Computers and Chemical Engineering 37 (2012) 48–63

Contents lists available at SciVerse ScienceDirect

Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng

An iterative modelling approach for improving the performance of a pulsed

electric field (PEF) treatment chamber
Kai Knoerzer ∗ , Pascal Baumann, Roman Buckow
CSIRO Food and Nutritional Sciences, Private Bag 16, Werribee, VIC, Australia

a r t i c l e i n f o a b s t r a c t

Article history: An important component of the pulsed electric field (PEF) technology is the treatment chamber in which
Received 21 March 2011 the food is exposed to high voltage pulses. Non-uniformity of the treatment, particularly with respect to
Received in revised form 8 July 2011 the electric field distribution, is a common problem in continuous PEF treatment chambers. A previously
Accepted 2 September 2011
developed and validated Multiphysics model of a pilot-scale PEF system with co-linear electrode config-
Available online 8 September 2011
uration was simplified and embedded into an iterative algorithm, automatically modifying the model’s
treatment chamber geometry and dimensions, and evaluating the simulated process with respect to a set
Keywords:
of performance indicators. The algorithm was capable of identifying configurations that were superior
PEF
Pulsed electric fields
to the standard co-linear treatment chamber configuration. A 3D Multiphysics model with the identified
Modelling geometrical properties was developed and a corresponding treatment chamber manufactured. The Mul-
Optimisation tiphysics model was validated by comparing measured and predicted temperatures at various process
Electric field strength conditions induced in NaCl solution and apple juice.
Treatment uniformity Crown Copyright © 2011 Published by Elsevier Ltd. All rights reserved.

1. Introduction The electric field strength E is commonly seen as the main pro-
cess variable and can be estimated by Eq. (1):
Pulsed electric field (PEF) processing is an innovative mild
V
treatment technology, which can be used for non-thermal pas- E= (1)
h
teurisation of pumpable foods at low or moderate temperatures.
The mechanism of microbial inactivation is based on the dis- where V is the applied voltage and h is the distance between high
charge of high voltage electric pulses (up to 70 kV cm−1 ) of a voltage and grounded electrode (typically in parallel plate config-
few microseconds into the liquid within the treatment zone, urations).
which is located between a grounded and a high-voltage electrode The dissipated energy into the liquid, Qspec , leads to an increase
(Angersbach, Heinz, & Knorr, 2000; Heinz, Alvarez, Angersbach, & in temperature due to ohmic heating and can be estimated by the
Knorr, 2001). During the treatment, the membrane potential of electric field strength applied to the product and its electrical con-
microbial, plant or animal cells is exceeded, which leads to the for- ductivity and flow rate:
mation of pores, causing a release of intracellular liquid and cell 
death (Zimmermann, Pilwat, Beckers, & Riemann, 1976). Unlike Qspec = ṁ−1 ·  · f · (x, y, z) · (E(x, y, z))2 dxdydz (2)
in thermal pasteurisation, where heat conduction is a time lim-
iting factor, the delivery of the lethal treatment in PEF processing where  is the temperature and, hence, location dependent elec-
is instantaneous. trical conductivity of the media,  the pulse width, f the pulse
The efficiency of the PEF treatment depends on a number of repetition rate (frequency), ṁ the mass flow rate and x, y, z the
process variables, particularly the electric field strength, the spe- spatial coordinates throughout the treatment chamber.
cific energy input, the treatment temperature and time, but also on The typical electric field strength required for liquid food pas-
the material properties, such as the electrical conductivity of the teurisation is in the range of 20–40 kV cm−1 (Toepfl, Mathys, Heinz,
treated product. & Knorr, 2006). To estimate the increase in temperature, Eq. (3) can
be applied:
Qspec
T = (3)
Cp (T )
∗ Corresponding author at: 671 Sneydes Road, Werribee, VIC 3030 Australia. Tel.:
+61 3 9731 3353; fax: +61 3 9731 3202; mobile: +61 4 2912 6767. where Cp (T) is the temperature dependent specific heat capacity of
E-mail address: [email protected] (K. Knoerzer). the treated product.

0098-1354/$ – see front matter. Crown Copyright © 2011 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2011.09.002
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 49

As many PEF process variables are almost impossible to deter- decreasing insulator bore diameter), it was possible to identify a
mine experimentally (at least without disturbing or influencing design superior in performance compared to the treatment cham-
the entire process with measurement equipment within the con- bers supplied by the manufacturer. The performance elements
strained space of the treatment zone), Multiphysics modelling can were implemented as features in the software routine in the form
be utilised to simulate and predict the electric field distribution, of a single dimensionless performance parameter (DPP). While this
the flow characteristics and the temperature distribution in PEF problem could also be addressed and solved with complex optimi-
processing. sation algorithms, an iterative algorithm solving a set of models
Several studies have shown that PEF leads to a considerable within a defined range of geometrical parameters was seen as
inactivation of microbial cells in different products (Heinz, Toepfl, preferable, given the simplicity of the approach.
& Knorr, 2003; Puertolas, Lopez, Condon, Raso, & Alvarez, 2009;
Toepfl, Heinz, & Knorr, 2007). A reliable and effective PEF treatment 2. Materials and methods
for microbial inactivation is highly dependent on the electric field
strength uniformity within the treatment zone of the PEF chamber. 2.1. Pulsed electric field system
Non-uniform electric fields lead to two major problems: Firstly,
occurrence of high electric field strength peaks causing potentially The Multiphysics PEF model was based on a Diversified
significant increases in temperature and, thus, over-processing in Technologies 25 kW Pulsed Electric Field System (Diversified Tech-
areas where the electric field strength peaks are located, and insuf- nologies, Inc., Bedford, MA, USA). The system consists of a PEF
ficient inactivation in regions where the threshold for microbial treatment enclosure and a modulator cabinet. The PEF treatment
inactivation is not reached. The second major problem of a non- chamber is connected to a liquid food handling system (two
uniform electric field is the erosion of insulators and electrodes due 150 L stainless steel supply and collection tanks), three tube heat
to arcing. Peaks in the electric field strength increase the risk of arc- exchangers (#PT 113-18-1500, Hipex Pty Ltd, Thomastown, Victo-
ing; hence, either the applied peak voltages have to be decreased, ria, Australia), a pump (#SLA 80A-4, CMG Pty Ltd, Rowville, Victoria,
which in turn causes lesser inactivation of microorganisms in areas Australia) and the control systems for all other settings like liquid
with lower electric field strength, or peaks must be prevented by flow rate, temperature and back pressure. The PEF system supplies
changing the chamber design and geometry (Fiala, Wouters, van mono-polar pulses of almost rectangular shape. The pulse widths
den Bosch, & Creyghton, 2001; Misaki, Tsuboi, Itaka, & Hara, 1982). (durations) can be set in a range of 1–10 ␮s and the frequency can
Recent studies on Multiphysics modelling of laboratory-scale be modulated up to 3,000 Hz. The maximum output voltage of the
PEF systems showed a drastic increase of temperature near the system is approximately 40 kV.
chamber walls and, particularly, at sharp electrode or insulator The temperature measurements for the validation of the model
edges (Gerlach et al., 2008). These temperature hotspots were with the improved chamber configuration were performed using a
explained by electric field strength peaks and insufficient mixing fibre optics system (#REFLEX 4 – RFX321A, Neoptix Inc., Quebec,
and recirculation of the liquid inside the chamber due to laminar QC, Canada). The fibre optic temperature sensors were inserted into
flow conditions. It was suggested to introduce turbulent flow with the second grounded electrode in a similar fashion as in the work
static mixing devices (Lindgren, Aronsson, Galt, & Ohlsson, 2002) of Buckow et al. (2010).
or grids (Jaeger, Meneses, & Knorr, 2009) to improve the effective-
ness and treatment uniformity of PEF pasteurisation processes. The 2.2. Pulsed electric field treatment chamber
application of such static mixing devices may limit PEF processing
to non-particulate liquid foods of low viscosity and can compli- 2.2.1. Previously validated model
cate cleaning and maintenance operations of the system. However, The geometry of the treatment chamber of the validated model
industrial-scale PEF processing usually provides liquid velocities developed by Buckow et al. (2010) is illustrated in Fig. 1. The cham-
that are sufficient to create turbulence, which, in turn, may avoid ber consists of two grounded electrodes made of stainless steel,
or at least reduce the occurrence of temperature hot spots (Buckow, which are located at the top and the bottom of the chamber, one
Schroeder, Berres, Baumann, & Knoerzer, 2010). high voltage electrode made of stainless steel, which is situated in
Another possibility to overcome the above mentioned issues is the middle between the grounded electrodes, and two polytetraflu-
to modify the insulator design by decreasing the bore diameter oroethylene (PTFE) insulators, separating the high voltage and the
with a rectangular or elliptical (inward concave) cross-section (see grounded electrodes.
Fig. 5). Such modifications can change the electric field strength The CFD model was designed in 3D and is a good approxima-
distribution and flow characteristics to ensure turbulent regimes tion of the (rotation symmetric) pilot-scale treatment chamber. The
in this critical section of the treatment chamber. two PTFE insulators have an outer diameter of 98 mm, an inter-
The objective of the work presented in this manuscript was to nal diameter of 16 mm and a total height of 43 mm. Serving as
develop an improved chamber design with respect to the shape, spacer between high voltage and grounded electrode (h = 6.3 mm),
configuration and dimensions of insulators and electrodes; thereby the internal diameter decreases to 5.3 mm centred along the height
improving the uniformity of the PEF treatment with respect to of the bore, creating a zone with high electric field strength during
electric field strength distribution (also preventing arcing), thus, processing.
reducing erosion of the chamber material and dielectric break- The three electrodes have an outer diameter of 16 mm and an
downs that may occur during the treatment of dielectric materials internal diameter of 5 mm.
(Gongora-Nieto, Pedrow, Swanson, & Barbosa-Canovas, 2003).
A software algorithm was programmed in MATLABTM (The 2.2.2. Justification to utilise 2D axis-symmetric models for
Mathworks Inc., Natick, MA, USA) interfacing to a COMSOL chamber improvement studies
MultiphysicsTM (COMSOL AB, Stockholm, Sweden) model, capa- As the computational demand to solve one 3D model of the dis-
ble of progressively modifying the shape and geometry of the cussed system, involving the solution of the conservation equations
treatment chamber as determined by the cross-section of the of mass, momentum, energy and charge and the associated equa-
insulator (e.g., decreasing diameter), electrode distance and tube tions on the model boundaries, is high (approximately 1.25 h on
diameter. By extracting the models’ solutions and evaluating pro- a workstation with two dual core processors (each 2.33 GHz) and
cess performance (in terms of electric field strength, associated 20 GB RAM), it was not feasible to solve a large amount of models
uniformity, treatment volume and pressure drop caused by the with different chamber configuration (various shapes, diameters
50 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

