TV = numberof cells in pneu-escalator pneu-escalator [ °]

P (N, m) = fraction of particles experiencing mixing m Qv = inclination angle of sectionalizing plate [°J
times through N cells pb = bulk density of particles [g/cc]
5 = cross-sectional area of sampling case a2 = variance of residence time of particles
t = discharge time within pneu-escalator [-]
t = average residence time of particles within <f> = dimensionless time defined by Eq.(lO) [-]
pneu-escalator
tc = average residence time of particles within
one cell Literature Cited
6 = dimensionless time denned by Eq.(5) 1) Shinohara, K. and T. Tanaka: J. Chem. Eng. Japan, 5, 279
Of = inclination angle of the filter of (1972)

ON DESIGN MARGIN FOR PROCESS SYSTEM WITH

PARAMETER UNCERTAINTY*

Takeighiro TAKAMATSU, Iori HASHIMOTO

ANDSUTEAKI SHIOYA
Department of Chemical Engineering, Faculty of Engineeringy
Kyoto University, Kyoto, Japan

A method of determining rational design margin of unit operation in a process is

developed by taking account of parameter uncertainties usually involved in actual
process system design. Since decision variables which are to be determined in process
system design consist of design variables and manipulating variables, design may be
estimated rationally by considering admissible ranges of manipulating variables. The
method proposed in this paper is applied to design of a simple chemical reaction pro-
cess with recycle, and the result obtained by this method is compared with that ob-
tained by the method proposed previously by the authors.

erformance index. At the end of this paper, the design

Introduction ofa chemical reaction process with recycle is described
as a numerical example of the proposed method.
In usual process system design, the nominal design
value is enlarged by a design margin or safety factor Design Margin in Concert with Control of Ma-
for compensation of undesirable effects of many uncer- nipulating Variable
tainties included in design procedure and in system
operation. Quantitative and rational evaluation of Description of problem
design margin, however, has not been discussed suffi- The system equations of a process system can be
ciently, especially for a large, complex process system. described by
p
The authors have proposed a method to estimate Mx, u,M,P)=O
quantitative design margin taking account of param-
eter uncertainty4'5). In this paper a method is devel- where, x, u, M, P,fx denote n-dim. vector of state
oped to determine design margin in concert with man- variables, r-dim. vector of manipulating variables,
ipulating variable. Since decision variables consist of m-dixn. vector of design variables, p-dim. vector of
design variables and manipulating variables design parameters and n-dim. vector function, respectively.
margin may be determined rationally by considering Here, design and manipulating variables are defined
ranges of the manipulating variables. The method pro- as follows; Design variable M represents the variable
posed in this report is simpler in calculation than the which cannot be altered after being determined. On
other method2'3), and also can give some important the other hand, manipulating variable u represents
information in design procedure, such as which manip- the variable which can be changed after the process is
ulating variable strongly affects output variables and constructed.
Output variables from the process /-dim. vector Z
is given by
Z=Ux,u, M, P)
* Received on February 5, 1972
〒606京都市左京区吉田本町
京都大学工学部化学工学教室 高松武-郎 wheref2 represents /-dim. vector function (/<r).

VOL 6 NO. 5 1973 453

 Output vector Z is desired to be a certain specified mum values u and M, the deviation of state variable
value, C. Ax is given from Eq.(l-) as

z=c dx=-i
where C is /-dim. constant vector. Nominal optimum xd
design means that decision variables u and M are where dfjdx represents nXn nonsingular matrix and
determined so that the value of a performance index 9/i/3tt,
dfxldM,
rices, dfJdP represent
respectively. nXr, nXm,
Superscript nXp mat-
- 1 indicates the inverse
/given by Eq.(4) takes the minimum value under the matrix. These matrices are evaluated at P, u, M. Sub-
conditions of given parameter value P and output stituting Eq.(5) into the equation for small perturba-
value C.
tion of Eq.(2), the deviation of output vector z/Z is
given
by the equation
J=g{x, u, M)

If the true value of parameter P is given a priori, \du dx \ax) ou)

