Ethylene furnace heat flux correlations

Equations are presented that correlate and predict heat flux as a function of
operating, burner and furnace parameters for all major ethylene-cracking
furnace configurations

Joseph Colannino John Zink Company LLC

T
his article presents —statistically, all are significant
mathematical heat flux contributors to y0 and zmax. This
models for ethylene highlights the need to work
furnaces in one of three with burner design engineers at
configurations: fired by floor Wall burners the earliest possible stage in the
(2 elevations)
burners only, fired by wall design of ethylene units to
burners only, and fired by predict the heat flux profile.
floor burners in conjunction
with one or more rows of Heat flux profile
wall burners (Figure 1). It is The distribution of heat flux
assumed that the floor burn- (the heat flux profile) is an
ers do not release their heat Ethylene important criterion for the
process
instantly, but over some tubes
performance of ethylene-crack-
distance. To determine the ing units. The heat flux profile
explicit functionality, we use is the fraction of radiant heat
an analogy from jet theory incident to a tube at a given
and a global energy balance elevation. Figure 2 shows a
on the furnace. For the wall- Floor burners typical heat flux profile for a
fired-only case, we presume (2 elevations) furnace that is fired from the
that the low heat release and floor. To understand why heat
short flame lengths allow us Figure 1 An ethylene furnace represent- flux profile is important, we will
to treat the radiant heat as a ing an elevation of one cell fired by floor briefly overview the cracking
point source. Additionally, burners and wall burners process.
we couple these two models
for the floor plus wall-fired case. We will see that Production of ethylene via thermal cracking
the model establishes similar conditions for field Ethylene production proceeds via the thermal
and test units, and may be used to generate real- cracking of a hydrocarbon feedstock. For exam-
time heat flux curves from flue-gas temperatures. ple, ethane and propane may crack in a complex
These results have been incorporated into a series of steps to ultimately form ethylene and
state-of-the-art configuration program, which is byproducts, as follows:
also described.
One aim of this discussion is to show that it is Ethane cracking: H3CCH3 " H2C=CH2 + H2
possible to characterise normalised heat flux Propane cracking: H3CCH2CH3 " H2C=CH2 + CH4
profiles as a function of two parameters: the
elevation at which the maximum heat flux occurs The amount of ethylene production normalised
(zmax) and the heat flux at the floor (y0). In turn, by the maximum possible yield is known as the
these two parameters are determined by operat- fractional conversion or conversion efficiency.
ing conditions and furnace and burner design Thermal cracking reactions such as the previously

www.digitalrefining.com/article/1001040 PTQ Q1 2008 1

 shown ethane and propane
cracking (to ethylene) are 100
highly endothermic and
require heat. The process
90
heat is supplied by special
burners in an ethyl-
ene-cracking furnace 80
operating at approximately
1200ºC. The reacting feed- 70
stock inside the tube will
abstract heat from the tube Figure 3 Technicians are measuring

Normalised elevation, Z
wall, thus cooling the tube 60 heat flux with a special probe on a
metal. The result is process test furnace
outlet temperatures (ie, coil 50
outlet temperatures – ment life. However, these are
COTs) that are hundreds of interrelated in complex ways. For
40
degrees lower than the Zmax example, higher temperatures can
surrounding furnace increase conversion but may also
temperature. However, if 30 increase coking, reduce run time
heat is applied non-uni- and shorten tube life. An appro-
formly, a side reaction may 20
priate heat flux profile balances
proceed to deposit carbona- and optimises these competing
ceous polymers on the influences.
inside tube wall known as 10

