AbstractData for heat transfer from packed beds are reexammed m the hght of new rnslghts Much of the data
mcludes a length effect, resultmg from a lugher heat transfer coetficlent near the mlet, makmg It unsmtable for use
m chemical reactor design, where the length IS so long that an asymptotic heat transfer coeliiclent IS desired The
data IS reexammed m order to exclude studies Influenced by the length effect and retammg data gtvmg an
asymptotic heat transfer coot&rent The asymptotic coefficient 1s correlated well over a large range of Reynolds
numbers (207000) The data also mdlcate that the Blot number decreases as the Reynolds number Increases, but IS
approximately constant for Reynolds numbers above 500, taking the value
Bld,e=027
R(l l )
1055
1056 CHIHSIUNG LI and B A F~NLAYSON
a' 0 08 0 15 0 21 0 23 0 20 0 18 0 14
ml
The asymptotic solution can then be approximated as squares sense Valstar[8] mmumzed the errors m tem
perature values throughout the bed, whereas De Wasch
T = {Z&(A,r) exp C~A,*zl}/CAIJ1~A~~l and Froment[9] used only exit profiles
(5) All of these methods have been extensively dlscussed
A 1x24048
m the literature, but usually without mentlomng the fact
that If k. and h, depend on length then each method may
and IS independent of BI, although It st~U depends on a,
yield a different value k. and h, even m the absence of
and thus k. Thus suggests that If the Blot number IS htgh.
expenmental errors The maJor emphasis of this paper IS
It would be dficult to measure the heat transfer co
that each of these methods of data analysis yields a
efficient, smce the temperature solution IS Independent of
different heat transfer coefficient, which IS not always the
Bt Of course the Blot number under those conddlons IS
asymptotic coefficleni, and the data should be compared
less cntlcal We find below that the BJ decreases as
mth caution
Reynolds number mcreases, so that the scatter of data for
h, and Bt at low Reynolds numbers (high BI) IS not LENGTFIEFFIET
surprismg, nor is it of great consequence Experimental evidence confirms the fact that the heat
Nearly all the heat transfer data IS obtamed m the transfer coefficient (h,) and the effective thermal con
same way, namely the radml temperature profile 1s ductivity (kJ depend on length , Figure 1 shows data
measured at several bed depths with the thermocouples from three mvestlgatlons dlustratmg how the effective
placed Just above the packmg The difference m studies thermal conductlvlty decreases as the length IS mcreased
results from the analysis of the data to determine k. and Most methods of data analysis (Methods 1, 2 and 4)
h, Four widely used methods are outlmed cause an error m h, d k. IS wrong, thus makmg h,
Method 1 The effective thermal conductlvlty IS depend on length, too DeWasch and Froment[9] (ther
determmed by solving eqn (1) for k,, and the temperature Figs 6 and 7 and our Table 2) and Paterson[l l] found
denvatlves are obtamed by dlfferentiatmg the tem that k. and h, decreased with mcreasmg bed depth
perature profiles Thus local value of k. IS averaged to Finally we anticipate the conclusions below Methods 3
obtam a constant k. used for the entire bed The h, IS and 4 do not grve asymptotic heat transfer coefficrents
found from eqn (4) with A, obtained from the slope of and should gve a coefficient above the asymptotic co
the stra& hne of log T vs z, which IS &Al2 This efficients of Method 2 Data analyzed by Methods 3 and
method was used by Coberly and Marshall[4] 4 do he generally above that analyzed by Method 2
Method 2 Fust A1 IS found from the radial tem Method 1 does not mve an asymptotic coefficient, but
perature profile at the exit of the bed can be either above or below the asymptotic value Data
analyzed by Method 1 follow tlus companson, also
T&z = 1) _ 2JL4) 22 I * 1 I
T(r = 0, z = 1) A,
.
The fluid temperature at the wall, T& wluch IS merent
from the coolant temperature, Tk, 1s estimated by ex
trapolatmg the measured bed temperature to the wall IO I I L I . .
