Scale-up of bioprocesses

(2 lectures)
Chapter 11

Contents to be discussed (2 lectures)

• Scale up
• What are really the problems?
• Mixing
• What is it - and why is it?
• Mixing time
• Impellers
• Fluid flow
• CFD - uses and problems
• Rheology - Newtonian and non-Newtonian fluids
• Power consumption for stirring
• Scale-up effects on
• Heat evolution and cooling
• Mass transfer
• The Oldshue table!

1
Why it is a good idea to sort out problems as
early as possible..
Relative cost of making changes
to the process design
pre-design phase 1
design phase 10
construction phase 100
plant in operation 1000

“Commit your blunders on a small scale

and make your profits on a large scale”
H. Baekeland, the inventor of Bakelit, 1916

What is a scale-up problem?

A scale-up problem is something that we do not see

in the small-scale experiment but occurs in the large
scale process.
This can be a surprise and is very often
disappointing..

2
Scale up problems are of different nature

Biological
• The number of cell divisions from cell bank to fermenter is larger for large
scale processes than lab-scale work. This may lead to loss of properties in
GMOs (e.g. loss of plasmids).
• The longer overall process time increases the risk of contamination.
Chemical
• Cheaper chemicals (= less pure chemicals) need to be used in production
than in development work
• Salts to be avoided for easier downstream, i.e. neutralizing agents or
nitrogen supply should preferably not be salts
• By-products need to be disposed of
Physical
• Mixing, pressure (solubilities)

Seed train

Meyer et al, Chapter 1 in Industrial

Biotechnology: Products and
Processes, Wiley, 2017

3
 Scale up problems

Less predictable..
Predictable
• product quality
• mixing times
• (e.g. partial protein
• oxygen transfer oxidation, misfolding,
• heat transfer degradation)
• power input • foaming
• aeration rates • by-product formation

There are no excuses for missing these ones…

The safest way to increase production capacity

is simply to multiply the production units!

one roller bottle

Many roller bottles

This is, however, often expensive, so you’d better hope that the customer pays...

4
 The choice of a suitable reactor is the first
major scale-up question..

Bioreactors
submerged cultures immobilized cultures

bubble packed bed fluidized bed

stirred tank columns reactors reactors
reactors
airlift reactors

A typical industrial stirred tank reactor

5
Task of the bioreactor

Function
Containment (ensurance of sterility)
Introduction of gaseous reactants (e.g. oxygen)
Introduction of liquid reactants (e.g. carbon source)
Removal of gaseous products (e.g. carbon dioxide)
Control of the physical environment (e.g. temperature, shear rate, pH)
Suspension (e.g cells, particulate matter)
Dispersion (two-phase systems)

6
 Scale-up
What is really the difference between growth in an
agar plate and in a bioreactor?

• Size (obviously) and price

• 3D instead of 2D
• Oxygen transfer
• Heat transfer
• Mixing

Some typical values

Lab-scale Production scale

(negative cash flow) (positive cash-flow)
5-15 l bioreactor 30 m3 bioreactor
• OTR • OTR
• 300-500 mmol/l h • 100 mmol/l h
• Heat transfer • Heat transfer
• 40-70 kW/m3 • < 20 kW/m3
• Power input • Power input
• 15-30 kW/m3 • 1-3 kW/m3

7
 Meyer et al, Chapter 1 in Industrial
Biotechnology: Products and
Processes, Wiley, 2017

Mixing

8
9
What is mixing?

Mixing is achievement of uniformity

The problem here is how to best characterize the mixing,

knowing beforehand that a complete characterization is
impossible

Julio Ottino

Purpose of mixing

Process Example
• Blending • Substrate, pH-control)
• Suspending • Sludge treatment
• Dispersing • Two-phase processes
• liquid/liquid • glycerol/water (P. pastoris)
• gas/liquid • Aeration!
• Heat transfer • Temperature control

10
How does mixing occur?

• Bulk flow convection (distribution)

• Stretching 1/ 4
  3 
• Turbulent dispersion (down to Kolmogorov l min   l 

size of eddies)  P/M 

• Molecular diffusion tD = lmin2 / (4 π2 DA)

11
A simple quantification of mixing

s

s(t)

s0
Degree of mixing, E

s (t )  s 0
t time
E
s  s0

Mixing time

Mixing is (approximately) a first order process, which gives

c(t )  c final
1 E   exp( k mixt )
c final  c0

Mixing time, tm, is the time needed for E to reach e.g 0.63 (or 0.95 or
0.99)

tmix.E = ln(1-E)

tm is related to tc (circulation time)

12
 Estimating mixing time

V
v pump  d s N tc 
3
v pump

Stirrer pump capacity Circulation time

As a first estimate, tm can be approximated to 4 tc (circulation time)

for fully turbulent flow. For a geometrically similar reactors with
the same impeller types, we thus get:
1
tm 
N

Propeller or disk turbine?

