Journal of Food Engineering 91 (2009) 183–196

Contents lists available at ScienceDirect

Journal of Food Engineering

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

Modelling the industrial production of vinegar in aerated-stirred fermentors

in terms of process variables
José-María González-Sáiz a,*, Diego Garrido-Vidal a, Consuelo Pizarro b
a
Chemical Engineering, Department of Chemistry, University of La Rioja, C/Madre de Dios 51, 26006 Logroño, La Rioja, Spain
b
Analytical Chemistry, Department of Chemistry, University of La Rioja, C/Madre de Dios 51, 26006 Logroño, La Rioja, Spain

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

Article history: The production of vinegar in an aerated-stirred reactor was modelled using a hybrid approach. The
Received 17 February 2008 growth rate was modelled according to the total biomass specific growth rate, lg m, which represents
Received in revised form 13 August 2008 the growth of an active biophase and the fraction of cells which are growing. A quadratic model based
Accepted 15 August 2008
on the process variables, i.e. hydrostatic pressure, ethanol concentration, aeration, agitation and temper-
Available online 7 September 2008
ature, was developed for lg m and the substrate and product kinetics were related mechanistically to the
growth rate, including the kinetics of acetoin and ethyl acetate production. This model was coupled with
Keywords:
the models for oxygen transfer and gas–liquid transfer, reported in a previous study, and the prediction
Bioprocess modelling
Continuous
capacity of the system of equations was validated by simulating semi-continuous and continuous real
Fermentation processes developed in a pilot fermentor.
Quadratic model Ó 2008 Elsevier Ltd. All rights reserved.
Semi-continuous
Vinegar production

1. Introduction anol consumption. However, it could only be applied to the

Mechanistic and pseudo-empiric kinetic models have been
widely applied to acetic fermentation (Bar et al., 1987; Caro
et al., 1996; Gómez and Cantero, 1998; Ito et al., 1991; Mesa
et al., 1986; Nanba et al., 1984; Ory et al., 1998; Park et al.,1991;
Romero et al., 1994), based on the general scheme of reactions
proposed by Sinclair and Kristiansen (1987). Most of the mathe-
matical expressions for the specific growth rate have accounted
for intracellular metabolism and divided the cells in terms of via-
bility (Sinclair and Topiwala, 1970). An empirical model was also
proposed using a different approach (Kruppa and Vortmeyer,
1999). These models could not satisfactorily explain the industrial
process of vinegar production mainly because an inhibitory effect
of acetic acid was considered (González-Sáiz et al., 2003). This
inhibitory effect was not observed in industrial fermentation.
In a previous study (González-Sáiz et al., 2003), the authors pro-
posed a new pseudo-empiric kinetic model. The parameters of the
specific growth rate equation were optimised by a genetic algo-
rithm applied to data obtained from fermentors in a vinegar man-
ufacturing plant in La Rioja. The data were obtained in batch
fermentors operating at constant temperature, oxygen supply,
hydrodynamic conditions and initial concentration. The model as-
sumed that the decrease in growth rate with time was due to eth-
* Corresponding author. Tel.: +34 941299634; fax: +34 941299621.
E-mail address: josemaria.gonzalez@unirioja.es (J.-M. González-Sáiz).
standard process conditions prevailing in the industrial fermentor
studied.
The effect of oxygen transfer has been studied in different bio-
systems (Çelik and Çalik, 2004; Elibol and Ozer, 2000; Tang and
Zhong, 2003), but in acetic fermentation the effects of oxygen con-
centration have only been considered in the context of bacterial
growth (Park et al., 1991; Romero et al., 1994). However, it is well
known that a failure in oxygen supply can dramatically reduce cell
population viability (Drysdale and Fleet, 1989; Muraoka et al.,
1982). The effect of agitation and aeration has also been studied
in the context of various biosystems (Croughan et al., 1987; El-
Enshasy et al., 2006; Evans and Liu, 2003; Zuo et al., 2006), but
not for acetic acid fermentation. A new model for acetic fermenta-
tion is required to explain the effects of oxygen transfer and hydro-
dynamic conditions on acetification rate.
Therefore, the main objective of this paper is the development
of a global model for the industrial process of acetic fermentation.
The kinetic model for cell growth, ethanol and oxygen consump-
tion, and acetic acid production has been proposed on the basis
of an hybrid approach, considering the results reported in a previ-
ous study (Garrido-Vidal et al., 2003), where the authors studied
the effect of hydrostatic pressure, ethanol concentration, aeration,
agitation and temperature, i.e. the process variables of an aerated-
stirred fermentor on acetification rate. Hybrid models have been
widely applied in several biosystems (Chen et al., 2000; Oliveira,
2004; Zorzetto et al., 2000). In these types of models, the structures

0260-8774/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfoodeng.2008.08.028
184 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196

Nomenclature

A acetic acid concentration in the bioreactor and in the rEA ethyl acetate formation rate due to the fermentation
effluent stream, g L1 process, g L1 h1
AMC acethyl methy carbinol (acetoin) concentration in the rE ethanol consumption rate due to the fermentation pro-
bioreactor and in the effluent stream, g L1 cess, g L1 h1
C i;L concentration of component i in liquid phase before ri rate of production or consumption of component i with-
mass transfer, g L1 in the fermentor, g L1 h1
Ci concentration of component i in fermentation medium r O2 oxygen consumption rate due to the fermentation pro-
and effluent stream cess, g L1 h1
C if concentration of component i in the feed stream rX t cell growth rate due to the fermentation process,
C O2 ;L concentration of oxygen in liquid phase, g L1 g L1 h1
C O2 ;L concentration of oxygen in oxygen-saturated liquid SM suspended matter concentration in the bioreactor and
phase, in equilibrium with partial oxygen pressure in in the effluent stream, gDW L1
the oxygen-saturated gas phase, g L1 SMf suspended matter concentration in the feed stream,
D dilution rate, h1 gDW L1
F1 volumetric flow rate of filtered vinegar stream, L h1 t fermentation time, h
F2 volumetric flow rate of purged vinegar stream, L h1 TL liquid-phase temperature, °C
FV volumetric flow rate of vinegar effluent stream, L h1 Vc volume of wine charged, L
FW volumetric flow rate of feed wine, L h1 Vd volume of vinegar discharged, L
E ethanol concentration in the bioreactor and in the efflu- VL liquid-phase volume, dm3 or m3
ent stream, g L1 vvm air-flow rate divided by the reactor working volume,
EA ethyl acetate concentration in the bioreactor and in the h1
effluent stream, g L1 Xt total biomass concentration in the bioreactor and in the
kA;G a volumetric acetic acid transfer coefficient in gas-side effluent stream, gDW L1
interphase, h1 X factors of the quadratic model
kAMC;G a volumetric acetoin transfer coefficient in gas-side inter- Y 0A=E acetic acid/ethanol yield factor, g acetic acid/g ethanol
phase, h1 Y A=E acetic acid/ethanol stoichiometric coefficient, 1.30 g
kE;G a volumetric ethanol transfer coefficient in gas-side inter- acetic acid/g ethanol
phase, h1 Y 0AMC=X acetoin/biomass yield factor, g acetoin/g biomass
kEA;G a volumetric ethyl acetate transfer coefficient in gas-side Y 0A=O acetic acid/oxygen yield factor, g acetic acid/g oxygen
interphase, h1 Y EA=A ethyl acetate/acetic acid stoichiometric coefficient,
ki;G a volumetric component i transfer coefficient in gas-side 1.47 g ethyl acetate/g acetic acid
interphase, h1 Y EA=E ethyl acetate/ethanol stoichiometric coefficient, 1.91 g
kL a volumetric oxygen transfer coefficient in liquid-side ethyl acetate/g ethanol
interphase, h1 Y 0EA=X ethyl acetate/biomass yield factor, g ethyl acetate/g bio-
kW;G a volumetric component i transfer coefficient in gas-side mass
interphase, h1 Y E=O ethanol/oxygen stoichiometric coefficient, 1.44 g etha-
LGTRi liquid–gas transfer rate for component i, g L1 h1 nol/g oxygen
N impeller rotation speed, rpm or s1 Y 0E=O ethanol/oxygen yield factor, g ethanol/g oxygen
O2ðoÞ oxygen effluent stream, g L1 h1 Y 0X=E biomass/ethanol yield factor, g biomass/g ethanol
O2ðiÞ oxygen feed stream, g L1 h1 Y 0X=O biomass/oxygen yield factor, g biomass/g oxygen
OTR oxygen transfer rate, g L1 h1 Xt total biomass concentration, gDW L1
OUR oxygen uptake rate, g L1 h1
P pressure in fermentation medium, atm Greek symbols
Pg aerated mechanical power input, w lg m total biomass specific growth rate, h1
R recycling rate lg overall specific growth rate, h1
rA acetic acid production rate due to the fermentation pro- lS superficial air velocity through the liquid medium,
cess, g L1 h1 m s1
r2 A adjusted correlation rate m viability factor
rAMC acethyl methy carbinol (acetoin) formation rate due to
the fermentation process, g L1 h1

