le) n§I lecture

CHEMICAL PROCESS SYNTHESIS

J. J. SIIROLA
Eastman Kodak Company
Kingsport, TN 37662

p ROCESS SYNTHESIS IS THE specification of

chemical and physical operations and the se-
lection and interconnection of equipment to imple-
ment these operations to effect desired chemical
processing transformations. Synthesis is the first
of an iterative set of process design activities
which also includes analysis, evaluation, and op-
timization. Significant progress has been made,
particularly since the advent of the digital com-
puter, in the development of a more scientific and
less empirical framework for modeling, simulat-
\
J. J. Siirola received his B.S. from the University of Utah and his
ing, and improving equipment and operating Ph.D. from the University of Wisconsin-Madison in 1970 where he
parameters for a given design. However the developed the AIDES process synthesis system and coauthored the
necessarily prerequisite invention of the design, introductory text, Process Synthesis. Besides a continuing involve-
ment in synthesis technique development, implementation, and ap-
the generation of the processing structure repre-
plication , his research interests also include simulation and optimiza-
sented as a flowsheet, remains largely a creative tion aspects of computer-aided design, non-numeric programming,
art. Economic competitiveness has demonstrated artifrcial intelligence, and technology assessment. He is currently a
the importance of correct structural choices in Research Associate in the Eastman Chemicals Division of Eastman Kodak
meeting process objectives. This realization has Company.

led to present efforts to investigate the possibility

of formalizing the synthesis activity. The ultimate dates only from the late 1960's. Some investi-
goal is not just to invent technically feasible de- gators studied the general overall invention of
signs, but to produce structural configurations complete flowsheets, while others concentrated on
that when analyzed, evaluated, and optimized will specific process synthesis subproblems such as
prove by design objective criteria to be superior reaction paths, reactor network configurations,
to (possibly all) other structural arrangements. separation train systems, energy recovery net-
Inventing flowsheets is, of course, not new and works, fault trees, and safety and control systems.
some guidance embodied in design rules-of-thumb Several excellent reviews with extensive bibli-
based on experience developed through trial and ographies [(Hendry, Rudd, and Seader (1973),
error has existed for many decades. However, Hlavacek (1978), Westerberg (1980), and
formal consideration of structural generation Nishida, Stephanopoulos, and Westerberg (1981) J
chronicle the developments and progress made
in these areas to the present time. Here we will
consider some of the underlying paradigms which
The ultimate goal is not just to have been suggested for continuous process syn-
invent technically feasible designs, but to
produce structural configurations that when analyzed,
thesis.
evaluated, and optimized will prove by design
objective criteria to be superior to ... SYNTHESIS IN DESIGN
other structural arrangements.
Process synthesis has been defined as the in-
vention of the process structure. This definition
© Copyright ChE Division, ASEE, 1982 can be extended to include invention of associ-

