Professional Documents
Culture Documents
Paper Reactores8 PDF
Paper Reactores8 PDF
Abstract
`Tracera or Residence time distribution (RTD) studies are commonly exploited as a means of developing an understanding of the
`mixinga status of vessels of various types. The e!ort involved in the setting up of such plant studies can be considerable and the
execution of the experiment its self is often a test of endurance. In the past full interpretation of the results has not been easy and as
a consequence super"cial treatments have been employed. This paper presents an alternative method for interpreting RTD data,
which is relatively easy to use and addresses some of the weaknesses of more conventional methods. An extention to the `tanks in
seriesa concept is presented (ETIS) and united with the `reactor networka formulation. The suitability and appropriateness of the
model is discussed and compared with the `closeda dispersion model 2000 Published by Elsevier Science Ltd. All rights reserved.
Keywords: Residence time; Tanks in series; `Closeda system; Networks; Comparison; Case studies
200-4340.
*C D *C *C
" ! , (1)
E-mail address: Alaistair.martin@OMIST.ac.uk (A. D. Martin). *h u¸ *z *z
0009-2509/00/$ - see front matter 2000 Published by Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 1 0 8 - 1
5908 A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917
where the dimensionless group boundary conditions in 1944. Their solution, published
in a dimensional form, is reproduced here in non-dimen-
D 1
"N " (2) sional form (Eq. (7)) to maintain consistency with Leven-
u¸ " Pe spiel and Smith's `opena solution (Eq. (4)). Yagi and
and Miyauchi reproduced the Thomas and McKee solution
in 1953 with an alternative condensation of the terms
t Q t ut (Yagi & Miyauchi, 1953).
h" " 0 " (3)
¹ < ¸
0 0
C L Pea e.C+\F?L >, Pe
E(h)" "2 L a cos a
The `opena boundary conditions from which this model C Pe(a#1)#4 L 2 L
draws its name de"ne the #ow condition at the reactor B L L
inlet and outlet. The `opena condition is physically Pe
#sin a , (7)
achieved when the #ow is undisturbed at the inlet and the 2 L
outlet. In 1957 Levenspiel and Smith published the ana-
where a is given by the positive roots of Eq. (8).
lytical solution to Eq. (1) for the `opena boundary condi- L
tions (Levenspiel & Smith, 1957). This solution is shown Pe 2a
in Eq. (4). tan a " L . (8)
2 L (a!1)
L
C 1 Pe \ .C\F Levenspeil (1972) published expressions for the mean and
E(h)" " e F
C 2 ph (4) variance of the `closeda system RTD though did not
B report the analytical solution for the RTD itself. Leven-
Where C is the concentration that would have been spiel's mean and variance can be shown to be equal to
B
obtained had the `dyea been evenly dispersed through- those of Eq. (7) thus con"rming their association. These
out the vessel under study. results are shown below Eq. (9) for the mean,
With mean and variance
K
2 L L
hM "1# , (5) L Pe(a #1)
Pe hM "4 L "1 (9)
! K
L L
2 8 L Pe(a #1)
p " # , (6) L
Pe Pe
and Eq. (10) for the variance.
It can be seen that the mean of this distribution is
K
a function of the dispersion and at high dispersion (low L L
Pe) the mean is substantially '1. This result appears to L Pe(a #1)
p "32 L !1
be inconsistent with the mass balance but may be ex- ! K
L L
plained by the di!usion of `dyea upstream of the injec- L Pe(a #1)
L
tion point. Thus, the pulse injected at the inlet is not the
same as the pulse which would have been measured at 2 (1!e\.C)
" 1! , (10)
the inlet. Casual observation of the inlet to and exit from Pe Pe
a high-dispersion reactor would also con#ict with the where K is given by Eq. (11)
assumed `opena boundary conditions. The rigorous L
mathematical de"nition of the `opena boundary condi- 2Pea
K " L . (11)
tions makes this model more di$cult to use than the L Pe(a#1)#4
tanks in series formulation as the user is often left in some L
doubt as to the degree to which the `opena condition is Satisfactory enumeration of Eq. (7) becomes increasingly
achieved in the system under investigation. di$cult at higher values of Pe. This di$culty derives
The **closed++ dispersion model treats the system in from the relative magnitudes of the early terms in the
exactly the same fashion as the `opena dispersion model. series with respect to the later terms and the "nal sum.