0 02 04 06 08 IO
z
(TT&,,=&;[T:T(J=R,r)]dr (8) Fig 1 The effective thermal conductlvlty vs bed depth (The k.
shown m this figure IS the average value of the local &.s across
the radrus of the bed )
This method was used by Fehx[7]
Symbol Ref (I Re,
Method 4 The k. and h, are found by comparing
I Cl21 0 28 48
calculated results usmg eqn (3) with the experunental 0 [41 0 56 506
data and adjusting k, and h, to give the best fit m a least A [13,141 0 16 238
Heat transfer m packed bedsa reevaluation 1057
Table 2 The k, and h,s reported by De Wasch and Froment[9] for dtierent bed depths Rep = 400, d, = 0 099 m,
d, =00057m
L Cm) k (kcal/m hr OC) h (kcal/m' hr % a' B1
Methods 2 and 4
Fust analyze the simulated data using Method 4 for a
bed depth of 1016 m The k. and h, are determmed to
ave an exact fit of the temperature profile at the bed
exit, but they account for the k, and h, throughout the
bed, mcludmg the entrance remon where k. and h,
depend on length In Method 2, however, only the shape
of the temperature profile at the outlet 1s used (not the
absolute value) to determme A, and then a (or k=) IS
.,L_
0 02 04
Z
06 08 0
I
c
Method 1 1s particularly susceptlbIe to the length effect
The k. and h, found by the different methods of data
analysis are grven m Table 4 We emphasize these dlff Id
ww
erences are entirely due to the method of data analysis,
w
... #
+
w,
so
@.
smce the simulated data were identical As predicted, w&v +
c+s
ww
Method 2 gives the asymptotic h,, Methods 3 and 4 gve
. l * ww. b
higher h, and Method 1 gives a lower h, The h, aven o;sIlB+ 40
B
o
ted to include a length dependence The same thing holds IO IO" IO3
R=P
true for cylmdncal packing (Fig 4) except now the data
Fig 4 Expenmental data for wall heat transfer coefficient with
analyzed by Method 1 by Pogorslu hes below the cyhndncal packmg
asymptotic coefficient, as predicted
It IS still unclear how long the bed should be for the
local k, and h, to reach the asymptotic values So far no
such experimental results have been recorded However,
several experlmental temperature profiles reported m the
literature (see Fig 5) indicate that whenever LYZ IS
greater than 0 2 the plot of log T vs z IS a straight hne
This IS consistent with the result derived above under the
assumption of constant k, and h, the temperature
1 1 23* 77 0 3 10
2 0 97 123 6 30
3+ 1 23* 137 5 53
4++ I 12 146 6 42
*
Using k e = the average of those in Table 2
+ 1  0 284m, z; = 1 016m
=1
++L = I 016111
Heat transfer m packed bed reevaluation 1059
temperature profile would not yreld a straight hne Based T= z+bbr2 r,sril
on this reasonmg we use a greater than 0 2 as a cnterlon
to determme when the entrance regon does not affect If we take r, = 0 and use this solution, we find that data
the k. and h, determined by Method 1 and Method 3 analyzed by Methods 1, 2 and 3 are unaffected by this
We next examme the effect of axial dlsperslon on /c. problem, provided az ~0 2 Method 4 IS affected,
and h, If an axial dlsperslon term, y a2Tlaz2, IS added however, and the error 1s
to the left hand side of eqn (2), the equation can be
solved assuming a, y and BI are constant If we use the
$$(S)(l$+&)
solutron for a> 0 2, when these assumptions are valid,
we can determine the error m k, and h, assumed by
analyzmg the data lgnonng axial dmperslon The detaded Forb=O2,az=O2andBr=1,theAa=O3a,gvmg
analyses are derived elsewhere[22] and show that for slgmficant error For BI = 5 the error ISreduced to ha =
methods 1, 2 and 3 the data analysis aves the same 0 la
result, We see that the data for h, 1s affected by the length of
the bed, and different methods of analysis ave different
Ak,
=_= Ah, lu, h, We do not have a welllusttied explanation of the
k. h, 1+a, (9) cause of the length effect, but several posslbdltles exist
Possibly the developing velocity and temperature pro
a,2 = 1+ 4ayA,2 (10) files influence the results In a packed bed the velocity
profile would be developed wlthm a few particle dl
where Ak, IS the error m k, due to the neglected axial ameters of the mlet, due to the presence of the packmg
dlsperslon effect Phillips et al [23], have proved eqn (9) In a nonisothermal bed, however, the axial velocity
for k, when using Method 1 For a, close to 1 (as profile can change because of a changmg temperature
assumed) eqn (9) can be further sunphfied to profile (see Schertz and Blschoff [26]), and, because of
contmnulty, anytime the axial velocity protile changes a
Ak. Ah
_=v= _ ayAgZ radial velocity 1s introduced This might affect the wall
(11) heat transfer coefficient, especially smce the radial flow
k, h,
would be more pronounced near the wail None of these