Characteristics Propeller Disk turbine

Flow direction axial radial
Gassing less suitable highly suitable
Dispersing less suitable highly suitable
Suspending highly suitable less suitable
Blending highly suitable suitable

13
 Impeller types

Fluid foil impeller Rushton turbine

Propeller Pitched blade turbine

14
 The Navier-Stokes equation

A full description of the flow pattern requires

solving the Navier-Stokes equation.
This equation is really nothing but Newton’s
2nd law of motion:
F = d(mv)/dt or rather d(mv)/dt = F
(however including also a convective transport
term for momentum)

Standard form of N.S.

Du
l   p   l  2 u   l g
Dt

Rate of
pressure forces
change of
p.u.v. shear forces gravity
momentum

15
The flow in a stirred tank fermentor is difficult to
simulate precisely due to

• Multiphase flow
• gas and liquid, and possible even more phases like cells or
particles
• Turbulence
• Non-newtonian rheology

Often CFD calculation are made using time

averaged flow

Calculation of a time averaged flow in a stirred reactor with one Rushton turbine
Lundén, 1994

16
 Injection and spread of a tracer

Lundén, 1994

Computational Fluid Dynamics Model for Predicting Flow of Viscous Fluids in a Large Fermentor with Hydrofoil
Flow Impellers and Internal Cooling Coils

Biotechnology Progress
Volume 14, Issue 2, pages 248-258, 5 SEP 2008 DOI: 10.1021/bp9701168
http://onlinelibrary.wiley.com/doi/10.1021/bp9701168/full#fig6

17
 Computational Fluid Dynamics Model for Predicting Flow of Viscous Fluids in a Large Fermentor with Hydrofoil
Flow Impellers and Internal Cooling Coils

Biotechnology Progress
Volume 14, Issue 2, pages 248-258, 5 SEP 2008 DOI: 10.1021/bp9701168
http://onlinelibrary.wiley.com/doi/10.1021/bp9701168/full#fig7

Computational Fluid Dynamics Model for Predicting Flow of Viscous Fluids in a Large Fermentor with Hydrofoil
Flow Impellers and Internal Cooling Coils

Biotechnology Progress
Volume 14, Issue 2, pages 248-258, 5 SEP 2008 DOI: 10.1021/bp9701168
http://onlinelibrary.wiley.com/doi/10.1021/bp9701168/full#fig8

18
 Computational Fluid Dynamics Model for Predicting Flow of Viscous Fluids in a Large Fermentor with Hydrofoil
Flow Impellers and Internal Cooling Coils

Biotechnology Progress
Volume 14, Issue 2, pages 248-258, 5 SEP 2008 DOI: 10.1021/bp9701168
http://onlinelibrary.wiley.com/doi/10.1021/bp9701168/full#fig9

Turbulent eddies can be resolved using LES

Derksen and van den Akker. Large Eddy Simulations on the Flow Driven by a Rushton Turbine.
AIChE J, 45:209-, 1999

19
 Modelling of multiphase flow

To resolve the flow one must

• Model phenomena on a smaller scale than the grid
• Turbulence goes down to the micrometer scale
• Relative velocity of bubbles, particles and liquid
• Collision, coalescence and break-up of bubbles

Many systems do not have stationary solutions

Rheology
dv y
Shear rate (s-1)  
dz

Shear stress (N m-2)   

Dynamic viscosity (kg m-1 s-1)

Newtonian fluids
“The resistance which arises from the lack of slipperiness originating in a
fluid, other things being equal, is proportional to the velocity by which
parts of the fluid are being separated from each other”
Isaac Newton (1687)

20
 Some values of viscosities

Fluid Temperature (C) Dynamic viscosity (kg m-1 s-1)

Water 0 1.793 10-3
21 9.84 10-4
100 5.59 10-4
Ethanol 20 1.200 10-3
Glycerol 60 0.98
Sucrose solution (20 wt%) 21 1.916 10-3
Sucrose solution (60 wt%) 21 6.02 10-2
Olive oil 30 8.40 10-2
Molasses 21 6.6

Sometimes the unit poise is used. 1 P is 0.1 kg m-1 s-1 (Pa s).
1 centipoise (cP) is 0.001 kg m-1 s-1 i.e. approximately the viscosity of water at
room temperature.

