Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Article

pubs.acs.org/IECR

Analysis of Hopf Points for a Zymomonas mobilis Continuous


Fermentation Process Producing Ethanol
Hangzhou Wang, Nan Zhang, Tong Qiu, Jinsong Zhao, Xiaorong He, and Bingzhen Chen*
Department of Chemical Engineering, Tsinghua University, 100084 Beijing, China

ABSTRACT: Zymomonas mobilis is one of the most promising industrial microorganisms for industrial ethanol production.
However, biomass, product and substrate oscillation behaviors have been observed under certain operating conditions in both
experiments and numerical simulations of continuous fermentations involving this organism. This article analyzed oscillation
behavior during fermentation based on Hopf bifurcation analysis, in which the Hopf singularity point and limit cycles were
introduced. All Hopf points in the operating parameter values panel were simultaneously located and identified for given sets of
operating conditions. Calculation of the distribution of Hopf points contributed to enhancing the stability of the fermentation
process, maintaining high product quality, and providing insights into system dynamics.

■ INTRODUCTION
Bioethanol is an important renewable and sustainable
minimizing ethanol inhibition.16 In addition, experimental and
theoretical studies also illustrated that variables such as dilution
alternative fuel source,1 both directly in the form of fuel rate and feed substrate concentration have an impact on static
ethanol and as a blend with gasoline. Agricultural feed reactants and dynamic behavior in continuous stirred tank fermen-
that are suitable for fermentation processes producing tors.17,18
bioethanol contain sugars, starches, and cellulose materials. The existence of Hopf bifurcations were reported to cause
The yeast Saccharomyces cerevisiae2−5 and the bacterium oscillatory behavior in fermentation processes with Z. mobilis.2
Zymomonas mobilis (Z. mobilis)3−6 are currently the most Optimal operating points and regions, calculated from
important microorganisms for industrial production of ethanol. numerical analysis of oscillating biosystems, can be used to
Z. mobilis produces less biomass than Saccharomyces cerevisiae, avoid the oscillatory behavior that unfavorably decreases the
while more carbon is funneled to the ethanol fermentation. It fermentation yield and declines product quality. Follow-up
was reported that the ethanol yield of Z. mobilis could be as design and control strategies can be developed to maintain
high as 97% of the theoretical yield of ethanol to glucose,7 stability and guarantee high productivity and high product
whereas only 90−93% can be achieved by Saccharomyces quality.6,18 With properly selected operating conditions, these
cerevisiae.6 Thus, Z. mobilis has been promoted as a more Hopf points can be eliminated to avoid oscillations, directly
promising microorganism than yeast for the industrial leading to a more stable and productive fermentation process.
production of ethanol.8 The advantages of numerical analysis are listed as follows:
Continuous fermentation using Z. mobilis is prone to 1. As the fermentation process is slow, its dynamic
dynamic instability,2,3,6,9 in which sustained oscillations of simulation through experiment may be inefficient and
biomass, substrate, and product concentrations are routinely unable to locate points corresponding to certain dynamic
observed, especially at low growth rate and high ethanol behaviors, such as bifurcation and oscillation.
concentrations.10 Oscillations are mainly caused by product 2. Because only a limited number of experiments can be
inhibition of enzymatic reactions, and experimental studies have performed, some dynamic behavior may be missed or
revealed that high temperatures may enhance such inhibitory neglected.
effects.11 Meanwhile, previous work has also showed that
oscillatory states are achieved at low dilution rates.10 These Existing experimental work has shown the existence of
oscillations increase the average residual glucose, which oscillatory behavior; researchers analyzed these dynamic
correspondingly decreases the ethanol yield and declines systems using single operation parameter analysis with all
product quality.6 A considerable amount of work2,3,6,7,10−15 other state variable changes monitored,1,2,14,16,18,19 detecting
has demonstrated oscillatory behavior in fermentation frequent bifurcation, chaotic behavior, and even force
processes both experimentally and numerically. With regard oscillations.1,14
to bioethanol fermentation systems utilizing Z. mobilis, several However, all previous work has suffered limitations because
experimental and theoretical studies focused on dynamic of the lack of ability to deal with two or three simultaneously
stability, which affects the industrial productivity of fuel changed operation parameters. In this article, the effect of
ethanol.3 operating condition values on product yield was analyzed, with
In alcoholic fermentation, it is assumed that ethanol inhibits
the maximal specific growth rate indirectly by inhibiting the Received: May 17, 2012
synthesis of the main intermediate, which in turn determines Revised: December 12, 2012
the overall growth rate.11 Therefore, the glucose conversion, Accepted: December 18, 2012
the ethanol yield, and the production rate will be favored by Published: December 18, 2012

