Lecture 20: The fixed bed catalytic reactor

The fixed bed catalytic reactor is one of the most widely used reactor types in the refining
and petrochemicals industry. In its simplest form, it is simply a tube filled with solid catalyst
through which gaseous (or, less frequently, liquid) reactants flow and are converted into
products. When a new process is to be developed, this is often the first type of reactor to be
considered. Froment and Bischoff present a list of large-scale refining and petrochemical
processes that are carried out in fixed bed reactors. On the pages following that list, they list key
factors that have led to improvements in these processes. Note that many of these are not due to
the kind of reactor engineering that we learn in this course, but on ‘common sense’ or
infrastructure improvements. An example of such an advance is the use of larger multitube
reactors, which was made possible by improved welding techniques for making them and by
increased shipping clearance for moving them to the plant site. Nevertheless, we will focus on
the kinetics and reactor modeling in such systems. Advances in these areas have also made
important contributions. Froment and Bischoff’s chapter on fixed bed reactors is quite good, and
we will follow the book more closely in this section.

One of the first questions that we must address in designing a fixed bed reactor is if and how
we will add or remove heat to and from the reactor. The simplest choice is to use an adiabatic
reactor. Since heat transfer is not an issue in this case, the reactor can be a single vessel of
relatively large diameter, will require no utilities during steady state operation, and only a single
catalyst bed will be required. Unfortunately, many reactions of interest cannot be successfully
carried out in a single adiabatic reactor. If the reaction is sufficiently endothermic, then the
temperature in the reactor will drop as the reaction proceeds, and the reaction may become
unacceptably slow before the desired amount of reaction has occurred. For the case of
exothermic reactions, the adiabatic reaction temperature may be higher than economically
reasonable reactor materials can withstand, or high temperatures may lead to unfavorable
equilibria or production of unwanted byproducts. The next-simplest choice is to use a series of
adiabatic reactors with interstage heating or cooling. Examples of such reactors are given on
pages 395 and 396 of Froment and Bischoff. This allows us to add heat to an endothermic
reaction or remove it from an exothermic reaction while still having separate heat exchangers
and reactors, or separate heat exchange and reaction sections within the same reactor. Finally,
we could have continuous heat addition or removal through the wall of the fixed bed reactor.
This will require that the reactor have fairly small diameter so that heat transfer in or out of it is
fast enough to avoid unacceptable temperature gradients in the radial direction.

A useful way to consider and compare these design alternatives is on a contour plot of
reaction rate as a function of conversion and temperature. This is similar to what is done on page
397 of Froment and Bischoff (figure 11.3-4). As another example, consider methanol production
from synthesis gas (H2 + CO) according to the reaction
CO + 2 H2 ↔ CH3OH.

The enthalpy and entropy change of reaction for this process are ∆Hrxn = -90.1 kJ mol-1 and ∆Srxn
= -219.2 kJ mol-1 K-1. For a given inlet composition, total pressure, and temperature, this
information allows us to compute the equilibrium composition. Note that the reaction is
exothermic and reversible, so we expect that there may be an optimal temperature or temperature
profile for the reactor. Suppose that for a particular catalyst, the forward rate of this reaction was
given by

where pCO and pH2 are the partial pressures of the reactants in bar. Note that the rate is expressed
per catalyst weight rather than per reactor volume. This is a common practice for catalytic
reactions. The net rate (even though this is not an elementary reaction) must therefore be given
by

where Keq is the pressure equilibrium constant, given by

where pref is the reference pressure for the standard state enthalpy and entropy of reaction, which
is 1 atmosphere. Therefore, we have

The reactant and product compositions are related by stoichiometry. For constant-pressure
operation, we can characterize the reaction by a single extent of reaction, x. If we have a
stoichiometric feed (2 moles of H2 for each mole of CO), the number of moles of each species is
given by NCO  No 1  x  , N H2  2 No 1  x  , NCH3OH  No x , and Ntot  No  3  2 x  , where No
is the initial number of moles of CO. The partial pressures are given by

So, the net reaction rate is

This allows us (for a given total pressure) to construct a contour map of reaction rate vs.
temperature and extent of reaction. This is shown below for a total pressure of 50 bar, where the
color indicates the logarithm (base 10) of the reaction rate in kgmol per kg catalyst per second.
The dark curve is the equilibrium line. Below that curve, we have net reaction in the forward
direction, and above that line, we have net reaction in the reverse direction. The white curve
shows the locus of temperatures that give the maximum rate for a given extent of reaction. This
curve follows the equilibrium curve quite closely. Following this path of maximum reaction rate
would give the minimum total residence time (minimum catalyst) for the reactor.