coke (ie, coking inside the Furnace configurations

tube wall). Coke sticks to 0 From a burner perspective,
the tube walls and insulates 0 50 100 ethylene furnaces come in three
Normalised heat flux, y
the process fluid from the basic varieties — floor fired, wall
furnace. This can result in fired, and floor plus wall fired –
local overheating of the Figure 2 A typical heat flux profile as previously noted. To begin
tube surface and ultimately, examining each configuration,
if neglected, tube rupture. After sufficient deposi- we will consider the floor-firing-only case.
tion, the tube may be seen to be glowing from the
furnace side. For this reason, tube surface Heat flux from floor firing
temperatures are carefully monitored. Consider heat released from floor burners along
During normal operation, coke accumulates at a the furnace length (qh) and being absorbed by
low rate on internal tube surfaces. Periodically, the process tubes, or otherwise removed from the
furnace temperature is drastically lowered and furnace (qp), and finally exiting the furnace into
some steam or steam and air are added to the the convection section (qc). Then, a simple heat
feedstock. This procedure interrupts the produc- balance for a two-dimensional furnace is as
tion of ethylene but removes the coke. Periodic follows:
applications of steam and air are known as a
steam-air decoke cycle. For decoking, you can qh - qp = qc.
write a generic reaction, as follows:
Typically, process engineers are interested in
C + a H2O + (2 – b – a)/2 O2 " (1- b) CO + b CO2 + a H2 the normalised heat flux (0 < y < 1); that is, the
heat flux (MW/m2) divided by the maximum heat
where a is the steam/carbon ratio of the influ- flux. A specialised heat flux probe (Figure 3)
ent, and b is the CO/CO2 ratio of the effluent. converts radiant heat flux to an mV signal. You
The time between decoking episodes is the run move the heat flux probe to various elevations
length. It is desirable to increase run length, along the centreline of the test heater and record
conversion efficiency, and preserve tube and equip- each mV reading. After the readings are complete,

2 PTQ Q1 2008 www.digitalrefining.com/article/1001040

you determine a normalised where a is a constant of
heat flux profile by dividing proportionality, and c is the
each mV reading by the maxi- ratio of the burner-related
mum mV reading. You may dimension to the furnace
use statistical procedures to height.
smooth the heat flux profile Presuming a constant heat
and estimate the true maxi- loss (b) along the length of the
mum with greater accuracy.1 furnace, Equation 1 becomes:
Generally, heat flux probing
is done in special test furnaces dy = a –b
designed for this purpose. dz 1 + cz
Figure 4 shows one such ethyl- (2)
ene furnace simulator – the
internal furnace dimension of This integrates to c = a1n(1 +
this particular simulator is 11m cz)– bz + c, where c is a
from floor to the beginning of constant of integration. Let y0
the convection section. be the normalised heat flux at
Now qh and qp vary with the the floor (z = 0). With this
normalised furnace height (0 substitution, we find c = y0.
< z < 1), where z = 0 is the Then the integrated equation
floor elevation and z = 1 is the becomes c = a1n(1 + cz)– bz +
top of the furnace. y0.
Accordingly, the following Setting Equation 2 to zero
differential equation repre- Figure 4 An ethylene furnace simulator gives the elevation (zmax) at
sents the heat flux as a (courtesy of John Zink Co LLC) which the normalised heat flux
function of elevation: is a maximum (y = 1). This
reduces to:
dy = q (z) – q (z)
h p
dz a = cz + 1
max
b (3)
Analogy from jet theory
In order to solve the previously noted differen- The difference of the normalised heat flux from
tial equation, the explicit functionality of qh and the floor heat flux is given by y – y0. The ratio of
qp must be known. An analogy with concentra- this flux difference to the maximum flux difference
tion profiles from jet theory can help determine is therefore:
this. Jet theory gives the centreline concentra-
tion of a jet along an axial centreline as C = 1/x’,
’
y – y0 = a1n(1 + cz )– bz (4)
where C is the centreline concentration normal- 1 – y0 a1n(1 + czmax )– bzmax
ised by the initial concentration (0 < C < 1), and
x’
‘
is the number of nozzle diameters down- Dividing Equation 4 by b and using Equation 3
stream.2 However, x’ is taken from a virtual to substitute for a/b, we arrive at:
origin one diameter upstream of the nozzle exit.
If we shift the origin to the nozzle exit, we obtain y – y0 = (1 + czmax )1n(1 + cz)– z (5)
x’ = x + 1, where x is the number of nozzle diam- 1 – y0 (1 + czmax )1n(1 + czmax )– zmax
eters from the nozzle exit rather than the virtual
origin, or C = 1/(1+x).’
If we presume that the For the data sets, we have examined c ≈ 1. This
heat release is proportional in some way to the is tantamount to saying that the actual heat flux
fuel concentration issuing from a burner-related distribution correlates with the furnace height
dimension, we obtain: rather than scales with a burner-related dimen-
sion. This could be an artefact of internal
qh (z) = a reflections within the enclosure, convective heat
1 + cz (1) patterns or some other furnace-related phenome-