Eq. (1) is solved with the `closeda boundary conditions. When h"0 the series itself is non-convergent with suc-
The `closeda boundary conditions relate to the physical cessive terms oscillating as the following.
situation in which the #ow approaches the inlet to the
reactor in idealised plug #ow (Pe"R), transforms to lim (S )"2(!1)L\e.C. (12)
L
dispersed #ow within the reactor and returns to idealised L
plug #ow at the exit. This situation is very closely ap- At a practical upper limit of Pe"33 a conventional
proximated in many real reactors even those which are double precision summation will yield a residual of less
themselves low dispersion devices. Thomas and McKee than 0.00005 at h"0.001 but will require some 300 terms
(1944) published the analytical solution with `closeda to achieve this.
A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917 5909
The tanks in series model seeks to describe the #ow in reaction and expanding the dispersion equation in
a reactor system by considering it to be discretised into of Taylor series. This analysis leads to the following equiva-
a strand of equal-sized hypothetical CSTRs. Each hypo- lence relationship for Pe:
thetical CSTR is independent of those preceding or fol-
Pe"2(n !1). (16)
lowing it. The number of tanks in series n describes the 2
2
dispersion with n "1 representing in"nite dispersion Kramers and Alberda (1953) also proposed this equiva-
2
and being equivalent to Pe"0. Integration of a simple lence. Fig. 2 shows quite clearly that for the variance
dynamic mass balance around the strand of reactors equivalence, n "2 and Pe"2.557, respectively,
2
readily yields the system RTD (Eq. (13)). Eq. (13) is also the RTDs are quite di!erent. A plot generated for the
the de"nition of the Erlang distribution. `Kramers, Alberda and Elgetia equivalence shows very
similar behaviour though the co-location of the peaks is
C nL2
E(h)" " 2 hL2 \e\L2 F. (13) slightly poorer than illustrated in Fig. 2. The `closeda
C (n !1)!
B 2 dispersion model exhibits a considerably higher peak
With mean and variance value than the tanks in series model. This di!erence
reaches a maximum at n "2, Pe"2.557. The `closeda
hM "1, (14) 2
2 dispersion model can also be seen to exhibit a region with
1 a positive second derivative at h(1 which is absent from
p " . (15) the tanks in series model. At values of n '15,
2 n 2
2 Pe'28.97 the di!erence between the two maxima is less
Conceptually, the development of this model is easy to than 5% and is likely to be di$cult to resolve experi-
follow. It also has the advantage that the precise de"ni- mentally. Thus for practical purposes the two models are
tion of the inlet and exit boundary conditions is not su$ciently similar to be regarded as the same. The tanks
required. Similarly concerns regarding the method of in series and `closeda dispersion models however di!er
`dyea injection and measurement do not arise. Phys- signi"cantly from the `opena dispersion model under the
ically, this model is at its best when the number of tanks same conditions.
in series is low and concerns over the appropriateness of
either `opena or `closeda boundary conditions are at
their height. This model however has a signi"cant draw- 3. Extended tanks in series
back when n is small due to the integer constraint.
2
Fig. 1 shows that the E curve for n "1 di!ers very If the tanks in series model is merely regarded as
2
signi"cantly from that resulting from n "2. Many real a residence time distribution function, whose form de-
2
CSTRs exhibit RTDs which lie in this range and are pends solely on the value of the parameter n , and is
2
therefore only characterised very approximately by the freed from the arti"cial need to have a physical manifes-
tanks in series model. Frequently in the interpretation of tation, it becomes possible to address the quantisation
RTD data n has been related to Pe or N via the issue arising in the analysis of high-dispersion systems.