Fig. 1. Dimensions and geometries of the electrodes and insulator of the former validated co-linear PEF treatment chamber (Buckow et al., 2010).

and gaps) and fully coupled charge, mass, momentum and energy where fi are the values from the 3D model, yi the values from the
conservation equations in 3D. 2D model and ȳ the mean of the values from the 2D model.
However, since the geometry of the discussed treatment cham- For this reason, R2 of greater than 0.85 suffices to justify a sim-
ber is rotation symmetric, a simplification of the 3D model to plification from 3D to a 2D axis-symmetric model without the risk
an axis-symmetric 2D model is possible, leading to a significant of obtaining inaccurate predictions, particularly for the electric
decrease in computational demand. To validate this approxima- field strength distribution, which was selected as the most rele-
tion, 2D and 3D models were developed with identical dimensions vant variable for improving the PEF treatment chamber (see Eqs.
(as per Buckow et al., 2010; Fig. 1), process conditions and material (24)–(29)). Furthermore, for the chamber improvement studies,
properties. The potential at the high voltage electrode was set to the Multiphysics models were decoupled, only solving the charge
20 kV, the chamber inlet temperature to 25 ◦ C, the liquid flow rate
to 5 L min−1 , the pulse repetition rate to 500 Hz, the pulse width
to 5 ␮s, and the liquid was assumed to have a constant electri-
cal conductivity of 5 mS cm−1 . Although it is known that electrical
conductivity increases with increasing temperature, for the com-
parison of predictions in 2D and 3D models, this dependency was
not seen relevant. The rationale behind this simplification was that
the temperature increase was expected to be identical in both sce-
narios (which was confirmed, see Fig. 3), hence, also the change
of the temperature-dependent electrical conductivity would be
identical, leading subsequently to identical predictions of the rel-
evant process variables. The predicted electric field strengths,
temperatures and velocities of both models were compared. The
comparison was performed in both treatment zones; in each treat-
ment zone 30 locations (covering an axis-symmetric plane with
a 3 × 10 matrix), at 3 different radial coordinates (symmetry axis,
1 mm distance from symmetry axis and 2 mm distance from sym-
metry axis) and 10 different heights (2 mm steps, from 6 mm before
the insulator region inlet to 6 mm behind the insulator bore outlet)
were selected (Fig. 2).
The results showed a very good agreement between the two
models, with coefficients of determination R2 of greater than 0.99
for temperature and electric field strength distribution and greater
than 0.85 for the velocity distribution (Fig. 3). The reason for the
lower R2 in the comparison of the velocities can be explained by the
fact that all velocity values are close (in the range of 3.9–4.4 m s−1 )
and, thus, due to the nature of the equation for R2 determination
(see Eq. (4)), the value for R2 becomes small (i.e., as all values of
yi are close to the mean ȳ, the denominator becomes small and,
therefore, also the value for R2 ).

 2
(yi − fi ) Fig. 2. 3 × 10 matrix, covering an axis-symmetric plane in the treatment zone of the
R2 = 1 − i 2
(4) 3D model (a) and the 2D model (b) for comparison of predictions and justification
(y − ȳ)
i i to utilise 2D axis-symmetric models in the optimisation study.
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 51

Fig. 3. Parity plots for temperature (a), electric field strength (b) and velocity (c) of 2D axis-symmetric and 3D models at an applied voltage of 20 kV, an inlet temperature of
25 ◦ C, a flow rate of 5 L min−1 , a pulse repetition rate of 500 Hz, a pulse width of 5 ␮s and an electrical conductivity of the processed liquid of 5 mS cm−1 .

conservation equation for predicting the electric field strength

distribution, to further decrease the computational demand and,
thus, increase model solving rate. It has been shown previously by
Buckow et al. (2010) that temperature changes also impact on the
electric field strength, which decreases with increasing tempera-
tures; this dependence, however, is only marginal with an increase
in temperature of around 10 ◦ C leading to a decrease of the electric
field strength of less than 10% (Buckow et al., 2010). Furthermore,
from a qualitative point of view, the distribution of the electric field
with, e.g., peaks at the edges where the insulators come in contact
with the electrodes, still remains very similar when temperature
changes. Hence, this further simplification is justified within the
scope of this study, which is the improvement of electric field uni-
formity and avoidance of electric field strength peaks throughout
the treatment zones.

2.2.3. Insulator configurations for performance improvement

studies Fig. 4. Comparison of the electric field distribution with four different insulator
As discussed earlier, the insulator shape and the geometric prop- shapes (30% inset).
erties (dimensions) of the treatment chamber have a pronounced
impact on the electric field distribution inside the chamber. Apart
from affecting the uniformity of the electric field, also the mag- equal to the electrode diameter (see Fig. 4a); “rectangular inset”,
nitude of the electric field strength and, thus, the associated i.e., the insulator bore had a smaller diameter than the electrodes
temperature increase is affected. Hence, the insulator shape and (see Fig. 4b); “elliptical inset”, i.e., the cross-section of the insulator
dimensions, the electrode gap and the electrode diameter were bore had an inward concave shape (see Fig. 4c); and “rectangular
chosen to be variables in the iterative algorithm for performance rounded edge inset”, i.e., the insulator had a smaller diameter than
improvement. the electrodes, with the edges chamfered by different radii (see
As possible candidates, four different insulator shapes were Fig. 4d). While more complex geometrical configurations in a co-
taken into consideration: “no inset”, i.e., the insulator diameter was linear electrode setup may have the ability of increasing the electric
52 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

fluid flow and electric fields. Such models require the simultaneous
solution of the partial different equations describing the conserva-
tion of mass, momentum, energy and electric charge.