including conservation laws such as material and/or energy bal-
ance equations are combined with non-parametric models for
the specific growth rate, such as artificial neural networks. These
types of models require several sets of data for validation and their
predictive ability is unreliable under extrapolation conditions, so
they are usually applied to well-established processes where data
is readily available or for batch-to-batch iterative optimisation
(Teixeira et al., 2006). Therefore, empirical models with secondary
order polynomials have been widely applied to study the effect of
variables on cell growth and optimisation of bioprocesses (Henri-
ques et al., 2006; Nikerel et al., 2006; Rita et al., 2003).
Another new feature of this research has been the inclusion of
a kinetic model of ethyl acetate and acetoin production. The glo-
bal model was coupled with a model for kL a and gas–liquid trans-
fer, reported in a previous study (González-Sáiz et al., 2008),
which enables dissolved oxygen concentration and the impact
of gas–liquid transfer on the concentration of the compounds to
be predicted. The predictive ability of the global model was dem-
onstrated by simulating an exhaustive set of fermentation pro-
cesses developed in a pilot fermentor. This validation was
undertaken for all important operating configurations of indus-
trial fermentors, i.e. continuous fermentor, continuous fermentor
with cell recycling and semi-continuous fermentor. The broad
applicability range of the global model developed in this study
has not been reported in literature on mechanistic acetic fermen-
tation models.
 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196 185

2. Materials and methods 2.2. Fermentation processes developed with the pilot fermentor

2.1. Pilot fermentation system The process variables controlled in each fermentation process
were hydrostatic pressure, ethanol concentration, aeration, agita-
The experimental study was developed in a New Brunswick tion and temperature. Table 1 shows the range of the process vari-
(BioFlo IV) pilot fermentor, with a maximum working volume of ables considered. The table reports also the concentrations of
10 L. The fermentor was an aerated-stirred reactor (ASR) acetic acid, ethyl acetate, acetoin and suspended matter. These
equipped with a gas distributor or sparger and a mechanical agi- parameters can be considered to be system responses because they
tation system, comprising two flat blade Rushton turbines, with depend on the kinetics of fermentation and wine concentration. In
six blades fixed onto the disk and three baffles. Agitation was fact, acetic acid concentration was stoichiometrically related to
controlled by PID within ±1 rpm and temperature was maintained ethanol concentration.
by circulating water through the jacket, whose temperature was
controlled by PID within ±0.1 °C. A condenser connected to the 2.2.1. Continuous processes
exhaust line was used to maintain a closed system. A FOSS NIR- In continuous fermentations, the process variables were kept
Systems 5000 Liquid Analyser was connected on-line to the vessel constant for at least 168 h. The feed stream was controlled by a
and a gas chromatograph Hewlett Packard 5890 series II was con-
nected to the gas outlet. The pressure in the headspace of the ves-
Table 1
sel was controlled by a valve fitted to the air exhaust line. Ranges of process variables and concentrations
Neglecting the pressure due to the height of the medium on ac-
Variable Range
count of the dimensions of the fermentor, the pressure of the
head space was assumed to be the pressure prevailing in all parts Pressure 1–2 atm
Ethanol concentration 1–50 g L1
of the fermentor. Fig. 1 shows a diagram of the pilot fermentation
Acetic acid concentration 57–132 g L1
system. Ethyl acetate concentration 0.02–4 g L1
The inoculum to start the fermentation process was obtained Acetoin concentration 0.75–1.50 g L1
from a commercial vinegar production facility in La Rioja. The wine Suspended matter 0.24–0.94 g L1
used as feed was also the same white wine used in the plant. vvm 3–37 h1
Agitation 200–1000 rpm
Inoculums and wine were introduced into the fermentor without Temperature 26–33 °C
any pre-treatment.