If the reactor is run adiabatically, then the temperature and extent of reaction are related by the
energy balance. When we assume constant heat of reaction and total specific heat (not such a
great assumption in this case, but the results will be qualitatively correct) then this relationship is

where Cp is the specific heat of the mixture, per mole of methanol (assumed constant). On the
above plot, this is a straight line, with slope (-∆H/Cp). For an exothermic reaction (like the one
considered here) that slope will be positive. H2 and CO both have a molar heat capacity of about
7/2 R = 29 J mol-1 K-1, while the molar heat capacity of methanol is about 44 J mol-1 K-1 at 298
K, increasing at higher temperatures. So, at x = 0, the mixture specific heat per mole methanol is
about 3×29 = 87 J mol-1 K-1. At x = 1, the mixture specific heat is 44 J mol-1 K-1 (at 298 K). So,
the heat capacity per mole methanol (which is proportional to the heat capacity per unit mass)
drops by almost a factor of 2 over the course of the reaction. Nevertheless, we will assume it is
constant, with a value of 60 J mol-1 K-1. Then, the adiabatic reaction paths on the above plot will
have a slope of 90000 J mol-1/60 J mol-1 K-1 = 1500 K. These adiabatic reaction paths are
superimposed on the reaction rate contour plot presented below.

In an adiabatic reactor, we necessarily follow one of these straight lines. As you can see, we
cannot follow any of them over a very large range of conversion without running into the
equilibrium line. However, we can move from line to line by using multiple reactors with
cooling in between. Thus, we might follow a zig-zag pattern on the above chart, as illustrated
below.

As you can see from the above picture, for a very exothermic reaction like this one a large
number of stages will be required to get to high extent of reaction. It may, therefore be more
economical to design a reactor with continuous cooling through the reactor wall. This could be
engineered to more closely follow the curve that gives the maximum reaction rate.
Once we have decided on the basic layout of the reactor (adiabatic, multiple adiabatic
reactors with interstage cooling, multitube continuously cooled, or some combination thereof),
we must model the reactor more quantitatively for doing actual reactor design. At the micro- and
meso-scale, we need to model the reaction kinetics and any transport limitations within or at the
surface of the catalyst particle. We learned how to do that in the first half of the course.
However, we must now couple those micro-and meso-scale models to a macro-scale reactor
model. The simplest way to do that is to separate the two. We assume that we can write the
overall reaction rate as a function only of the gas phase concentrations and temperature, and we
only model the gas phase. This is called a pseudohomogeneous model. We can further sub-
classify the pseudohomogeneous models based on the types of transport that are included. If we
have only convection (no mixing in the radial direction, no diffusion or dispersion in the axial
direction) then we have the ideal plug flow reactor model. If we also consider transport by
diffusion or dispersion in the axial direction, then we have the plug flow model with axial
dispersion. If we explicitly consider mixing by diffusion or dispersion in both the radial and axial
directions then we have a 2-dimensional reactor model. We can write the same 3 analogous
models as heterogeneous models, where we explicitly consider differences between the bulk
fluid phase composition and the composition at the catalyst surface.

For the ideal PFTR model with no axial mixing, the governing equations are (in Froment
and Bischoff’s notation)

where us is the superficial velocity (volume of gas per reactor cross-sectional area per time), CA is
the concentration of reactant (A) in the gas phase, rA is the rate of production of A per unit time
per unit mass of catalyst, ρB is the overall bed density (mass of catalyst per volume of reactor), cp
is the specific heat (per unit mass) of the gas phase, T is the gas phase temperature, ∆H is the
heat of reaction, U is an overall heat transfer coefficient for heat transfer through the reactor
wall, dt is the tube diameter, Tr is the temperature of the heating or cooling fluid outside the tube,
p is the pressure, f is a friction factor (dimensionless), and dp is the equivalent particle diameter
(defined as the diameter of a sphere with the same external surface area per unit volume as the
catalyst particle). The first two of these equations are the plug flow reactor model that we have
considered extensively already. The third equation gives the pressure drop across the packed bed.
Usually, we want the pressure drop across the bed to be small, because we don’t want to spend
money buying and running compressors. Various correlations are available for the friction factor.
The most popular correlation for the pressure drop is the Ergun equation. For it, the friction
factor is calculated from
where ε is the void fraction of the bed, G is the superficial mass flow velocity (mass per reactor
area per second = ρus), µ is the dynamic viscosity, and av is the external particle surface area per
unit reactor volume. Other correlations are given in Froment and Bischoff and references therein.