2.3.1. Governing equations for electrostatics

Based on the law of charge conservation, the governing equation
for the electric potential is:

∇ ((T ) · ∇ · V − J) = 0 (5)

where J is the density of the electrical current.

Assuming static conditions, i.e., no generation of electromag-
netic forces, the relation of electric field and potential can be
described by:

Fig. 5. Geometry variables for the different types of insets ((a) “rectangular inset”, E = −∇ V (6)
(b) “elliptical inset”, (c) “rectangular rounded edge inset”).

where E is the electric field strength.

field uniformity to an extent which is not possible with the four
selected insulator shapes, the authors believe that from a manufac- 2.3.2. Governing equations for turbulent flow and heat transfer in
turing, processing and cleanability perspective the selected shapes the system
are preferable. The thermo- and fluid-dynamic behaviour of the system is
Four models were developed with 30% inset, i.e., a minimum described by conservation equations of mass, momentum and
internal insulator diameter of 70% that of the electrodes (except for energy (Chen, 2006). The development of a flow field is governed
the “no inset” model), with an applied voltage of 10 kV at the high by the continuity equation (mass balance):
voltage electrode (Fig. 4). Comparison of the electric field distribu-
tions clearly shows that the insulator shape significantly influences ∂
+ ∇ · (v) = 0 (7)
the uniformity of the electric field and the associated strength of ∂t
the electric field peaks. The “no inset” model shows the highest
where v is the velocity vector.
electric field strength of approximately 64 kV cm−1 at the interface
Due to the high flow rates investigated in this study, associ-
of insulator and electrode, which was about four times the value of
ated with high velocities in the treatment chamber, the fluid flow
the maximum electric field strength of the “elliptical inset” model
was turbulent with Reynolds numbers clearly exceeding 2,300
prediction, showing a peak strength of 16 kV cm−1 (Fig. 4c).
(the onset of turbulence in pipe flow). Turbulence was solved
Apart from the shape, other variables were also taken into con-
by applying the k–ε model which included an additional “tur-
sideration in the performance improvement studies. Four different
bulence viscosity” and “turbulent thermal conductivity” in the
geometry parameters were varied: The internal diameter d of the
equations for conservation of momentum and energy, respectively,
electrodes ranging from 2 mm to 20 mm, the height h of the elec-
to take the contributions of turbulent eddies into account (COMSOL
trode gap ranging from 1 mm to 30 mm, a total inset ins (i.e., the
Multiphysics, 2006; Nicolaï, Verboven, & Scheerlinck, 2007). The
internal diameter reduction of the insulator) in a range of 0–90%
turbulent viscosity T is given by:
of the electrode diameter d, and for the “rectangular rounded edge
inset” models also the chamfer radii rad ranging from 0 to 40% of the k2
diameter reduction ins (Fig. 5). The electrode length at both ends T = C (8)
ε
of the insulator region was kept constant at 20 mm. These parame-
ters were selected as they all affect the treatment. As simple design where C = 0.09 (Launder & Spalding, 1974), k is the turbulent
guidelines the following can be stated: The electrode gap h dictates kinetic energy and ε the dissipation rate of turbulence.
mainly the maximum achievable average electric field intensity Here, the momentum equation (extended according to COMSOL
according to Eq. (1), while the internal electrode diameter d impacts Multiphysics, 2006) gives the following expression:
on flow conditions (turbulent or laminar), pressure drop and treat-  
ment volume, but most importantly on treatment uniformity with ∂v
 + (v · ∇ )v = −∇ P + ∇ · (( + T ) · ∇ v) + g (9)
increasing diameter leading to more pronounced non-uniformity. ∂t
The inset of the insulator region, i.e., the decreased internal diam-
eter of the insulator, has very similar effects on process variables where v denotes the average velocity, P is the pressure,  repre-
and treatment uniformity as the electrode diameter. In addition, sents the dynamic viscosity of the fluid, and g represents the gravity
the greater the difference between electrode and insulator diam- constant.
eter, i.e., the greater the inset, the more uniform the electric field In addition to the continuity equation, the k–ε closure includes
distribution will be. The radii of the chamfered edges of the insula- two extra transport equations solved for both k and ε using empiri-
tor inset mainly affects the electric field uniformity, with increasing cal model constants (COMSOL Multiphysics, 2006). The k–ε closure
radii generally leading to a wider spread and, therefore, reduction equations were coupled with the energy conservation equation
of the field peaks at the edges. for heat transfer through convection and conduction, assuming
non-isothermal flow. This equation was modified (from Kowalczyk,
2.3. Governing equations Hartmann, & Delgado, 2004 and extended according to COMSOL
Multiphysics, 2006) by including the turbulent thermal conductiv-
As indicated earlier, the fully coupled scenarios, as utilised for ity kT (with kT = CP T /PrT ):
the comparison of 2D and 3D model predictions, as well as for the  
∂T
model of the improved treatment chamber, are Multiphysics prob- Cp + v · ∇ T = Qeff + ∇ · ((k1 + kT )∇ T ) (10)
∂t
lems, involving concurrent phenomena of heat transfer, (turbulent)
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 53

where k1 is the thermal conductivity of the liquid domain, and PrT where k1 and k2 are the thermal conductivities of the two respective
is the turbulent Prandtl number. The source term Qeff arises from subdomains.
electric energy dissipation in the liquid and can be expressed as:
2.4.3. Fluid inlet and outlet
Qeff = ϕ ·  · E 2 (11)
The boundary at the liquid inlet was defined as a flat velocity
where ϕ is a factor for the time-averaged potential (to account for profile in streaming direction; the full turbulent velocity profile
the pulsed potential in a stationary solution), which can be esti- was found to be developed within less than 10 mm upstream:
mated for ideal rectangular pulses as the product of pulse repetition
rate f and pulse width . v = v0 (18)

As the inlet velocity v0 changes for an adjusted volume flow at

2.4. Boundary conditions a temperature of 20 ◦ C after exiting the heat exchanger at different
set temperatures, this had to be taken into consideration, following
For each physical phenomenon, a set of boundary conditions the expression:
was defined to account for the interactions at solid–liquid and
solid–solid interfaces (between subdomains). V̇ (20 ◦ C) · ((20◦ C)/(T0 ))
v0 = (19)
A
2.4.1. Electrical interaction between subdomains with V̇ (20 ◦ C) as the volume flow rate at 20 ◦ C, (T0 ) as the density
The boundaries of inlet, outlet and PTFE insulator were assumed of the liquid at inlet temperature T0 , (20 ◦ C) as the density of the
to be electrically insulated: liquid at 20 ◦ C and A as the cross-section area of the inlet.
n·J =0 (12) A pressure of 1.8 bar was defined at the treatment chamber out-
let.
The same boundary condition applies to the symmetry axis in At the fluid inlet, a temperature boundary condition was
the 2D axis-symmetric model. applied:
The boundaries of the high voltage electrode were set as electric
potential with: T = T0 (20)

V = V0 (13) A convective flux boundary was applied at the outlet. This condi-
tion states that the heat transfer across this boundary is convection
with V0 as the maximum output voltage during a pulse. dominated and, therefore, radiation is negligible:
The other electrodes are grounded, i.e., the potential at the
boundaries is zero −n · (−k∇ T ) = 0 (21)

2.4.2. Fluid-dynamic and thermal interaction between 2.5. Material properties

subdomains
Fluid-dynamic boundaries had to be defined for the k–ε clo- The thermophysical properties of the salt (NaCl) solution and
sure between the solid the liquid domain only, whereas thermal apple juice (see Section 2.6) were assumed to be similar to pure
boundaries were defined for the energy balance for all domains. water, except for the electrical conductivity. Therefore, density,
For the 2D axis-symmetric models, symmetry boundary condi- specific heat capacity, thermal conductivity and viscosity were
tions were assumed on the symmetry axis and can be simplified to taken from the NIST/ASME database (Harvey, Peskin, & Sanford,
the following expressions: 1996) as functions of temperature. The temperature dependent
electrical conductivity of the NaCl solution was taken from the
n · (−k∇ T ) = 0 and n · v = 0 (14) previous study (Buckow et al., 2010):
A logarithmic wall function condition as described by Buckow (20 ◦ C)

