You are on page 1of 11

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/249969103

OPTIMIZATION OF FERMENTATION PROCESS


USING A MULTI-SCALE KINETICS MODEL

Conference Paper · July 2010

CITATIONS READS

2 354

3 authors:

Emily W.T. Liew Jobrun Nandong


Curtin University Sarawak Curtin University Sarawak
10 PUBLICATIONS 14 CITATIONS 81 PUBLICATIONS 155 CITATIONS

SEE PROFILE SEE PROFILE

Yudi Samyudia
Curtin University Sarawak
75 PUBLICATIONS 285 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multi-scale contol for process industry View project

Renewable fuels from wastes View project

All content following this page was uploaded by Jobrun Nandong on 03 September 2017.

The user has requested enhancement of the downloaded file.


The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

OPTIMIZATION OF FERMENTATION PROCESS USING A MULTI-


SCALE KINETICS MODEL
Emily Wan-Teng Liew, Jobrun Nandong, and Yudi Samyudia

Department of Chemical Engineering


Curtin University of Technology, Sarawak

Corresponding Author’s E-mail: {emily.liew; yudi.samyudia}@curtin.edu.my

ABSTRACT

Process modelling technique has frequently been adopted in bioreactor design,


optimization and control but one major challenge is the need to overcome the deviation from
ideal mixing behaviour. The integration of mixing phenomena into bioreactor modelling is
vital, whereby the common approach in bioreactor design, optimization and control has
always relied on kinetics of fermentation which assumes well-mixing conditions. In this
paper, a novel modelling approach is proposed by incorporating both aeration rate and stirrer
speed into the fermentation kinetics in order to describe the non-ideally mixed behaviour in a
bioreactor. Optimization of bioreactor is performed to determine the aeration rate and stirrer
speed which maximize the ethanol productivity and yield in batch fermentation with the
implementation of the novel modelling approach. Aeration rate and stirrer speed are treated
as decision variables in the optimization problem. Optimal solutions are generated to analyze
different modeling approaches as compared to experimental results.

Keywords: Bioreactor; Optimization; Mixing; Aeration Rate; Stirrer Speed.

1. INTRODUCTION
Bioreactors are widely used in process industries for mixing and blending of liquids for
biochemical reactions, which are equipped with an impeller operated in turbulent flow regime
[Harvey and Rogers, 1996]. The flow around the rotating impeller blades interacts with the
stationary baffles and generates a complex, recirculating turbulent flow. The presence of such
parameters often make the task of optimization difficult [Ranade, 1997]. Thus, the awareness
is growing on the non-uniformity distribution of the intensity and quality of flow, turbulent
kinetic energy, turbulent eddies and concentrations of species involved throughout the
bioreactor to ensure efficient operation of bioreactors, not only to achieve good yield but also
consistent product quality [Venneker et al., 2002].
From engineering perspective, mathematical modeling is one of the most successful
scientific tools available to improve the metabolic capabilities of relevant microorganisms, to
assist in the evolutionary process of genetic manipulations of the cell metabolism as well as
to improve the bioprocess conditions [Wiechert, 2002]. Hence, it is expected that modeling
will become more pervasive in the design of bioreactors [Jiang et al., 2002]. The common
approach in bioreactor design, optimization and control has always relied on kinetics of
fermentation which assumes well-mixing behaviour. Previous studies, e.g. Liew, et al (2009)
showed that the deviation from ideal mixing behaviour could lead to severe loss in yield and
changes in microbial physiology. The integration of mixing phenomena into bioreactor
1
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

modeling is therefore vital, but it is a challenging task because the detailed description of the
turbulent flow field, in combination with other transport equations, need to be described for
the interactions of mixing and fluid flow [Jenne and Reuss, 1999]. The mixing is particularly
important in the case of complex operations such as fermentation processes where an
accurate modeling is essential at both the design and operational stages including for the
derivation of optimal control strategies [Bezzo et al., 2003]. On the other hand, it is also
important to take note that such a fermentation process can be highly sensitive to certain
variables, such as batch time, liquid volume and initial nutrient concentrations due to their
complex effects on cellular metabolism. Model-based optimization is therefore a vital tool in
determining the batch operating strategies [Hjersted and Henson, 2006]. The optimization of
bioreactors now implies the manipulation of both microbial culture and the environmental
factors involved, i.e. a multivariable optimization of the process [Konde and Modak, 2007].
In this paper, the objectives of this research are two folds. First is to investigate the
inclusion of aeration rate (AR) and stirrer speed (SS) as the way of capturing the mixing
mechanism within a bioreactor so that its effect on the bioreactor’s performance, i.e. yield
and productivity, can be studied. Second is to determine the optimal conditions of AR and SS
for batch fermentation process where Response surface methodology (RSM) will be applied
by maximising the yield and productivity. In this study, we propose two modeling approaches
for captuirng the mixing phenomena. Prediction results using these models will then be
compared with experiment results from which we examine the application of different
modeling approaches for optimization and control studies.

