Fluid flow and mixing in

bioprocesses (BENG0005)
Fluid flow in columns (packed beds) Lecture
Objectives of this section

We are working towards the following case study:

• Case study 2: Fluid flow in columns

Learning outcomes
• Estimate pressure drop down a packed chromatography column
• Analyse and design a complete chromatography system based on fluid flow
Lecture Outline
• Packed beds in bioprocessing (chromatography)
• Column design
• Fluid behaviour in packed beds
• Derivation of Kozeny-Carman (K-C) equation
• Application of K-C equation
• Critical velocity and deviations from K-C model
Packed beds in bioprocessing
What is a packed bed?
Packed bed in bioprocessing

• Packed bed is any porous structure made up of particulates

compressed together
• Chromatography is prime example, units composed of two essential
parts:
• Column is usually steel or glass tube, filled with;
• ‘matrix’ – essentially porous beads

• Column is ‘packed’ to compress beads together, reduce space

between

• Chromatography is essential purification step for bioprocessing –

need to be able to characterise
Chromatography column

• Pump
• Column – Packed
• Valves
• Detector
• Connections
Chromatography column
Chromatography column

• Solid phase (VS)

Beads/matrix
• Mobile phase (moves through
void volume)
Buffer
Sample for separation
Chromatography column
• Long tortuous path through varying cross-
sections
• Typically slow flow in vertical direction, little to
no spread of fluid in horizontal plane
• Distribution of liquid across width is therefore
important – particularly if using adsorptive
matrix, ensure full utilisation of capacity

• Hagen-Poiseuille: if cross-section is relatively

small compared to flow rate, pressure will be
large
Fluid behaviour in packed beds

• Long tortuous path through varying cross-sections

• Typically slow flow in vertical direction, little to no spread
of fluid in horizontal plane
• Distribution of liquid across width is therefore important –
particularly if using adsorptive matrix, ensure full
utilisation of capacity

• Hagen-Poiseuille: if cross-section is relatively small

compared to flow rate, pressure will be large
Calculating pressure drop

Pressure we can apply will be limited by compressibility of matrix and structure of column
What will it be affected by?
• Column length
• Fluid viscosity
• Flow rate
• Porosity of bed

To design process we need to know what the pressure drop will be across given bed design
Since cross-section of a void space is so small, flow will be laminar, so we can use H-P
But we don’t have a regular cross-section as in a pipe
Kozeny-Carman equation
Kozeny-Carman equation
• Kozeny realised that a porous material of cross-sectional area A
and length l could be modelled as a series of small capillary
tubes, where each tube has a diameter of D
• Only a certain amount of the material (the tubes themselves)
would be open to fluid transfer. This is equivalent to the cross-
sectional area of all the tubes, A’

• The porosity, or voidage, of the material relates the two areas:

𝐴′
∈! =
𝐴

• Inside each tube, diameter would be so small that laminar flow

must apply, so Hagen Poiseuille equation can be applied
Kozeny-Carman equation

• Carman later modified this assumption to

account for tortuous path of fluid
• The thickness of the layer is l, but mean length
of the tubes themselves (describing actual fluid
path) is l’
• The actual velocity of fluid inside tubes must
also be higher to have same superficial
velocity, u – call this interstitial speed, u’
Kozeny-Carman equation

Starting from Hagen-Poiseuille and expression for voidage:

!"#$% )*
∆𝑝 = and 𝜖( =
&² )

Kozeny found that:

𝑢𝐴 𝑢
𝑢𝐴 = 𝑄 = 𝑢 * 𝐴* → 𝑢 * = =
𝐴′ 𝜖(
Carman found that:
𝑢 𝑙′
𝑢* = *
𝜖( 𝑙

So the Hagen-Poiseulle equation for the ‘tubes’ that make up the void spaces is:
32𝜇𝑢𝐿 32𝜇𝑢𝐿𝑙′
∆𝑝 = =
𝐷² 𝜖( 𝐷²𝑙
 Kozeny-Carman equation
Now we need an expression the characteristic diameter, D, which is the diameter of the ‘tubes’.

First we need the void volume in terms of solid volume and voidage. Starting from:
𝐴′ 𝑉!
𝜖! = =
𝐴 𝑉"

𝑉! = 𝜖! 𝑉"

The solid volume will be equal to the total volume less the void volume, so:
𝑉# = 𝑉" − 𝑉! = 𝑉" − 𝜖! 𝑉" = 𝑉" (1 − 𝜖! )

𝑉#
𝑉" =
(1 − 𝜖! )
Substituting this into 𝑉! gives:
𝜖! 𝑉#
𝑉! =
(1 − 𝜖! )
Kozeny-Carman equation
Returning to the ‘tubes’ that make up the spaces through which fluid can flow in our packed bed:
• The tubes will have a certain internal volume, given by 𝑉(
• and they will also have an internal surface area, which we will call S
• Both of these properties will be dependent upon the characteristic diameter of the tubes, D. We can exploit this
to find an expression for D in useful terms.
Start by dividing both sides of the 𝑉( expression by S:

𝑉( 𝜖( 𝑉+
=
𝑆 𝑆(1 − 𝜖( )
The total internal volume of our ‘tubes’ will be given by: Where:
𝑛𝜋𝐷²𝑙′ n is the number of tubes
𝑉( = D is the diameter of the tubes
4
l‘ is the length of the tubes
While the total internal surface area will be given by
𝑆 = 𝑛𝜋𝐷𝑙′
Kozeny-Carman equation
$!
Substituting the terms for Ve and S into the %
equation gives:
𝑛𝜋𝐷²𝑙′ 𝜖! 𝑉#
=
4𝑛𝜋𝐷𝑙′ 𝑛𝜋𝐷𝑙′(1 − 𝜖! )

𝐷 𝜖! 𝑉#
=
4 𝑛𝜋𝐷𝑙′(1 − 𝜖! )

4𝜖! 𝑉#
𝐷=
𝑆(1 − 𝜖! )
Which is a general equation for the characteristic diameter of void spaces in a packed bed.
Our bed is made up of spheres however, with nominal diameter ds. If we let ns be the number of spheres, then our solid
volume will be given by:
𝑛# 𝜋𝑑# ³
𝑉# =
6
And the total surface area of the void spaces (which is the same as the surface area of the beads) will be:
𝑆 = 𝑛# 𝜋𝑑# ²
Kozeny-Carman equation
Substituting these into the 𝐷 equation gives:
4𝜖( 𝑛+ 𝜋𝑑+ ³ 2𝜖( 𝑑+
𝐷= =
6𝑛+ 𝜋𝑑+ ² (1 − 𝜖( ) 3(1 − 𝜖( )

Putting this back into Hagen Poiseuille equation gives:

"
32𝜇𝑢𝐿𝑙′ 3 1 − 𝜖( 72𝜇𝑢𝐿𝑙′ 1 − 𝜖( ²
∆𝑝 = * =
𝜖( 𝑙 2𝜖( 𝑑+ 𝑑+ ²𝜖( ³𝑙

It has been shown empirically that l’/l = 2.5 for a packed bed, so:
180𝜇𝑢𝐿 1 − 𝜖( ²
∆𝑝 =
𝑑+ ²𝜖( ³
Or:
𝜇𝑢𝐿 𝑑+ ²𝜖( ³
∆𝑝 = Where β =
𝛽 180 1 − 𝜖( ²
Kozeny-Carman equation

Assumptions:
• Particles are uniformly spherical
• Particles themselves are not porous
• Particles are incompressible

N.B.: It is reasonable to assume a voidage of 0.3 - 0.4. If specific properties of the mobile
phase are not given, you may assume those of water.
Practice 1
A packed column with a diameter of 0.2 m and bed length of 0.6 m is filled with matrix which has a
sphere radius of 175 µm. The voidage of the bed is 0.3 and the flow rate is 0.57 m³/h. What is the
pressure drop across the column?
Practice 2
You have a chromatography system for which you need to estimate the pressure drop. The
following information is known:

L = 0.5m
∈e = 0.3
ds = 150 × 10-6 m
u = 0.01 ms-1
µ = 0.001 N s m-2
ρ = 1000 kg m-3

What is the pressure drop?

Reflections - Implications of equation

• What happens if we double the flow rate or bed length?

• What happens if we double the particle diameter?

Practice 3
A packed column with a diameter of 300 mm and a bed height of 0.15 m is operated at a flow rate
of 80 l/h. At this flow rate, the pressure drop is measured at 2.5 bar. Assuming a voidage of 0.3,
what is the radius of the particles in the bed?
Deviation from model
• We assume in K-C that the matrix is incompressible,
but this is not the case

• It will compress slightly under pressure, up to a point

• Beyond this point the wall and bead to bead frictional

support is less than the force exerted by drag and
gravity

• Void spaces are closed, column impassable

• Pressure decouples from flow rate, increases

exponentially

• Irreversible, matrix unusable

Deviation from model
• Interested in optimisation – max flow rate
without risk of trashing expensive matrix
(£1600/L is a cheap example!)
• There is a point immediately before the bed
collapses which is maximum flow rate we can
apply
• We define the superficial velocity at this flow
rate as ucrit
Critical velocity, ucrit
Beyond the ucrit, the relationship between Q and Δp
becomes non-linear

What impacts ucrit?