Using the approxlmatlon from orthogonal collocatlon[24] effects can at present be conclusively shown to be the
that cause of the length effect, but are possibdlfies which
are consistent vvlth the data The importance of dfieren
6B1 tlatmg between overall coefficients, applicable to the
A12=BI+3 (12) whole bed, and asymptotic coefficients, is, however,
clearly estabhshed Method 2 aves an asymptotic h,, or
and Pe,, = 2 gives at least the best estnnate of it, compared to Methods 1, 3
and 4
Ak,
== Ah,  1 $ 2 3B1 Anally we examme the crucial question which k. and
Pe,, 0 R B1+3
(13) h, should be used for chemical reactor deslgnv Should
k, hw
the h, be one which fits the heat transfer data of the
This expression can be used to estrmate the per cent whole bed or only the one representing the asymptotic
error caused by axial dispersion For Method 4 the result local heat transfer rate? Table 6 lists typical values of
was derived by Young1251 bed lengths, particle diameters and tube diameters for
1060 CMHSIUNG LI and B A FINLAYSON
Table 6 Design parameters for the packed bed reactors operated m the chemical Industry
T

Reaction L/R d&/R a Reference
I
Methanol oxidation 70
0 04  0 25* ,0 36 lZ1
Benzene hydrogenation 60
mdustnal reactors The beds are so long that a%0 2 for made measurements for f by f m cyhndncal packmg, but
most cases This means that If a heat transfer expenment plots of log T vs z &d not reach a straight hne, so this
were done m such a bed the asymptotic heat transfer data mdudes a length effect and 1s not used Data for
coefficient would be measured If the data were analyzed 2 by 4m cyhndrlcai packmg was orlgmally analyzed
usmg any of the methods (wlthm expenmental error) To by Method 1 Here It IS reanalyzed by Method 2 to
reproduce data m these long beds, the asymptotic heat elunmate the length effect, and this recalculated data IS
transfer coefficients would have to be used When heat used below Hawthorn et al 1163, did experiments m
effects due to chemical reaction are present, the shape of very long beds (a> 7) for constant wall heat flux A
the temperature profile could be changed so that even the mathematical model was used to calculate the radial
asymptotic value was mappropnate However, m the temperature profile and the calculated temperature drop
absence of mformatlon on the effect of chenucal reaction from the center of the bed to the wall was subtracted
on heat transfer coefficient, the best we can do IS use the from the measured temperature drop from the center of
asymptotic heat transfer coefficient the bed to the coohng medium These numbers were
The asymptotic heat transfer coefficient 1s lower than compared, and the difference was ascribed to the tem
any average h, for the entue bed Thus a reactor design perature drop very near the wall, which IS modeled by
based on the asymptotic h, IS conservative the actual the heat transfer coefficient Due to the length of bed the
reactor would be less likely to exhlblt a runaway length effect 1s neghglble This 1s one of the few sources
situation than the design would mdlcate of heat transfer data derived from a chemically reactmg
system at high temperature We reanalyzed Hawthorns
EXAMINATION OF HEAT TRANSRER DATA data usmg Pe,, = 8 rather than 10, and the revised cal
Expenmental data was exammed m detail to decide culatlons are presented below (the effect of Pe,, ISsmall)
which data was Influenced by length effects or axial Data for annular beds 1s reported by Yagl and
conductlon In many cases the ongmal thesis was ex Kunu[31], Baddour and Yoon[32], and Kunu and
ammed, and m some cases the data was reanalyzed Suzuh[33] A constant temperature dtierence was
usmg Method 2 A complete tabulation 1s avdable[22] mamtamed between the Inner and outer surfaces along
We are Interested m only the asymptotic heat transfer the entire bed depth Thus the temperature vaned only
coefficient radially and there 1s no length effect The data, however,
Data obtamed by MaedaC51, Phllhps et al (231, Yagl apphes to an annular bed, and for large d,lR the packmg
and WakaoE61, Hashlmoto ef al [lS], and Kunu et al [17], at the Inner surface may be different from that of a
1s accepted smce the h, and k, were determined by cyhndncal bed Thus the data must be put m a separate
Method 2 Data obtamed by Fehx[7] was analyzed by classticahon
Method 3, and It IS accepted If the testing sectlon &d not Unfortunately a large number of data must be rejected
fall m the entrance region (I e If the region over which because they mclude a length effect and thus do not ave
h, 1s deduced IS for az 2 0 2) t Coberly and Marshall [4] an asymptotic h, Valstar[8], De Wasch and Froment [9J,
