Professional Documents
Culture Documents
T h e treatment of variational problems has received con- can affect the temperature profile only indirectly through
siderable attention in the chemical engineering literature the mechanical design of the heat transfer system-and
in the past decade. These can be classified as control the process-variable-based objectives do not always corre-
or design problems. I n chemical engineering, one would spond to economic objectives. McEwan and Beveridge
expect the most useful results to come from the solution (1965) offer an exception. Hence, it was decided to formu-
of the design rather than the control problems, because late the problem in terms of design variables and economic
in controlling a process one can, through feedback, correct objectives.
and control by specifying the control algorithm (the func- Annable (1952) compared the performance of an
tional relationship between corrective action and error) autothermal ammonia synthesis reactor (Figure 1) with
and not the control action function-e.g., valve stem posi- the maximum yield that could be obtained if one had
tion us. time. The solution to the variational control prob- direct control of the temperature profile. He found that
lem would yield the control action function, and while
there are cases-e.g., batch processes-in which this would
be useful, in most process control problems we are free
to exert corrective action after the process has started
and error has been measured. This is obviously not true
in design problems; the configuration is first set and then
the process is operated. This situation is more compatible
to the nature of variational problem solutions. Thus it
was decided to study variational problems that arise in
design.
Beginning with Temkin and Pyzhev (1960) and Denbigh
(1944, there has developed a considerable literature (Aris,
KFJ
1961) on the processing conditions under which a chemical
reaction should be carried out in order to obtain maximum Heat Exchanger
yield or selectivity. Although this work is of great value,
there are two shortcomings: The process variables such
as temperature, pressure, and concentration are not under
direct control of the designer-for example, the designer
Product Gas 4 Feed G o s
2 0 0 -350AT M
N H 3 = I 7-22Oo/
' Present address, Nippon Light Metal Co., Ltd., Tokyo, Japan.
' Present address. University of Natal, Durban, South Africa. Figure 1 . Ammonia synthesis reactor
Ind. Eng. Chern. Process Des. Develop., Vol. 9, No. 4, 1970 503
conversion could be increased by 14%. Obviously, one \
\
does not have direct control of the temperature profile, \ Equil. Temp
but it could be affected by the configuration of the heat
transfer surface-e.g., added insulation and/or fins.
Indeed, Dyson (1965) considered the general problem
of determining the heat transfer coefficient us. length func-
tion that would maximize the yield in an autothermal
reactor.
In this paper, the Maximum Principle is applied to
the design of a variable heat transfer coefficient, ammonia
synthesis reactor. Economic factors are included in the
analysis. A simplified computational procedure is de-
veloped and results are presented in graphical form to
permit design of reactors of this class with little additional
computation. I t was found that the yield could have
been increased by 5.4%,had the TVA reactor, as described
by Baddour et al., (1965) and Eymery (1964), been de-
signed with a variable heat transfer coefficient.
In addition to being a contribution to the specific field ) IA IB Jc
of ammonia synthesis reactor design, it is hoped that
the paper contributes to the more general field of economic
)
- Distance
Bottom
autothermal reactor design. Figure 2 . Schematic of temperature profiles and NHI mole
Simulation fraction in TVA NH3 synthesis reactor
504 Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4,1970
where
-1H = heat of reaction, kcallkg mole of Nz
S? = cross-sectionalarea of catalyst zone, m
R = reaction rate, kg moles of N r / (hr m)
C,. = heat capacity of reacting gas
hlaterial Balance. Referring to Figure 3, consider the
incremental distance in the catalyst zone. Performing a
Y 2material balance yields
dX i
GI=- =1
ds
where the variable, s, is to be viewed as a general distance
where N\ = mass flow of nitrogen per unit area of catalyst along the path or extent of the phenomena (In this
zone, kg moles/ (m. h r ) . particular problem it is identified, as shown in Equation
On a semiempirical basis, the Temkin-Pyzhev (1960) 9, with the dimensionless reactor length) and where
equation for the ammonia synthesis is given as
Kc = L G
K1 = S1.L W * C p ,
K? = Si*LW.Cpf
where Kj = ( - A H * S ? ) / ( C . C p , ) ( LTo)
f = catalyst activity
Results. The temperature profile computed is compared
P H p\,
9
with the measured plant profile in Figure 4. The data
P L H= partial pressures of H?,N r ,and N H
used in the simulation are presented in Appendix A.