• Matrix compressibility – more compressible material or more
porous, lower ucrit
• Matrix particle size – larger particles tend to be more
compressible, therefore same effect
• Matrix age – over time beads become more compressed,
reduces permeability therefore lowering ucrit
• Liquid viscosity – increase results in greater liquid entrainment 𝑄,-./
in solid phase. Thus a decreased permeability and a lower ucrit 𝑢,-./ =
𝐴
• Wall support – greater frictional support from column wall means
higher ucrit
More practice - Critical velocity, ucrit
From the following data for a 16 mm diameter column, which is the
critical flow rate and therefore what is the critical velocity?

Q Δp
(mL/min) (bar)
3.6 0.5
6.9 0.75
10.8 1.00
13.5 1.25
17 1.50
19.9 2.60
Critical velocity, ucrit
3
Q Δp
2.5
(mL/min) (bar)
3.6 0.5 2

Δp (bar)
6.9 0.75 1.5
10.8 1.00
13.5 1.25 1

17 1.50 0.5
19.9 2.60 0
0 10 20 30
Q (ml/min)

17𝑚𝑙 1𝑚 ! 1𝑚𝑖𝑛
𝑄,-./ × 0 ×
𝑚𝑖𝑛 10 𝑚𝑙 60𝑠
𝑢,-./ = = = 1.41×101! 𝑚/𝑠
𝐴 1𝑚
𝜋×(8𝑚𝑚× ! )²
10 𝑚𝑚
Column diameter and ucrit

• Under pressurised flow, structure of matrix is reliant

upon frictional forces – particle/particle and particle/wall
• Particle/particle friction not as great as particle/wall

• Particle in centre of column will be less supported

• Particle at wall of column will be more supported

• Particles further from the wall have greater freedom of
displacement
Column diameter and ucrit

• Increasing column diameter

means much smaller proportion
of particles have wall support
• Frictional forces will be lower
overall, therefore matrix is more
susceptible to drag of fluid flow
• ucrit decreases – inversely
proportional

Going back to previous example, what would happen if we tested a

greater column diameter?
Practice 4
Using the following data estimate the value of ucrit for the each of the column diameters tested. From
this provide an estimate as to the flowrate that a 80mm (ID) column should be operated under.

16 mm (ID) 26 mm (ID) 50 mm (ID) 100 mm (ID)

Q Δp Q Δp Q Δp Q Δp
(mL/min) (bar) (mL/min) (bar) (mL/min) (bar) (mL/min) (bar)
3.6 0.5 1.0 0.5 16.5 0.7 145 0.2
6.9 0.75 7.0 0.9 59.5 1.4 210 0.4
10.8 1.00 16.9 1.6 84.0 2.1 270 0.7
13.5 1.25 30.4 2.6 104.5 2.8 300 1.1
17 1.50 37.2 3.3 121.0 3.5 330 1.4
19.9 2.60 39.0 4.1 124.0 3.9 335 1.8
Problem 4
16 mm (ID) 26 mm (ID) 50 mm (ID) 100 mm (ID)
Q Δp Q Δp Q Δp Q Δp
(mL/min) (bar) (mL/min) (bar) (mL/min) (bar) (mL/min) (bar)
3.6 0.5 1.0 0.5 16.5 0.7 145 0.2
6.9 0.75 7.0 0.9 59.5 1.4 210 0.4
10.8 1.00 16.9 1.6 84.0 2.1 270 0.7
13.5 1.25 30.4 2.6 104.5 2.8 300 1.1
17 1.50 37.2 3.3 121.0 3.5 330 1.4
19.9 2.60 39.0 4.1 124.0 3.9 335 1.8

Qcrit highlighted in red (beyond these values Δp increases

disproportionately). Extrapolate from these to 80mm value.
Problem 4 ucrit vs. ID
0.0016
0.0014
0.0012 y = 0.0089x-0.647
0.001
ucrit (m/s)

0.0008
0.0006
0.0004
0.0002
0
0 20 40 60 80 100 120
ID (mm)

𝑢"#$% = 0.0089×80&'.)*+ = 5.225×10&*

𝑄"#$% = 5.225×10&*×𝜋× 0.04𝑚 , = 2.626×10&)𝑚³/𝑠

𝑄"#$% = 157.6𝑚𝑙/𝑚𝑖𝑛
Practice - Considerations for scaling

• If we maintain superficial velocity, and increase diameter to increase throughput (greater Q) –

do you foresee any problem with this approach?

• If greater adsorptive capacity is required (e.g. protein A column) will need greater VS and VT
overall – how do we achieve this given the above?
Practice 5
You have 100g of protein to separate. The capacity of your matrix is ~ 10 mg
protein/ml of matrix. The pressure limit for the material is 3 bar. Assuming the column
diameter to be one third the height will the following design be acceptable? If not,
what will you alter and by how much?

ds = 30 µm
µ = 0.003 N s m-2
ρ = 1.1 x 103 kg m-3
∈ = 0.4 Rule of thumb, capacity
should be 5-10x load weight
dc = 0.4 m
Q = 190 L/h