GAS CHROMATOGRAPH

Oxygen Air outlet Air inlet

Volatiles
FW Wine feed

Ethanol Sample
Acetic acid
Biomass
Ethyl acetate
Acetoin

FV Vinegar effluent Recycle

NIR

TANGENTIAL FILTER

F1 F2
Filtered vinegar Purge
Fig. 1. Diagram of the pilot fermentation system.
186 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
PID controller to keep the working volume and the ethanol con-
centration at the desired level. When constant values of ethanol,
acetic acid, biomass concentrations, dissolved oxygen and flow
rate were reached, a steady state was assumed to prevail, and
the conditions were maintained for 72 h more, monitoring the
concentrations of ethanol, acetic acid, ethyl acetate, acetoin and
biomass every hour, and the remaining variables (pH, dissolved
oxygen, temperature, pressure in the fermentation medium, aera-
tion rate and agitation speed) every 15 min. At steady state, the
concentration of the compounds did not depend on time and
the rate of consumption or production of a component i within
the fermentor was calculated as
FV ðF W  F V Þ
ri ¼  ðC i  C if Þ   C if
V V
where F/V was D, the dilution rate, representing the holding time or
processing rate of the continuous reactor. When the condenser was
installed and the fermentation volume was constant, F V and F W
were equal.
The oxygen consumed or oxygen uptake rate (OUR), was calcu-
lated using the following balance:
OUR ¼ O2ðoÞ  O2ðiÞ
O2ðiÞ was the mass flow rate of oxygen per unit volume of the
feed, calculated on the basis of oxygen concentration in air at
1 atm and 25 °C. O2ðoÞ was the same parameter in the exit stream,
measured by the gas chromatograph fitted on-line.
The oxygen transfer rate was defined as
OTR ¼ kL a  ðC O2 ;L  C O2 ;L Þ
At steady state, C O2 ;L was constant and kL a was calculated
according to the balance between OTR and OUR
dC O2 ;L
¼ OTR  OUR
dt
kL a  ðC O2;L  C O2;L Þ ¼ O2ðoÞ  O2ðiÞ
C O2 ;L was measured as a system response but it was not used as
a variable. A continuous process with cell recycling was developed
by passing the effluent stream through a tangential filter. A fraction
of the concentrated biomass was returned to the fermentor and the
remaining biomass was purged, as shown in Fig. 1. The recycle rate
was established at the purge valve and corresponded to the flow
rate of filtered vinegar compared with the total exit stream
F1
R¼
FV
2.2.2. Semi-continuous processes
In each semi-continuous process, the following parameters
were fixed: final ethanol concentration for discharge of medium,
initial ethanol concentration, working volume, hydrostatic pres-
sure, and aeration, agitation and temperature conditions. When
the cells consumed ethanol and its final set concentration was
reached, a volume of fermentation medium was discharged and,
after that, a volume of wine was charged. The volumes of medium
discharged and wine charged were calculated to recover initial eth-
anol concentration and working volume. It was considered that a
semi-steady state had been reached when the final concentrations
and the volumes of wine charged and medium discharged were
constant in 10 batches.
2.3. Analytical methods
Ethanol, acetic acid, ethyl acetate, acetoin and suspended mat-
ter concentrations were monitored on-line by near-infrared spec-
troscopy using a FOSS NIRSystems 5000 Liquid Analyser,
connected on-line. A previous study described the regression
models developed for this system (Garrido-Vidal et al., 2004).
The suspended matter concentration included the bacteria cells,
in suspension in the medium, and all the suspended solids and
colloidal matter contained in the wine. The NIR system measured
all the suspended matter, without any differentiation. The
amount of bacterial cells in wine was considered to be negligible,
taking into account wine storage conditions. Therefore, the in-
crease in suspended matter in the bioreactor was due to cell
growth. Air-flow measurements were provided by a mass flow
sensor and pressure in the head space of the fermentor was mea-
sured using a barometer. Dissolved oxygen concentration was
ð1Þ monitored using an Ingold sterilizable dissolved oxygen electrode,
while pH was controlled by a pressurized pH electrode. On-line
data for air flow, rpm, temperature, pH and dissolved oxygen
were obtained by the probes and stored in the database of the
control software provided by New Brunswick.
A gas chromatograph Hewlett Packard 5890 series II was con-
nected to the gases outlet to measure on-line the concentrations
of oxygen, ethanol, acetic acid, ethyl acetate and acetoin in the ex-
haust line. The compounds were analysed simultaneously by a sys-
ð2Þ tem with two chromatographic columns, two detectors and one
electro valve. Firstly, the samples were introduced in a 2 mL loop,
prior to the column, with a 150 °C pre-heating programme to avoid
condensations. After injection, the samples passed through a
molecular sieve (13  45/60 mesh), where oxygen and nitrogen
were retained, and through a packed column, with support carbo-
pack B 80/20, to separate ethanol, acetic acid, ethyl acetate and
ð3Þ acetoin. Oxygen was quantified using a TCD detector and polar
compounds were quantified using a FID detector. Laboratory air
was used as a standard, i.e. the same air that was introduced in
the fermentor by a compressor.
ð4Þ
2.4. Experimental design
ð5Þ
Cell growth kinetics, substrates and main product kinetics, and
secondary product kinetics were modelled using data from 35 con-
tinuous fermentation processes, according to a Doehlert design
(Doehlert, 1970; Massart et al., 1994) applied to pressure, ethanol
concentration, aeration, agitation and temperature, with 5, 7, 7, 7
and 3 levels, respectively. These data were previously reported
by the authors in another study (Garrido-Vidal et al., 2003). An-
other previous study (González-Sáiz et al., 2008) described the
experimental methodology developed to calculate the models for
ð6Þ kL a and liquid–gas transfers.
Once the global model had been calculated, the applicability of
the model was studied by simulating the following processes,
which were developed in the pilot fermentor for this study:
– A continuous process with condenser, head space pressure
1.5 atm, air flow 132 L h1 (20.31 h1), agitation 600 rpm, tem-
perature 29.5 °C, ethanol set point 20 g L1. It was used to dem-
onstrate that the simulations predicted the same steady state
independently of the initial concentrations.
– A semi-continuous process with condenser, head space pressure
1.5 atm, air flow 132 L h1 (20.31 h1), agitation 600 rpm, tem-
perature 29.5 °C, ethanol set point for discharge 20 g L1, and
initial ethanol concentration point for charge 40.14 g L1. It
was used to demonstrate that the simulations predicted the
same steady state independently of the initial concentrations.
– The 35 experiments of the Doehlert’s design used to calculate
the model were simulated as a calibration set.
The validation was developed by testing the prediction capac-
ity of the model by simulating a set of fermentation processes
 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
performed with the pilot fermentor for this study. These fermen-
tation processes were very different from the processes devel-
oped to calculate the models and the set included the types of
fermentation processes that can be used in an industrial plant,
i.e. continuous or semi-continuous, cell recycling, open or closed
systems:
– A set of six continuous processes without cell recycling, without
condenser and with process conditions different from the Doehl-
ert design of continuous experiments.
– A set of 10 continuous fermentation processes with condenser,
head space pressure 0.85 atm, air flow 220 L h1, agitation
500 rpm, temperature 29.5 °C, ethanol set point 20 g L1 and
varying the recycling rate within range [0–0.9].
– A set of 21 semi-continuous fermentation processes without
condenser, ethanol set point for discharge 20 g L1, initial etha-
nol concentration point for charge 40.14 g L1 and initial concen-
trations of each experiment equal to the final concentrations of
the previous experiment. Hydrostatic pressure, aeration, agita-
tion and temperature corresponded to the values of a Doehlert
design applied to these variables.
– A set of four semi-continuous processes with headspace
pressure 0.85 atm, air flow 220 L h1, agitation 500 rpm, tem-
perature 29.5 °C and closed system, i.e. with condenser. The
initial ethanol concentration point for charge was 40.14 g L1
and initial concentrations of each experiment were 40.14
g L1 ethanol, 82.73 g L1 acetic acid, 2 g L1 ethyl acetate,
0.3 g L1 acetoin and 0.021 g L1 suspended matter. Ethanol
set points for discharge were 4.36, 8, 12 and 16 g L1 for each
process.
3. Results and discussion
3.1. Structure of the model
In a previous study (Garrido-Vidal et al., 2003), the authors re-
ported the results of 35 continuous processes, based on a Doehl-
ert design applied to pressure, ethanol concentration, aeration,
agitation and temperature, with 5, 7, 7, 7 and 3 levels, respec-
tively. In each steady state, the rates of substrates consumption
and product secretion and cell growth rate were calculated as ex-
plained in Section 2.2. The exhaust line was connected to a con-
denser and the other volatile compounds in the gas outlet after
the condenser were assayed by gas chromatography, although
these values were practically negligible. The consumptions due
to ethyl acetate formation were calculated by the ethyl acetate
production rate and the stoichiometric coefficients Y ðEA=EÞ (1.91)
and Y ðEA=AÞ (1.47). These consumption levels were taken into ac-
count when calculating the rates of consumption of ethanol
and production of acetic acid, acetoin and ethyl acetate due
exclusively to biological reactions. In that study, the authors
demonstrated the linear relationship between the growth rate
and the kinetics of substrates and products, and robust and reli-
able values for the yield factors were calculated: 1.29 ± 0.02 for
Y 0A=E , 1.9 ± 0.2 for Y 0A=O , 1.5 ± 0.1 for Y 0E=O , 0.0061 ± 0.0006 for Y 0X=E
and 0.009 ± 0.001 for Y 0X=O . These results demonstrated that bac-
teria used the energy obtained by ethanol oxidation with oxygen
for growth, and other consumptions such as maintenance energy
or cell respiration were ruled out (González-Sáiz et al., 2003;
Gómez and Cantero, 1998). Acetic acid was obtained as a result
of oxidation. Finally, the structure of a kinetic model with a
growth-associated kinetic of products and the biophase charac-
terized by X t , the total biomass concentration, was proposed as
a conclusion in that study
187
!
1 dX t
rE ¼   ð7Þ
Y 0X=E dt
!
1 1 dX t
r O2 ¼    ð8Þ
YE=O Y 0X=E dt
1 dX t
rA ¼ Y A=E  ð9Þ
Y 0X=E dt
In steady state, the total biomass concentration in the fermen-
tation medium and the effluent stream was the difference between
the suspended matter in the wine and the suspended matter in the
bioreactor
dX t
¼ D  ðSM  SMf Þ ¼ D  X t ð10Þ
dt
In the previous study mentioned above (Garrido-Vidal et al.,
2003), the dilution rate was considered to be the specific growth
rate and response surfaces were build to study the effects of the
process variables on D. The data obtained in the previous work
have been revised for this contribution to calculate a kinetic mod-
el for acetic fermentation, based on the structure of the model re-
ported previously. The mechanistic part of the model, i.e. the
growth-associated kinetic and the yield factors were assumed
and the aim was to develop a model for cell growth. X t can be as-
sayed in industrial plants by direct assays of dry weight, UV–vis
or NIR. This technique allows total biomass to be measured to-
gether with the other components of the fermentation medium
(Garrido-Vidal et al., 2004). The main premise of the cell growth
model is the fact that there is a fraction of total biomass capable
of cell division, which has been defined as viability, m, together
with the frequency of cell multiplication, which has been defined
as the specific growth rate, lg
dX t
dt
Xt  m ! Xt ð11Þ
1 dX t
lg ¼  ð12Þ
X t  m dt
Rearranging Eq. (12) allowed us to define a new variable, the to-
tal biomass specific growth rate, lg m
dX t
¼ lg m  X t ð13Þ
dt
lg m represents the growth of an active biophase, including the
fraction of cells that are growing and the frequency of the multipli-
cation of these cells. Both components depend on the process
variables.
Therefore, the dilution rate obtained in each steady state of the
study by Garrido-Vidal et al. (2003) corresponded to the value of
lg m for each set of process variables, comparing Eqs. (10) and
(13). The effects of the variables on lg m were synergetic and very
complex, and it became difficult to obtain a deterministic model
with a mathematical expression capable of accurately explaining
biological complexity. However, the surface of the Taylor expan-
sion curve can explain the effect of the variables on the response.
Second-order Taylor equations, i.e. quadratic models, are the most
applied, and higher order models are only used when quadratic
models show a clear lack of fit. The quadratic model is an approx-
imation of the unknown function of the response and explains the
process for a wide variety of operating parameters (Vogel and Tod-
aro, 1997). This study proposes a quadratic model for lg m. The
range of the process variables was sufficiently limited to allow
the quadratic model to fit the data properly and approximate the
response function, but broad enough to cover all the situations of
an industrial process, bearing in mind that the model would offer
a worse explanation of the behaviour of fermentation out of this
range. The calculated model was as follows:
188 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196

