches to robust process control design have generally
used bounds around the parameters or frequency response
of a nominal plant model to describe model uncertainty.
The control system is then designed to minimize the effects
of a worst-case situation. Current design approaches for
robustness are described by Morari and Zafhiou (1989).
Process control applications of these design techniques
include those of Agamennoni et al. (1988)and Skogestad
et al. (1988).Advantages of existing design techniques for
robustness include the following: (1)closed-loopstability
is guaranteed over the entire range of model uncertainty
(robust stability); (2) an upper bound on a given perform-
ance measure is guaranteed (robust performance). Because
the controllers are generally designed for worst-case situ-
ations that may have a low probability of occurring, the
resulting robust controllers may be very conservative for
more typical operating conditions that have a much higher
probability of occurring.
In this paper, a new approach to robust process control
design is developed in which model uncertainty is char-
acterized by probability distributions. This approach
allows closed-loop performance tradeoffs to be analyzed
as a function of the likelihood of controller performance;
that is, performance can be characterized by a probability
measure for all situations between nominal and worst-case
conditions. The result is a more complete analysis strategy
that can result in better controller design.
In the subsequent development, a general linear repre-
sentation of the plant description is used in which mod-
eling error is described by probability distributions.
Modeling error due to both parameter uncertainty and the
linear approximation of a nonlinear plant can be included
within this probabilistic framework. It should be noted
that the error resulting from the approximation of a non-
linear system by a linear model may be greater than any
model parameter uncertainty. For example, this situation
* Present address: Department of Electrical Engineering,
Stanford University, Stanford, CA 94305.
0 1992 American Chemical Society
process does not exist or is too complex for controller
design, and consequently, an empirical linear model (e.g.
a transfer function model) is developed from experimental
data. In this instance, the parameters of the linear ap-
proximation can be represented by probability distribu-
tions.
Although we describe some methods and examples of
approximating this type of modeling error, it is not the
intent of this paper to provide a well-formulated descrip-
tion of how to identify model uncertainty descriptions.
However, we note that probabilistic descriptions of mod-
eling error can be developed from a wide variety of sources,
including statistical information on phenomenological
model parameters, empirical model parameters, or fre-
quency response (Cloud and Kouvaritakis, 1987;Correa,
1989; Goodwin and Salgado, 1989; Stengel and Ryan, 1989).
Also, process knowledge is usually available in the form
of engineering heuristics and information about the range
of operating conditions. The probabilistic model de-
scription is sufficiently general to capture such prior
process knowledge and incorporate it within the design
procedure.
In addition to the development of a general probabilistic
framework, a statistical measure of closed-loop disturbance
rejection capabilities is introduced for process control
applications. A disturbance rejection measure is generally
more appropriate for process control applications because
the set-point remains constant for long periods of time.
In the development of this measure, it is important to note
that performance specifications for outputs or inputs can
be formulated in terms of statistical moments. For ex-
ample, a typical product specification is expressed in terms
of a mean and standard deviation (also referred to as root
mean square). Well-known control design strategies have
been developed to minimize statistical moments of the
outputa and inputs. These controller design strategies
include qlassical methods such as minimum variance
control (Astrijm, 1970;Box and Jenkins, 1976;Kucera,
1979),in addition to current methods such as robust linear
quadratic Gaussian (LQG) control strategiea (Stengel, 1986;
Bernstein and Haddad, 1990) and constrained minimum
variance control (Makila et al., 1984;Hotz and Skelton,