Non-newtonian fluids

For a non-newtonian fluid, the viscosity is changing with rate of shear

(or with time). Typical examples are
• Dilatant fluids (viscosity  as )
• Pseudoplastic fluids (viscosity  as )
• Bingham plastic fluids (exhibits a yield stress – e.g. Ketchup)

Bingham fluid Dilatant fluid

shear stress (N m-2)

Newtonian fluid

Pseudoplastic fluid

shear rate (s-1)

21
Rheological models for non-newtonian fluids
K = consistency index, n = power law index

  K 
Power law model (or Ostwald- de n
Waele model):

Bingham fluid    0  

A yield stress to be exceeded before flow starts

   0  K 
n
Herschel-Bulkley fluid
i.e. Power law with a yield stress

Determination of viscosities are made in

rheometers

Vane - Cup Vane - Vane

22
Determination of viscosities are made in rheometers

Fermentation rheology

• Single cell cultivations of yeast and bacteria are often

Newtonian
• Cultivations involving filamentous fungi or excretion of
polysaccharides are often pseudoplastic and sometimes
exhibit a yield stress.

23
Rheology of a fiber suspension
(steam pretreated spruce)

Shear rate dependent viscosity, which is strongly affected by WIS content

M. Wiman et al. Biotechnol. Bioeng. 108, 1031–1041, 2011

Power consumption for stirring

Power input required for stirring is affected by

•Type of stirrer
•Number of stirrers
•Aeration
•Stirring rate
•Properties of the liquid..

24
 Power consumption in stirred tank reactors
Power number (Newton number) Reynolds number (defined
based on the stirrer
diameter)

N p  f (Re s ,..)
𝑃 𝑙 𝑁 𝑑𝑠 2
𝑁𝑝 = 𝑅𝑒𝑠 =
𝑙 𝑁3 𝑑𝑠 5


Stirrer rate diameter of stirrer

 𝑁 𝑑𝑠 2
Kinematic viscosity = 𝑅𝑒𝑠 =
𝑙 

Dimensional analysis

We assume that the power consumption (given a certain geometry of the system) will be determined by
the physical parameters liquid density, l and viscosity and the design parameters stirrer diameter and
stirrer speed. We thus assume a correlation of the form

P     N  d s (1)

where the , , , and  are unknown exponents.

Eq. 1 must be dimensionally correct. Using the notation mass =M, length =L and time = t, P has
dimensions M L2 t-3, l has dimensions M L-3,  has dimensions mass M L-1 t-1, N has dimensions t-1, and
ds ha dimension L Thus, we get 3 linear constraints on the exponents, which can be written
 
 1  1 1 0 0  
 2    3  1 0 1    (2)
    
  3  0  1  1 0  
 

This leaves 2 degrees of freedom for the five variables, and we should be able to write the correlation in
terms of 2 dimensionless groups. The three constraints can be written

25
1   (3)
2  3     (4)
 3     (5)

Expressed in terms of  we get:

  1 
  3 
  2    3(1   )  5  2

P   
P 1  
 N 3 
d 5 2   
s
N d s  Nd s2 
3 5

Nd s2
Re s  (Reynolds number based on stirrer diameter)
N p  Re s 

P
Np  (Newton number or power number)
N 3 d s5

Power number vs. Reynolds number

100
laminar
region
Power number, N p

10
transition turbulent
region region

Np is constant at
turbulent conditions!
1
1 10 10 2 103 104 10 5
Reynolds number, Re s

26
 The choice of impeller affects the power consumption

Laminar Turbulent
region region
Power number, Np

Turbine impeller,
100
blade width = 1/5 of turbine diameter
50
Pitched blade impeller,
blade width = 1/8 of turbine diameter
20