lg m ¼ 0:06874  0:005261  X 1 þ 0:043798  X 2 þ 0:03401  X 3
þ 0:031077  X 4  0:000232  X 5  0:000595  X 1  X 1
 0:025606  X 2  X 2  0:040034  X 3  X 3  0:038633  X 4  X 4
 0:009619  X 5  X 5  0:0077  X 1  X 2  0:02373  X 1  X 3
þ 0:044161  X 2  X 3  0:013122  X 1  X 4 þ 0:031656  X 2  X 4
þ 0:009065  X 3  X 4 þ 0:011637  X 1  X 5  0:022271  X 2  X 5
 0:017904  X 3  X 5 þ 0:011003  X 4  X 5 ð14Þ
where X 1 , X 2 , X 3 , X 4 and X 5 were the codified values for the process
variables
P  1:5
X1 ¼ ð15Þ
0:5
E  20:681
X2 ¼ ð16Þ
27:39
vvm  19:917
X3 ¼ ð17Þ
20:084
N  600
X4 ¼ ð18Þ
505:9
T L  29:5
X5 ¼ ð19Þ
4:52
Codification encoding has prevented variables with large figures
from becoming more important than the model, disguising other
variables. The robustness of the values obtained by the continuous
processes assess assayed the robustness of the model calculated in
this work study for lg m. The adjusted correlation rate value, 0.984,
demonstrated the suitability of the quadratic model. The individual
and combined effects of the process variables were explained pre-
viously by using biological and physicophysical–chemical argu-
ments (Garrido-Vidal et al., 2003).
In terms of oxygen transfer, volumetric oxygen transfer, kL a,
was critical in the oxygen balance of fermentation. In a previous
study (González-Sáiz et al., 2008), a model was calculated from
the kL a values obtained by the Doehlert design of continuous pro-
cesses and by applying genetic algorithms
 0:363
Pg
kL a ¼ 1:0386  104  ð1:319ÞP  ð1:107ÞT L   ðlS Þ1:439
profile, especially in terms of flavour. As regards Spanish regula-
tions, acetoin concentration must be higher than 0.04 g L1 for vin-
egar to be sold for consumption. Concentrations of 1 g L1 can be
reached although extremely high concentrations are not desirable,
since acetoin can also be produced by secondary oxidative routes
related to alterations of vinegar.
Acetoin or acetyl methyl carbinol is formed by the condensation
of two molecules of acetaldehyde, as an intermediate of the metab-
olism of intracellular oxidation of ethanol with oxygen to produce
acetic acid. Some intracellular enzymes, such as pyruvate descarb-
oxilase, are involved in the production of acetoin, which also de-
pends on the concentration of certain compounds such as lactic
acid and pyruvate. Considering that acetoin is related to the oxida-
tive mechanism of bacteria, it is logical to suggest that it depends
on cell growth.
Fig. 2a shows the ethyl acetate production rate vs the growth
rate and Fig. 2b presents the acetoin production rate vs the growth
rate, calculated from data obtained in the previous study (Garrido-
Vidal et al., 2003). A clear linear tendency was observed, as demon-
strated by the good values of r2 A for the linear models. Linearity
demonstrated that ethyl acetate and acetoin production were asso-
ciated with cell growth. The slope of linear regressions provided
the yield factors Y 0EA=X and Y 0AMC=X , with the values 6.6 ± 0.7 and
2.2 ± 0.3, respectively. The kinetic models deduced are
!
1 dX t
rEA ¼   ð21Þ
Y 0EA=X dt
!
1 dX t
rAMC ¼  0  ð22Þ
Y AMC=X dt
Finally, one of the main causes of the reduction in yields during
acetic fermentation is the loss of ethanol and acetic acid due to li-
quid–gas transfers. In the case of ethyl acetate and acetoin, these
losses prompt a depletion of the sensorial features of the final
product. In the case of water, these losses have an important effect
0.30
Ethyl Acetate formation rate

y = (6.6 ± 0.3) x + (-0.001 ± 0.008)

VL 0.25 2
R A= 0.9778
ð20Þ
0.20
Apart from the main product of acetic fermentation, i.e. acetic
(g/L*h)

acid, various secondary products are important for vinegar. These 0.15

products have not been studied in such depth in literature. Esters

0.10
are the mean compounds for vinegar flavour, ethyl acetate being
the most important. Ethyl acetate is formed by the esterification 0.05
of acetic acid with ethanol. The sensorial effect of ethyl acetate is
typical of vinegars manufactured by traditional methods, and 0.00
which are characterized by high levels of residual ethanol, slow 0.00 0.01 0.02 0.03 0.04

acetification rates and low volatile loss. These features favour the Cell growth rate (g/L*h)
formation of ester during long fermentation processes or during 0.09
Acetoin formation rate (g/L*h)