(T ) (T − 20 ◦ C)
et al. (2010) and COMSOL Multiphysics (2006) is assumed at all = · 0.024234 · ◦C
+1 (22)
mS cm−1 mS cm−1
fluid-solid boundaries, accounting for the thin laminar boundary
layer which is not resolved by the mesh. Similarly, a ther- with (T) as the electrical conductivity at any given temperature T
mal wall function was defined at the solid–liquid interfaces. and (20 ◦ C) as the electrical conductivity at a reference tempera-
Instead of assuming continuity of the temperature across the thin ture of 20 ◦ C.
(non-resolved) laminar layer, the thermal wall function is used, The electrical conductivity of apple juice (Westcliffe Apple
accounting for a sudden temperature change between the solid sur- Juice, ALDI, NSW, Australia) was measured in the temperature
face and the liquid due to the omitted laminar layer (for a complete range of 4–80 ◦ C, using an electrical conductivity meter (#WP-
description of the logarithmic and the thermal wall function, the 84 Conductivity-TDS-Temp. Meter, TPS Pty Ltd, Springwood, QLD,
reader is referred to COMSOL Multiphysics, 2006). Australia) and was accurately described (R2 > 0.99) by Eq. (23):
For the outer wall, a heat flux condition was applied. The default (T )
 (T − 20 ◦ C)

form of the heat flux boundary condition is given as: = 2.048 · 0.02593 · ◦C
+1 (23)
mS cm−1
−n · q = q0 + hheat (Tinf − T ) (15) Eq. (22) or (23) was implemented into the models, describing the
with n as the normal direction to the boundary, hheat as the heat temperature dependent electrical conductivity of the NaCl solution
transfer coefficient, Tinf as the external temperature and q = − k ∇ T and apple juice, respectively.
(k is the thermal conductivity, which is replaced by k1 + kT in the The material properties for the insulators (PTFE,
k–ε closure): k = 0.24 W m−1 K−1 ,  = 2200 kg m−3 , CP = 1050 J kg−1 K−1 ) and
the electrodes (stainless steel, k = 44.5 W m−1 K−1 ,  = 7850 kg m−3 ,
−n · (−(k1 + kT )∇ T ) = q0 + hheat (Tinf − T ) (16) CP = 475 J kg−1 K−1 ) were taken from the material library of
COMSOL MultiphysicsTM .
Continuity of heat flux is assumed at all solid-solid boundaries:
Both liquids, NaCl solution and apple juice, were assumed to be
n · (k1 ∇ T ) = n · (k2 ∇ T ) (17) incompressible and Newtonian.
54 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

 Eav

2.6. Computational methods
MV = (27)
V0 /hmin
The partial differential equations (PDE) describing the Multi- n
av±10%
physics scenario, i.e., the coupled equations for conservation of UV = (28)
ntotal
mass, momentum, energy and electric charge were solved with the
commercial software package COMSOL MultiphysicsTM . The con- and
sidered problems were discretised with the finite element method  E
av
(FEM), utilising an unstructured mesh with tetrahedral elements EPV = (29)
Emax
for the 3D model and triangular elements for the axis-symmetric 2D
model, respectively. Although being computationally more expen- where Vzone is the volume of the treatment zone (insulator region)
sive, an unstructured mesh was selected due to its ability to of the respective scenario, Vmax the volume of the largest treatment
conform onto complex geometries in some of the geometry config- zone (i.e., at d = 20 mm, h = 30 mm, and “no inset”), Eav is the aver-
urations. A mesh independency study was performed and the mesh age electric field strength of the insulator region, V0 the applied
of the base models, particularly in critical regions inside the treat- potential, hmin the minimum electrode distance (gap) of all sce-
ment chamber, such as the insulator gap and the edges caused by narios investigated (hmin = 1 mm), nav±10% the number of elements
the diameter difference of insulator and electrodes, refined until no (in the MATLAB matrix) in the treatment zone with electric field
further noticeable change in the predicted process variables could strengths within 10% of the average electric field strength, ntotal
be observed (results not shown). the total number of elements in the treatment zone and Emax the
All computations were carried out on a workstation running the maximum electric field strength in the respective scenario.
64 bit OS Windows 2003 server. Two dual-core processors (each Each of the weighable parameters is made dimensionless by
2.33 GHz) and 20 GB RAM allowed for solving the fully coupled 3D relating the expression in the numerator to their respective maxi-
models in approximately 75 min, the fully coupled axis-symmetric mum values. Therefore, all parameters yield values between 0 and
2D models in about 10 min and the axis-symmetric 2D models, only 1, and, thus, also the DPP returns values between 0 and 1, with 1
solving for the electric field distributions, in less than 2 s. being the optimum achievable.
The first variable (TVV, treatment volume variable) expresses the
2.6.1. Algorithm for improving the performance of the treatment volume inside the treatment zone related to the volume of the
chamber largest treatment zone investigated in this study, where the diame-
The four basic insulator shape geometries (“no inset”, rectangu- ter of the insulator bore is equal to the electrode diameter (20 mm),
lar inset”, “rectangular rounded edges inset”, and “elliptical inset”) and the electrode gap is 30 mm. The second variable (PDV, pressure
were set up in COMSOL MultiphysicsTM , and the following settings drop variable) is a measure of the pressure drop caused by the diam-
applied to the models: eter reduction of the insulator bore which has been derived from
a simplified Bernoulli equation (Eq. (30)), not taking into account
 = 4 mS cm−1 (constant), V0 = 10 kV. the level difference part.
The geometrical dimensions of these default models were defined  · v21  · v22
as being variable. An interface to the COMSOL MultiphysicsTM mod- + p1 = + p2 (30)
2 2
els was programmed in MATLAB 7.6TM (The Mathworks Inc., Natick,
MA, USA). Based on this interface, a software routine progres- The third parameter (MV, magnitude variable) of the DPP
sively modified the models’ geometrical properties according to expresses the electric field strength magnitude of the analysed
the parameters discussed in Section 2.2.3. Overall, 103,170 models model, relating the average electrical field inside the insulator
were automatically generated, solved and the solution stored for region to the maximum electric field strength achievable in a par-
further analysis. The next step in the algorithm included the conver- allel plate configuration (V/hmin ) with the minimum electrode gap
sion of the COMSOL MultiphysicsTM data into MATLABTM matrices of 1 mm. The fourth variable (UV, uniformity variable) indicates the
with a spatial resolution of 0.01 mm, followed by a performance electric field uniformity in the gap region. The algorithm counts
evaluation of the respective model predictions. all elements inside the treatment zone that have an electric field
strength within a 10% range of the average electric field strength
and relates it to the total number of elements in this area (in the
2.6.2. Performance evaluation and identification of the optimum
MATLAB matrix); hereby, the elements are a defined fraction of the
For improving the co-linear treatment chambers frequently
treatment chamber volume, hence, the uniformity variable rep-
utilised in PEF processing, it was seen important to take a number
resents the volume ratio in which the condition of electric field
of dependent and independent parameters into consideration: The
strength being within 10% of the average electric field strength in
treatment zone volume as a measure for throughput, the uniformity
the treatment volume is fulfilled; i.e., the volume ratio of uniform
of the electric field distribution, including the prevention of areas
and effective treatment. The last parameter in the DPP equation
with overly high electric field strengths (peaks), and also the pres-
looks at the electric field strength peaks (EPV, electric field strength
sure drop caused by the diameter differences between insulator
peak), indicating the magnitude of electric field strength peaks in
and electrode bores. Five weighable and dimensionless parame-
the treatment zone by relating the average value of the electric
ters were selected and an equation was developed to express the
field strength to the maximum value in the respective configura-
performance of each modelled scenario with one dimensionless
tion.
parameter, the DPP (dimensionless performance parameter):
The exponents a1 –a5 , can be adjusted and, thus, the different
DPP = TVV a1 · PDV a2 · MV a3 · UV a4 · EPV a5 (24) parameters weighed according to their relevance; increasing the
exponent leads to a stronger emphasis on the respective parameter.
with:
V The weight of each parameter is hereby strongly dependent on the
TVV =
zone
(25) particular application of PEF processing.
Vmax As for this study, special emphasis was put on the uniformity
 4
 of the electric field strength distribution and prevention of peaks;
(d − ins)
PDV = (26) therefore, the exponents of MV, UV and EPV were set to values
d4 greater than the ones for TVV and PDV.
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 55