10

1
1 10 102 103 104 105

Reynolds number, Res

Aeration lowers the power consumption

1.0
Number of cavities
increase from 1 to 6

Cavities increase
Vortex trails in size
and small bubbles Flooding
Np,g/Np

0.5
Increasing
Froude number

0
0.01 0.02 0.03 0.04 0.05
Aeration number, NA

Figure 11.5 Np,g/Np sketched as a function of NA for a 6-bladed Rushton turbine. Adapted from
Ekato Handbook of Mixing Technology (www.EKATO.com))

27
 Heat production in bioreactors

The main sources of heat production are:

• Metabolic heat
• can be estimated from the oxygen consumption rate in
aerobic processes
• 𝑄 = 460𝑞𝑂 kJ m-3 h-1
• Dissipated stirring power
• For a large scale reactor this may be 1-5 kW m-3

Heat transfer
c p
Pr 


0.14
  
Nu  C Re 2 / 3 Pr1 / 3  


 w

where Nu = i df/
correction for viscosity changes
close to reactor wall

28
Heat transfer is affected by scale

The heat evolution increases with volume, but the

reactor wall heat transfer area increases only with
volume2/3 (unless internal coils are added)

Reactor volume (m3) Surface area/ volume (m-1)

0.001 55.5
0.1 11.9
1 5.5
50 1.5
100 1.2

Scale-up related effects on mass transfer

• Superficial gas velocity

• Non-constant power dissipation
• The mass transfer is higher close to the
impellers – the reactor can no longer be
regarded uniform
• Surface aeration
• Important in small scale, but not in large
scale

29
Superficial gas velocity changes with scale

Reactor volume (L) Superficial gas velocity (m s-1)

10 0.006
100 0.013
1000 0.028
10000 0.063
Superficial velocity as a function of scale for a fixed aeration rate of 1 vvm
(volume per volume per min) in a cylindrical stirred tank reactor with h/dt = 1.

You likely have to decrease the “vvm” used in lab-scale to

avoid the risk of flooding (or change reactor or impeller
design). See example 11.4

Not everything can be maintained constant

during scale-up!

Property Pilot scale Plant scale (12.5 m3)

(100 L)
P 1 125 3125 25 0.2
P/V 1 1 25 0.2 0.0016
N 1 0.34 1 0.2 0.04
ds 1 5 5 5 5
upump 1 42.5 125 25 5
tc 1 2.94 1 5 25
Nds 1 1.7 5 1 0.2
Res 1 8.5 25 5 1

Oldshue, 1966

30
 It is not possible to maintain all process
parametes constant during scale-up!

Property Pilot scale Plant scale (12.5 m3)

(100 L)
P 1 125 3125 25 0.2
P/V 1 1 25 0.2 0.0016
N 1 0.34 1 0.2 0.04
ds 1 5 5 5 5
To maintain same mixing time, the
upump 1 42.5 125 specific power5 input would need to
25
tc 1 2.94 1 5 drastically 25
be increased!
Nds 1 1.7 5 1 0.2
Res 1 8.5 25 5 1

Oldshue, 1966

The mixing time, will for instance, be substantially

longer in large scale at realistic power input

Example
Use of a compartment model to predict mass
transfer

Bubble regions

Well-mixed
regions

Coalescing case Non-coalescing case

Oosterhuis et al.

31
Use of a compartment model to predict
mass transfer

0.4
 Pg 
kl am  2.6  10 u  
2 0.5
s
Coalescing
V 
0.7
 Pg  Non-coalescing
kl am  2  10 u  
3 0.2
s
V 

Bubble region
kl ab  0.32us0.7

Use of a compartment model to predict

mass transfer

Unstructured
3 stirred tank

Oxygen
transfer
2
rate (arb. 2 compartment model
units)

1
Region of measured data

0.01 0.02 0.03

Superficial gas velocity (m s-1

32
Rules of thumb?
Possible simple scale-up criteria

• Constant blending time (tm)

• Is the blending of a substrate (base) critical to the
process?
• Constant (mean) kLa
• Is the oxygen transfer rate limiting for the process?
• Constant power input/volume
• Is the stirring power a main cost factor? A constant
power input of gives similar scale-up as constant kLa

Use your imagination and make sound

engineering approximations of key values
when scaling up!

Identify important Calculate Need for change of

physical processes Need for scale-down characteristic reactor geometry or
and microbial experiments? response times type of reactor?
reactions

Construct a simple
Decide on a scale-up model allowing
strategy exploratory simulation

33