© 2012 American Chemical Society 1645 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

rigorous Hopf points location in the operating region. the oscillatory phenomenon is one of the most significant. In
Moreover, the synergism between three multivariate operating the oscillatory situation, state variables experience periodic
parameters was discussed. changes. Theoretically, oscillation corresponds to Hopf points
As stated, numerical analysis methods can enable complete pertaining to singular points. A supercritical bifurcation occurs,
and thorough analysis of dynamic systems. In this article, we with orbital limit cycles, when the reaction system is situated
first introduce the numerical methodology for locating Hopf near this type of point. A pair of conjugate eigenvalues in
point in steady state solutions as well as the limit cycles and mirror symmetry, with at least one nonzero real part in the
then apply the proposed methodology to the Z. mobilis other n−2 eigenvalues, can indicate this phenomenon. Periodic
fermentation process. systematic oscillations will be detected near these Hopf

■ INTRODUCTION TO HOPF POINTS AND


NUMERICAL ANALYSIS METHODS
singularity points, which clarifies the significance of locating
Hopf points. Currently, most research on Hopf points
emphasizes accurate search for and location of Hopf points
In general, chemical processes are assumed to be described by and limit cycles.
the following equation:
⎧ dx
⎪ = F (x , α )
■ HOPF POINT DETECTION AND LOCATION
The unique feature of a Hopf point lies in its special pair of
⎨ dt conjugate pure image eigenvalues associated with the Jacobian

⎩ x(0) = x 0 (1) matrix at the steady state point. By the Hopf bifurcation
theorem, Hopf point detection can be achieved by calculating
where x represents state variables in the dynamic system such eigenvalues. A more common way is to design a detection
as temperatures, pressures, components concentration, etc. (x function to identify Hopf points and the corresponding unique
∈ Rn) and α is a parameter representing the operating stable steady state. Consider the function described in eq 3
condition value or design variable, such as flow rate, initial
feedstock concentration, reactor volume, and so on. Equation 1 ψH(x , α) = ∏ (λi(x, α) + λj(x, α))
describes the dynamic behavior of a set of state variables under i>j (3)
specified operating conditions, providing access to better
understanding of system behaviors and optimization. The where λi are the eigenvalues of the Jacobian matrix at a specified
solution satisfying eq 2 is the equilibrium solution representing steady state point. According to theoretical analysis, this
a set of practical state variable values under steady operating function value will vanish when encountering a Hopf
conditions. bifurcation point, associated with a pair of conjugate pure
image eigenvalues, λ1,2 = ± iω0. Note that paired real
F (x , α ) = 0 (2) eigenvalues λ1 = κ, λ2 = −κ can be a special case of ψH = 0,
As the main operating points, a steady state solution set which corresponds to a neutral saddle situation, rather than the
inevitably contains singular points that are worth looking into Hopf point we are primarily concerned with. In that sense, the
to enable smooth operation. A thorough study of singularity Hopf point must be carefully identified from neutral saddle
helps to screen and evade these distinct points, at which rich points when the zero point of detection function is found. At a
dynamic behavior exists and stability is practically difficult to generic Hopf bifurcation point, the function ψH is expected to
maintain. Meanwhile, the accurate location of singular points, be real, smooth, and have a regular zero value, with a
including the clear distribution of Hopf points, will no doubt nonsingular Jacobian of the proposed dynamic system assured
contribute to safer decisions when a change in operating simultaneously at this specific point. Newton’s method can be
conditions must be made, because a number of specific singular applied, and to avoid tedious explicit computation of all the
points may possibly appear when switching across the operating eigenvalues of A, the current work used a method based on the
region. Here, the matrix Fx(x,α), which is the Jacobian matrix of bialternate product to compute ψH.
F(x,α), is introduced to assess stabilities of steady state The procedure of Hopf point detection based on the
operation points. If x = x* is the solution of F(x,α) = 0, then bialternate product is expressed in eq 4:
under the operating condition governed by α, the system will
ψH(x , α) = det(2Fx(x , α) ⊙ In) (4)
maintain in the steady state when all state variables take a full
set values of x. In general, steady state points have different Here, ⊙ represents the bialternate product of the matrix and
stability characteristics and the stability of a system under formulated as
certain operating conditions and state variable values is
determined by the eigenvalues of the Jacobian matrix. apr aps δpr δps
Specifically, the system is stable if all eigenvalues are located (2A ⊙ In)(p , q),(r , s) = +
on the left of the complex panel, otherwise the system is δqr δqs aqr aqs (5)
unstable. In other words, when every real part of all the
Jacobian matrix eigenvalues is negative, re(λi) < 0, i = 1,2, ...,n, where
then the system is stable when a minor disturbance is ⎧1, i = j