www.digitalrefining.com/article/1001040 PTQ Q1 2008 3

 the heater is enough in conjunc-
tion with zmax to reconstruct the
1.0
heat flux profile. Indeed, you can
calculate y0 if y is known at any
elevation by calculating:
0.5

y0 = yk – zk*
Reduced heat flux, y*

r2 = 98.4%
0.0
1 – zk*

where yk is the heat flux at a

–0.5 particular elevation zk, and

z* = (1 + zmax )1n(1 + zk )– zk
–1.0 k (1 + zmax )1n(1 + zmax )– zmax

–1.5
–1.5 –1.0 –0.5 0.0 0.5 1.0 Heat flux from temperature
Reduced elevation, z* measurements
Radiative heat flux is propor-
Figure 5 Reduced heat flux plot. The plot comprises nearly 1000 heat flux tional to temperature to the
measurements taken from a variety of fuels, operating conditions, and fourth power.3 If the heat flux
burner and furnace designs, and shows an excellent correlation between were proportional to the flue gas
y* and z* temperature at a given elevation,
(T/Tmax)4 would therefore give
non. Whatever the case, this simplifies Equation 5 the normalised heat flux. With such a substitu-
to: tion, Equation 5 becomes:

y – y0 = (1 + zmax )1n(1 + z )– z (6) y* = z* = T* (7)

1 – y0 (1 + zmax )1n(1 + zmax )– zmax
where T* is the reduced temperature
For convenience, we let:
T* = T4 – T0 4 ;
4 4

y* = y – y0 and T max – T0
1 – y0
it comprises T (the absolute flue-gas tempera-
z* = (1 + zmax )1n(1 + z )– z ture), T0 (the flue gas temperature at the floor) and
(1 + zmax )1n(1 + zmax )– zmax Tmax (the maximum flue gas temperature [meas-
ured at zmax]). This is not strictly correct; flue gas
which we refer to as the reduced heat flux and and wall emissivities are very different and floor
reduced elevation respectively. Therefore, firing involves combustion against the wall by
Equation 6 reduces to y* = z*, and a plot of y* convention of practice. Thus, the wall and flue gas
vs z* should yield to a straight line. Figure 5 temperatures are not identical. Nevertheless
shows a plot of nearly 1000 heat flux measure- beyond the vicinity of the floor, there is some hope
ments taken from a variety of fuels, operating that the gas radiation dominates, making Equation
conditions, and burner and furnace designs. 7 approximately correct. Comparing around 50
Despite the simplifications, the figure shows a temperature-derived and actual heat flux profiles,
striking correlation. Moreover, y* and z* are the average correlation coefficient was greater than
functions of y0 and zmax alone, meaning that if 95%, with virtually all of the points having r2 >
you know the elevation of the maximum heat 90%.
flux and the initial heat flux (heat flux at the
floor) you may determine the entire curve. Heat flux from wall-only firing
Actually, knowing the heat flux at any point in Most wall firing is done with premixed burners,

4 PTQ Q1 2008 www.digitalrefining.com/article/1001040

although there are exceptions.
At least for premixed fuel issu- 1.0

ing into an 1100ºC furnace,

Normalised furnace height, z

0.8
combustion is rapid and flames
are short — so much so that we
0.6
may treat them as a point
source. A two-dimensional 0.4
model should well approximate
the wide heaters commonly 0.2
employed, except perhaps near
the end walls. With these 0.0
0.0 0.5 1.0 0.5 1.0 0.5 1.0 0.5 1.0 0.5 1.0 0.5 1.0
assumptions, the heat flux from Normalised heat flux, y
a row of burners located at the
kth elevation (zk) will be given as: Figure 6 Floor plus wall firing, selected cases, comparison of model with
actual heat flux. The horizontal bars show the normalised elevation of the wall
s burners; their widths indicate the relative fractions of wall/floor firing (qW,k). The
yk = closed symbols are the actual data, the solid curve is the composite floor plus
s2 + (zk – z )2
wall heat flux curve given by Equation 6
where yk is the normalised
Heat flux factors
heat flux generated from the kth
burner elevation, and s is the Operating factors Furnace factors Burner factors
width/height (inverse aspect 1. Fuel pressure 7. Furnace height 10. Fuel port geometry
2. Air-preheat temperature 8. Furnace width 11. Tile geometry
ratio) of the furnace.
3. Furnace bridgewall temperature 9. Heat absorption surface 12. Degree of FGR
The heat flux from all burner 4. Excess oxygen in flue gas 13. Number of stages
contributions will be given by: 5. Heat release of floor burners 14. Burner type
6. Hydrogen content in fuel
sqW ,k
y=1
Ymax k
s + (zk – z )2
2