It may be enough for practical design to calculate the
nominal optimum decision values u, M based on the
nominal parameter value P. Value of parameter P ^{dM dx\dxl dMJ
included in Eq.(l) and Eq.(2), however, involves an
dp dx \dx dP\
inevitable error of estimation, and also frequently
deviates from the nominal value P during operating +\
period. Usually, a design margin or safety factor is
applied to the nominal design value for compensation where
3/a/9*, dfjdu, df2/dM and 9/2/3P are_/xrc,
of undesirable deviation of output variable or system /xr, /xm, /x/> matrices evaluated at P, u, M, re-
performance caused by such parameter uncertainties. spectively. The range of parameter deviation AP
The problem investigated in this paper is to estimate a is assumed to be given by p-dim. constant vector yj
reasonable design margin AMwhich must be added to from experience or experiment. Moreover, the allowa-
the nominal design value M. ble range of output deviation AZ is assumed to be
Determination of reasonable design margin given by JCL^JZ^//Q, where ACL and ACjj
It is assumed that only the range of parameter devia- express the lower and the upper limits for an allowable
tion, that is, upper and lower limit of parameter devia- ange of AZ. Therefore the rational design margin
tion, is known, and that the probability density func- can be determined based on the relation
r
tion of parameter deviation is not given a priori. If the
probability density function is given, a different meth-
,p. od from that proposed here, such as the statisitical deci- i>'"'UAPk=±r)k, Au<=U} k=\, 2,. . (7)
sion method, may be applied. The equality constraint where Urepresents a domain of admissible control Ju,
Eq. (3) means practically an inequality constraint, that nd AZmust be subjected to Eq.(6).
is, the output Z is allowed to deviate only in a specified The sign ofAPk which is a component ofAP should
direction or range. Moreover the direction of devia- be chosen so that the deviation ofAPk causes an undesi-
a
tion of output Z which has a bad effect on the perform- rable effect on AZt. For example, let us consider the
ance index may usually be known. These assumptions case where the following equations can be written for
might be reasonable in practical process design. Eq.(6),
Then the problem investigated here is to determine AZ1 =a1Au+b1AM+d1AP1 +elAP2
the value of design variable, M(=M-\-JM), so as to AZ2 =a2Au+b2AM+d2AP± +e2AP2 (6) '
fulfill a specified performance under the condition of
parameter devation in a given range. The principle of and where the allowable ranges of AZ± and JZ2 are
given by
determining design variabe M(-M-\-dM) is to mini-
mize the performance index which is evaluated when
the process is designed so as to satisfy a specified per- (7)'
formance under any given conditions of parameter
deviation. ACL may usually take negative value for many cases.
This problem can be solved as follows by using a Moreover, the constraint for AZ sometimes needs
first-order approximation which is convenient to eval- either the upper or the lower bound only. The sign of
uate the effect of parameter deviation on the perform- AP is uncertain, and it cannot generally be known
ance index and the output variables. The first-order hether AP takes negative or positive value. Only the
approximation is evaluated at nominal value ofparam- absolute value of /IP can be estimated. In some cases,
eter,.P~and design margin AM is determined as below t may be clear that P changes dynamically in one
based on P. direction from the nominal value of P.
When parameter deviates by AP from the nominal Assuming that d± and e± take positive values and that
value P, and decision variables u and M are also moved AZX must not be over the positive value of ACU±, the
by Au and AM respectively from the nominal opti- sign ofAP1 or AP2 must be positive. Assuming that d2
w
i
454 JOURNAL OF CHEMICAL ENGINEERING OF JAPAN
 and e2 take negative value and positive value, respec-
Reduction Stream;
tively, and that AZ2 must be greater than the negative Mo,Cap Fi ^R
value ofJCL2, the sign ofJP± and AP2 must be positive
and negative, respectively. Practically it does not hap-
FnshFed I I r1^ 'FiXR
pen that AP2 takes simultaneously positive value for 1
AZXand negaive value for AZ2, but the above treat- Reactor 1 Faq-F/ Separa tor
ment gives a result in the direction of safety. In general, RecycleStream ^^
0 represents a negative effect for adding some margin
to the nominal design values. Here, 0 is defined as the contains noR X Y.A B
first variation of the performance index J due to Au Fig. 1 A simple chemical reaction process with recycle
and AM. The physical meaning of this definition is to
determine the margins of design and manipulating
variables so that the deviation ofJ caused by AM and k IkY
Au becomes as small as possible. Then 0 is defined by
_dg(df, f dg
~\dM dx\dx
X Fig. 2
Y
Denbigh's reaction