K , , K , = rate constants; K , = f 6exp. ( - E , I R T )
Reduction of Differential Equations to Dimensionless Design Problem
Form. Using the dimensionless groups Objective Function. Let:
V1(X0,X I ) = value of the product, considered t o be a
function of the composition, X n , and the
temperature, XI, only, the pressure
being regarded as fixed
V 2 ( X 2 )= value of the feed gas, considered to be
a function of the temperature, X r , only,
the pressure and composition being
where
regarded as fixed
C; = total flow at top of reactor, kg molesim hr C , ( X , ) = annual cost of the reactor, considered to
T,, = reference temperature be a function of the length, X i , only,
L = reference length the pressure being regarded as fixed
The system equations can be written as The following objective function is used in this study.
2001
To P
I :o 2:o 3.. 0 410 i.0
Bot tom
Distance ( M I
Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970 505
J = [Vi - V?- C,]/n Z 2 = unit value of
(10)
feed gas
I dZ2
- = -[Z1K1 + Z?Ke]u(l4c)
where n is a normalization constant to be evaluated a t \ ds
a later step. temperature
Formal Statement. MAXIMIZE. J = [VI - V 2 - C , ] / n
By optimal choice of
Zn = unit value of dZ,
-=1
reactor length f ds
u(s) = heat transfer coefficient as a function of axial
position
According to the Maximum Principle (Rozonoer, 19691,
X L ( S ) = inlet feed gas temperature
if J is to be maximized, H ( s ) must be extremized with
X , ( S ) = reactor length
respect to u(s) a t each point, 0 5 s 5 S.
Subject to Equations 6 through 9 For convenience we define
0 5 u(s) 5 u* (104
K H [Z1K1 + Z?K?] (15)
From Equation 13 H is linear in u and the coefficient
Xi(0) = X2(0);feed and reacting gas of u is KH(Xi - X2).
temperatures must be equal at top (lob) Since ( X , - X , ) > 0 , 0 5 s 5 S, the general design
X,,(O)given (10c) policy is as follows:
If
Xi(S) = kiX*(S)+ h2[Xo(0) - Xo(S)];
KH < 0, then uopt = u*
over-all enthalpy balance (10d)
KH > 0, then uopt = 0 (16)
KH = 0, then uopt is not specified by the maximum
where principle-Le., extremization of H .
This last case, in which KH = 0, is known as the singular
case. I t is of interest only if it persists. Z1 = Z2 = 0
is sufficient to cause K H = 0. For this condition to persist,
it is sufficient to have dZ,/ds = d Z 2 / d s= 0. From Equation
14, d R / a X , = 0 is sufficient. This is a particularly
interesting condition, since it defines the ideal temperature
discussed above. This condition allows one to evaluate
Adjoint System and General Design Policy uOptwhen KH = 0.
The corresponding solutions t o Equations 6 t o 9 and
For convenience we define the so-called adjoint varia-
14 must satisfy the eight boundary conditions. Three of
bles,
these are imposed on the state variables:
ax,(s)'
From Equations 6 to 9 and 11,
where J,,, is the maximum value of J .
H = -[ZlKi + Z?K2][Xi- X ? ] U- Now, a t s = 0 the above constraints apply. Treating
[ZoKo - ZiK?]R + 23 (13) these constraints by the method of Lagrange multipliers
From Equations 12 and 13, (Converse, 1970) we form the Lagrangian
L = Jmax [ X ( O ) ]- xi[Xn(O) - ao] -
20 = unit value of 1 dZn A?[XI(O)- XZ(O)]- X,X,(O)
N2flow ,f ds=
506 Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970
evaluated from dRdXl = 0. The following algorithm can
now be followed:
or 1. Guess si.
Z,(O) = -A, = -Z(O)-i.e., Z?(O)is set by Z I ( 0 ) 2 . Guess X , ( S ) .
(17) 3. Integrate the state and adjoint equations until XO
iiL dJ,,, = X,(O), which is fixed. This identifies the top of the
- xi = 0 reactor. During this integration u ( s ) is chosen so that
3x1 ax, dR ax, = 0 in the region s1 s 5 S. For s s s1 it
or Z1(0)= hi-i.e., is unknown. is set equal u" or .u, depending on the sign of K H .
At s = S there are no explicit state constraints. The 4. Is Xl(0) = X2(0)?
over-all enthalpy balance does apply, but this is taken Yes: X o ( S )is satisfactory.
into account through the integration of the state equations. No: Reset X o ( S )and return to step 3.
Hence 5. Iterate on s1until J is maximum.