Table 1
where qW ,k is the fraction of
the heat released at zk normalised by the heat kth elevation normalised by the heat released at
released at some reference elevation, and Ymax is the floor.
the maximum heat flux value occurring at zmax. If Equation 9 introduces no new adjustable param-
the heat release is identical for each elevation eters. Figure 6 shows a panel of results for various
Ymax will occur at the average row height z–. The floor and wall burner combinations. Equation 9
equation then reduces to the following: does a great job of predicting the addition of wall-
fired heat to the overall heat flux curve. The
y= k
[s2 + (zk – z )2 ]-½ deviation at the top of the furnace is thought to be
due to reflection from the ceiling of the test
k
[s2 + (zk – z– )2 ]-½ (8) furnace, which is not included in the model.

Floor plus wall firing Determining y0 and zmax

For the floor plus wall-firing case, we will use So far, we have related heat flux to zmax, y0 and
the floor as the reference elevation for qW,k. T. These provide useful correlations. However,
This leads to the following equation for floor neither zmax, y0 nor T are known prior to meas-
plus wall firing: urement of the heat flux curve. We need to know
(9) how to predict heat flux as a function of various
operating, burner and furnace factors for two
p – bz + y0
sqW ,k
y = af1 + k
main reasons. First, the earlier in the design
s2 + (zk – z )2 process you can specify heat flux, the more influ-
ence you have to achieve it. Second, engineers
where qW,k is the ratio of heat released at the must predict heat flux performance (not merely

www.digitalrefining.com/article/1001040 PTQ Q1 2008 5

 correlate it after the fact). Therefore, prediction correlate heat flux with elevation or temperature
is more important than mere correlation. in test and field units
The author has correlated y0 and zmax from • The model has been validated with nearly 1000
more than 65 floor-fired heat flux tests as a func- heat flux measurements
tion of the factors listed in Table 1. • The model accurately predicts the heat flux
Table 1 comprises operating, furnace and burner from the addition of premix wall burners to the
factors. All factor categories were found to be floor-firing case
statistically significant at the 95% confidence level • John Zink Company has incorporated these
or greater. Once determined, these were correlated results into its Fort NOx program, which calcu-
in a semiempirical model to predict the influence lates NOx, CO and heat flux, among other
various factors have on y0 and zmax. In turn, y0 and important design and performance outputs.
zmax were used to predict overall heat flux curves.
These models were coded into John Zink Fort NOx is a mark of the John Zink Company LLC.
Company’s proprietary program, Fort Nox, for
burner design. The program not only predicts heat References
flux, but also emissions such as NOx and CO. In 1 Colannino J, Modeling of Combustion systems — a Practical
addition, it calculates all fuel and flue gas proper- Approach, CRC Taylor and Francis, Boca Raton F L, 2006, 491–497.
The author maintains known errata and information at www.
ties, and takes account of the effects of furnace
combustion-modeling.com, last accessed 1 November 2007.
type, elevation, burner spacing, fuel composition
2 Cheremisinoff N P, Gupta (eds) R, Handbook of Fluids in Motion,
capacity and performance curves.
Ann Arbor Science, Ann Arbor Michigan, 1983, 260.
All burner design and performance calculations 3 Siegel, R, Howell J, Thermal Radiation Heat Transfer, 4th Edition,
comprise state-of-the-art semi-emprical models Taylor and Francis, New York, 2002, 565.
for compressible or incompressible flow, coupled
with kinetic models for emissions. These models Joseph Colannino is director of research and development, John
have been thoroughly validated with test and field Zink Company LLC.
data. Based on this discussion, the conclusions
can be summarised as: Links
• Equations have been presented that correlate
and predict heat flux as a function of operating, More articles from: John Zink Hamworthy Combustion
burner and furnace parameters for all major More articles from the following categories:
ethylene-cracking furnace configurations: floor Combustion Engineering
fired, wall fired, and floor plus wall fired Emissions Control
• This article has provided similarity relations to

6 PTQ Q1 2008 www.digitalrefining.com/article/1001040