vinegar aging.
0.08 y = (2.2 ± 0.1) x + (- 0.001 ± 0.003)
Esterification is catalyzed enzymatically by esterases (EC 3.1.1.) 2
0.07 R A= 0.9676
or chemically by inorganic acids. Ethyl acetate is produced by Ace-
tobacter aceti during acetic fermentation by a type of intracellular 0.06
esterase (Kashima et al., 1998, 2000). Extra-cellular formation of 0.05
ethyl acetate can be ruled out during fermentation, although it is 0.04
the only mechanism during vinegar aging. It is logical to suggest 0.03
that the intracellular production of ethyl acetate depends on cell
0.02
activity. Similar mechanisms are related to other esters, which
0.01
are also important for vinegar flavour and also very volatile. There-
fore, ethyl acetate concentration during fermentation can be used 0.00
0.00 0.01 0.02 0.03 0.04
as an indicator of ester concentration.
Cell growth rate (g/L*h)
Acetoin is another important secondary product. The presence
of acetoin demonstrates the chemical and microbiological integrity Fig. 2. Secondary products formation rates vs the cell growth rate: (a) ethyl acetate
of vinegar and constitutes a positive contribution to its sensorial and (b) acetoin.
 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
on the concentration profiles and total working volume. A previous
study (González-Sáiz et al., 2008) proposed a model for liquid–gas
transfers in acetic fermentation
LGTRi ¼ ki;G a  C i;L
ki;G a was defined as the volumetric component i transfer coeffi-
cient in the gas-side interphase. The following models were calcu-
lated in that study for each component using the values obtained
for ki;G a with a Doehlert design of batch processes without fermen-
tation applied to pressure, aeration, agitation, temperature and the
concentration of each component and genetic algorithms
Ethyl acetate : kEA;G a ¼ 565:2  ð0:3730ÞP  ð1:008ÞT L
 0:019
Pg
  ðlS Þ1:106
VL
P TL
Ethanol : kE;G a ¼ 0:590  ð0:4346Þ  ð1:060Þ
 0:030
Pg
  ðlS Þ0:910
VL
P
Acetoin : kAMC;G a ¼ 0:061  ð0:6006Þ  ð1:087Þ TL
 0:154
Pg
  ðlS Þ1:251
VL
P TL
Acetic acid : kA;G a ¼ 0:173  ð0:5126Þ  ð1:087Þ
 0:167
Pg
  ðlS Þ1:404
VL
P TL
Water : kW;G a ¼ 0:892  ð0:3028Þ  ð1:070Þ
 0:102
Pg
  ðlS Þ1:335
VL
In Eqs. (20), (23)–(27) and (28), air superficial velocity mea-
sured at the exhaust line of the fermentor was corrected for the ef-
fect of pressure and temperature of the medium on the air flowing
through the fermentation medium, to obtain lS .
3.2. Applicability of the model
A simulation environment was developed in Matlab (The Math-
Works, Inc.) to evaluate the predictive ability of the model. The
core of the simulation environment was the solution of the differ-
ential equation system inferred from the models reported in Sec-
tion 3.1, by the fourth order Runge–Kutta algorithm. A term was
added for each component to consider the effect of the volume of
wine introduced, V c , and the volume of vinegar discharged, V d .
The simulation environment calculated V c and V d to keep ethanol
concentration and volume within the set point. The effects of the
liquid–gas transfers on the concentrations were considered to sim-
ulate open systems and not closed systems, i.e. those with a con-
denser. In the case of ethanol and acetic acid, the consumptions
due to ethyl acetate synthesis were taken into account. The recy-
cling rate with respect to volume discharge was introduced to cal-
culate the biomass concentration in continuous processes with cell
recycling. Fermentor volume was predicted by the volumes
charged and discharged, together with the effect of water loss
due to liquid–gas transfer, ruling out the effects of other losses.
The signal provided by the oxygen probe was estimated from the
C O2;L values, considering that the signal was proportional to the
oxygen partial pressure in the gas phase in equilibrium with C O2;L
and that the 100% signal was calibrated by saturation of the fer-
mentation medium with air at 30 °C and 1 atm.
The differential equation system predicted the concentrations
along the whole fermentation medium and the predictive ability
improved when mixing conditions were closer to the ideal perfect
mixed reactor. When differences between the medians and the
averages of a set of parameters over a period of time (in the case
189
of continuous processes) or during a number of batches (in the case
of semi-continuous processes) were less than certain limits, the
model considered that a steady or semi-steady state has been
reached. A batch was the period of fermentation time of a semi-
ð23Þ
continuous process between the beginning of vinegar discharge
and the time when the ethanol set point for discharge was reached
again. The initial batch concentrations were the concentrations
after the charge of wine; and the final concentrations were those
when ethanol reached the set point of discharge. The values of
the limits, shown in Table 2, were lower than the significant figures
of the real parameters.
Finally, it was mathematically possible to obtain negative values
for oxygen and ethanol concentration. A condition was introduced
to stop simulation when ethanol was consumed. In the case of oxy-
ð24Þ gen, a negative value indicated that oxygen consumption was high-
er than the oxygen transfer rate. This situation could be achieved in
processes with cell recycling when the accumulation of biomass
exceeded the oxygen transfer capacity of the system. In a real pro-
ð25Þ
cess, the growth cell decreases to reach a state with 0% of dissolved
oxygen, i.e. a limit-by-oxygen state. This state cannot be explained
by models based on dissolved oxygen concentration (Park et al.,
1991; Romero et al., 1994) since oxygen is a response due to the
ð26Þ
balance between oxygen consumed and oxygen transferred. It
can be inferred that in limit-by-oxygen states, cell population via-
bility decreases because there is no oxygen for all the cells. This
ð27Þ effect was simulated by introducing a correction factor over lg m,
calculated inside the Runge–Kutta algorithm to balance growth cell
and oxygen consumption with oxygen transferred to the medium
and keep oxygen concentration above zero.
ð28Þ The first step in the study of the applicability of the global model
was to evaluate the impact of the value of the initial concentrations
of biomass and the other compounds on the predictions of the mod-
el. Fig. 3 shows the predicted concentration profiles and rates of
simulated continuous processes with condenser, head space pres-
sure 1.5 atm, air flow 132 L h1 (20.31 h1), agitation 600 rpm, tem-
perature 29.5 °C, ethanol set point 20 g L1 and different initial
concentrations of suspended matter, total biomass, ethanol and
acetic acid. All the predicted processes reached the same concentra-
tions, wine and vinegar flows and acetification rates in the steady
state, calculated with 72 h of simulation. Table 3 shows the initial
concentrations and results of the predicted processes, together with
the results of the semi-steady state reached in the pilot fermentor,
with the initial concentrations reported in Table 3. As shown in Ta-
ble 3, the predicted values were close to the experimental ones.
The same study was developed with a semi-continuous process.
Fig. 4 shows the predicted concentration profiles and rates of simu-
lated semi-continuous processes with condenser, head space pres-
sure 1.5 atm, air flow 132 L h1 (20.31 h1), agitation 600 rpm,
temperature 29.5 °C, ethanol set point for discharge 20 g L1, and
initial ethanol concentration point for charge 40.14 g L1 and differ-
Table 2
Parameters and limits used to evaluate the steady and semi-steady states
Parameters Limits
Continuous (steady state Semi-continuous (final batch
concentrations and wine concentrations and volumes
and vinegar streams) charged and discharged)
Vinegar 102 102
Wine 102 102
Acetic acid 104 104
Total biomass 105 105
Suspended matter 105 105
Dissolved oxygen 102 102
Ethyl acetate 105 105
Acetoin 105 105
Ethanol – 104
190
Fig. 3. Simulated concentration profiles and rates of continuous processes with condenser, head space pressure 1.5 atm, air flow 132 L h1 (20.31 h1), agitation 600 rpm,
temperature 29.5 °C, ethanol set point 20 g L1 and different simulated initial concentrations of suspended matter, total biomass, ethanol and acetic acid.
Table 3
Predicted vs experimental values of continuous processes with different initial
simulated concentrations
Initial suspended matter
concentration
Initial total biomass
concentration
Initial ethanol
concentration
Initial dissolved oxygen
Initial acetic acid
concentration
Initial ethyl acetate
concentration
Initial acetoin
concentration
Wine flow in steady stated
Vinegar flow in steady
stated
Acetification rate in steady
stated
Suspended matter in
steady state
Total biomass in steady
state
Ethanol concentration in
steady state
Dissolved oxygen in
steady state
Acetic acid concentration
in steady state
Ethyl acetate
concentration in steady
state
Acetoin concentration in
steady state
J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
ent initial concentrations of suspended matter and total biomass.
All predicted processes reached the same batches times, initial
and final batch concentrations, and charged and discharged vol-
Simulations Experimental umes in the semi-steady state, calculated with 10 batches. The rates
Different values within range 0.04 g L1 of the wine charge pump and the vinegar discharge pump in the
[0.04–0.63] g L1 simulation processes were the same as in the real process,
Initial suspended matter- – 10 L h1 and 50 L h1, respectively. Table 4 shows the initial con-
suspended matter in wine
centrations and results of the predicted processes together with
Different values within range 31.58 g L1
[3.948–47.38] g L1 the results of the semi-steady state reached in the pilot fermentor,
100% 100 ± 3% with the initial concentrations reported in Table 4. As shown in Ta-
Different values within range 75.17 g L1 ble 4, the predicted values were close to the experimental ones. The
[55–110] g L1
batch initial concentrations of ethanol in predicted and real pro-
2.00 g L1 2.00 g L1
cesses were slightly lower than the initial ethanol concentration
0.30 g L1 0.30 g L1 for charge. This was because the discharge of vinegar had to be fast
so the next batch could be started as soon as possible, but the charge
0.4430 L h1 0.4253 ± 0.02 g L1 of wine had to be slower in order to avoid the negative effect on bio-
0.4430 L h1 0.4253 ± 0.02 g L1
mass due to an abrupt change in concentration. Therefore, ethanol
5.81 g L1 h1 5.53 ± 0.4 g L1 in the medium was being consumed during the charge of wine.
These results showed that biomass concentration and viability
0.43 g L1 0.43 ± 0.04 g L1 reached in the simulated steady or semi-steady state were unique
for each set of process variables, independently of the initial con-
0.40 g L1 –
centrations, and corresponded to the values of the experimental
19.91 g L 1
19.11 ± 2 g L1 process. Therefore, the model was able to predict any steady or
semi-steady state for a set of process conditions, although the via-
4.66% 4.71 ± 0.14% bility of the initial cell population was unknown.
Finally, the experiments with the Doehlert design of continuous
88.44 g L1 89.56 ± 7 g L1
processes achieved to develop the models for kL a, the cell growth
2.84 g L1 2.70 ± 0.4 g L1 kinetics, substrates and main product kinetics and secondary prod-
uct kinetics were reproduced by simulation. The simulated pro-
cesses were performed in the same order as the experimental
1.18 g L1 1.22 ± 0.2 g L1
ones, and the initial concentrations of each experiment were the fi-
nal concentrations of the process simulated before. The volatile