Murase (1968) found that because of instability in the
numerical integration of the adjoint equations, the itera-
tion on s I did not converge; the following general design
policy seemed to apply.
For s l S s g S. choose u ( s ) so that aRlaX1= 0.
The above four equations plus Equation 17 constitute
For s25 s 5 sl, set u ( s ) = u*.
the five boundary conditions on the adjoint equations.
The normalization constant, n, is chosen to make For 0 5 s 5 sz,set u ( s ) = 0.
Z,,(S) = 1 (18) Intuitively, this is a satisfying policy. At the top, the
This vields heat transfer coefficient is set a t its lowest value in order
t o increase the reacting gas temperature. This is followed
by a section in which the coefficient is large, in order
to bring the reacting gas temperature back down to the
ideal temperature. At this point, the coefficient is set
in a variable manner, so that the ideal temperature profile
is followed in the rest of the reactor. This has intuitive
appeal when one is trying to maximize the yield. When
one is trying to maximize economic return as in this
case, this policy is correct only because ZI(S) N Z , ( S )
I t is important to recognize the role played by the N 0; else the singular policy a t the bottom of the reactor
adjoint variables. They really are evaluators and the
would be incorrect.
external economics influence the design through the bound-
ary conditions of the adjoint variables. The above partial
Results
derivatives are unit prices-Le., interstage transfer prices.
The economic analysis presented in Appendix B Murase (1968) used the above design policy to solve
indicates that Zl(0) N 0 and Z z ( 0 ) N= 0. The effect of problems in which the conversion is fixed and the reactor
this approximation was studied by Roberts (1969) and length is to be minimized.
found to be negligible. From Equation 14d, Z3(s) is con- The state equations were integrated from s = 0 to
stant. In determining the optimal ( u ) s we are therefore s = S in accord with the following algorithm.
free to select &(S) to be any value, since it merely changes
1. Set X o ( 0 )and X 3 ( 0 )to known values.
the value of H by a constant value and therefore does
2. Guess X I(0).
not affect the design policy-Le., u ( s ) . In computing the
optimal length of the reactor, dC,/aX,j is important and 3. Set X 2 ( 0 )= Xl(0).
the optimal length could be determined from Equations
4. Guess s2.
11 and 21, recognizing that aC,/aX, is a function of X.3.
5. Integrate over 0 5 s 5 s2 with u ( s ) = 0.
6. Integrate from s = s2 with u ( s ) = u" until an ideal
Although the adjoint boundary conditions are known
temperature is reached-Le., the temperature a t which
a t s = S, three of the four state boundary conditions
are known a t s = 0. Formally, one could guess three
aR dXl = 0. Call this point sl.
of the state variables a t s = S (setting the fourth in 7. Integrate from s1 until Xo = known final value,
with u ( s ) set so as to cause a R l a X , = 0.
accord with Equation 10d), and integrate the state and
8. Iterate on s2 until reactor length is minimum.
adjoint equations to s = 0, iterating the guessed values
9. Iterate on Xl(0) until reactor length is minimum.
until the boundary conditions at s = 0 are met. Such
mixed boundary value problems are difficult to solve Murase (1958) carried out these calculations for 0.10
numerically, and we therefore seek t o limit the number 5
- X o ( S ) 0.22. The results for X o ( S )= 0.20 (the value
of boundary conditions to be matched. obtained from the simulation of the TVA reactor) are
Since X i is absent from all G,, its absolute value is presented in Figure 5 . The loci of sz,sl, and X s ( S ) are
unimportant. The length of the reactor is given by XI(S) shown as a function of the top temperature. From this
- Xi(0) and the value of X , i ( S ) can be chosen to be figure the optimal conditions are: top temperature =
any value. If one guesses a value for X o ( S ) and assumes 451"C, si = 0, s1 = 2.17 meters, and minimum length
that the singular case will persist for a finite region con- = 4.29 meters. The corresponding bottom temperatures
nected to s = S, say si 5 s s S, then XI(S) can be were found to be 474" and 249C for the reacting and
Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970 507
600
*\.
5501
'.
'*,/Ideal Temp Profile
50C
0
0
v
Q,
L
3
I
z 400
0
Q
300- , lL4.fS ,
E
I .O 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 IC
) ?