Australia), replacing the treatment chamber setup as supplied by

the manufacturer.
The process conditions were adjusted to an inlet volume flow
rate (at 20 ◦ C prior to entering the heat exchanger) of 8 L min−1
for the trials with the NaCl solution (conductivity at 20 ◦ C of
2.6 mS cm−1 ) and the apple juice (measured conductivity at 20 ◦ C of
1.8 mS cm−1 ) to ensure turbulent liquid flow (simulated Reynolds
numbers ranging from 20,000 to 60,000). The inlet temperature
was set to 25 and 45 ◦ C, and the rectangular pulses were regulated
at the modulator cabinet to pulse widths of 2 and 4 ␮s, and pulse
repetition rates of 500 and 700 Hz at an applied peak voltage of
38–39 kV.
Temperature measurements were performed in the second
grounded electrode (in the centre of the liquid stream at the out-
let of the second treatment zone) by insertion of an electrically
inert fibre optic probe. Inlet and air temperature in the treat-
ment chamber cabinet were measured for further use as boundary
conditions in the Multiphysics model. The measured temperature
was recorded when the temperatures inside the treatment cham-
ber reached a stationary state after approximately 5–10 min. All
experiments were duplicated. Frequency, pulse width, pulse shape
and voltage were recorded with an oscilloscope (#GDS-1102, GW
Instek, Taipei, Taiwan) attached to the output ports of the PEF
system. Time-averaged potentials were related to the maximum
potential to determine the factor ϕ (see Eq. (11)), while the maxi-
Fig. 6. Flowchart of the optimisation algorithm (initial step: model generation out- mum potential was used as boundary condition at the high voltage
side algorithm), including the generation of all scenarios, data conversion from electrode.
COMSOL MultiphysicsTM to MATLABTM and determination of optimum.

As TVV and PDV were assumed to be equally important, their 3. Results and discussion
exponents were set to the same value (a1 = a2 = 1). MV and EPV, i.e.,
the concentration of the electric field in the treatment zone, ensur- 3.1. Results of the iterative algorithm for performance
ing a safe and efficient treatment, and the prevention of pronounced improvement
electric field peaks, which may lead to over-processing and erosion
of electrodes and insulators, respectively, were seen equally impor- The algorithm created more than 100,000 2D axis-symmetric
tant and, therefore, set to the same values a3 = a5 = 1.5. The main models and the distribution of the electric field was numer-
objective to improve the performance of the treatment chamber, ically predicted using the implemented MATLABTM –COMSOLTM
however, was the uniformity of the electric field strength within interface. For some selected models, the predicted electric field dis-
the treatment zone. Hence, the highest exponent (weight) was used tributions were compared to illustrate the impact of the insulator
for the uniformity variable UV (a4 = 2). It has to be noted that the geometries on the distribution of the electric field (Fig. 7).
values for the weights were arbitrarily chosen and no sensitivity As already reported by other authors (e.g., Mastwijk et al., 2007),
analysis was performed. However, as long as the trend of impor- the numerical simulations from this study also showed that the
tance (weights) is kept the same, the algorithm will return the same electric field strength is highest in close vicinity of sharp insulator
(or very similar) geometrical properties for an improved treatment edges. For example, such edges are created by the diameter differ-
chamber. ences of insulator bore and electrodes, but also interfaces between
As all parameters return values between 0 and 1, also the DPP electrode and insulator for small “elliptical insets” and in the cen-
will yield values below 1 and, thus, the maximum DPP from all tre along the height of an elliptical insulator shape for larger insets
simulated scenarios is seen to determine the geometry with the exceeding 50% of the electrodes’ internal diameter (Fig. 7). Towards
best performance for this particular study. the centre line (symmetry axis) of the treatment zone, the electric
Fig. 6 illustrates the algorithm for identifying appropriate geo- field strength at inlet and outlet of the insulator region (treatment
metrical properties of the treatment chamber with improved zone) decreases, improving radial uniformity with larger insulator
performance in a flow chart, highlighting the different steps from insets.
model generation, over model solving, data extraction and perfor- For the determination of the treatment chamber performance
mance evaluation of the respective scenarios, to identifying the features (treatment volume, pressure drop caused by the insulator
geometrical dimensions of the co-linear treatment chamber con- inset, and magnitude and uniformity of the electric field strength),
figuration with enhanced performance. the MATLAB algorithm loops through all solved models, returning
geometric dimensions (shape, electrode diameter, insulator bore
2.7. Experimental setup for validating the Multiphysics model of inset and insulator height) and DPP values. For the chamber geome-
the improved treatment chamber tries without an insulator inset (i.e., the “no inset” configuration),
the varying geometry features were the internal diameter of the
The improved treatment chamber with the shape and electrodes and the electrode gap and, thus, the DPP values can be
dimensions as determined by the algorithm for performance illustrated in a 2D plot as a function of the tube radius and insulator
improvement (exact geometries shown in Section 3, Figs. 11 and 12) height (electrode gap) (Fig. 8).
was built to proof the validity of the model. The treatment cham- The greatest DPP value for the “no inset” scenario was found
ber was incorporated into the pilot-scale PEF system (see Section at a tube diameter of 2 mm and an electrode gap of 4 mm
2.1) at CSIRO Food and Nutritional Sciences (Werribee, Victoria, (DPP ≈ 2 × 10−4 ). As the emphasis of this study was to improve
56 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