Given these equations and some feed conditions (temperature, pressure, and composition), it
is straightforward to integrate the initial value problem given by the above three equations. For
exothermic reactions, a critical determination to be made from the model is whether any ‘hot-
spots’ will develop in the reactor, and whether ‘runaway’ or ‘extreme parametric sensitivity’ is
possible. For an irreversible exothermic reaction (for example, most complete or partial
oxidation reactions), the reaction rate will increase very sharply with conversion, due to the
temperature rise caused by the heat release from the reaction and the strong temperature
dependence of the rate parameters. This can lead to localized hot spots in the reactor, and in the
worst case can lead to reactor runaway – generation of heat sufficient to damage the reactor. If
the adiabatic reaction temperature for the reaction under consideration, plus any others that
would be initiated at high temperatures, is above the maximum allowable operating temperature
for the reactor, then it is essential that the development of hot-spots be avoided. Froment and
Bischoff discuss analytical criteria that have been developed for determining whether hot-spots
will be a problem. A more straightforward approach is to simply solve the one-dimensional
model presented above for a range of intitial conditions and see whether and for what conditions
unacceptable temperatures are attained.

The application of this model to multiple catalyst beds in series, with heating or cooling
between them is straightforward. We simply apply this model to each reactor bed, following the
flow through the system. However, deciding how many beds to use and how to size them (given
the fact that we can model the system) is a more difficult problem. This is discussed fairly
extensively in Froment and Bischoff.

A technique that is used to recover heat released by exothermic reactions in a fixed bed
reactor is to use the hot product to heat the cold feed. If this can be done without any additional
heat input, it is called autothermal operation. It is usually economically favorable to recover the
heat of reaction, and this offers a way of doing so without using an intermediate heat transfer
fluid (i.e. steam). It complicates the reactor modeling task, because it couples the inlet reactor
temperature to the outlet temperature. It also increases the difficulty of reactor start-up, since the
heat produced by reaction is used to initiate the reaction. Analysis of this reactor configuration is
an important topic, and is discussed in the text, but since we will not have time to cover it in
class, I will not reproduce that discussion here.

The next level of sophistication in modeling of fixed bed reactors is to consider mass
transport by dispersion in addition to mass transport by the overall flow. We are still assuming a
uniform velocity, and uniform properties across the bed cross-section. As discussed previously,
the importance of axial dispersion can be assessed based on the Peclet number, defined as
 us L
Pe  , where us is the superficial velocity, L is the length of the reactor, and Dea is the
Dea
effective dispersion coefficient in the axial direction. The model equations, with this axial
dispersion term included are

where λea is the effective thermal conductivity in the axial direction. It depends on the catalyst
properties and the catalyst packing as well as the flow, since much of the heat conduction will be
through the solid catalyst. The effective transport coefficients, in principle, capture mixing and
heat conduction in the axial direction from all sources, including non-uniformities in the flow
field. Mathematically, we now have a boundary-value problem rather than an initial value
problem. In principle, this leads to the possibility of multiple steady states in the reactor.
However, it is unlikely that for any realistic conditions that would be present in a fixed-bed
catalytic reactor, these multiple steady states could be achieved. You know how to integrate
these equations numerically. Unless the bed is unusually shallow or axial mixing is unusually
strong, this model usually only results in slight changes from the ideal plug flow model.

For reactors with heat addition or removal at the wall, it may be necessary to consider
gradients of temperature and composition in the radial direction. As was the case for axial
dispersion, effective transport parameters that capture mixing due to an assortment of physical
mechanisms are used. With these terms, the equations become

Mathematically, these are now partial differential equations, in two dimensions (the cylindrical
coordinates r and z). As written, they are a boundary value problem in two dimensions – that is,
the PDEs are 2nd order in both the r and z coordinates. Sometimes, it will be possible to neglect
the axial mixing term in this formulation, in which case the equations become
These are now an initial-boundary value problem – equivalent to a time-dependent one-
dimensional boundary value problem. Mathematically, they can be handled much more easily
than the full 2-dimensional boundary value problem. Since we are now explicitly considering
transport in the radial direction, the term corresponding to overall heat transfer from the heating
or cooling medium to the reactor has disappeared from the equations. Heat transfer within the
bed is now being treated explicitly in the equations, and heat transfer across the wall is included
in the boundary condition at the wall. At the wall, we have the boundary conditions

where Tw is the temperature of the (inner surface of) the wall and αw is a heat transfer coefficient
from the wall to the bed. This assumes that the wall temperature is known. Of course, one might
replace the wall temperature with the temperature of a heating or cooling fluid and replace the
heat transfer coefficient with one that represents heat transfer from that fluid to the bed. Various
correlations for the effective conductivity and for the heat transfer coefficient are given by
Froment and Bischoff. If we wish to accurately predict radial temperature gradients, it is
essential that we use an accurate value for the effective conductivity. For given heat generation,
it determines the temperature rise within the reactor. Froment and Bischoff (beginning on p.
457) give a fairly detailed presentation of the application of this model to catalytic oxidation of
o-xylene to phthalic anhydride on vanadia catalyst. They show that there is strong parametric
sensitivity to the feed temperature, and how dilution of the catalyst bed with inert material can
reduce this sensitivity. They also compare this to an equivalent one-dimensional model, and find
that the predicted temperature for runaway differs by 5°C from the prediction of the 2-D model.