encountered. On the contrary, if there is only one eigenvalue δij = ⎨
⎩ 0, i ≠ j

that consists of a positive real part, the whole system would fall
into an unstable situation under which even a small disturbance
would cause extensive system changes. Stable steady state In the matrix, rows are labeled by the multi-index (p, q) (p = 2,
operating conditions are desirable in both operation and design, 3, ..., n; q = 1, 2, ..., p − 1), and columns are labeled by the
but some specific unstable steady situations also have multi-index (r, s) (r = 2, 3, ..., n; s = 1, 2, ..., r − 1). The result of
potentially practical and theoretical applications, among which the determinants is enumerated in eq 6:
1646 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

Figure 1. Schematic of Hopf analysis procedure.

⎧−aps ,

⎪ apr ,
r=q
r ≠ q, s = q
■ LOCATION OF LIMIT CYCLES
Bifurcation theory suggests that limit cycles are intrinsically of
⎪ concomitance with Hopf singularity points. The shooting
⎪ apr + aqs , r = p, s = q method, a widely accepted algorithm for calculating limit cycles,
(2A ⊙ In)(p , q),(r , s) =⎨ first converts the proposed question to a periodic boundary-
⎪ aqs , r = p, s = q
⎪ value problem for the cycle continuation in eq 7 as follows:
⎪−aqr , p=s
⎧ dx
⎪ T
⎩ 0, otherwise ⎪ = F (x , α )
(6) ⎨ dt 2π

Theoretically, the detection function equals zero when the ⎩ x(0) = η (7)
Hopf bifurcation occurs, but in practice, we first identify the
After parameter α is preset, we search for an initial value η
sign of the test function as positive or negative. If ψH(xi,αi)- and a period T satisfying eq 7 to obtain the periodic solution
ψH(xi+1,αi+1) < 0, this reveals a Hopf singularity point standing for the problem.
between (xi,αi) and (xi+1,αi+1); thus, the rigorous Hopf point Fortunately, software packages such as AUTO, 20−22
(xH,αH) location is subsequently found through a gradually CONTENT,23 MATCONT,24,25 DsTool,26 PyDSTool,27
reduced search space. XPPAUT,28 and some previous research by the authors’
1647 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

Figure 2. Fermentation process.