+|du dx\Bx) 3aj K' Table 1 Given parameter values and boundary
conditions
where dg/dx, dg/dM and dg/du are n-dixn., m-dim. and
r-dim. row vectors evaluated at u, M and P. In Eq.(8)
mol//,
the coefficients multiplied to AM and Au are not in kB=OA= hr-i,
F^0 100 mol/hr,
C1 =70 mol/hr,

do=0.1 hr-i,
kx=0.02
C2=0.005mol//,
$6=0.01 &//,
general equal to zero. Eq.(7) subjected to Eq.(8) is
reduced to a linear programming (L.P.) problem be- Table 2 Nominal optimum design
cause 0 is a linear function of AM and Au. Hereafter, a Fi=320, 7= 1.27 x01*, S=0.921, /3=0.278
solution of the L.P. problem is assumed to exist. After ^=0.179, xb=0.552, *x=0.020, *F=0.030
iciz=0.219, J=154$
ll, a reasonable design margin AMmay be determined
from the solution of the L.P. problem given by Eqs.
(6)-(8). The design margin obtained by this method partially recycled at the ratio of a to the amount of
(denoted by Method I) may be more rational than that A and B in the inlet flow to the separator, and Xand Y
by the method proposed previously by the authors4'5) are also recycled at the ratio ofp. Denoting the frac-
(denoted by Method II) because the manipulating tion of molar concentration Q [gm°ty^] of each compo-
a
variables admissible in a certain specified range are nentito CA0 (Q=C^5 i=A, B, R, X, Y) by xi9 the
considered. In fact, if the deviation Au ofa manipulat- hourly molar flow rate of fresh feed by FA0 [mol/hr],
ing variable is sufficient to compensate for the undesira- the flow rate of the inlet flow into the reactor which
ble parameter uncertainties and also minimizes 0, the consists offresh feed and recycle flow by Ft [mol/hr]
addition of a design margin to the nominal design ( (volumetric flow rate) X CA:o=Fi), and the volume of
variable is not always necessary. Moreover, it is also the reactor by V, the process equations for steady state
shown from this consideration which manipulating are given by
variable has a strong effect on the values of output FAa+aFiXA-FiXA- V{kB+kx)CA0 xA=0
variable and performance index.
aFiXB-FiXB+ VCAa {kBXA- {kR+kY)XB} =0
Application to Design of a Simple Chemical pFtxz-FiXx + VCA0kxxA=0
Reaction Process with Recycle
PFtXy-FiXy + VC^ kyX^O , (9)
Nominal optimum design at steady state -FixB+ VCMkBxB=0
Let us consider the design margin of the reactor
XA+Xb+Xr+Xx+Xy=1
in a simple chemical process with recycle shown in
Fig. 1 as an example of the method proposed in this where kt (i=B, R, X, Y) [1/hr] shows the reaction rate
paper. The reactor is assumed to be a single C.S.T.R. constant for each elementary reaction. These equa-
with perfect mixing and the type of reaction is assumed tions correspond to Eq.(l) for general expression.
to be Denbigh's reaction, which comprises first order Output state Zx and Z2 are defined as
elementary reactions as shown in Fig. 2. R is the desired Z1 =FixB
product, and this is produced from fresh feed contain- Za= (xX+Xy)CM (10)
ing only A (molar concentration CAo [gmol//]). The
effluent flow from the reactor containing A, B, R, X Zx is the amount ofR produced hourly and Z2 is the
and Y is assumed to be separated into a product flow molar concentration of undesired products X and Y.
containing only R and another flow containing A, B, t is assumed that the values ofZx and Z2 are desirable
Zand 7. Zand Fare undesired products. A and B are to be maintained as
I
VOL 6 NO. 5 1973 455
 Table 3 Rational design margin for parameter uncertainty
Design margin
Allowable ranges of AZ±, AZ% --s :
by the method proposed here by the p
case £t $2 £3 AV * Aa Ap AJ AV w