68 CHEMICAL ENGINEERING EDUCATION

 ated control and safety systems, development of
steady state and transient (start-up and shut-
Often more than merely
down) operating procedures, and other similar the invention of a feasible design,
design specification activities. Synthesis should be process synthesis involved an optimization
differentiated from 'flowsheeting,' a term which over structure.
refers to the simulation, often by computer, of an
existing structure.
The number of tasks that must be performed efforts. These include functional and morphologi-
to convert available raw materials into desired cal analyses (mechanical schemes for forcing
products and environmentally acceptable by- associations among desired design goals), imita-
products may be quite large. Even given the con- tion of nature, analysis of models of other designs,
cept of unit operations, the wide range of existing and brainstorming. These methods and aids rely
technologies capable of accomplishing these tasks heavily on analogies with or adaptations of past
and the multitude of ways in which these tech- working designs and none have explicit provisions
nologies can be interconnected to achieve the de- for generating optimal structural arrangements.
sign objectives with varying efficiency leads to Are there attributes specific to the synthesis
synthesis problems characterized by discrete de- problem that should be exploited and developed
cisions and combinatorial difficulties. Further- into solution procedures not necessarily re-
more, although the existence of some feasible pro- sembling techniques now employed by successful
cessing structure is usually not in question, the designers? Are there aspects of 'artificial intel-
multivariable and sometimes subjective nature of ligence' research, a relatively new branch of com-
the design objective criteria (safety, reliability, puter science concerned with understanding and
operability, environmental impact, feedstock se- simulating intelligent problem solving, that might
curity, quality, resilience, familiarity, experience, contribute to a formalism for process invention?
as well as capital investment, operating expense, Early in systematic process synthesis research, op-
etc.) makes it difficult to quantify the concept of timization, evolution, and decomposition emerged
'optimum.' Often it is desirable to generate a as potential invention approaches. These remain
series of structural alternatives judged to be the dominant framework for synthesis develop-
nearly optimal by some economic criteria for ment today.
further subjective evaluation.
Despite the fundamental importance of syn-
OPTIMIZATION APPROACHES
thesis in chemical process design, few existing
design texts give more than passing reference to Often more than merely the invention of a
the necessity for first establishing a workable feasible design, process synthesis involves an op-
manufacturing process for producing the desired timization over structure. A natural approach,
product, preferring instead to concentrate on the and in fact one of the earliest proposed, isi to
more deductive and quantitative aspects of combine structural optimization with equipment
analysis, evaluation, and (parameter) optimiza- and operating parameter optimizations now
tion. When mentioned at all, more is noted about routine because of recent advances in computer-
such attributes as experience, self confidence, con- ized numerical methods. In fact, the synthesis
structive discontent, open mindedness, and demon- problem has been formally defined as a nonlinear
strated intuition and common sense of engineers mixed integer and continuous variable optimiza-
who appear to be successful at synthesis than tion problem. Equipment selection, interconnec-
about invention techniques themselves. However, tion, and other discrete decisions are represented
some qualitative concept discovery aids do seem by integer variables (often zero or one) and
to be used by successful designers. The most basic constraints among these variables. The difficulty is
of these is historical information from such that no code exists to solve this generally exceed-
sources as engineering textbooks, handbooks, cata- ingly large mixed optimization problem.
logs, proprietary company reports, commercially One approach is to invent a superstructure in
available comparative process analyses, and open which all technologies under consideration and all
scientific and patent literature, experience, and their possible interconnections are embedded. As-
rules-of-thumb. Design concepts have also been sociated with each branch of an interconnection
discovered through individual or group creative leading from one item of equipment to several