and Zldkowskl[Zl] all used Method 4 to analyze their
data and it thus IS affected by the length Campbell and
tFehxs testmg sectron usually began at I. = Sm We have
appbed the strmgent crltenon that az, 2 0 2, but a less stnngent Huntmgton134,3.5) used Method 1 to analyze a constant
cnterlon could be argued Unfortunately we have no basis on which wall heat flux expenment, even though the Method as
to decide, and so have used the number 02 sumes a constant wall temperature In some experiments
Heat transfer m packed bedsa reevaluation 1061
and are not used below Calderbank and Pogorskl[ 13,141 The data which 1s accepted as being free from length
used Method 1 but most of the beds are so short (a < effects 1s put to two additional tests We require that the
0 2) that the expenment mcludes a length effect re error due to axial dispersion effects be less than 5%, as
gardless of the method of data analysis Thts data IS not calculated by eqn (13) We also discard data for Bl> 12,
used below smce then the temperature profile IS msensltlve to h,,
Aerov and Urn&[361 report data which 1s bemg determined pnmarlly by k. For Bi > 12, less than
frequently [ 1,2] reported as following the correlation 20% of the total thermal resistance exists at the wall, and
an expenmental error of &lo% m the total rate of coolmg
leads to an error m h, of 250%
h4 _
k,
o 155
((?d,>
P
pr~/3
(15) The data for h, IS correlated m terms of the Nusselt
number Spherical and cyhntical packmg @ve dtierent
The orlgmal Russian report, however, gves the correct results and are correlated separately The best cor
correlation as relation for spherical packmgs IS
0 05 5 d,ldt s 0 3
eqn (16) 1s about 45% higher than Yaa and Wakaos
data, whereas eqn (15) is 35% lower Aerov and Ummk 20sRe, 57600
did not measure h,, they determined it from thetr
measured k, and correlations for the overall heat transfer Constant wall temperature, spherical packmg
coeffictent (see below) Unfortunately these correlations
include a length effect so that the data 1s not usable This correlation 1s compared to the expenmental data m
Qumton and Storrow[20] performed a careful ex Fig 6 and predtcts the data with an average devlatlon of
periment keeping the wall flux constant The boundary 140/o,and a modtied correlation coefficient of @ = 0 98
condition m eqn (2) IS replaced by from the linear regression analysis The @ represents
the fraction of the vanation of data which can be ex
platned by the correlation Other forms of correlations
cz
Q (17) tied are listed m Table 7 The form of the correlation
T=1+2aQz+Q(++) (18)
1 Nu = a Re; 0 17 0 79 14 0 98
3 Nu = a Rez 0 029 0 94 21 0 97
m
4 Nu a(Re Pr)' 33+b Re" ' Pro 4 1 33 0 14 38 0 98
5 Nu = a Re; 0 16 0 93 33 0 85
6 Nu = a+b Re Pr 5 21 0 126 62 0 76
P
7 Nu m = a Remb 0 03 1 06 39 0 05
10 lJ*d,/k,  aReb 7 13 0 47 17 0 92
14 U*dt/kf = a Re b 1 a4 0 76 35 0 79
P
15 OJ*d,lk,hc~~4 6 dp/dt) = a Re b 1 39 0 90 27 0 88
P
16 (U*dt/kf)exp(6 dp/dt) = a Re b 1 26 095 27 0 89
P
preferred by Beek[l], form 4 m Table 7, IS less suc This correlation gves an average deviation of 33% and
cessful m predlctmg the data, with an average deviation B2 = 0 85 and IS compared to the data m Fig 7 There 1s
of 38% clearly more scatter m the data for cylmdrlcal packmgs,
We note that all the expemnents were done with au, and the heat transfer coefficients are larger
so there IS no Prandtl number dependence A reasonable For annular beds there IS very little data avadable for
extension would be to regard the constant 0 17 = 0 17 the heat transfer coefficient on the inner wall Yam and
(PrlO 7)3, but ths IS not proved by these data The
correlation (17) IS very close to that proposed by Yagl
and Wakal[6], but here It correlates their data as well as r
that by Kunu et al [ 171 The data of Fehx [7] and Plautz
and Johnson[lS, 191, which has a slight length depen
dence at high Reynolds numbers, IS slightly above this
correlation At lower Reynolds numbers the Fehx and
Plautz data have more length dependence and the data 1s
stgmficantly above the correlation Thus we see again
that the reason for the large scatter m heat transfer data
IS partially due to a mixing of overall and asymptotic
h,
The best correlation for cylmdrlcal packmgs 1s
h,d,=a+O041RePr P (24)
kf
combmatlons a and BI The Blot number, m particular, Fig 10 Blot number vs Reynolds number
1s the ratio of h, to &JR The data m Fig 9 indicate that
the Blot number decreases as the Reynolds number which predicts the data with *25% The Blot number 1s
increases Figure 10 shows data for the Blot number as a the ratio of two experimentally measured quantities, k.