1 1.0 2 x 10-4 10.75 x 10-4 1097 -0.018 -0.278 1.26 565 71

2 1.0 2 x 10-4 20 x 10-4 233 0.0022 -0.278 0.0095 565 71
3 1.0 2 x 10-4 22.5 x 10-4 0 0.0076 -0.278 -0.326 565 7.<
4 1.0 4x 10-4 20 x 10-4 233 0.0022 -0.278 0.0095 565 71
5 1.5 2 x 10-4 20 x 10-4 150 -0.0008 -0.278 -1.01 365 6.6

where \AkB\ ^0.08, [Akx\ ^0.04, AZ±^-£±, ^^AZ^>-^Z

Z2=C2 (15)
Eqs.(lO) and (ll) correspond to Eqs.(2) and (3), The design margin AV can be determined by solving
respectively. For simplicity the following objective he L.P. problem, which is composed of Eqs.(6)-(8)
function is assumed for the numerical calculation. and Eqs.(13)-(15). The result is shown in Table 3
t
j__/volume ofVunit cost \ , f/recycle rate\ for various allowable ranges of AZ± and AZ2. The
J \reactor /\ofreactor/ ' (\ofAandB) design marging calculated by the method presented
_j_/recycle rateN) /unit cost \ in the literature4) (Method II) which does not take
^VofXand Y ))Ifor recycle/
account of operation of manipulating variables, are
" =V$b+ {aFi(xA+xB)+^Fi(xx+xY)} $r (12) also shown in the right column of Table 3, In the case
Practical performance index Jmight be more complex, by Method II, design margin AV cannot compensate
but the above simplification may be enough to illust- for undesirable deviation AZ2 because the coefficient
trate a numerical example of design margin problem. multiplied to AVin the equation of AZ2 is positive and
This form ofJis referred to in the literature1). AZ2 is greater than <J2 even ifAVis equal to zero. That
Nominal optimum design is to determine the values is, by Method II, there is no feasible solution of the L.
of design variable V and operating variables a and ft P. problem composed of Eqs.(6)-(8) and Eqs.(13)-
which minimize the value of performance index Jgiven (15). Then the design margin AVby Method II is de-
by Eq,(12), subjected to Eqs.(9), (10) and (ll) under termined so as to satisfy only the constraint ofAZ±.
the specified values of FA0, CAo, Cx and C2. The From this calculated result the following points are
values of parameters and the boundary conditions concluded, i) Comparing the method proposed in
used in the numerical calculation are given in Table 1. this paper (Method I) with Method II, the design
The results of numerical calculation are shown in margin z/Fand the increment AJ of performance index
Table 2. The following can be recognized from the Jby Method I are smaller than those by Method II in
results; i) p is not equal to zero. Partial recycle ofX each case except for case 1. For example, in case 2,
and Yis more effective than no recycle ofX and Y be- AVby Method I is 233 /, which corresponds to about
cause the output molar concentration ofX and 7, that 2% of nominal value V. On the other hand, AV by
s, Z25 is specified, ii) The optimum recycle rate is 320 Method II is 565 /, which corresponds to about 4.5%
moles/hr in this example, and the optimum retention f nominal value. AJby Method II is about 350 times
time in the reactor is about 4 hr based on CAo. as much as that by Method I. In case 1, AJby Method
Design margin for parameter uncertainty II is about 3 times that by Method I, in spite of the fact
i
For simplicity, the rate constants kB and kx are that AVby Method II is smaller than that by Method
assumed to involve the maximumestimation error of I. Moreover, the deviation of output Z2 due to uncer-
7]° % for absolute values of the nominal values kB and ainty ofkB, kx cannot be compensated by Method II,
kx, respectively. and the deviation AZ2 is about 3 times f2. These re-
sults suggest that Method I is more reasonable than
\JkB\^kB,
100 MM,ssi*
Method
II. i ) It becomes clear by comparing case 2
and case 3 that manipulating variable /3 can sufficient-
The value ofrj° is.taken as 20 in this example. It is ly compensate for the effect of AkB and Akx so that
assumed that the allowable ranges for deviation of out- AZ2 is smaller than <f2. It also becomes clear by com-
t
o
put variables Zx and Z2 are given by paring case 2 and case 5 that the estimated design
margin AV decreases if fx becomes smaller, iii) If£3 is
4Zi^-£i, f2^jz2^-f3
greater than or equal to 22.5 X 10"4, the design margin
The movable ranges of Aa, Aft (deviation of manipu- AVis zero, as shown in case 3. Thatis, a and /3 can suf-
lating variables a and /3) and AV (margin for design ficiently control JZ2 within the specified allowable
variable V) are constrained by ranges of AZ2.