Minimum Reactor Length ( M I 30C
(23)
where Mi molecular weight of the feed. Figure 7. Ammonia mole fraction profile
508 Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970
0,25t- - - ----I
9
N
H Figure 8. Ammonia mole
i fraction v s . reactor length
I
a n d optimal heat transfer
- 500: \
coefficient v s . reactor length
Y
v
c
a
0
3
I I
0.05J L o
1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
M i n i m u m Reactor Length ( M I
Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970 509
this can be achieved in a number of ways (Equation Consideration of an economic objective function
22). indicates approximately the same general design policy
I. Increase u above the switch point. as a minimum length objective function subject to fixed
11. Increase SIabove the switch point. conversion. The solutions to this problem are presented
111. Decrease S2 above the switch point. graphically, permitting design without numerical integra-
I1 and I11 can be achieved by increasing the number tion of the state differential equations. The procedures
of cooling tubes above the switch point, whereas in our whereby the results of this latter problem can be used
initial problem only I was considered. to solve the original economic problem are described above.
0
Case E
2000
0
E 1
Figure 1 1 . Variable U , ,
3 c I I I I
-;
._
.-
o*'ol/ \ 'Case D
Case E
"
0
0.05'
Y I I I I
5.O 10.0 15.0 20.0 25.0 ?
(+J)
t = s LS x IO (m3hr/kg moles)
c 0.254
Q
0
the Bottom
0.05-
450 500 550 600
Optimal Temperature o f t h e Reacting Gas ("CHat the Bottom)
510 Ind. Eng. Chem. Process Des. Develop., Vol. 9, No. 4, 1970
Figure 14. Ammonia mole fraction v s . optimal temperature of feed
gas a t bottom
a w m m a
a a
m 0 $ 02
% 0 $ In
0
0 0 00 s 0
0
Figure 15. Ammonia mole fraction vs. maximum temperature of reacting gas a n d
its location
:i
Reactor basket cross-section area 0.95 m2
Catalyst zone cross-section area, S2 0.78 m'
Number of tubes
Tube outside diameter, 0 2
Tube inside diameter, D1
Cooling Tubes
84
50.8 mm
38.1 mm
i
aE 0 500 1000 1500 2000
REACTION
KINETICS Equation B3 gives the capital cost of the reactor. For
El = 20,800 kcalikmole of NP, E? = 47,400 kcal/ an annual cost, an econ?mic life of 10 years and an interest
kmole of K 2 (Annable, 1952) rate of 10% are pssurnd, such that
f i = 1.78954 X lo", f2 = 2.5714 X 10'' Annual reactor cogt = [A + ( B + CX,)'
'ID (B4)
-AH = 26,600 kcalikg mole of N a where D = 0.16275 (Taylor, 1964).
f = 1 OBJECTIVE FUNCTION. From Equations B1, B2, and
B4,
Appendix B. Determination of Economic
Obbctive function and Partial Derivatives VI ( S ) G K , - K,X,(S) + K6X1 ( S ) 035)
where
Value of Product Gas. By assumption, the value of the
product gas is dependent on its temperature and ammonia Kq = 2GS2NiCAvF ( 0 ) - Cp,,. T,WLLTiCHXi
content-i.e., first-order approximations are used: C H ,cost Ki = 2GSzNiCA
of heating, and C.4, cost of ammonia, are considered fixed,
whereas in fact they vary with temperature and ammonia Ks = Cp3,ToWNiC~
concentration, respectively. V J ( S )= K ; - K i X , ( S ) (B6)
VI = V , ( X ! ,XI)
,
By assertion
.W Ni CH
(kgihr) (hr/yr) ($/kcal)
/
Heating value, $/yr
Ni C.4
Dimensionless ammonia hr/yr $/kg N H I
flow, kg NH:i/kg
K : + K , = 11.9877
Ki = 55.063
where Kb = 0.2964
K - = 0.2944
Ka = 3.475 x lo-
Kg = 1.987 x l o -
Then from Equation 3 and above Acknowledgment
J , = /KI + K ; ]- K , X , ( S )+ K & I ( S ) - Grateful acknowledgment is made for partial financial
K - X ? ( S )- [ K , + KsXi(S)] (B8) support provided by National Science Foundation Grant
62-7326, by the South African Council for Scientific and
ECONOMIC DERIVATIVES Industrial Research, and the hippon Light Metal Co.,
From Equation B5, Ltd., Tokyo, Japan. The work is obviously dependent
on the fine facilities of the Kiewit Computation Center,
Dartmouth College, Hanover, N. H. Tapes of programs
used by Murase (1968) can be obtained a t cost from
From Equation B6, the center.
Literature Cited