The next level of sophistication in modeling a fixed bed reactor is to explicitly account for
non-uniform axial velocity. Up to this point, we have assumed plug flow in all of the fixed bed
reactor models, even when we were considering radial variations in other quantities. This is a
good approximation across most of the bed, but near the wall there are variations in velocity due
to variations in the void fraction near the wall. This can be included in the model by modifying
the equations as shown in Froment and Bischoff on p. 463-464. This is still not a complete
hydrodynamic model, in the sense that there is no attempt made to accurately simulate the flow
patterns in the reactor. Rather it simply takes into account the effect of the wall and of the
changes in void fraction near it on the velocity in the reactor.
 All of the above models were categorized as pseudohomogeneous models. Only equations
describing the fluid phase were considered, and it was assumed that the reaction rate could be
written as a simple function of the gas phase concentrations. This is not necessarily the case. For
all of the above model types, we could also explicitly model the solid phase and any heat and
mass transfer processes within it and in the boundary layer adjacent to it. As we saw earlier when
we were considering intraparticle and interfacial gradients alone (without an overall reactor
model), we can write for a single (spherical) catalyst particle

where Cs is the local concentration within the particle (a function of position), Ts is the local
temperature within the particle, ξ′ is the radial position within the particle, rA(Cs,Ts) is the local
reaction rate within the particle, per unit particle volume, De is the effective diffusion coefficient
within the particle, λe is the effective thermal conductivity within the particle, kg is the mass
transfer coefficient for mass transfer between the external surface of the particle and the bulk
fluid, hf is the heat transfer coefficient for heat transfer between the external surface of the
particle and the bulk fluid, Css is the local concentration at the surface of the particle, Tss is the
temperature at the surface of the particle, C is the concentration in the bulk gas flow, and T is the
temperature in the bulk gas flow. This is a boundary value problem that we can solve for
particular values of the bulk concentration and temperature. It is coupled to the reactor model by
replacing the pseudohomogeneous reaction rate terms with terms that give the mass and heat
transfer to or from the particles. For the ideal plug flow model of the gas phase, the model
equations then become

The extensions of this to include axial and/or radial dispersion are exactly as in the
pseudohomogeneous model. There are two possibilities for treating this coupled model. The
most straightforward approach is to simply solve the boundary value problem for the particle at
each temperature and concentration predicted by the overall reactor model. It may be more
efficient to solve the equations for the individual particle for a grid of temperature-concentration
pairs and then parameterize these results or interpolate between the points in this table for use in
the overall reactor model. For each set of conditions, we would calculate the whole concentration
profile within the pellet, but all we need to keep for use in the reactor model is the global
effectiveness factor.

As we saw earlier in the semester, when the reaction is exothermic there can be multiple
possible steady states for the catalyst particle. Therefore, even for the ideal plug flow overall
reactor model, there can be multiple steady states for the whole reactor due to the possibility of
multiple steady states within the individual particles. Usually, due to heat transfer between the
particles, all of them will be in the same (either the upper, high conversion, high temperature or
the lower, low conversion, lower temperature steady state) steady state. Therefore, the whole
reactor is effectively either in a high conversion or low conversion steady state. If multiple
steady states are possible, then transient computations have to be performed to determine which
will be achieved. This is conceptually identical to what we did in constructing a phase plane for
a stirred tank reactor. Obviously, implementation of the transient modeling will be substantially
more difficult here than for the CSTR, since we have ordinary or partial differential equations for
the steady state conditions rather than algebraic equations. Application of heterogeneous
modeling for a PFTR is illustrated with two nice examples in Froment and Bischoff.

After completing your study of these lecture notes and the associated homework, you
should be able to:

(1) Construct a contour plot of reaction rate vs. conversion and temperature and use it to aid in
the design of the heat addition/removal system for a fixed bed catalytic reactor
(2) Write the governing equations for a fixed bed catalytic reactor at various levels of
approximation and explain the approximations that are being made in each set of
equations
(3) Use the Ergun equation and related correlations to find the pressure drop through a packed
bed
(4) Correctly couple the pseudohomogeneous reactor modeling equations to the governing
equations for reaction and diffusion within a catalyst pellet