function of Reynolds number We see that we can use and h, If the overall heat transfer rate 1s measured
accurately, but k, IS, say lower than it should be due to
expenmental error, then h, IS larger than rt should be
(25) Errors m Bt (= h&/k.) are thus magmfied Thus a scatter
of the data of *25% 1s not surprlsmg Some data m Fe 8
1s not included m Fig 5 or 7, it 1s included m Fe 8
because it IS Just at the hmit of acceptability as to the
length effect, and it provides the void fraction depen
I ! I I I, I dence, whch 1simportant This is also the reason there 1s
600 
no void fraction dependence m the correlation (22) the
accepted data &d not cover a wide range of e values
Equation (25) IS thus a new result which can be used
along with Pe,, = 810 for high Reynolds numbers We
note that a Br constructed from k. and h, has a slight
Reynolds number dependence, but at these high Rey
NU
too
nolds numbers, the variation of X31with Re IS w&m the
scatter of the data
The solution 1s
Let us define an overall heat transfer coefficient 0 such The difference of At2kJdr m eqn (30, 33) LSdue to the
that the average temperature out of the packed bed 1s the fact that A12k, depends on length The additional term m
same m the one and twodlmenslonal models Then eqn (30), however, 1s the effect of the temperature profile
being different at different positrons, and this effect IS
less important m long beds, but IS usually important m
0 =  GzIn T,(l) (29)
experlmental studies
To illustrate the difference m u and U*, consider the
Puttmg eqn (28) mto eqn (29) aves example m Table 2 There we have
Then we obtain
ue _ Ay: (33)
I
(34)
for Ia*,and kfm eqn (34) The best correlations are then dlmenslonless radial coordinate, f/R
radius of bed
modtied Reynolds number, GdJ[p( 1  E)]
y exp (6dJd,) = 2 03 Re,0* (35)
Reynolds number, Gd,/p
surface area of packmg particle
20 I Re, 5 7600
temperature
005sd,ld,zsO3 dlmenslonless temperature, (T  Tk)/(T:  Tk)
dimensionless center temperature
Spherical packing inlet temperature
mean temperature
dunenslonless mean temperature
fluid temperature at the wall
3 exp (6d,/d,) = 1 26 Rep095 coolant temperature
(36)
k, overall heat transfer coefficient
volume of packmg particle
20sRe, ~800
axml coordmate m bed
dlmenslonless axial coordinate
[26] Schertz W W and Ehschoff, A I Ch E J 1969 15 597 [33] Kunu D and Suzuki M , Intern Heat Trans Conf 1966 V
1271Carberry J J and White D , Id Ertg~g Chem 1969 61(7)
35 1341 izpbell J M and Huntmgton R L , Petrol Ref 1952 31
[ZS] Hougen 0 A and Watson K M , Chemrcal Process Pnn 123
crples, Vol 3, 1st Edn, p 1007 John Whey, New York 1949 1351 Campbell J M , Ph D Thesis, Umverslty of Oklahoma
[29] Froment G F, Ind Engng Chem 1%7 59(2) 18 (1951)
[30] Goleblowskl A and Wasala T, Intern Chem Engng 1973 1361 Aero; M E and Ummk N N , J Tech Phys 1951 21 1351
13(l) 133 1371 Cruder J
E and Foss A S . A ICh E J 1965 11 102
[31] YaB S and Kunu D , A I Ch E J 1960 6 97 i38j Leva M ,
Wemtraub M , Grimmer M and Clark E L , Znd
1321 Baddour R F and Yoon C Y , Chem Engng Prog Symp Engng Chem 1948 40 747
Ser 1961 57(32) 35 [39] Verschoor H and Schult G , Appl Scr Res 1952 A2 97