Fig. 4. Simulated concentration profiles and rates of semi-continuous processes with condenser, head space pressure 1.5 atm, air flow 132 L h1 (20.31 h1), agitation
600 rpm, temperature 29.5 °C, ethanol set point for discharge 20 g L1, initial ethanol concentration point for charge 40.14 g L1 and different simulated initial concentrations
of suspended matter and total biomass.
Table 4
Predicted vs experimental values in semi-continuous processes with different initial
simulated concentrations
Initial suspended matter
concentration
Initial total biomass
concentration
Initial ethanol concentration
Initial dissolved oxygen
Initial acetic acid
concentration
Initial ethyl acetate
concentration
Initial acetoin concentration
Wine charged in each batch
Wine charged in each batch
Batch time
Initial/final batch suspended
matter concentration
Initial/final batch total
biomass
Initial/final batch ethanol
concentration
Initial/final batch dissolved
oxygen concentration
Initial/final batch acetic acid
concentration
Initial/final batch ethyl
acetate concentration
Initial/final batch acetoin
concentration
J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196 191
compounds in the gas outlet after condenser were excluded and
the system was considered as perfectly closed for simulation.
Fig. 5 shows the predicted vs measured plots of the variables in
Simulations Experimental
the steady state. The plots displayed a good linear correlation:
95% confidence intervals of slopes included 1, 95% confidence
Different values within range 0.07 g L1
[0.021–1.02] g L1
intervals of y-intercepts included 0 and the adjusted correlation
Initial suspended matter- – rates were close to 1. In the case of ethanol, the y-intercept did
suspended matter in wine not include 0, and showed that predicted values were slightly
40.14 40.14 g L1 biased, since the simulation environment maintained the ethanol
100% 100 ± 3%
concentration equal or under the set point. However, the slope
82.73 g L1 82.73 g L1
and adjusted correlation rate showed almost perfect linearity. All
2.00 g L1 2.00 g L1 in all, the plots demonstrated the good correlation of the models
with the experimental data, considered as a calibration set.
0.30 g L1 0.30 g L1
1.59 L 1.59 ± 0.02 L
1.58 L 1.58 ± 0.02 L 3.3. Evaluation of the predictive ability of pilot processes
4.28 h 4.35 h
0.40/0.51 g L1 (0.40 ± 0.03)/ A set of six continuous processes was developed in the pilot fer-
(0.52 ± 0.04) g L1
1 mentor without cell recycling, without condenser and with process
0.38/0.49 g L –
conditions different to the Doehlert design of continuous experi-
39.97/19.86 g L1 (37.22 ± 3)/ ments reported in Section 3.2. Fig. 6 shows the predicted vs mea-
(21.27 ± 2) g L1 sured plots of the variables in steady state. The plots showed a
0/0% (0.15 ± 0.005)/
good linear correlation: 95% confidence intervals of slopes included
(0.27 ± 0.008)%
81.58/105.63 g L1 (81.38 ± 6.51)/ 1, 95% confidence intervals of y-intercepts included 0 and the ad-
(98.36 ± 7.87) g L1 justed correlation rates were close to 1, demonstrating the predic-
1.22/1.53 g L1 (1.22 ± 0.66)/ tive capacity of the models with an external validation set.
(1.62 ± 0.13) g L1 Another set of 10 continuous fermentation processes was devel-
1.12/1.38 g L1 (1.11 ± 0.09)/
oped in the pilot fermentation system with condenser, head space
(1.35 ± 0.11) g L1
pressure 0.85 atm, air flow 220 L h1, agitation 500 rpm, tempera-
192 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196

D (1/h) Acetification rate (g/L*h) Kla (1/h)

0.12 9 800
y = (0.94 ± 0.09) x - (0.003 ± 0.005) 8 y = (1.02 ± 0.09 )x + (- 0.3 ± 0.4) 700 y = (1.04 ± 0.07) x + (- 16 ± 24)
0.10 2
R A= 0.9291 2 2
7 R A= 0.9443 600 R A= 0.9616
0.08 6

Predicted
Predicted

500

Predicted
5
0.06 400
4
300
0.04 3
200
0.02 2
100
1
0.00 0
0 0 100 200 300 400 500 600 700 800
0.00 0.02 0.04 0.06 0.08 0.10 0.12
0 1 2 3 4 5 6 7 8 9
Experimental Experimental Experimental

Suspended matter (g/L) EthanoL (g/L) Dissolved Oxygen (%)

1 55 200
50 175 y = (1.2 ± 0.2)x + (-2 ± 11)
y = (0.97 ± 0.05) x + (0.02 ± 0.03) y = (0.999 ± 0.001) x + (- 0.03 ± 0.02) 2
2 45 2 R A=0.8469
0.8 R A= 0.9757 R A= 1 150
40

Predicted
125
Predicted

Predicted
35
0.6 30 100
25
75
20
0.4 15 50
10 25
0.2 5 0
0.2 0.4 0.6 0.8 1.0 0 0 25 50 75 100 125 150 175 200
Experimental 0 5 10 15 20 25 30 35 40 45 50 55 Experimental
Experimental

Acetic Acid (g/L) Ethyl Acetate (g/L) Acetoin (g/L)

120 4 1.2
3.8 y = (1.1 ± 0.6) x + (- 0.1 ± 0.2)
110 y = (0.9 ± 0.1) x + (8 ± 9) 2 y = (1.2 ± 0.2) x + (-0.2 ± 0.2)
2 3.6 R A= 0.8634 1.1
R A= 0.9163 R2 A= 0.8366
100 3.4

Predicted
1
Predicted
Predicted

90 3.2
3 0.9
80 2.8
2.6 0.8
70
2.4
60 0.7
2.2
2 0.6
50
50 60 70 80 90 100 110 120 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Experimental Experimental
Experimental

Fig. 5. Predicted vs measured plots of the Doehlert design of 35 continuous processes used to calculate the kinetic model.

D (1/h) Acetification rate (g/L*h) Kla (1/h)

0.14
11 700
0.12
y = (1.1 ± 0.3 )x + (-0.01 ± 0.02)
10
R2 A= 0.9641 9
y = (1.1 ± 0.3) x +(-1 ± 2) 600 y = (1.1 ± 0.2)x + (- 83 ± 90)
0.10 8 R2 A= 0.9420 2
500 R A= 0.9755
Predicted

Predicted

Predicted

7
0.08 400
6
0.06 5
300
4
0.04 3 200
2
0.02 1 100
0 0
0.00 0 1 2 3 4 5 6 7 8 9 10 11
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0 100 200 300 400 500 600 700
Experimental Experimental Experimental

Suspended matter (g/L) EthanoL (g/L) Dissolved Oxygen (%)

0.6 55 100
0.6 50 y = (0.999 ± 0.004)x +(- 0.1 ± 0.1) y = (1.1 ± 0.2)x + (-8 ± 10)
y = (0.8 ± 0.3) x + (0.1 ±0.1) 2
45 R A= 0.1 2
0.5 R2 A= 0.9365 75 R A= 0.9693
40
Predicted

0.5
Predicted

35
Predicted

0.4 30
50
25
0.4
20
0.3 15
25
0.3 10
5
0.2
0 0
0.2 0.3 0.4 0.5 0.6
0 5 10 15 20 25 30 35 40 45 50 55 0 25 50 75 100
Experimental Experimental Experimental

Ethyl Acetate (g/L) Acetoin (g/L)

Acetic Acid (g/L)
110 1.7 1.2

y = (1.0 ± 0.1) x + (-0.5 ± 9) 1.5 y = (1.1 ± 0.2) x + (- 0.1 ± 0.2) 1.1 y = (0.9 ± 0.4) x + (0.0 ± 0.3)
100 2
R A= 0.9930 R2 A= 0.9839 R2 A= 0.9082
1.3 1
Predicted

90
Predicted
Predicted

1.1 0.9
80 0.8
0.9
70 0.7 0.7

0.5 0.6
60
0.3 0.5
50 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
0.3 0.5 0.7 0.9 1.1 1.3 1.5 1.7
50 60 70 80 90 100 110 Experimental
Experimental
Experimental

Fig. 6. Predicted vs measured plots of the 6 continuous processes without condenser and without cell recycling, used for external validation.
 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196 193

ture 29.5 °C, ethanol set point 20 g L1 and varying the recycling
rate within range [0–0.9]. As shown in Fig. 7, the predictions of
the model were close to the experimental results. Total biomass
was enhanced exponentially with recycling rate and it would
therefore have been reasonably an enhancement of the acetifica-
tion rate. However, the acetification rate increased with recycling
rates under 0.6, but from 0.6 to 0.9 it was constant. Dissolved oxy-
gen decreased with the recycling rate within range [0–0.6], and a
zero value was obtained with a recycling rate of 0.6. Therefore, it
was concluded that the oxygen supplied was insufficient to satisfy
cell population demand with recycling rates above 0.6, and cell via-
bility decreased to adapt the multiplication rate to oxygen supply,
reaching limit-by-oxygen states with zero dissolved oxygen con-
centrations. The simulation environment was able to predict these
states thanks to the correction factor over lg m, which was intro-
duced to avoid negative values in dissolved oxygen.
In terms of semi-continuous processes, a set of 21 experiments
was developed without condenser, ethanol set point for discharge
20 g L1, initial ethanol concentration point for charge 40.14 g L1
and initial concentrations of each experiment equal to the final
concentrations of the previous experiment. Hydrostatic pressure,
aeration, agitation and temperature corresponded to the values
of a Doehlert design applied to these variables. The rates of the
wine charge pump and the vinegar discharge pump in the simula-
tion processes were the same as those in the real process, 10 and
50 L h1, respectively. Fig. 8 shows the predicted vs experimental
plots of the averages in the last 10 batches, with the values of
the variables shown in Table 2 being maintained; these were used
as criteria to evaluate the semi-steady states in both experimental
and predicted processes. As inferred in Fig. 8, the predicted values
were close to the experimental ones, considering the values of
slope, y-intercept and adjusted correlation rate. The other concen-
trations did not have enough variability to build a predicted vs
experimental plot because in each batch the initial and final con-
centrations were almost the same, which is logical bearing in mind
that an ethanol concentration set point was established for dis-
charge and for charge. In the case of ethyl acetate, variability was
due to the impact of liquid–gas transfer, which was less important
for the other variables. Table 5 shows the ranges obtained in exper-
iments and predictions for these variables. It was obvious that the
experimental and predicted results were close. The initial batch
concentrations of ethanol in predicted and real processes were
slightly lower than the initial ethanol concentration for charge,
as explained in Section 3.2. Some conditions with extremely low
agitation and/or aeration values were unable to start fermentation,
neither the experimental processes nor the predicted ones. These
zero values were not represented in plots to avoid forced linear
correlation.
Finally, Fig. 9 shows the acetic acid concentration profiles of 3
batches of the semi-steady state of four semi-continuous processes
predicted by the models, together with the experimental concentra-
tions. The process conditions were headspace pressure 0.85 atm, air
flow 220 L h1, agitation 500 rpm, temperature 29.5 °C and closed
system, i.e. with condenser. The initial ethanol concentration point
for charge was 40.14 g L1 and initial concentrations of each exper-
iment were 40.14 g L1 ethanol, 82.73 g L1 acetic acid, 2 g L1 ethyl
acetate, 0.3 g L1 acetoin and 0.021 g L1 suspended matter. Total
biomass in simulations was considered as 0.001, i.e. as starting con-
ditions. Ethanol set points for discharge were 4.36, 8, 12 and
16 g L1 for each process. The rates of the wine charge pump and
the vinegar discharge pump in the simulation processes were the
same as in the real process, i.e. 10 and 50 L h1, respectively. The cri-
teria used to evaluate the semi-steady states in both experimental
and predicted processes were the same as those shown in Fig. 8.
As inferred in Fig. 9, the models offered good predictions of the real
processes.