electrode and insulator overlap and, therefore, create a region of

greater uniformity compared to larger systems where distinct dif-
ferences between electric field strength peaks at interfaces and the
bulk of the treatment zone can be observed.
The “rectangular” and “elliptical insulator inset” comprised a
further variable for performance improvement: The minimum
diameter of the insulator bore. Therefore, the DPP values can be
illustrated in form of a 3D distribution, with electrode radius,
electrode gap and difference between the electrode diameter and
the minimum diameter of the insulator bore (i.e., “inset”) as axes
(Fig. 9). The highest DPP value for the “elliptical inset” configu-
ration was found at an electrode inner diameter of 20 mm, an
electrode gap (insulator height) of 11 mm and an insulator inset,
i.e., diameter difference, of 25% of the electrodes’ inner diameter
(DPP ≈ 2.12 × 10−4 ).
As the “rectangular inset” configuration still exhibits pro-
nounced electric field strength peaks at the edges of the insulator
bore, similar to the “no inset” configuration, the total volume of
the treatment area was found to be smaller for achieving more
uniform treatment. Therefore, the highest DPP value was deter-
mined for an electrode diameter of 5 mm, an insulator height of
9 mm, but with an inset of only 5% of the electrodes’ diameter
(DPP ≈ 2.43 × 10−4 ).
For the “rectangular rounded edge” configuration, the radii of
the chamfered insulator edges had to be taken into consideration
as a fourth parameter. Therefore, visual illustration of the DPP val-
ues was not possible in a single (2D or 3D) plot. For each radius of
the chamfered edges, a separate 3D DPP distribution was plotted.
Fig. 10 shows the DPP distributions for three chamfer radii, namely
10, 25 and 40% of the diameter difference between the electrode
and the insulator bore (i.e., insets). Some slices in these plots appear
Fig. 7. Examples for the simulated distribution of the electric field strength in the incomplete, particularly for increasing insets and increasing cham-
cross-section of the treatment zone for “elliptical” (a), “rectangular” (b) and “rect- fer radii at low electrode gaps. This is caused by combinations of
angular rounded edge” (c) insulator shape insets (insets 0–60% of the electrodes’ insulator height (electrode gap), electrode radius, insets and cham-
internal diameter and radii of 50% of total inset for the “rectangular rounded edge”
fer radii, which are physically not possible, i.e., when the insulator
configuration) with a constant internal electrode diameter of 10 mm, an electrode
gap of 15 mm, a constant electric conductivity of 4 mS cm−1 and an applied peak
height is not sufficient for incorporating chamfer radii of a certain
potential of 10 kV. percentage of the inset.
The results indicate best performance and, thus, highest DPP
value (DPP ≈ 3.53 × 10−4 ; clearly higher than the values of all other
shapes investigated) for an electrode internal diameter of 12 mm,
an insulator height (electrode gap) of 20 mm, an inset of 15% of the
electrode diameter (i.e., 0.9 mm), and a chamfer radius of 40% of
the total inset (i.e., 0.36 mm).
The comparison of the maximum DPP values for all insula-
tor configurations, at the weighing exponent settings as discussed
earlier, shows that the “no inset” configuration gave the lowest
global DPP value. Surprisingly, the “rectangular inset” configu-
ration yielded a higher maximum DPP value than the “elliptical
inset”, but at a lower treatment volume. Nonetheless, the “rectan-
gular rounded edge” configuration yielded the highest global DPP
value, indicating superior performance over all other configura-
tions investigated.
The analysis of the parameters in the DPP equation for the “rect-
angular rounded edge” scenario gave the following values for the
improved chamber geometry:
The treatment volume variable (TTV) had a value of 0.3405 which
relates to 34.05% of the maximum treatment volume of this study
(i.e., diameter of 20 mm, electrode gap of 30 mm and “no inset”
configuration). Compared to the configuration of the treatment
Fig. 8. DPP value distribution for all “no inset” model geometries.
chamber supplied by the manufacturer, the treatment volume was
increased by more than 600%, allowing for increased throughput or
the electric field strength uniformity and magnitude, the per- prolonged exposure times, respectively.
formance evaluation unveiled that, for a configuration without The pressure drop variable (PDV) gave a value of 0.522. From this
insulator inset, only small treatment volumes allow a relatively value alone, it is not possible to draw conclusions on the actual pres-
uniform distribution of the electric field. This is likely caused by the sure drop in the system; the fully coupled 3D model (see Section
fact that the electric field strength peaks at the interface between 3.3), however, predicted a pressure drop of 0.02 bar in the entire
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 57

Fig. 9. DPP value distribution of all model geometries for “elliptical” (a) and “rectangular inset” (b).

treatment chamber, whereas the total pressure drop in the origi- The magnitude variable (MV) of the improved design resulted
nal system was only 0.004 bar if water is used as treatment media. in a value of 0.0412, which, in fact, was low compared to the value
Although the original system performs better with respect to the of the chamber supplied by the manufacturer. However, this was
pressure drop, a total pressure drop of 0.02 bar, caused by the insu- expected as the MV is determined by dividing the predicted average
lator inset, is negligible compared to the overall pressure drop along electric field strength through the maximum electric field strength
the piping of the entire system, including the peripheral setup, such that could be achieved in this study. As the electric field can be
as liquid handling system and heat exchangers. estimated with Eq. (1), smaller gaps, as in the original chamber

Fig. 10. DPP value distribution for all model geometries for “rounded rectangular edge inset” configuration for radii of 10% (a), 25% (b) and 40% (c) of the difference of electrode
and insulator bore diameter (inset).
58 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

Fig. 11. Dimensions and geometries of the electrodes and insulator of the improved co-linear PEF treatment chamber.

configuration, lead to higher electric field strengths and, therefore, a rounded edge radius of 0.36 mm. The three electrodes made of
greater values for MV. The MV value of the original system was stainless steel for the improved chamber have an outer diameter of
found to be 0.101, i.e., the average electric field strength is about 16 mm and an inner diameter of 12 mm.
2.5 times higher than for the improved chamber design.
The penultimate, and in this study most important (i.e., great- 3.3. 3D model of the improved treatment chamber
est weight) parameter of the DPP equation, the uniformity variable
(UV) of the improved design yielded a value of 0.887, which A fully coupled (i.e., simultaneously solving the conservation
indicates that 88.7% of the treatment zone exhibits electric field equations of mass, momentum, energy and charge) 3D model of
strengths within the ±10% range of the average electric field the improved chamber was developed in COMSOL MultiphysicsTM ,
strength of this chamber configuration. In comparison, the origi- for simulating PEF processing of NaCl solutions and apple juice
nal treatment chamber yields a value for UV of only 0.3202, i.e., (Fig. 13). The results presented in this section for electric field,
less than one third of the treatment zone provides electric field turbulent flow and temperature distribution are all based on
strengths within the predefined 10% range of the average electric the following process settings: Electrical conductivity at 20 ◦ C of
field strength. 2.6 mS cm−1 for the NaCl solution, applied peak voltage of 40 kV,
The electric field strength peak variable (EPV), showed a value pulse width of 4 ␮s, pulse repetition rate of 700 Hz, flow rate of
of 0.4193 for the improved design. This indicates that the average 8 L min−1 (adjusted at 20 ◦ C, prior to entering the heat exchanger),
electric field strength of the treatment zone has approximately 42% inlet temperature of 25 ◦ C, and external air temperature of 20 ◦ C.
of the strength of the highest electric field peak of the treatment
chamber. This value is significantly better than the one of the orig- 3.3.1. Electric field distribution
inal setup, which had an EPV value of only 0.1099. Accordingly, the Setting a potential of 38.5 kV at the high voltage electrode leads
peak electric field strength at the sharp edges of the original config- to an average electric field strength of approximately 17 kV cm−1 in
uration is approximately 9 times higher than the average electric the two treatment zones (Fig. 14). Along the centre line of the tube,
field strength of the chamber. Such electric field peaks are likely to the electric field strength only fluctuates in a small range between
cause temperature hotspots, over-processing of the treated liquid, 13 and 17 kV cm−1 . In the vicinity of the insulator edge (0.1 mm
arcing and associated erosion of the electrodes and insulators. from the edge) the electric field strength peaks at 28.7 kV cm−1
As discussed above, the improved design is superior in most (Fig. 15). This is much closer to the average value of 17 kV cm−1
parameters seen relevant for PEF processing; i.e., treatment vol- (Fig. 15) than in the original chamber design (Fig. 1), where the
ume, uniformity and avoiding of peak electric field strengths. Two maximum electric field strength was found to be nine times higher
of the discussed parameters of the original design yield higher than the average value (Buckow et al., 2010).
values: the pressure drop, and the average electric field strength
caused by greater electrode distances in the improved design, 3.3.2. Flow characteristics
which can be overcome by applying higher voltages. As sufficient mixing of the treated liquid in the treatment
zone is important to prevent (or minimise) the formation
3.2. Design of the improved treatment chamber of temperature hotspots, turbulent flow and, therefore, the
Reynolds number and turbulent kinetic energy are of utmost
Figs. 11 and 12 illustrate the new design and dimensions of relevance. The highest turbulent kinetic energy was found at
the improved treatment chamber. The new PTFE insulators mea- the outlets of the two treatment zones with up to 0.36 m2 /s2
sure an outer diameter of 98 mm, an internal diameter of 16 mm (Fig. 16), as obstacles enforce a change of flow direction. The
and a total height of 53 mm. The spacer between high voltage and Reynolds number was found to be between 20,000 and 60,000
grounded electrode has a height of 20 mm and the internal diame- and, thus, turbulence can be assured within the treatment
ter decreases to 10.2 mm centred along the height of the bore with zone.
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 59

Fig. 12. Close-up of the design and dimensions of the improved insulator.

Fig. 13. Three dimensional configuration of the improved PEF treatment chamber with co-linear electrode configuration and magnified view on the treatment zones.

