Steady State Chemical Process Simulation
Steady State Chemical Process Simulation
SCOPE
The objective of this study is to investigate the different a rational topology supplied by the user, sequences the
techniques for process simulation. It may provide valuable computations, and supervises data input and output of
introductory information to potential users, prospective results as well as the iterative recycle computations.
developers, or beginning research workers in the field. A complete simulator combines expertise from three
The development of simulation applied to chemical areas. The preparation of accurate mathematical models
processes began in the mid 1950’s and the first simulator of chemical processing units and good correlations for
was published in 1958. In the mid 1960’s the most popu- thermodynamic and physical properties is an engineering
lar current structure for process simulation, namely, the problem. The design of efficient input and output pro-
modular program, was crystallized. According to this cedures and program/library organization is a data proc-
approach each chemical processing scheme can be repre- essing problem and the iterative calculations are numerical
sented by linking together a set of standard mathematical problems.
unit modules in arbitrary and flexible networks.
The main components of a modular simulation pro- Several simulation programs are in routine use in service
gram are: the unit module library of subprograms, the bureaus and largely within companies which have devel-
thermodynamic and physical property estimation library oped them. In the field of education they are used to
(TP), and the executive program with its library of teach process simulation, in support of teaching other sub-
services. The unit module subroutines compute physical jects such as stoichiometry and unit operations, and in
outputs for each standard processing operation given the the preparation of design projects.
physical inputs and the design (or decision) parameters. Recently, most research has concentrated on the areas
The necessary TP data are supplied automatically to the of dynamic simulation and the optimization of process
unit modules with some user control over the estimation simulations. The research literature of the last three years
optiom. The executive program manages the data libraries is reviewed and the directions for further progress are
(data bases), connects between unit modules according to noted.
Classification and Scope The design of a process simulation program has three
Since the modular steady state process simulation pro- major aspects, which may be classified as follows:
gram is the most widely used, the discussion is limited to 1. Data processing. All of the tasks dealing with the
this type. In addition, as we are interested in the problems input data itself, the best methods to test, store, and re-
of preparation and use of these progams,'we do not give trieve them, and the possibility of changing the data in an
details of their historical development nor the charac- interactive environment are in this category. The Struc-
teristics of existing program. Discussion on these subjects ture of Simulators discusses various aspects of data proc-
may be found in the general references summarized in essing.
Table 1. This table may also be useful for someone who 2. Engineering models. The preparation of an accurate
wants to be more familiar with the terminology, the struc- mathematical model of the chemical process units and
ture, and the techniques which are used in process simula- preparation of good correlations for thermodynamic and
tion programs. physical properties requires in-depth knowledge of the
In a typical process simulator the input data contains chemical processes. In Unit Module Algorithms and Ther-
the process topology, design and operating parameters of modynamic and Physical Properties the different ap-
the units, and the composition and state of the process proaches to engineering models are discussed.
feed streams as well as perhaps an estimate of any re- 3. Numerical procedures. These situations appear in
cycle streams. It is read and checked by the input sub- the computation-ordering algorithms, in the recycle cal-
routine of the process simulator. The input data is trans- culations, and also in the iterative solution of the more
ferred to the executive program which determines the complicated unit modules. This aspect is discussed in
order of calculation. Then the executive program calls the section on Computational Efficiency.
the unit module subroutines according to this order, and The user of a process simulation program may face a
passes the required input information to these sub- number of difficulties. Sometimes the physical property
routines. When one or more recycle streams are detected, data for some of the components used in the simulation
the composition of which cannot be known in advance, are lacking. In addition, the different word sizes, operat-
the executive program gives an estimated initial value ing systems, and compilers used on different computers
for the composition and state variables of these streams may lead to difficulties in using the same simulation pro-
and proceeds to obtain the correct values for these streams gram on these different computers even when the pro-
by some iteration method. gram is written in a high level language.