2. MODELING APPROACH

The majority kinetics of ethanol fermentation utilize a formal (macro) approach to describe
microbial growth, whereby they are empirical and based on either Monod’s equation or on its
numerous modifications which take into account the inhibition of microbial growth by a high
concentration of product and/or substrate [Starzak et al., 1994]. Models that were proposed so
far only explained the effect of ethanol inhibition via the mechanism of non-competitive
inhibition of a simple reversible enzymatic reaction without taking into consideration of the
mixing mechanism occuring inside the bioreactor, i.e. well-mixing behaviour. Deviation from
ideal mixing behaviour could lead to severe loss in yield and changes in microbial physiology
as studied in our previous work (Liew, et al (2009)). In this study, the mixing phenomena will
be taken into account for the bioreactor kinetics modeling. Two modeling approaches were
proposed, i.e. kinetics hybrid and kinetics multi-scale model, from which we will study how
the mixing in ethanolic fermentation process could influence the yield and productivity. This
study will be conducted for the following range of experimental work: 1-1.5LPM AR and
150-250rpm SS, which are constrained by our bioreactor equipment.
2.1 Kinetics Hybrid Model
In the model development, we use a set of experimental data of biomass, product and
substrate concentrations for different AR and SS to predict kinetics parameters, k1, k2, …, k6
using the Herbert’s concept of endogenous metabolism. Herbert’s concept was chosen since
it has been used in numerous studies to describe the kinetics of ethanol fermentation [Starzak
et al., 1994]. An optimization approach is formulated for the identification of the kinetic
parametrs.

2
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

The following linear regression model will then be used for a set of identified kinetics
parameter data for different AR and SS:

(r − r ) ( R − R)
Variable = β1 + β2 + β3 (1)
∆r ∆R

where the variable represents predicted k1 to k6, r and R denote the variables taken into
account, i.e. AR and SS, whereas r and R represent the baseline values for AR and SS. β1,
β2 and β3 values will be obtained through the least squares optimization.
The Herbert’s kinetics model embedded with the linear kinetic model of (1) will then be
combined with a macro-scale bioreactor model to produce the so-called kinetics hybrid model
as depicted in Fig. 1. Clearly, in this approach we integrate the mixing by including both AR
and SS in the model development.
Predicted Yield
Macro-Scale Bioreactor Model

Predicted Productivity
AR
Kinetic Model (Herbert’s Concept of
Endogenous Metabolism & Linear
Regression Model)
SS

Figure 1: Schematic Diagram of Kinetics Hybrid Model

Herbert’s kinetics model


The kinetics parameter data was estimated using the experimental data of biomass, product
and substrate concentrations for different AR and SS. For this purpose, we apply the
Herbert’s concept as follows: It was assumed that the observed rate of biomass formation
comprised of the growth rate and the rate of endogenous metabolism:

rx = ( rx ) growth + ( rx ) end (2)

where
k1 XS
(rx ) growth = exp(− k 5 P)
k2 + S (3)

It was also assumed that the rates of substrate consumption and product formation are
proportional to the biomass growth rate:

rs = (rs ) growth = − k 3 (rx ) growth


(4)
r p = ( r p ) growth = k 4 ( rx ) growth
(5)

3
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

The rate of growth due to endogenous metabolism by a linear dependence is shown below:

(rx ) end = − k 6 X (6)

Given the intial values of the kinetic parameters obtained from the literature data, an
optimization problem can be formulated to predict the kinetic parameters for each AR and SS
conditions. The obtained kinetic parameters are then used to obtain the linear regression
model of (1).

Macro-scale bioreactor model


A macro-scale bioreactor (Eqns. 7-9) will be used to compare the prediction of yield and
productivity using the kinetics hybrid model in Fig 1 with the experiment data:

Biomass formation: dX / dt = rx (7)


Substrate consumption: dS / dt = rs (8)
Product formation: dP / dt = rp (9)
P
Yield = × 100% (10)
S0 − S
P
Pr oductivity = × 100% (11)
BT

The S0 is the initial substrate concentration (g/L) of the medium and BT is the batch time (hrs)
allocated for the fermentation process.