Fig. 14. Simulated distribution of the electric field strength in the treatment chamber, including a close-up of both treatment zones. The settings were: electrical conductivity
at 20 ◦ C of 2.6 mS cm−1 , applied peak voltage of 40 kV, pulse width of 4 ␮s, frequency of 700 Hz, flow rate of 8 L min−1 , inlet temperature of 25 ◦ C, and external air temperature
of 20 ◦ C.
60 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

Fig. 15. Axial distribution of the electric field strength along the tube centre and at
0.1 mm distance from the insulator wall. Fig. 17. Axial temperature profile of the liquid passing through the first and second
treatment zone of the improved co-linear treatment chamber (model settings as
discussed earlier).
3.3.3. Temperature distribution
The simulation also showed a significant increase of the liq-
uid temperature in both treatment zones of the PEF chamber,
which is caused by ohmic heating in the regions of high elec-
tric field intensities. In the first chamber (i.e., first treatment
zone after fluid entering the chamber, between ground and high
voltage electrodes), the liquid heats up by about 7.5 ◦ C and in
the second chamber (i.e., second treatment zone) by approxi-
mately 10.5 ◦ C (Fig. 17). The difference in the heating extent in the
two identical treatment zones can be explained by the increase
of the electrical conductivity with increasing temperature and,
therefore, greater conversion of electrical into thermal energy.
The results show that only marginal radial temperature gradi-
ents occur in the improved design, with the highest difference
being less than 2 ◦ C at the outlet of the second treatment zone
(Fig. 18). Fig. 18. Radial difference of temperatures of the liquid from the maximum value at
The electrodes exhibit a uniform rise of temperature due to the the outlet of the first and second treatment zone of the improved system (model
steel’s high thermal conductivity. On the other hand, the insu- settings as discussed earlier).

lators only experience a significant temperature increase in the

regions close to the liquid as their thermal conductivity is low
(Fig. 19).
The measured temperatures at the outlet of the second
3.3.4. Model validation treatment zone (in the centre of the liquid stream) were com-
Numerical simulations were performed according to the set- pared to the numerically simulated data for the NaCl solution
tings described in Section 2: An applied voltage of approximately ((20 ◦ C) = 1.8 mS/cm) and the apple juice ((20 ◦ C) = 2.6 mS/cm),
38.5 kV, two different pulse lengths of 2 and 4 ␮s, pulse frequencies respectively. As shown in Table 1 and Fig. 20, the experimental data
of 500 and 700 Hz, a flow rate of 8 L min−1 (adjusted at 20 ◦ C prior were well predicted by the Multiphysics model, with the largest
to entering the heat exchanger), and two inlet temperatures of 25 deviation found to be approximately 1 ◦ C. The parity plot of Fig. 20
and 45 ◦ C, respectively. The energy dissipated into the liquid was in visually illustrates the good agreement between the simulation and
a range of 16–98 kJ kg−1 , and process temperatures were in a range measurement, yielding a coefficient of determination R2 close to
of 25–71 ◦ C. one.

Fig. 16. Simulated distribution of the turbulent kinetic energy in the treatment chamber, including a close-up of both treatment zones (model settings as discussed earlier).
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 61

Fig. 19. Simulated temperature distribution in the treatment chamber for liquid and solid parts, including a close-up of both treatment zones (model settings as discussed
earlier).

Table 1
Measured and predicted temperatures at the outlet of the second treatment zone in the improved PEF treatment chamber for NaCl solutions and apple juice at different
settings.

Product  (20 ◦ C) (mS/cm) V (kV)  (␮s) f (Hz) V̄ (L min−1 ) T0 (◦ C) Tmeasured (◦ C) Tsimulated (◦ C)

Apple juice 2.6 38.08 2 500 8 25 30.8 ± 0.14 31.06

2.6 37.92 2 500 8 45 52.9 ± 0.14 53.28
2.6 38.4 4 500 8 25 38.6 ± 0.14 38.85
2.6 38.08 4 500 8 45 63.0 ± 0.07 63.03
2.6 38.08 2 700 8 25 33.3 ± 0.28 33.52
2.6 37.92 2 700 8 45 55.8 ± 0.00 55.98
2.6 38.08 4 700 8 25 45.5 ± 0.35 44.56
2.6 38.08 4 700 8 45 71.7 ± 0.49 71.23
Salt solution 1.8 39.04 2 500 8 25 29.1 ± 0.00 29.22
1.8 39.04 2 500 8 45 49.9 ± 0.07 50.87
1.8 39.04 4 500 8 25 34.2 ± 0.07 34.28
1.8 39.04 4 500 8 45 57.6 ± 0.00 57.46
1.8 39.04 2 700 8 25 30.9 ± 0.14 30.78
1.8 39.04 2 700 8 45 53.2 ± 0.14 52.94
1.8 39.04 4 700 8 25 38.3 ± 0.07 38.44
1.8 39.04 4 700 8 45 63.3 ± 0.28 63.48

liquid foods containing highly heat labile components. However,

to ensure a homogeneous and, thus, safe treatment, and to prevent
partial over- or under-processing of the food and pronounced
erosion of parts of the treatment chamber, it is important to avoid
peaks of the electric field strength by improving the PEF treatment
chambers’ designs and configurations.
The purpose-developed software algorithm presented in this
work was able to identify insulator configurations of a co-linear
treatment chamber from more than 100,000 possible chamber
geometries superior in performance to the original treatment
chamber. According to the output of the algorithm, a 3D Multi-
physics model was developed and a new treatment chamber was
manufactured. The model was found to accurately predict the sta-
tionary temperature rise in the improved PEF treatment chamber,
yielding a coefficient of determination of close to 1 (Fig. 20), when
comparing the measurements and the simulations of the process-
ing of a salt solution and apple juice at varying process conditions.
Therefore, the conclusion can be drawn that the model also predicts
other important process variables, such as the electric field strength
Fig. 20. Parity plot of predicted and experimentally determined temperature values distribution and the flow characteristics, at good accuracy.
as shown in Table 1. The improved chamber design was capable of preventing
extreme electric field strength peaks, creating a more uniform dis-
tribution of the electric field, while significantly increasing the
4. Conclusions and outlook treatment volume of the original system and keeping the associated
pressure drop caused by the diameter reduction in the treatment
PEF processing is claimed to be highly effective for inactivation zone low. Due to the enhanced uniformity of the electric field
of vegetative microorganisms in liquid foods at moderate tem- strength distribution, the product will be treated more uniformly;
perature; therefore, it is commonly referred to as a non-thermal hence, there is less need for over-processing to ensure that areas
pasteurisation treatment. It is mainly for this reason, that PEF is with lower electric field strength are treated sufficiently; this in
considered a practicable preservation method, particularly for turn directly translates into better product quality. Furthermore,
62 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63

with increased treatment uniformity, product variability will be R2 coefficient of determination