SPRING 1982 69
oth@rA 11 a 11tructural param@t@r,' a continuous ram@tc:!rs, lfoww0r, inoquality con1tnints on d0-
variabl@ which r@pr@A@nt§ th@ fraction of :tlow to sirn variubl@s may ro8ult in a discontinuous
b@ §plit to @ach of th@ altc:!rnaw d@8timition8, A§ f0,u1iblo rorion boundary noce88itatins- di~cr0tc:!
fraction§, 8tructural param@t@r§ Ar@ con§trttin@d docision8 tho structural paramotor mothod wati
to b@ non=n@rtttiv@ and to §Um to unity for all dovelopod to circumvont, Inoquality con8traints
branch@§ from a junction, Th@§@ 8tructural pa= a11Mciatod with oquipmont dolotc:!d from tho 1upor-
ram@t@r§ aro includod with othor oquipmont and 11tructuro during structural paramotor optimiza-
oporatinr paramotor§ in an ovorall do§ig"n optia tion can forco con1train0d bohavior in tho romain-
mization, Should a 8tructural paramotor bo inr fftructuro, For many problom1 a complmoly
optimizod to zoro, tho corr{!§pondinr intorconn@c- continuous formulation m9.,y not bo po11dblo,
tion 18 dolotod from tho 8upor8tructuro, and §hould Tho optimal a1111ignmont formalfam from
all :tlow§ A§§Ociatod with an itc:!m of oquipmont oporation r@soarch is anothor optimization ap-
bo zoro, it too 18 dolotod from tho §tructuro, In proach that has boim appliod to tho 11ynthosis of
offoct, tho di§croto doci§ion maldnr a§poct§ of ho,it intogration notworks, Th11 synthesis 8Ub-
problom concorns tho hoatinu and coolins- tasks
which must bo porformod within a proceH. Tho
Contldoroltlo 11vln11 may problm statoment includes a complete\! doscription
rotult If polrt of complomont1ry hooting of tho streams to be heat0d or cooled, de~criptions
and coo!lng t11k1 con Ito m1do to drlvo ooch of auxiliary utiHties capabl0 of performinu tho r0-
othor r1thor than roqulro tho uto
of 1uxlll1ry utllltlot, quirod tasks, and descriptions and d0siS'fl equa-
tions for equipm0nt (furnaces, reboilers, con-
don10r8, and other hoat 0xchan10r!l, 0tc,) usoful
proco§ff synthosis ar@ roplac0d by a continuouff for implmnenting the tasks, Th0 obj0ct ls to
nonlinoar con§trainod optimization for which syntho!lize a scheme to accomplish th0s0 tasks in
solution tc:!chniquos 0xist, Mme optimal mannor, for oxamplo at minimum
Unfortunatoly, tho m0thod do0s not provid0 net preHnt cost, Comlidorablo SAvings may rosult
ruidanco for tho initittl ronoration of th0 sup0r= if pairs of complom0ntary hoatins- and coolins-
structuro, Tho largor th0 suporstructur@, th0 ta1ko can be made to driv0 each othor rather than
groawr will bo th0 numbor of structural pa- roquiro the UH of auxiHary utiliti011,
ramot0rs and tho sizo of th0 oorr0spondinr Of.I• Tho di1croto procoH of selecting th0 oequonce
timization, Th0 problom of assuring that th0 of eneriy intorchangos amons- 1treams and utili-
optimal structur0 has bo0n includod within a ties to bo heated and cooled is accompli!lhed by
givon solution formulation is ono which ari10s r0- dividinr the heat loads of thoso streams into suit-
poatodly in procoss synthe8is ro§oarch, In the ably 1mall, 0qual-siz0 el0montt1. Subject to thermo-
pres0nt ca10, human oxporionc0, rul0s=of-thumb, dynamic temp@rature con1traintt1, a cmit can be
and othor dosign aids ar@ used to sol0ct the ltlli!Ociatod for the 'allocation' (effected by a tiny
oquipmont and intercmnnoction8 to be includod in exchanger) of heat from any element of a 1tr0am
tho 8Uf.lorstructur0, or ut111ty to be cooled to a suitable element to be
Bocau8o of tho ffiH of th0 optimizations l'O= hoated. If it is as1mmed that theH costs are inde-
quirod, tho structural parametor method has b0en pendent of each other, then linoar pros-rammins-
appliod primarily to smallor synthmds sub- mm bo employod to perform an optimal allocation,
probloms (gonorally fowor than ten int0rconnoc- Later, if multiplet adjacent elements for two
tion8) tmch 11 roactor and di8tillation networks, streams aro found to bo paired, tho tiny 0x-
Soveral 8olution technique§ have been investigated changer1 are morgod into a larger cotmtorcurront
including 1imultaneom1 and soquential optimiza- unit. In this mannor, tho amount of heat to be
t1cm of tho 1tructural paramotor8 and othor do- tran1forrod in each exchanifer in the structure
1ign variablos by a varioty of codH. The foaaiblo has also boon made di1crot0,
rogion is ofton multimodal and 10arch mothod1 Differont ldnd1 of structuro1 can be synthe-
1tarting from diiforont points may bo requirod to 1ized dopending on whether the oritrinal streams
ffnd a global optimum. The convergonce rato of 1ro divided into elemonts sequentially, or partly
nonlinoar con1trained optimization is often 8low, in parallol re1ulting in stream splitting, and also