The unit module subroutines calculate the physical In Experience with Simulation Programs, we discuss the
output from the units when the physical input and the uses of chemical process simulation in industry and edu-
unit design parameters are known. For this calculation, cation, and we explore the problems which limit further
the unit module subroutines needs physical and thermo- extension of their use.
dynamic data. These data are obtained from the thermo- In Future Development, we propose some possible solu-
dynamic and physical property package which contains tions to the problems raised in this paper, and we forecast
estimation parameters for the required properties. (Esti- the direction of future development in the field of chem-
mation parameters are those pure component parameters ical process Simulation.
used in correlation equations for physical property rep-
resentations, for example, the Antoine coefficients for va- STRUCTURE OF SIMULATORS
por pressure, critical properties for use in equation of General simulation programs have been constructed in a
states, etc.) After the chemical process has been com- number of ways. Although they all use unit modules for
pletely calculated, the output unit converts the large their basic calculations, they can be viewed as having one
amount of information obtained into the form of an easily of two basic internal structures-fixed structure or vari-
readable technical report. able structure. In fixed structure programs, the executive
The preparation of a computer simulation is usually program is exactly the same no matter what is being simu-
controlled by the requirement that the program be able to lated. The matching of streams to unit modules and the
solve most process problems with tolerable accuracy for path of calculations through the program is determined by
a reasonable cost in user and computer time. The state of data supplied by the user. In variable structure programs,
the art is such that it is often possible to perform a large a different executive program is written or generated for
each different flow sheet. If the executive is written by
number of simulations and still achieve both of the above the user in FORTRAN, this will generally consist mostly
requirements. However, there are process simu1at:on situa- of calls to the various unit modules or subprograms. Al-
tions where this is quite difficult. Still, the judicious use ternately, a problem oriented language (POL) may be
of a simulation package can greatly increase engineering used which is interpreted directly into executable code, or
productivity, although at higher cost per person-hour. the POL may in turn generate a procedure-oriented lan-
The cost per project and the elapsed time, however, can guage (often FORTRAN) which in turn is compiled,
be substantially lower. linked, and executed.
-A-
lem being solved.
A variable structure program that uses a POL and
vq!)
generates a FORTRAN main program has a number of ad-
vantages. Only those modules actually needed are loaded,
FORTRAN statements may be added directly by the
user, and data storage space may be allocated only as
needed by the simulation. In return for the extra over-
head that the translation, compilation, and linkediting ~
Maximal system
Authors 5Pe Programmed solved
A better estimate for 2 according to the Quasi-Newton where E is a small tolerance which must be supplied by
method will be the user.
- -
- + tZif( Some authors have found (Orbach and Crowe, 1971;
zi+l = zi Zi) (5.6) Friedman and Pinder, 1972; Shacham and Kehat, 1973)
- that none of these tests are satisfactory. The user gen-
where t is a scalar damping factor and fl an approxima-
- erally knows what accuracy he needs for engineering pur-
tion to the negative inverse of the Jacobian matrix $/az. - -
poses, but he cannot specify this - as -the error tolerance
The iteration process is usually begun with an identity because the value of f (zi)= ~i - zi is a function not
matrix for?? and t = 1, that is, the direct iteration method only of the real error but also of the rate of convergence
- of the iterative method. The penalty for specifying inac-
(5.1). In every subsequent iteration, the H matrix is curate values of the error tolerance is either excessive
corrected according to 5.7 computation time or inaccurate answers.
- - - - -
- -
Hi+l = Hi - (ti% + -- -s T -H ~ -/s --
~ i y i ) ~ ~ l i y i(5.7)
Klaus and Moore (1972) and Shacham and Kehat
(1973) have found ways of bounding the solution and to
where measure the exact error independent of the solution
- method, but only for problems with one variable or one
yi = f ( z i + l ) -TCZi)
- =-
Pi Zi+l
-
- Zi
tear stream. Orbach and Crowe (1971) have used an
estimation of the maximal eigenvalue of the iteration ma-
trix for improving the accuracy of the error test.
and S T is an arbitrary vector. More research is needed to ensure that the user will
Broyden (1965) has chosensT = TTbut other choices obtain answers with exactly the required degree of ac-
are possible (Kehat and Shacham, 1 9 7 3 ~ ) .The ,Quasi- curacy without excessive computation time.