2 "
variances of the RTDs (Eqs. (10) and (15)). This is tanta- Simply introducing the concept of a non-integer number
mount to assuming that the tanks in series model and the of hypothetical tanks in series will achieve the desired
`closeda dispersion model are equivalent. Elgeti (1996) result. The exit age distribution or E curve of the
develops an alternative relationship between the two
RTD forms by following the progress of an arbitrary
Fig. 1. Exit age distributions of the tanks in series model for 1 and Fig. 2. Comparison of the tanks in series model with the closed disper-
2 tanks in series. sion model with a common variance of 0.5.
5910 A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917
extended tanks in series (ETIS) model is given by a subset These components serve to de"ne the volume elements
of the gamma distribution family (Eq. (17)). of the reactor and their connectivity, respectively. The
`threadsa receive #ow from an up stream or source
nL2
E(h)" 2 hL2 \e\L2 F. (17) `knota and discharge to a down stream or sink `knota.
!(n )
2 The characteristics of the `threadsa are de"ned by the
When n is integer it is identically equal to the Erlang #ow through the thread its hypothetical volume and exit
2 age distribution (Q, <, E). The `knotsa receive #ows from
distribution (Eq. (11)). As with the Erlang distribution the
mean and variance are given by Eqs. (14) and (15). The source `threadsa and distribute the summed #ow to the
ETIS model removes the problem of quantisation which sink `threadsa and are de"ned as zero-volume blender
occurs as n tends to 1 in the tanks in series model. splitters. The `knotsa are characterised by a single-#ow
2 split fraction parameter ( f ). A number of assumptions
Q
are implicit in this basic de"nition of the network.
4. Reactor network structure 1. Individual `threadsa are assumed to be fully seg-
regated from each other.
The models described in the previous sections are all 2. The exit age distribution of an individual `threada is
one dimensional in their nature and are consequently given by the convolution of the composite E curve
unable to describe gross structure in the #ow within passed by the source `knota.
a reactor. To describe such large scale structures in the 3. The composite E curve passed at any `knota is given
#ow pattern it is necessary to introduce appropriate by the linear #ow weighted sum of the E curves of the
large-scale structure to the model. Many model struc- contributing `threadsa.
tures have been proposed to describe a range of physical 4. There is no dispersion through `knotsa.
#ow con"gurations (Levenspeil, 1972; Monteith
& Stephenson, 1981; Smith, Elliot & James, 1993). To Assumptions 1 and 4 are likely to be the most conten-
date the fundamental building blocks or `threadsa of tious, particularly when considering reaction kinetics in
these reactor networks have been limited to either ideal conjunction with the RTD. They are however critical to
CSTRs or ideal PFRs. Algebraic expressions for the the formulation of a manageable problem. Clearly within
C and E curves of some of the simple con"gurations have the context of assumption 4 the choice of model for the
been derived (Levenspeil, 1972; Monteith & Stephenson, characterisation of dispersion is restricted to the `closeda
1981). The technique for generating the model RTD for dispersion model, the TIS model and the ETIS model.
the majority and the more complex networks involves Despite the reservations expressed above the ETIS model
the numerical integration of the dynamic CSTR mass is used to illustrate the concept of the network.
balance Eq. (18) (Smith et al., 1993). Fig. 3 shows a network of three `threadsa connected
together in to two #ow paths or `strandsa by three
dC (C !C) `knotsa. This type of network is a common result of
" G (18)
dt ¹ interpreting the RTD of a large shallow packed bed
reactor. This type of network di!ers from those employed
Clearly to describe real DPFRs using simple CSTR or
by previous authors (Levenspeil, 1972; Monteith
PFR building blocks requires large numbers of the fun-
& Stephenson, 1981; Smith et al., 1993) in that all
damental unit and the numerical integration of reactor
`threadsa accommodate dispersed #ow.
network models becomes inordinately cumbersome as
the extent of the network grows. The use of the CSTR as
5. Experimental data interpretation
the basic building block also constrains the network
description to integer values of n . Employing the ETIS
2 Data Gathering. There are four general types of experi-
model to characterise the basic building block signi"-
mental protocol for the conduct of `dye tracera
cantly reduces the complexity of network required whilst
simultaneously relaxing the integer n constraint. To-
2
gether these advantages dramatically improve the tracta-
bility of the data analysis problem. The reactor network
structure is developed further in the following sections to
facilitate the description of `reala vessels exhibiting com-
plex composite dispersion behaviour with bypassing and
stagnant zones.