2.2 Kinetics Multi-Scale Model


The kinetics multi-scale model is developed using a slightly different approach from the
kinetics hybrid model. We still use the developed kinetic model (Eqns. 1-6), but the macro-
scale bioreactor model (Eqns. 7-9) is replaced with a general balance over an element of
reactor volume model so that it allows us to include a mixing model with it. The resulted
model is illustrated in Fig. 3.

Mixing Model (k-ε


Predicted Predicted turbulence model &
Yield Productivity General Balance
Over An Element of
Multi-scale Bioreactor Reactor Volume)
Model
Macro-Scale
Bioreactor Model

Kinetic Model

SS AR

Figure 2: Schematic Diagram of Ethanol Fermentation Multi-Scale Model

4
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

Mixing model
The mixing model is developed based on k-ε turbulence model (governing turbulence)
[Dubey et al., 2006]. These models could be used to describe the mixing mechanism in the
bioreactor. The k- ε turbulence model is usually used in order to describe the mixing behavior
and to compute turbulence in the bioreactor. Standard k-ε turbulence model is used since it is
proven to be most successful in past works [Dubey et al., 2006]. The following is the
standard k-ε turbulence model (Eqns. 12-16):
The energy dissipation can be expressed as:

ε = (∆pFu ) / m = (∆pu ) /( xρ ) (12)

where ∆p denotes the pressure drop, m the mass, F the tube cross-section and x the axial
coordinate.

The fluid flow equations to be solved for a constant density fluid are [Bode, 1994]:

div ( ρu ) = 0 (Continuity Equation) (13)


µ eff
div ( ρuk ) = div ( grad _ k ) + G − ρε (Transport Equation) (14)
σk

µ eff ε
div ( ρuε ) = div ( grad _ ε ) + (C1G − C 2 ρε ) (Transport Equation) (15)
σε k

k2
uT = C µ ρ (Eddy Viscosity) (16)
ε

whereby G is the dissipation function τijτij/(2µeff); Cµ = 0.09; C1 = 1.44; C2 = 1.92; σk = 1.0;


σΓ = 1.3.

Reactor model

The general balance over an element of reactor volume model [Ranade, 2002] is adopted as
reactor model as follows:

∂ ( ρφ ) ∂ ( ρU iφ ) ∂ ∂φ
+ = (Γφ ) + Sφ
∂t ∂xi ∂xi ∂xi (17)

where ρ is the density of fluid, ø is the concentration of any component, Ui is the local
velocity in the xi direction, Γø is the effective diffusivity of ø and Sø is a volumetric source
term (rate of production of ø per unit volume) of ø.

The source term will be equal to the rate based on intrinsic kinetics, i.e. there are no
concentration or temperature gradients within the volume element under consideration. We
use the Computational Fluid Dynamics (CFD) software to solve these models (Eqn 12-17)
from which the yield and productivity can be predicted using (12-17) and (10-11).

5
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

3. METHODS AND MATERIALS

3.1 Microorganism and Inoculum


The bioreactor used is BIOSTAT A plus 2L, MO-Assembly. Industrial Baker’s yeast is
utilized as the inoculum culture with glucose as the substrate. Inoculum was grown in a
250mL conical flask and was allowed to stand under room temperature for 8 hours.
3.2 Growth and Medium
1.5L of fermentation medium is prepared by adding 75g glucose, 7.5g yeast, 3.75g NH4Cl,
4.37g Na2HPO4, 4.5g KH2PO4, 0.38g MgSO4, 0.12g CaCl2, 6.45g citric acid and 4.5g sodium
citrate. The medium culture is sterilized at 121oC for 20 minutes and then cooled down to
room temperature. 40mL of inoculum is added to the fermentation medium. Temperature and
pH conditions are maintained and controlled at 30°C and pH 5 respectively. The batch
process is stopped after approximately 72 hours and the samples ae taken in every 2-3 hours
for glucose and ethanol concentration analysis. Experiments are repeated by various
conditions of AR and SS, within the range of 1-1.5LPM AR and 150-250rpm SS.
3.3 Glucose and Ethanol Concentration
Samples were clarified by filtration and analyzed using R-Biopharm test kits and UV
spectrophotometer. All samples were tested under room temperature for standardization.

4. RESULTS AND DISCUSSION

4.1 Kinetics Hybrid Model


Table 1 summarizes the experimental data for different conditions of AR and SS. Based on
the Herbert’s model and the experimental data of X, S and P, we obtained a set of kinetics
parameter data. Linear regression analysis was then performed for the experimental data of
AR, SS, and a set of kinetics parameters, k1, k2, …,k6, from which the values of β1 , β2 and β3
in (1) were obtained.