Newton method may be very efficient for the solution of
recycle systems (Genna and Motard, 1973; Batstone et al., EXPERIENCE WITH SIMULATION PROGRAMS
1971) but there are also counter examples (Batstone et al.,
Probably the most important experience gained in the
-
1971). The main drawback of this method is the large
storage requirement for the matrix H . This method is also
last few years is that process simulation is not an end in
itself. It is merely a new tool for the chemical engineer
more di5cult to program than the other two. since it cannot replace engineering judgement. Some ef-
Table 7 summarizes the different simulation programs fort is required to make the best use of this tool.
and their method of convergence acceleration. This table So far, the degree of acceptance and application of proc-
also shows where one may find examples using the different ess simulation in industry has varied widely. Industrial
methods. users are generally stubborn, uninterested in partial solu-
More research is needed to improve the Successive tions, and require systems that are flexible and reliable
Over Relaxation (SOR) type methods (Wegstein, Domi- using familiar techniques. Further, not all process engi-
nant Eigenvalue) . The Quasi-Newton methods must be neering jobs require a simulation program-some merely
require refurbishing an old solution manually.
carefully investigated and require further refinement be-
fore they can justifiably replace the SOR-type methods. In education, the major barriers to wide acceptance of
simulation seems to be the availability of reasonable data
Convergence Testa and cost.
Accepted methods for testing the convergence of the Industry
recycle system are the absolute error test [Equation The development and acceptance of process simulation
(5.8)1, and the relative error test [Equation (5.9)] by the chemical industry is a continuous process which
started in the middle 1950’s. Di Bella (1974) divides the
(5.8) history of process simulation into five-year periods.
The period 1955-1959 was characterized by the de-
(5.9) velopment of computer programs for individual unit op-
erations. The developer of these programs was required
Simulation
program Reference Maximal system solved Integration method
We would expect substantial developments in the area computer programs for equipment design will continue to
of dynamic simulation in the near future using the more be independent. However, the transition to using one set
sophisticated integration methods and faster digital com- of programs to the other will be made simpler.
puters to reduce the cost of dynamic simulation. The most important directions in hardware development
will continue to be increased execution speed of the com-
Computer Software and Hardware Developments
puter and a reduction in cost of different components.
One aim of computer software development is to solve The first of these developments will expand the areas of
more complicated problems with maximal efficiency, with application of process simulation, the second will increase
minimal effort and knowledge from the user. For this the number of users. In particular, the reduction in cost
reason, higher level languages (FORTRAN, PL/1) were of main memory will probably obsolete the current em-
developed and are much easier to use than Assembly phasis on virtual storage technology for large applica-
Languages. Subroutine packages and problem oriented tion programs and in the future the user will have avail-
languages (POL) have been developed for efficient solu- able a very large memory to handle large simulations
tion of specific problems. The user of CSMP (IBM, 1969) efficiently. The reduction of the price of terminals and of
need not know either programming or numerical analysis graphic units makes these tools available to more and
for the solution of complicated systems of differential more engineers. Direct-view storage tube graphics is re-
equations. volutionizing the cost-effectiveness benefits of interactive
The development of process simulation programs has graphics, although its use in process simulation has yet
just begun to move in the same direction. So far, two to be demonstrated.
POL based simulation programs have been developed There are several other developments in computer
(Nott, 1971; Hanyak and Klaus, 1973). A program for hardware whose influence on process simulation have not
converting material and energy balance equations into yet been investigated. Such development; are in parallel
a FORTRAN program is available (hlah and Rafal, 1971; processing CPU units, minicomputer and microcomputer
Soylemez and Seider, 1972). The FLOWTRAN system enhancement, microprogrammed machines, and intelligent
includes a precompiler which translates the process struc- terminals. Other developments like distributed process-
ture supplied by the user into a FORTRAN program. ing, network computing, and certified scientific software
In the future, the user of simulation programs will have are bound to have an impact, but more research is needed
to know less about computing methods, languages, and to exploit these areas.