The network dexnition: The reactor network structure is
constructed from two component types.
1. Threads,
2. Knots. Fig. 3. Thread and knot concept for the tractor network model.
A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917 5911
experiments. The classical experimental technique in- the mean residence time when the `closeda dispersion,
volves the injection of a Dirac d function pulse of `dyea TIS or ETIS models are appropriate.
followed by the measurement of its concentration as it The estimates obtained from Eqs. (19) and (20) have
emerges in the outlet stream over an extended period of often been related to the dispersion number (N ) or the
"
time. This method produces a readily interpretable result number of tanks in series (n ) via Eqs. (10) and (15) with
2
and maximises the potential for resolving the detailed little regard to the overall shape of the RTD. A number of
structure of the experimental exit age distribution. additional empirical relationships have also been de"ned
A common alternative technique employs a `stepa input. to characterise the dispersion in reactors:
This technique generates the reactor F curve, which may
t
be di!erentiated to yield the E curve. Direct di!erenti- g " , (21)
& tM
ation of experimental data results in the ampli"cation of
noise arising from sampling and analytical methods. t
Curve "tting to the integrated form or F curve may also I " , (22)
+ t
lead to poorer estimates of the parameters. This arises
because detailed features are less distinctive on the t
I "1! K . (23)
F curve than on the E curve. A third experimental proto- tM
col involves the injection of a `blocka input. This method
Smith et al. (1993) examined these indices and found
is considerably inferior to the two previous methods. The
them to be mutually inconsistent when used to describe
experimental exit age distribution obtained from this
their data. These indices may be consistently related to
technique is a hybrid lying between the classical E and
the variance of the RTD only in circumstances when the
F curves. The duration of the `blocka tends to produce
distribution is symmetrical, i.e. very low dispersion or
a broad peak and damp out the detailed structure, which
plug #ow. This situation rarely coincides with the occa-
would have been revealed by a d function pulse input.
sions when the departure from plug #ow is important.
A "nal technique involves the analysis of a natural ran-
dom signal in the input stream. This technique has the
advantage of not altering the process stream but su!ers
6. Selected case studies
from the same disadvantages as the `blocka input tech-
nique. The AWWA guide (Teefy, 1996) o!ers advice on
The ETIS model has been used in the following case
the conduct of `dye tracera tests in water and waste-
studies to provide the quanti"cation of the dispersion
water treatment plants particularly with respect to the
within an individual `threada. This has been done be-
selection of suitable `dyesa.
cause of the ease of application and despite concerns
Data processing. Historically the interpretation of
regarding strict validity at intermediate values of Pe. The
experimental RTD data has been rather super"cial
curve-"tting process has been conducted on Microsoft
(Tomlinson & Chambers, 1979). Often this interpretation
EXCEL 97威 using the built in SOLVER and mathemat-
has been limited to extraction of the mean and variance
ical functions. The built in functions were also supple-
of the data set (Eqs. (19) and (20)).
mented with visual basic for applications (VBA) code
C t *t developed by the author.
¹
tM " L L L, (19)
C C C *t Simulated very large CSTR. The reactor under study
L L consisted of a nominal 1500 m cuboid vessel with
C t*t a multi-ported inlet manifold located at the bottom of the
p
L L L !tM . (20)
C C *t C vessel and an outlet weir located on the opposite wall.