and transformations have boon 1ugg01tod to on how exchangers are merged, In fact, networks
oliminato con1traints on tho 1tructural pa- that involve minimum total exchanror area can be
70 OHIMIOAL llNQl)fl!lUINO l!lDUOATION
synthesized graphically without linear pro- mercial, laboratory, or literature scheme for the
gramming by recognizing that maximum thermo- same or similar product or process, or it may have
dynamic efficiency for the total network results been obtained by the application of a series of
when the temperature driving force is mini- design rules or by some other synthesis technique.
mized in individual exchangers. Some methods use cost and functional analysis or
In addition to problems related to choosing the thermodynamic analysis to detect weaknesses and
size and division of the heat load elements, the suggest structural modifications, whereas others
principal difficulty with the optimal assignment concentrate on systematic reordering of design
approach is that the synthesis problem is not rules, or processing tasks. Two-level Lagrangian
linear and the match costs are not independent of analysis can confirm the advantage of an evo-
each other. Large merged exchangers cost less lutionary change, sometimes without requiring a
per unit area than small ones. Near-minimum complete optimization of the revised design.
area networks generated by this method are Evolutionary methods for structural improve-
characterized by a large number of small ex- ment are quite appealing, possibly because they
changers, whereas minim1un cost networks tend resemble much historical process development.
to have much fewer exchangers. Furthermore, They have been applied to almost all synthesis
some versions of the method automatically ex- subproblems including heat integration, separa-
clude the warmest portions of the streams to be tion systems, and even entire :flowsheets. These
heated and the coolest parts of the streams to be methods, like the embedded superstructure OP-
cooled from integration, assigning them im- timization approaches, are concerned with the
mediately to an auxiliary heating or cooling utility structural optimization aspects of process syn-
which, however, might not be the optimal policy thesis. Their success depends critically, as do
where multiple utilities of varying cost are avail- most gradient-type searches in highly irregular
able.
Recently, it has been suggested that process
synthesis problems be solved by mixed-integer
Evolutionary methods for strudural
linear programming techniques. Discrete vari- improvement are quite appealing, possibly
ables not only indicate the existence of specific because they resemble much historical
equipment and interconnections but also make process development.
possible very good approximations of many non-
linearities inherent in real systems such as fixed
plus variable operating costs. Mixed-integer linear domains, on the nature of the starting point.
programming codes, which generally use branch Given a poor initial design, they will rarely evolve
and bound algorithms to handle the discrete vari- to a truly novel globally optimal :flowsheet.
ables, exist which can handle quite large problems
of this type. The question remains whether it is SYSTEMATIC GENERATION APPROACHES
more efficient to let such a generalized algorithm
blindly find appropriate values for numerous dis- To find new, less prejudiced, or more creative
crete variables or whether specific features of the solutions for synthesis problems, it may not be
synthesis problem and other external procedures desirable to use existing designs singly as the
should be utilized to set combinations of variables starting point for evolutionary approaches or in
known in advance to be feasible and expected to combination for embedded optimization ap-
be near-optimal. proaches. Yet efforts to invent new designs face
enormous combinatorial difficulties arising from
the large number of processing tasks that may be
EVOLUTIONARY APPROACHES required and technologies available to perform
Another class of process synthesis approaches them. Recognizing that it is generally not possible
attempts to systematize the methods and aids that to consider all possible equipment arrangements
many successful engineers are thought to use to simultaneously, a synthesis approach based on a
improve upon initial design concepts. These ap- decomposition was proposed whereby any design
proaches assume the existence of a feasible design problem for which a solution is not known is
and examine it for evolutionary improvement op- broken into smaller subproblems. Such a de-
portunities. The initial design may be a com- Continued on page 96.