11 140
130
Acetification rate (g/(L *h))

10 120
Dissolved Oxygen (%)

110
9 100
90
8 80
70
7 60
50
6 40
Predicted 30 Predicted
5 Experimental 20 Experimental
10
4 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Recycling rate Recycling rate

1.0
4.0
0.9
3.5
0.8
Total Biomass (g/L)

Correction factor

3.0 0.7
2.5 0.6
2.0 0.5

1.5 0.4
0.3
1.0 Predicted
Experimental 0.2
0.5
0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Recycling rate Recycling rate

Fig. 7. Experimental and predicted values of the 10 continuous processes with condenser and recycling rates from 0 to 9.
194 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196

Dissolved Oxygen initial batch concentrations (%) Dissolved Oxygen final batch concentrations (%)
150 175
y = (1.1 ± 0.1)x + (- 1 ± 7) 150 y = (1.0 ± 0.1)x + (2 ± 9)
125 2 2
R A= 0.9650 R A= 0.9519
125
Predicted

Predicted
100
100
75
75
50
50
25 25
0 0
0 25 50 75 100 125 150 0 25 50 75 100 125 150 175
Experimental Experimental

Ethyl acetate initial batch concentrations (g/L) Ethyl acetate final batch concentrations (g/L)
2.0 2.5
y = (1.0 ± 0.1) x + (- 0.0 ± 0.1) y = (0.95 ± 0.08) x + (0.1 ± 0.1)
2 2.0 2
1.5 R A= 0.9701 R A= 0.9749

Predicted
Predicted

1.5
1.0
1.0

0.5
0.5

0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 2.5
Experimental Experimental

Batch time (h)

14
12 y = (1.04 ± 0.08)x + (- 0.3 ± 0.6)
2
R A= 0.9803
10
Predicted

8
6
4
2
0
0 2 4 6 8 10 12 14
Experimental

Fig. 8. Predicted vs measured plots of the Doehlert design of 21 semi-continuous processes without condenser and with set point for discharge 20 g L1 ethanol, used for
external validation.