significantly minimised. Linking the Multiphysics model describ- T temperature [K, ◦ C]
ing electric field, flow and temperature distribution with kinetics T0 inlet temperature [K, ◦ C]
for the prediction of microbial inactivation, enzyme modulation or Tinf reference bulk temperature [K, ◦ C]
other chemical reactions will then allow for showing the extent Tmeasure temperature measured in validation experiments [◦ C]
of improved treatment uniformity not only from the viewpoint of Tsimulated temperature predicted for model validation [◦ C]
process variables but also, and more importantly, from the perspec- t time [s]
tive of target process outputs. This is currently being investigated at U velocity for the 2D–3D comparison [m s−1 ]
CSIRO Food and Nutritional Sciences (Werribee, Victoria, Australia) v,v1 ,v2 velocity [m s−1 ]
but was not within the scope of the presented study. v0 inlet velocity [m s−1 ]
As the emphasis of the presented work was particularly put v velocity vector [m s−1 ]
on a good interaction of electric field uniformity and magnitude, V̇ volume flow rate [m3 s−1 ]
the improved design identified by the model-based algorithm is V voltage [V]
only one example, representing the optimum for one viewpoint V0 maximum potential of a pulse [V]
of emphasis. The weighing exponents of Eq. (24) have a strong Vmax maximum treatment zone volume [m3 ]
impact on the outcome of the algorithm; hence, for each improve- Vzone volume of treatment zone of the different chamber con-
ment study of this kind, depending on objective process targets figurations [m3 ]
and food properties, the first step must be the definition of the x, y, z spatial coordinates [m]
weights (importance) of the variables in the DPP equation. The soft- yi predicted values from 2D model
ware algorithm, including the interface between MATLABTM and ȳ mean of the predicted values from 2D model
COMSOL MultiphysicsTM , was developed as a platform, therefore,
allowing further studies for improving the design of PEF treatment
Greek letters
chambers, targeting varying emphases on the parameters in the
ε dissipation rate of turbulence energy [m2 s−3 ]
DPP equation. Alternatively, the definition of further parameters
 dynamic viscosity [Pa s]
that may be of relevance for specific industrial applications of PEF
T turbulent viscosity [Pa s]
processes is possible.
 density [kg m−3 ]
 electrical conductivity [S m−1 ]
Notation
 pulse width [s]
ϕ factor for time-averaged potential of PEF pulses
A cross-sectional area of treatment chamber inlet [m2 ]
a1–5 exponents (weights) of the DPP equation within the Abbreviations
chamber improvement algorithm 2D two dimensional
CP specific heat capacity [J kg−1 K−1 ] 3D three dimensional
C model constant [m4 kg−4 ] CFD computational fluid dynamics
d electrode diameter [m] DPP dimensionless performance parameter
E electric field strength [V m−1 ] EPV electric field strength peak variable
Eav average electric field strength [V m−1 ] FEM finite element method
Emax maximum electric field strength [V m−1 ] HVE high voltage electrode
f pulse repetition rate (frequency) [Hz] MV magnitude variable
fi predicted values from 3D model NaCl sodium chloride
g gravity constant [9.8 m s−2 ] NIST National Institute of Standards and Technology
hheat heat transfer coefficient [W m−2 K−1 ] OS operation system
h electrode distance (gap) [m] PEF pulsed electric field
hmin minimum electrode distance (gap) of all scenarios inves- PDE partial differential equation
tigated [1 mm] PDV pressure drop variable
ins total insulator inset, i.e., diameter difference between PTFE polytetrafluoroethylene
electrode and insulator bore [m] RAM random access memory
J density of the electrical current [A m−2 ] TVV treatment volume variable
k turbulent kinetic energy [m2 s−2 ] UV uniformity variable
k,k1 ,k2 thermal conductivity [W m−1 K−1 ]
kT turbulent thermal conductivity [W m−1 K−1 ]
Operators
ṁ mass flow rate [kg s−1 ]
∂ partial differential
n normal direction
 difference
nav±10% number of matrix elements within a certain band of the
O nabla operator (vector differential operator)
average electric field strength in the treatment zone of a
PEF chamber
ntotal total number of matrix elements in the treatment zone Acknowledgements
(liquid domain)
P pressure including a fluctuating term [Pa] The authors gratefully acknowledge the Victorian State Govern-
p pressure [Pa] ment for the Science and Technology Infrastructure grant enabling
PrT turbulent Prandtl number this work to be carried out.
Qeff heat source [W m−3 ] Furthermore, we would like to thank Mr. Phil Muller and Mr.
Qspec dissipated energy [kJ kg−1 ] Piotr Swiergon for their support in setup and maintenance of
q, q0 heat flux [W m−2 ] the PEF system, and the manufacture of the improved treatment
rad, R rounded edge (chamfer) radius [m] chamber.
 K. Knoerzer et al. / Computers and Chemical Engineering 37 (2012) 48–63 63

References the inactivation of E. coli and milk alkaline phosphatase. Innovative Food Science
& Emerging Technologies, 10(4), 470–480.
Angersbach, A., Heinz, V., & Knorr, D. (2000). Effects of pulsed electric fields on cell Kowalczyk, W., Hartmann, C., & Delgado, A. (2004). Modelling and numerical simu-
membranes in real food systems. Innovative Food Science & Emerging Technolo- lation of convection driven high pressure induced phase changes. International
gies, 1(2), 135–149. Journal of Heat and Mass Transfer, 47(5), 1079–1089.
Buckow, R., Schroeder, S., Berres, P., Baumann, P., & Knoerzer, K. (2010). Simulation Launder, B. E., & Spalding, D. B. (1974). The numerical computation of turbulent
and evaluation of pilot-scale pulsed electric field (PEF) processing. Journal of flows. Computer Methods in Applied Mechanics and Engineering, 3(2), 269–289.
Food Engineering, 101(1), 67–77. Lindgren, M., Aronsson, K., Galt, S., & Ohlsson, T. (2002). Simulation of the tempera-
Chen, X. D. (2006). Modeling thermal processing using computational fluid dynamics ture increase in pulsed electric field (PEF) continuous flow treatment chambers.
(CFD). In D. W. Sun (Ed.), Series: Thermal food processing: New technologies and Innovative Food Science & Emerging Technologies, 3(3), 233–245.
quality issues, food science and technology Thermal food processing (pp. 133–151). Mastwijk, H. C., Gulfo-Van Beusekom, K., Pol-Hofstad, I. E., Schuten, H., Boonman, M.,
Boca Raton: Taylor & Francis. & Bartels, P. V. (2007). Definitions and guidelines for reporting on pulsed electric
COMSOL Multiphysics. (2006). Chemical engineering module. Stockholm, Sweden: field experiments. In H. L. M. Lelieveld, S. Notermans, & S. W. H. De Haan (Eds.),
COMSOL AB. Food preservation by pulsed electric fields (pp. 320–345). Woodhead: Cambridge.
Fiala, A., Wouters, P. C., van den Bosch, E., & Creyghton, Y. L. M. (2001). Coupled Misaki, T., Tsuboi, H., Itaka, K., & Hara, T. (1982). Computation of three-dimensional
electrical-fluid model of pulsed electric field treatment in a model food system. electric-field problems by a surface-charge method and its application to opti-
Innovative Food Science & Emerging Technologies, 2(4), 229–238. mum insulator design. IEEE Transactions on Power Apparatus and Systems, 101(3),
Gerlach, D., Alleborn, N., Baars, A., Delgado, A., Moritz, J., & Knorr, D. (2008). 627–634.
Numerical simulations of pulsed electric fields for food preservation: A review. Nicolaï, B. M., Verboven, P., & Scheerlinck, N. (2007). Modeling and simulation of
Innovative Food Science & Emerging Technologies, 9(4), 408–417. thermal processes. In P. Richardson (Ed.), Thermal technologies in food processing
Gongora-Nieto, M. M., Pedrow, P. D., Swanson, B. G., & Barbosa-Canovas, G. V. (2003). (pp. 91–112). Boca Raton, FL: CRC Press.
Impact of air bubbles in a dielectric liquid when subjected to high field strengths. Puertolas, E., Lopez, N., Condon, S., Raso, J., & Alvarez, I. (2009). Pulsed electric fields
Innovative Food Science & Emerging Technologies, 4(1), 57–67. inactivation of wine spoilage yeast and bacteria. International Journal of Food
Harvey, A. H., Peskin, A. P., & Sanford, A. K. (1996). NIST/ASTME—IAPSW standard Microbiology, 130(1), 49–55.
reference database 10, version 2.2. Toepfl, S., Heinz, V., & Knorr, D. (2007). High intensity pulsed electric fields applied
Heinz, V., Alvarez, I., Angersbach, A., & Knorr, D. (2001). Preservation of liquid foods for food preservation. Chemical Engineering and Processing, 46(6), 537–546.
by high intensity pulsed electric fields—Basic concepts for process design. Trends Toepfl, S., Mathys, A., Heinz, V., & Knorr, D. (2006). Review: Potential of
in Food Science & Technology, 12(3–4), 103–111. high hydrostatic pressure and pulsed electric fields for energy efficient and
Heinz, V., Toepfl, S., & Knorr, D. (2003). Impact of temperature on lethality and energy environmentally friendly food processing. Food Reviews International, 22(4),
efficiency of apple juice pasteurization by pulsed electric fields treatment. Inno- 405–423.
vative Food Science & Emerging Technologies, 4(2), 167–175. Zimmermann, U., Pilwat, G., Beckers, F., & Riemann, F. (1976). Effects of external
Jaeger, H., Meneses, N., & Knorr, D. (2009). Impact of PEF treatment inhomogeneity electrical fields on cell membranes. Bioelectrochemistry and Bioenergetics, 3(1),
such as electric field distribution, flow characteristics and temperature effects on 58–83.