456 JOURNAL OF CHEMICAL ENGNEERIING OF JAPAN

 u = r-dim. column vector of manipulating variable
u = nominal optimum value of u
Conclusion Au =ådeviation of u from u
V = tank volume [/]
A method to determine rational design margin tak- V = nominal optimum value of V [/]
ing account of parameter uncertainty is developed for AV ==design margin of V [/]
xi - molfraction of each component i to Cao
practical process system design. The design margin is [i=A, B, X, Y, R] [-]
determined by minimizing the deviation of a per for- X = undesired product
mance index under the allowable limitation of the Y = undesired product
$& = hourly unit cost of reactor [$///hr]
deviation of output variables. An altered form of the $r = unit cost of recycle ofA, B, Xand Y [8/mol]
performance index, however, may have to be consid-
ered in the future. The method proposed here is ap- a = recycle ratio ofA and B [-]
plied to design of a simple chemical reaction process jS = recycle ratio of X and Y [-]
£i = allowable upper or lower limit of output variable
with recycle. Numerical calculation shows that this inEq.(14) (i=l,2, 3)
method is more rational and practical. 7] =p-dim. vector of the range of parameter deviation
This work was partially supported by The SAKKO- 7j ° = relative range of parameter deviation in Eq.(13)
rjic = k-\h.element of vector r)
KAI FOUNDATION.
<Subscript>
A = reactant
Nomenclatur e B = intermediate product
R = desired product
A = reactant X = undesired product
B = intermediate product Y = undesired product
JCl = allowable lower limit of the deviation < Superscript >
of output vector - 1 = inverse matrix
ACu = allowable upper limit of the deviation
of output vector
d = molarconcentration of z'-th component Literature Cited
(i=A, B, R, X, Y)
Fao - moleflow rate of fresh feed 1) Levenspiel, O. : "Chemical Reaction Engineering", John Wiley
Fi = output flow rate from reactor Son., 136-137 (1962)
hi =.reaction rate constant (i=B, X, Y, R) 2) Rohrel, R. A. and M. Sobral,Jr.: IEEE Trans, on AC, AC-10, 1,
M =m-dim.column vector of design variable 43-48 (1965)
M =nominaloptimum value ofM ) Rudd, D. F. and G. C. Watson: "Strategy of Process Eng.",
AM = deviation of design variable from M Chapter 1 1-1-5,John Wiley Son (1968)
P = jfr-dim. column vector of parameter ) Takamatsu, T, I. Hashimoto and H. Ohno: /. E. C. Process
P = nominal value ofP Des. & Dev., 9, 3, 368-378 (1970)
AP = deviation of parameter from P 5) Takamatsu, T., I. Hashimoto and S. Shioya: Systems and Control
R = desired product of Japan, 15, 2 (1971)
3
4
Short Communications
A NOTE ON THIN FILM EVAPORATION
PREDICTION OF HEAT TRANSFER RATES*

V. Narayana MURTHY and P. K. SARMA

Department of Mechanical Engineering, Andhra University, Waltair, India

»l =3.8x lO"3^0-^0-65
Introduction and Analysis
The data obtained are also compared to the analysis
Chun and Seban2) experimentally investigated the of Dukler3) and it is observed that the deviation be-
falling-film evaporator under turbulent flow condi- tween the experimental results and the theoretical
tions of the liquid film. The data obtained are corre- analysis is considerable. Dukler3) obtained heat trans-
lated by the following non-dimensional equation for fer rates in falling films by numerical integration, and
the turbulent flow of the film. the analysis did not offer any straightforward equation
to facilitate ready computation. The present note at-
* Received on April 2, 1973
K. Sarma
tempts to establish an equation for heat transfer coeffi-
Department of Mechanical Engineering, College of Engineering, cient in explicit form making use of some of the availa-
Andhra University, Visakhapatnam-5300 1 3, India ble hydrodynamic aspects of falling liquid films. The
P
VOL. 6 NO. 5 1973 457