L L
Fig. 4 shows a schematic elevation of the vessel which
The sequential nature of the data gathering guarantees
was nominally 19.5 m in the axial direction by 19.5 m
the collection of biased data sets. So whilst for randomly
wide by 4 m deep. The inlet manifold directed the feed
gathered data fairly modest sample sizes are su$cient to
#ow across the #oor of the reactor underneath a grid of
obtain good estimates of the population mean and vari-
aeration equipment. The process feed rate was
ance this is not the case for RTD data. It can be seen that
0.0523 m/s. The reactor was simulated using a commer-
Eq. (19) is acutely `tail sensitivea, thus early truncation of
cial CFD package and an `experimentala RTD
the data set leads to serious under estimation of the
generated using a particle tracing technique. Several ap-
population mean or achieved mean residence time. To
proximations were made in the construction of the CFD
overcome this problem protracted measurement periods
model of this reactor:
are usually prescribed. Typical measurement periods are
chosen to be of the order of 3}5 times the mean residence 1. The vessel was assumed to be semi-in"nite (normal to
time with the need for longer periods coinciding with the direction of #ow).
highly disperse systems. Furthermore tM is an estimate of 2. The feed manifold was modelled by a continuous
F
the mean of the RTD. It is therefore only an estimate of axially pointing slot at the foot of the feed wall.
5912 A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917
7. Discussion
log (p)
was however published by Thomas and McKee in 1944.
u"1#Int ¹ #¹ , (24)
log 1! The rigorous handling of the inlet and outlet boundary
¹
0 conditions is a signi"cant bene"t relative to the ETIS
model and unlike the `opena dispersion model does not
where the numerical subscripts refer to the `threada add to the di$culty of use. The series solution presented
numbers in Fig. 13. in their paper is however numerically intensive in use and
Analysis of the experimental data resolved the peri- becomes practically non-convergent at low values of
odicity in the early period to be 11.55 min (0.061¹ ). h and high values of Pe. Frequently in the past the tanks
0
However the expected "rst peak at 4.75 min (h"0.025) is in series model has been assumed equivalent to the
almost completely absent from the experimental data. `closeda dispersion model. The equivalence has been
The "rst strong peak is detected at 16.3 min (h"0.086). established by equating the respective variances (Eqs. (10)
This result has two possible explanations. and (15)). This assumption has been shown to be false
except under the condition n "1 and Pe"0. Practic-
1. The sampling frequency at the outlet was insu$ciently 2
high to resolve the peak. ally however the di!erences become small enough for the
distributions to be regarded as the same when n '15
2. The dispersion normal to the direction of #ow was 2
incomplete. and Pe'28.97. The di!erences between the two distri-
butions are such that, for a chosen variance or equivalent
A simple test to check the possible sample frequency pair of n and Pe values, the reactor design based upon
2
explanation indicates that at least two samples would the ETIS model yields a slightly larger volume than that
have been gathered which contained signi"cant concen- based on the `closeda dispersion model. This arises from
trations of `dyea. This test indicates that the samples the more `peakya nature of the `closeda dispersion
collected at 4 and 6 min (h"0.0211 and 0.0316) are those model RTD. The `closeda dispersion model predicts that
a!ected and the expected concentration of `dyea is likely signi"cantly more material is discharged from the reactor
to be double that observed in any of the samples prior to with an exit age between 0.25 ¹ and ¹ .
0 0
the "rst signi"cant peak. Whilst this test is not conclusive The ETIS model as applied in the three case studies
the result suggests that incomplete normal dispersion is has provided very satisfactory "ts to the experimental
the more likely explanation. Based upon these deduc- data. There are few areas in which the application of the
tions the derived reactor network structure and para- more numerically intensive `closeda dispersion model
meter values are given in Fig. 13. The aggregated or late would practically bene"t the interpretation of the data.
period behaviour of this model is as expected equivalent Thread 3 of the simulated very large CSTR may describe
to a CSTR (n "1) with a mean residence time of the data better if the `closeda dispersion model were
2
¹ "190 minutes. The convolution of the long tail of the used. This `threada, though, accounts for less than 3% of
0
input trace with the CSTR characteristics describes the the reactor throughput and the value of n (22) is toward
2
late period behaviour of the experimental RTD very the upper end of the range in which the ETIS
satisfactorily. and `closeda dispersion models di!er signi"cantly.