SPRING 1982 71
CHEMICAL PROCESS SYNTHESIS technologies which sharply split a single multi-
Continued from page 71. component feed stream into two outlet streams,
composition of the total flowsheet generation one containing some subset of the feed com-
problem leads to the identification of many of the ponents and the other containing the remaining
synthesis subproblems mentioned before such as components. Differences in various physical
heat integration, separations sequencing, etc. in properties of the components such as volatility or
which much progress has been made. The de- solubility are exploited by various technologies to
composition principle might also be applied re- effect the desired separations. If S different
cursively to the solution of each of these sub- separation technologies are available (such as
problems. A special case arises if the decomposi- simple, azeotropic and extractive distillation, ex-
tion is done in such a manner that at least one of traction, fractional crystallization, etc.), the
the resulting subproblems is recognized im- number of different design sequences for the com-
mediately as solvable by some available technology. plete separation of an N-component mixture is
If this process is repeatedly and successfully on the order of SN-1 (2N-2) !/ (N ! (N-1) !) . •
applied until no unsolved subproblem remains,
EDITOR'S NOTES The concluding section of Dr.
the result is equivalent to the systematic specifica-
Siirola's lecture will be published in the next issue
tion of a finite sequence of technologies which to-
of CEE (Summer, 1982).
gether form a feasible design for the original pro-
cessing problem.
The heat integration synthesis problem de- REVIEW: TRANSPORT PHENOMENA
scribed previously is amenable to such a de- Continued from page 93.
composition strategy. At each stage, a stream to chemical reaction. This subject is only given a
be heated and another to be cooled ( either two very brief treatment by Treybal. The theoretical
process streams or a process stream and a utility) discussion in the present work draws heavily from
for which required tasks remain are selected and, results obtained in studies of chemical absorption.
if thermodynamic or other constraints are met, Although there are a few data for liquid-liquid
simultaneously solved with immediately recog- systems, there should be more. The presentation
nizable technology (an exchanger, furnace, etc.). generally falls short of the state-of-the-art in this
Should this heat exchange match fail to perform area.
the required task completely for a process stream Topics covered by Treybal but not discussed in
(reach the desired final temperature or phase the present work include calculational methods for
state), the remaining task, or 'residual', is simply multicomponent extraction, methods for labora-
included among the other as yet unsolved process tory and small-scale extractions, economics of ex-
heat transfer tasks for consideration in successive traction processes, and the competing factors in-
stages. With appropriate available utilities, this volved in the selection of a solvent. In general,
systematic generation scheme will produce a Treybal's book has the same theoretical basis, but
feasible design in a finite number of steps. Differ- a more practical, process-oriented flavor than the
ent designs result from alternative streams or present work.
possible portions of streams selected for each The subject material is probably too specialized
match and from how strictly the various task for the book to find much use in the undergraduate
specifications and other constraints are met. The chemical engineering curriculum. It seems better
number of such designs for heat integration suited as a reference for students and industrial
among M streams is pn the order of ( (M/2) !) 2 • practitioners with a special interest in liquid ex-
Specification of separation sequences for traction. It would have been more valuable in this
multicomponent mixtures is another synthesis regard had the authors included more references
subproblem for which systematic generation ap- to the recent literature. Only 2 % are from the
proaches have been proposed. The problem arises period 1971-1978, with the most recent of these
in reactor feed preparation, product purification, being from 1974.
by-product recovery, waste treatment and other A final minor criticism: the printing quality in
situations where the tasks of increasing concentra- the review copy was noticeably inferior to that
tion or component isolation are specified. In a found in most technical books printed in the U.S.
simplified form, a mixture is to be separated into The type was uneven and generally too light, and
each of its components using a sequence of pages were often slanted from the verticle. •

· 96 CHEMICAL ENGINEERING EDUCATION