the structure of the simulation programs. A brief train- We may expect that many of the traditional duties of the
ing period will be sufficient to provide the chemical en- engineer will be fu!filled by the computer, and develop-
gineer with the ability to use process simulation. The ment in this direction has already begun (Leesley, 1973;
development and maintenance team will have to be much Tsubaki, 1973). Growing out of the present state of proc-
more expert in numerical and optimization methods, in ess simulation and design is the complete automation of
programming and job control languages, in addition to the the process engineering effort. The movement into design
classical chemical engineering subjects (thermodynamics, automation for the production of process hardware specifi-
unit operation). In this way, the simulation program will cations, bills of material, plant layout, and final engineering
remain up to date with other developments in the compu- construction drawings is gaining considerable momentum
tational field. in the chemical engineering industry in contrast to the
I n the industrial process simulation field, a major effort aerospace and computer or electronic engineering indus-
will probably concentrate on extending correlations for the tries where it has already had considerable impact.
physical properties handled by the simulator. Present
simulators handle nonpolar fluid components adequately. ACKNOWLEDGMENT
Future developments will include the ability to handle
rigorously solids processing (including polymers), solids Preparation was supported in part by National Science
handling, and polar components. Such developments in Foundation Grant GK-36525X.
physical property correlations will of course spawn nu-
merous developments in adding unit operations modules, NOTATION
for example, solids handling, etc. a = parameter
Although process simulators will continue to add at f = function
least rudimentary equipment design functions, the con- H = estimate of the negative inverse Jacobian matrix
sensus is that computer programs for process design and i = subscript index
Separation Calculations by the Modified Form of Newton’s Wells, G. L., and P. ‘M. Robson, Computation for Process En-
Method,” Chem. Eng. J., 3, 176 (1972). gineers, Wiley, New York ( 1973).
Orbach, O., and C. X I . Crowe, “Convergence Promotion in the Westerberg, A. W., “Decomposition Methods for Solving
Simulation of Chemical Processes with Recycle-The Dom- Steady State Process Design Problems,” p 379 in D. M.
inant Eigenvalue Method,” Can. J. Chem. Eng., 49, 509 Himmelblau ( e d . ) , Decomposition of Large Scale Problems,
(1971). North Holland/American Elsevier, Amsterdam ( 1973).
PACER 245 User Manual, Digital Systems Corp., Hanover, , and C. J. Debrosse, “An Optimization Algorithm for
N. H. (1971). Structured Design Systems,” AIChE J., 19, 355 (1973).
Parmelee, R. P., T. I. Peterson, C. C. Tillman, and D. J. Hat- , and F. C. Edie, “Computer Aided Design, Part 1,
field, “Virtual Storage and Virtual Machine Concepts,” I B M Enhancing Convergence Properties by the Choice of Output
Systems I., 11, No. 2, p. 99 (1972). Variable Assignments in the Solution of Sparse Equation
Perry, R. H., and C. H. Chilton, Chemical Engineers’ Hand- Sets,” Chem. Eng. J., 2, 9 (1971a).
book, 5th Edit., pp. 2-85, McGraw-Hill, New York ( 1973). , “Computer Aided Design, Part 2, An Approach to
Peters, N., and P. E . Barker, “An Appraisal of the Use of Converqence and Tearing in the Solution of Sparse Equation
PACER, GESICS and CONCEPT for Chemical Plant Sim- Sets,” ibid., 2, 17 (1971b).
ulation and Design,” Chem. Eng., (London), 283, 149 Wilson, G. M., “Calculation of Enthalpy Data from a Modified
(1974). Redlich-Kwong Equation of State,” Ado. Cryogenic Engi-
SCOPE
I n recent years, developments such as sophisticated The expected value is not always suitable for evaluating
optimization methods and process synthesis techniques chemical processes since adverse deviations from the
have significantly advanced the procedures of process de- average might mean unacceptable or even hazardous per-
sign. However, the final stage of design, that of specifica- formance. Also, these methods have only been applied
tion of overdesign or safety factors, is still largely a sub- to a few processes where the statistical behavior of a
jective procedure based upon experience. process performance criterion is known or can be obtained
analytically.
I n general, safety factors are increased as the uncer-
The purpose of this study is to demonstrate how over-
tainty of the process variables increases; a procedure that
design can be achieved by the use of a process depend-
can be expensive if safety factors are specified too small ability criterion and stochastic simulation procedures. A
or too large. Recent studies to quantify overdesign repre- simple process involving a reactor and separator was
sent the uncertain parameters as random variables and chosen for study. This process has both variable conditions
compute the overdesign based upon the expected value and uncertain data, with two design variables, reactor
criterion. volume and number of separator stages.