5916 A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917
Application of the `closeda dispersion model to `threada sion model is also found to have the advantage of
1 of the packed bed reactor network may improve the "t rigorously de"ned inlet and exit boundary conditions.
to the experimental data in the region 0.2(h(0.3 and Combination of the ETIS model with the reactor net-
h"0.75. However, as can be seen from Fig. 9 the scope work structure permits a considerable increase in the
for improvement is small. versatility of networks with out the concomitant increase
Dispersion at knots. A basic assumption of the reactor in numerical intensity which characterises networks of
network structure is that there is no dispersion through CSTRs.
the `knotsa. This assumption is completely consistent
with the application of the `closeda dispersion model to
characterise the interconnecting `threadsa. Such a high Notation
level of consistency cannot be claimed for the use of the
ETIS due to the lack of rigour in the de"nition of the inlet C dimensionless concentration Dimensionless
and outlet boundary conditions. The implied assump- D eddy di!usion coe$cient L t\
tions for the tanks in series model relate to the inlet and E(h) exit age distribution function Dimensionless
outlet of a CSTR and would indeed be consistent with e base of natural logarithms Dimensionless
the no dispersion through the `knotsa assumption. The (2.718...)
same implicit assumption is made in the ETIS model but f #ow fraction Dimensionless
is conceptually di$cult when n is not an integer. I `mixinga index Dimensionless
2
The extent of dispersion through the `knotsa is only ¸ characteristic length L
likely to be signi"cant where both source and sink N dispersion number Dimensionless
`threadsa are themselves characterised by high disper- "
n number of tanks in series Dimensionless
sion. The qualitative e!ect of removing the central `knota 2
p precision Dimensionless
joining 2 CSTR `threadsa is shown in Fig. 2. By analogy Pe peclet number Dimensionless
with the previous discussion regarding the suitability of Q liquid feed rate L t\
the ETIS model for the characterisation of individual S a term in the series solution Dimensionless
`threadsa it may be deduced that the no dispersion of the closed dispersion
through `knotsa assumption holds within experimental model
accuracy when n '15 and Pe'28.97 in the more plug mean residence time t
2 ¹
#ow `threada. Quantitative assessment of the breakdown T Hydraulic mean residence t
of the no dispersion through the `knotsa assumption is time
not reported here but might be approached via consid- t elapsed time t
eration of the Kurtosis of the RTDs. tM mean of the residence time t
Ordering. Using the simple conservative `dyea -tracing distribution
methods reported here it is not possible to order the u velocity L t\
`threadsa in a reactor network without additional struc- < volume L
tural information. To make progress with respect to v volume fraction Dimensionless
ordering of model elements in the absence of structural z axial displacement L
information it is essential to employ a non-conservative
`dyea -tracer method with a second-order decay process.
The application of such techniques will not be discussed Greek letters
here.
! gamma function
8. Conclusions g E$ciency
p 3.141592
The ETIS model in conjunction with the reactor net- # strand dimensionless
work structure has been shown to be a versatile method residence time
of describing the characteristics of a small but diverse h dimensionless time Dimensionless
group of reactors. The ETIS model has been compared hM mean of the dimensionless Dimensionless
with the conventional tanks in series approach and has RTD
been found to be superior due to the elimination of the p variance of the dimensionless Dimensionless
quantisation which is inherent in the latter approach. RTD
The ETIS model has also been compared with the
Thomas and McKee `closeda solution to the dispersion Subscripts
model. The `closeda dispersion model has been shown to
di!er signi"cantly from the ETIS model in the range 10 pertaining to the recovery of
0(Pe(28.97 and 1(n (15. The `closeda disper- 10% of the injected `dyea
2
A. D. Martin / Chemical Engineering Science 55 (2000) 5907}5917 5917