Table 1: Summary of Experimental Data At Different AR and SS Conditions

Experiment AR (LPM) SS (rpm) X (g/L) S (g/L) P (g/L)


1 1.0 150 30.0 4.67 6.33
2 1.5 150 37.4 4.32 6.80
3 1.0 250 39.9 2.59 6.39
4 1.5 250 34.3 1.99 9.91
5 1.25 200 37.0 4.75 9.10

The linear regression results are given as:

kˆ1 = 1.4085 − 0.2852 X 1 + 0.3692 X 2 (18)


kˆ2 = 0.0010 (19)
kˆ = 0.6631 − 0.0148 X + 0.0220 X
3 1 2 (20)

6
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

kˆ4 = 0.1040 + 0.0142 X 1 + 0.0128 X 2 (21)


kˆ5 = 0.7558 − 0.1019 X 1 − 0.0211X 2 (22)
kˆ = 0.0143 − 0.0001X − 0.0019 X
6 1 2 (23)

( AR − 1.25) ( SS − 200)
where X 1 = and X 2 = .
0.25 50

By using the kinetics model of (18-23), the Herbert’s model of (2-6) as well as the macro-
scale bioreactor model of (7-9), we predict the substrate and product concentrations. These
predictions are then validated against the experimental daat for another set of AR and SS
conditions, i.e. 1.2LPM and 175rpm respectively. As seen in Figs. 4 and 5, it is shown that
the kinetics hybrid model reasonably fits the experiment data.

50
Model
40 Experiment

30
S (g/L)

20

10

0
0 10 20 30 40 50
Batch Age (hr)

Figure 3: Model Fitting for Substrate Concentration


8

6
P (g/L)

2 Model
Experiment
0
0 10 20 30 40 50
Batch Age (hr)

Figure 4: Model Fitting for Product Concentration

Table 2 summarizes the experimental and predicted yield and productivity using the kinetics
hybrid model. As shown in Table 2, the kinetics hybrid model does not provide good
predictions of yield and productivity for all range of AR and SS values. Thus, this result
shows the needs to include the mixing phenomena.

Finally, we use this model to find the optimum AR and SS. Response Surface Method (RSM)
was employed by maximizing the yield and productivity. The optimum values of AR and SS
were obtained as 1.43LPM and 250rpm with the maximum yield and productivity of
21.15% and 0.146g/L.hr respectively.

7
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

Table 2: Summary of Experimental and Predicted Results

Experimental Predicted
AR SS Experimental Predicted
Experiment Productivity Productivity
(LPM) (rpm) Yield (%) Yield (%)
(g/L.hr) (g/L.hr)
1 1.0 150 14.788 12.006 0.099 0.083
2 1.5 150 16.392 13.011 0.106 0.086
3 1.0 250 15.105 13.270 0.102 0.092
4 1.5 250 24.040 21.099 0.160 0.144
5 1.25 200 21.500 17.214 0.180 0.121

4.2 Kinetics Multi-Scale Model


Table 3 shows the summary of experimental data and predicted yield as well as productivity
using the kinetics multi-scale model for different AR and SS conditions.

Table 3: Summary of Experimental and Simulated Results

Experimental Simulated
AR SS Experimental Simulated
Experiment Productivity Productivity
(LPM) (rpm) Yield (%) Yield (%)
(g/L.hr) (g/L.hr)
1 1.0 150 14.788 18.600 0.099 0.139
2 1.5 150 16.392 17.000 0.106 0.110
3 1.0 250 15.105 15.900 0.102 0.113
4 1.5 250 24.040 23.100 0.160 0.153
5 1.25 200 21.500 21.700 0.180 0.200

The results in Table 3 show that for high SS values (250 rpm), the kinetics multi-scale model
predicts the yield and productivity much better than the kinetics hybrid model. This implies
that the mixing phenomena as modelled in (12-17) has more significant effect, especially
when SS is high. But this multi-scale model is not good for lower SS. These results also
indicate that the turbulence phenomena indeed occur at high steering speed.
Like the kinetics hybrid model, we use the multi-scale model to find the values of AR and SS
that maximise the yield and productivity. The optimum values of AR and SS were obtained
as: 1.32LPM and 250rpm with the maximum yield and productivity of 23.34% and
0.207g/L.hr respectively.
By examining the results in Table 3, it can be seen that the yield and productivity predictions
using the kinetics multi-scale model are closer to the experiment data at the range of 1.0-
1.5LPM and 200-250 rpm (experiments 3 to 5). Since the optimum values of AR and SS lies
within this range, the kinetics multi-scale model could predict the optimum yield and
productivity better than the kinetics hybrid model. This demonstrates that mixing phenomena
indeed should be included in bioreactor modeling. Thus, it is suggested that multi-scale
model approach is good for predicting the yield and productivity of fermentation process in
this study.