group29 have implemented the method described and laid the theses. In this work, to decrease inhibition by ethanol, a
foundations for the current work. removal membrane was positioned at the bottom of the
The strategy for locating Hopf points in a dynamic system fermentor to remove ethanol produced during fermentation.
was summarized in Figure 1. Beginning with a model of the Continuous Ethanol Removal. A coupled membrane
fermentation process, this procedure was successively con- separation system for ethanol removal was proposed to prevent
ducted by calculating the steady state solution of the dynamic ethanol inhibition and increase the productivity and efficiency
system as parameters were changed, then identifying the steady of the fermentation process. The membrane has selectivity
state solutions to determine whether or not there are Hopf toward ethanol within the reaction environment, with a gas/
points. If there is a Hopf point and the first Lyapunov liquid sweep stream on the permeate side to remove product
coefficient is negative, orbital stable limit cycles in the system from the membrane surface.
and corresponding oscillatory behavior are sure to exist in the With the removal of ethanol, inhibition would be prevented
dynamic system. Finally, parameters were varied to find all and cell activity maintained at a high level, making it possible to
Hopf singularity points by this ergodic process, providing a use a higher glucose feed concentration, resulting in increased
fermentor cell density. The increased cell activity and cell
valuable reference for operating point optimization.
density contribute to higher volumetric productivity and a
Fermentation Process Flowchart. There are several
smaller fermentor volume. Additionally, the use of a
reported models10,11,13−15 describe the Z. mobilis fermentation
concentrated glucose feed will decrease the amount of process
process. Figure 2 shows a diagram of the continuous flow water.19
fermentor, while the data used in the numerical simulation is Model Assumptions. The model adopted in this paper was
listed in Table 1. The kinetics developed by Jobses et al. for based on the following simplified assumptions:30
production of bioethanol by Z. mobilis was adopted.1
Product Inhibition. Ethanol production by Z. mobilis is 1. The fermentor is considered to be a well-mixed system
subjected to end-product inhibition. The product alters the cell and is not controlled by diffusion.
2. The feed contains only nutrient (substrate).
membrane composition and inhibits enzymatic reactions such
3. The specific growth rate is defined by a Monod-type
as carrier-mediated transport processes and metabolic syn-
equation.
4. The maximum specific growth rate is proportional to the
Table 1. Parameters in the Fermentation Process Model concentration of the key component (e).
5. k1, k2, k3, Ks, ms, mp, Ysx, and Ypx are constants.
param. value unit
Model Development and Discussion. One of the most
Cs0 140 kg/m3
generally acknowledged models of fermentation processes is
DMin 0.5 h−1
AM 0.24 m2
the maintenance model, for which substrate consumption
Din 0.04 h−1
equation is as follows:
Ks 0.5 kg/m3 ⎛ μ ⎞
k1 16 h−1 rs = ⎜ + ms⎟Cx
k2 0.497 m3/kg·h
⎝ Ysx ⎠
k3 0.00383 m6/kg2·h where μ is the specific growth rate. The first term represents
ms 2.16 kg/kg·h the growth rate and the second term represents the
mp 1.1 kg/kg·h maintenance. Similarly, the product generation rate equation
Ysx 0.0244498 kg/kg is defined as
Ypx 0.0526315 kg/kg
Cp0 0 kg/m3 ⎛ μ ⎞
Ce0 0 kg/m3 rp = ⎜⎜ + mp⎟⎟Cx
Cx0 0 kg/m3 ⎝ Ypx ⎠
CpM0 0 kg/m3 The growth rate of biomass rx has its classical definition and
ρ 789 kg/m3 can be expressed in the form:
VM 0.0003 m3
VF 0.003 m3 rx = μCx
Pm 0.1283 m/h The rate of the key compound re is formulated as
1648 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

re = f (Cs)f (Cp)Ce where the subscripts Din, Dout, DMin, and DMout represent their
respective dilution rates, and Ysx and Ypx refer to the yield
The formation rate of the key component is a function of the factors of biomass on substrate and product; k1, k2, k3 are
substrate concentration, inhibited by ethanol, while its activity empirical constants; V stands for volume; mS and mP are
is expressed in terms of concentration only.11 Subscript s maintenance factors based on substrate requirements and
represents the substrate, while subscript p refers to the product. product formation. The membrane side and the fermentation
f(Cs) is given by the Monod equation: side are distinguished by subscripts M and F.