Table 5 biomass with low ethanol concentrations. As shown in the Figure,

Predicted vs experimental range values of the Doehlert design of semi-continuous the acetification rate declined with ethanol concentrations below
processes with set point for discharge 20 g L1 ethanol 10 g L1, as explained by the model for lg m. Viability was not
Simulations Experimental recovered immediately when initial batch concentration was
Wine charged [1.69–1.71] L 1.69 L
reached again, and the effect in the pilot fermentor was observed
Vinegar discharged 1.69 L 1.69 L as a ‘‘re-start” condition at the beginning of each batch, which
Suspended matter initial [0.35–0.37] g L1 [0.32–0.38] g L1 could be explained by the models with the decrease in the initial
Suspended matter final [0.47–0.48] g L1 [0.43–0.51] g L1 batch biomass concentrations. Furthermore, it is not useful for
Ethanol initial [39.14–39.96] g L1 [39.38–39.96] g L1
the vinegar industry to work in semi-continuous with ethanol con-
Ethanol final [19.81–19.95] g L1 [19.71–19.97] g L1
Acetic acid initial [79.23–84.46] g L1 [79.07–82.34] g L1 centrations for discharges below 15 g L1, and 20 g L1 is the most
Acetic acid final [103.1–106.3] g L1 [101.7–108.5] g L1 used value. Lower set points are only applied when no more than
Acetoin initial [1.027–1.063] g L1 [1.003–1.086] g L1 one fermentor is available in the industrial plant, but the most
Acetoin final [1.280–1.315] g L1 [1.242–1.350] g L1 usual system consists of at least two fermentors, one of them
working in semi-continuous and periodically producing a regular
volume of active fermentation medium to be completely ex-
However, in case of processes with ethanol set points of 4.36 hausted in the other fermentor. The correction factor was experi-
and 8 g L1, correction factors had to be introduced of 0.516 and mental and it had to be calculated for each set of process variables.
0.88, respectively. The correction factors decreased the initial bio- In conclusion, it was shown that the models reported in Sec-
mass of each batch but not suspended matter concentration. It was tion 3.1 successfully explained the industrial process of vinegar
necessary for low ethanol set points due to the negative effect of production and built a global model to predict the type of fer-
the concentrations gradient change to reach again the batch initial mentation processes that can be developed in an industrial plant,
concentration and, above all, due to the low viability state of the i.e. continuous or semi-continuous processes, cell recycling, open
 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
Fig. 9. Predicted (lines) and experimental (dots) acetic acid concentration profiles of semi-continuous processes with condenser and with different set ethanol concentration
set points for discharge: 4.36, 8, 12 and 16 g L1.
or closed systems. These results also showed that biomass con-
centration, total biomass viability and specific growth can be con-
sidered as responses of the fermentation system for a set of
process variables. Therefore, the substrate consumption and prod-
uct excretion rates and dissolved oxygen concentration also be-
come responses. In the case of semi-continuous processes,
ethanol concentration changed throughout the batch, while the
other process variables remained constant. When the semi-steady
state was reached, the concentration profile of each batch and the
volumes charged and discharged at the end of each batch were
constant. In this state, the concentration profile depended on
the adaptability of bacteria to changes in ethanol concentration
and it can be considered that each point of the batch corre-
sponded to the specific growth rate and viability of the bacteria
in a steady state. Each steady state concentration in continuous
can be considered as a snapshot of the batch cycle with the same
process variables.
The wide applicability range of the global model developed in
this study has not been reported for mechanistic acetic fermenta-
tion models described in literature and it is an invaluable tool for
the design of industrial processes with aerated-stirred fermentors.
4. Conclusions
A model for industrial acetic fermentation processes was built
on the following premises:
195
– The growth rate was described by total biomass, X t , and the total
biomass specific growth rate, lg m, which represented the growth
of an active biophase and the fraction of cells which were grow-
ing. lg m was modelled by a quadratic model based on the pro-
cess variables, i.e. hydrostatic pressure, ethanol concentration,
aeration, agitation and temperature.
– Substrate consumption, ethanol and oxygen, main product syn-
thesis, acetic acid and secondary compounds production, acetoin
and ethyl acetate, were related to the growth rate by a growth-
associated kinetic.
– Oxygen transfer and liquid–gas transfers of the volatile com-
pounds were explained by the mass transfer models reported
in a previous study.
It was shown that the viabilities and cell multiplication rates
achieved in each simulated steady or semi-steady state were un-
ique to each set of process variables, independently of the initial
concentrations. The model successfully explained the steady and
semi-steady states reached in the pilot fermentor, within a broad
range of process conditions, although the viability of the initial cell
population was unknown, as demonstrated by simulation of the
continuous and semi-continuous processes developed in the pilot
fermentor, with and without condenser and including experiments
with cell recycling.
The model contains the effects of oxygen transfer and the
hydrodynamics of the fermentation medium, since these depends
196 J.-M. González-Sáiz et al. / Journal of Food Engineering 91 (2009) 183–196
on the variables involved in both phenomena, i.e. hydrostatic pres-
sure, agitation, aeration and temperature. Therefore, it is very use-
ful for the scale-up of fermentation processes. A future study will
demonstrate the scale-up capacity of the model and the simulation
environment will be applied to design processes with industrial-
scale fermentors.
Acknowledgements
The authors thank the Spanish Government (Ministerio de Edu-
cación y Ciencia, Project No. AGL2003-09138-C04-04), the Autono-
mous Government of La Rioja (Plan Riojano de I+D+I, Projects No.
ACPI-2004/02 and ACPI-2005) and the University of La Rioja for
their financial support.
References
Bar, R., Gainer, J.L., Kirwan, D.J., 1987. An unusual pattern of product inhibition:
batch acetic acid fermentation. Biotechnology and Bioengineering 29, 796–798.
Caro, I., Macías, M., Cantero, D., 1996. Optimum operating conditions in closed-
system industrial acetifiers (batch operation): a study by computer simulation.
Chemical Engineering Journal & Biochemical Engineering Journal 62, 183.
Çelik, E., Çalik, P., 2004. Bioprocess parameters and oxygen transfer characteristics
in b-lactamase production by bacillus species. Biotechnology Progress 20, 491–
499.
Chen, L., Bernard, O., Bastin, G., Angelov, P., 2000. Hybrid modelling of
biotechnological processes using neural networks. Control Engineering
Practice 8, 821–827.
Croughan, M.S., Hamel, J.F., Wang, D.I.C., 1987. Hydrodynamic effects on animal
cells grown in microcarrier cultures. Biotechnology and Bioengineering 29,
130–141.
Doehlert, D.H., 1970. Uniform shell designs. Applied Statistics 19, 231–239.
Drysdale, G.S., Fleet, G.H., 1989. The growth and survival of acetic acid bacteria in
wines at different concentrations of oxygen. American Journal of Enology and
Viticulture 40, 99–105.
El-Enshasy, H., Kleine, J., Rinas, U., 2006. Agitation effects on morphology and
protein productive fractions of filamentous and pelleted growth forms of
recombinant Aspergillus niger. Process Biochemistry 41, 2103–2112.
Elibol, M., Ozer, D., 2000. Influence of oxygen transfer on lipase production by
Rhizopus arrhizus. Process Biochemistry 36, 325–329.
Evans, G.M., Liu, G., 2003. Optimising activated sludge growth rate by intensifying
hydrodynamic forces. Journal of Chemical Technology and Biotechnology 78,
276–282.
Garrido-Vidal, D., Esteban-Díez, I., Pérez-del-Notario, N., González-Sáiz, J.M.,
Pizarro, C., 2004. On-line monitoring of kinetic and sensorial variables of the
acetic fermentation by nirs. Journal of Near Infrared Spectroscopy 12, 15–27.
Garrido-Vidal, D., Pizarro, C., González-Saiz, J.M., 2003. Study of process variables in
industrial acetic fermentation by a continuous pilot fermentor and response
surfaces. Biotechnology Progress 19, 1468–1479.
Gómez, J.M., Cantero, D., 1998. Kinetics of substrate consumption and product
formation in closed acetic fermentation systems. Bioprocess Engineering 18,
439–444.
González-Sáiz, J.M., Pizarro, C., Garrido-Vidal, D., 2003. Evaluation of kinetic models
for the industrial acetic fermentation: proposal of a new model optimised by
genetic algorithms. Biotechnology Progress 19 (2), 599–611.
González-Sáiz, J.M., Pizarro, C., Garrido-Vidal, D., 2008. Modelling gas–liquid and
liquid–gas transfers in vinegar production by genetic algorithms. Journal of
Food Engineering 87, 136–147.
Henriques, A.W.S., Jessouroun, E., Lima, E.L., Alves, T.L.M., 2006. Capsular
polysaccharide production by Neisseria meningitides serogroup C:
optimisation of process variables using response surface methodology.
Process Biochemistry 41, 1822–1828.
Ito, T., Sota, H., Honda, H., Shimizu, K., Kobayashi, T., 1991. Efficient acetic acid
production by repeated fed-batch fermentation using two fermentators.
Applied Microbiology and Biotechnology 36, 295–299.
Kashima, Y., Iijima, M., Nakano, T., Tayama, K., Koizumi, Y., Udaka, S., Yanagida, F.,
2000. Role of intracellular esterases in the production of esters by Acetobacter
pasteurianus. Journal of Fermentation Bioengineering 89 (1), 81–83.
Kashima, Y., Iijima, M., Okamoto, A., Koizumi, Y., Udaka, S., Yanagida, F., 1998.
Purification and characterization of intracellular esterases related to
ethylacetate formation in Acetobacter pasteurianus. Journal of Fermentation
Bioengineering 85 (6), 584–588.
Kruppa, R.K., Vortmeyer, D., 1999. Transient growth and product formation kinetics
of acetic acid bacteria. Bioprocess Engineering 20, 545–551.
Massart, D.L., Vandeginste, B.G.M., Buydens, L.M.C., De Jong, S., Lewi, P.J., Smeyers-
Verbeke, J., 1994. Handbook of Chemometrics and Cualiametrics, Part A.
Elsevier, London.
Mesa, M., Caro, I., Cantero, D., 1986. Viability reduction of Acetobacter aceti due to
the absence of oxygen in submerged cultures. Biotechnology Progress 12 (5),
709–712.
Muraoka, H., Watabe, Y., Ogasawara, N., 1982. Effect of oxygen deficiency on acid
production and morphology of bacterial cells in submerged acetic fermentation
by Acetobacter aceti. Journal of Fermentation Technology 60 (3), 171–180.
Nanba, A., Takumura, A., Nagai, S., 1984. Synergistic effects of acetic acid and
ethanol on the growth of Acetobacter sp.. Journal Fermentation Technology 6
(62), 501–505.
Nikerel, I.E., Öner, E., Kirdar, B., Yildirim, R., 2006. Optimisation of medium
composition for biomass production of recombinant escherichia coli cells using
response surface methodology. Biochemical Engineering Journal 32 (1), 1–6.
Oliveira, R., 2004. Combining first principles modelling and artificial neural
networks: a general framework. Computer Chemical Engineering 28 (5), 755–
766.
Ory, I., Romero, L.E., Cantero, D., 1998. Modelling the kinetics of growth of
Acetobacter aceti in discontinuous culture: influence of the temperature of
operation. Applied Microbiology Biotechnology 49, 189–193.
Park, Y.S., Toda, K., Fukaya, M., Okumura, H., Kawamura, Y., 1991. Production of a
high concentration acetic acid by Acetobacter aceti using a repeated fed-batch
culture with cell recycling. Applied Microbiology and Biotechnology 35, 149–
153.
Rita, R.C.L.B., Felipe, M.G.A., Almeida e Silva, J.B., Vitolo, M., 2003. Response surface
methodology for xylitol production from sugarcane bagasse hemicellulosic
hydrolyzate using controlled vacuum evaporation process variables. Process
Biochemistry 38, 1231–1237.
Romero, L.E., Gómez, J.M., Caro, I., Cantero, D., 1994. A kinetic model for growth of
Acetobacter aceti in submerged culture. Chemical Engineering Journal and
Biochemical Engineering Journal 54, B15–B24.
Sinclair, C.G., Topiwala, H.H., 1970. Model for continuous culture which considers
the viability concept. Biotechnology and Bioengineering 12, 1069–1070.
Sinclair, G.C., Kristiansen, B., 1987. Fermentation Kinetics and Modelling, 1a ed.
Open University Press, London.
Tang, Y.J., Zhong, J.J., 2003. Role of oxygen supply in submerged fermentation of
Ganoderma lucidum for production of Ganoderma polysaccharide and ganoderic
acid. Enzyme Microbiology Technology 32, 478–484.
Teixeira, A.P., Clemente, J.J., Cunha, A.E., Larrondo, M.J.T., Oliveira, R., 2006.
Bioprocess iterative batch-to-batch optimisation based on hybrid parametric/
nonparametric models. Biotechnology Progress 22, 247–258.
Vogel, H.C., Todaro, C.L., 1997. Fermentation and Biochemical Engineering
Handbook. Principles, Process Design and Equipment. Noyes Publications,
Westiwood, New Jersey.
Zorzetto, L.F.M., Maciel Filho, R., Wolf-Maciel, M.R., 2000. Process modelling
development through artificial neural networks and hybrid models. Computer
Chemical Engineering 24, 1355–1360.
Zuo, K., Cheng, H., Wu, S., Wu, W., 2006. A hybrid model combining hydrodynamic
and biological effects for production of bacterial cellulose with a pilot scale
airlift reactor. Biochemical Engineering Journal 29 (1-2), 81–90.