8
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

5. CONCLUSIONS

In this paper, two modeling approaches have been proposed, i.e. kinetics hybrid and kinetics
multi-scale models, and also validated against experimental data. In both proposed models,
the aeration rate (AR) and stirrer speed (SS) were considered as a way of integrating the
mixing process in the bioreactor. Furthermore, in the kinetics multiscale model, a more
detailed mixing phenomena was embedded for bioreactor model. The optimum AR and SS
was determined using Response Surface Methodology (RSM) by maximising the yield and
productivity. Our study suggested that the kinetics multi-scale model gave a better prediction
of both the maximum yield and productivity for high steering speed.

ACKNOWLEDGEMENTS

The author would like to express sincere appreciation and gratitude to all the lecturers who
have provided tremendous assistance throughout the research. This paper is part of the work
supported by MOSTI (grant no. 02-02-07-SF0001).

REFERENCES

1. Bezzo F., Macchietto S. and Pantelides C. C. General Hybrid Multizonal/CFD Approach


for Bioreactor Modeling, AIChE Journal, 49, 2133-2148, 2003.
2. Dubey H., Das S. K. and Panda T. Numerical Simulation of a Fully Baffled Biological
Reactor: The Differential Circumferential Averaging Mixing Plane Appraoch,
Biotechnology and Bioengineering, 95, 754-766, 2006.
3. E.W.T. Liew, J. Nandong and Y. Samyudia, Experimental Investigation on the Impact of
Aeration Rate and Stirrer Speed on Micro-Aerobic Batch Fermentation, Journal of
Applied Sciences, 9 (17):3126-3130, 2009.
4. Harvey A. D. and Rogers S. E. Steady and Unsteady Computation of Impeller-Stirred
Reactors, AIChE Journal, 42, 2701-2712, 1996.
5. Hjersted J. L. and Henson M. A. Optimization of Fed-Batch Saccharomyces cerevisiae
Fermentation Using Dynamic Flux Balance Models, Biotechnol. Prog., 22, 1239-1248,
2006.
6. Jiang Y., Khadilkar M. R., Al-Dahhan M. H. and Dudukovic M. P. CFD of Multiphase
Flow in Packed-Bed Reactors: II. Results and Applications, AIChE Journal, 48, 716-730,
2002.
7. Konde K. S. and Modak J. M. Optimization of Bioreactor Using Metabolic Control
Analysis Approach, Biotechnol. Prog., 23, 370-380, 2007.
8. Ranade V. V. An efficient computational model for simulating flow in stirred vessels: a
case of Rushton turbine, Chemical Engineering Science, 52, 4473-4484, 1997.
9. Ranade V. V. Computational Flow Modeling for Chemical Reactor Engineering, Process
Systems Engineering, UK, 2002.
10. Starzak M., Krzystek L., Nowicki L. and Michalski H. Macroapproach kinetics of ethanol
fermentation by Saccharomyces cerevisiae: experimental studies and mathematical
modelling, The Chemical Engineering Journal, 54, 221-240, 1994.
11. Venneker B. C. H., Derksen J. J. and Van den Akker H. E. A. Population Balance
Modeling of Aerated Stirred Vessels Based on CFD, AIChE Journal, 48, 673-685, 2002.
12. Wiechert W. Modeling and simulation: tools for metabolic engineering, Journal of
Biiotechnology, 94, 37-63, 2002.

9
The 5th International Symposium on PSE ASIA 2010
Design, Operation and Control of Chemical Processes July 25-28, 2010, Singapore

Brief Biography of the Presenter

Professor Samyudia joined Curtin University of Technology, Sarawak Campus in 2005 after
spending twelve years as academics and researcher at McMaster University (Canada), Delft
University of Technology (The Netherlands), Purdue University (USA), Murdoch University
(Australia) and Institut Teknologi Bandung. He received his Ir. (Ingenieur) degree from
Institut Teknologi Bandung in 1989 and PhD in Chemical Engineering from The University
of Queensland, Australia in 1996. He has extensively been involved in industrial research in
the area of process systems engineering (e.g. control, modeling and design) by securing a
number of competitive research grants and industrial funding i.e. AKZO NOBEL, Shell,
DOFASCO. He has supervised thirteen post-doctoral and postgraduate (PhD and Master)
students and twenty honors students. He is currently supervising six postgraduate students.
He has published more than 90 publications in reputable book series, international journal
and conference proceedings.

10

View publication stats

You might also like