f (Cs) =
Cs
K s + Cs
■ HOPF POINTS IN THE SYSTEM AND THE LIMIT
CYCLES
The function f(Cp) is defined in polynomial form: The set of steady state points in the investigated process of
fermentation is shown in Figure 3. In the case of Cs0 = 150.3
f (Cp) = k1 − k 2Cp + k 3Cp2

Thus, the equation for re can be substituted into the


following expression:
k1 − k 2Cp + k 3Cp2
re = CsCe
K s + Cs
The base set of parameters in the current investigation is
listed in Table 1.19
The dynamic model for the continuous fermentation process
for substrate (s), microorganism (x), and the key compound
(e) are given by the following set of ordinary differential
equations:
dCx
= μCx + DinCx0 − Dout Cx
dt
dCs ⎛ μ ⎞
= −⎜ + ms⎟Cx + DinCs0 − Dout Cs
dt ⎝ Ysx ⎠
k1 − k 2Cp + k 3Cp2 Figure 3. Changes in product concentration with dilution rate.
dCe
= CsCe + DinCe0 − Dout Ce
dt K s + Cs
kg/m3, a Hopf point was detected at Din = 0.052 h−1, which
The product (ethanol) on the fermentor side (p), the coincides exactly with reported literature values.2,16 At the Hopf
product (ethanol) on the membrane permeate side (pM), and point, we have a product concentration Cp = 58.050 kg/m3,
the mass balance for the product (ethanol) over the fermentor which approaches the maximum point of product yield of Din =
are given by 0.046 h−1, Cp = 58.103 kg/m3. This fact highlights the
⎛ μ ⎞ importance of Hopf point investigation in the present dynamic
dCp
= ⎜⎜ + mp⎟⎟Cx + DinCp0 − Dout Cp system, because it is so close to the maximum conversion point
dt ⎝ Ypx ⎠ that we seek. In Figure 3 the Hopf singularity point is marked
as a red star, with the first Lyapunov coefficient l1 value of
A P
− M m (Cp − CpM) −0.00153. The negative sign means the point corresponds to a
VF supercritical Hopf bifurcation and the limit cycle is orbitally
stable. In Figure 3, the abscissa denotes the feedstock dilution
The mass balance for the product (ethanol) over the rate, ranging from zero to a reasonable maximum. As the
membrane is given by variable indicating the degree of feedstock volume diffusing into
dCpM AM Pm the reactor volume, the dilution rate has a relatively small value
= (Cp − CpM) + DMinCpM0 − DMout CpM in practice. Therefore, only a range of the dilution rate worthy
dt VF of consideration is discussed. As mentioned, the Hopf points
where the AM is the permeation area and Pm is the permeability are theoretically very close to a risky but high conversion point,
of the membrane for ethanol. and associated limit cycles are depicted in Figure 4. Each limit
The exit fermentor dilution rate can be obtained as follows: cycle is generated by a very small disturbance from the initial
operating parameter value at the Hopf point.
AM Pm According to bifurcation theory, experimentally observed or
Dout = Din − (Cp − CpM)
VFρ mathematically simulated oscillations in a dynamic system must
start and end at certain critical points in the system. It has been
The corresponding exit membrane dilution rate becomes proven that static and dynamic bifurcation behaviors, over a
AM Pm wide range of operating parameters, abound in the Z. mobilis
DMout = DMin + (Cp − CpM) fermentation process. In the current work, the fermentor
VMρ dilution rate, the membrane dilution rate, the initial substrate
1649 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

potential oscillatory operation points can also be measured to


better estimate the operating risk as a probability.
Case 1: Cs0 and Din. In this case, the dilution rate of the
feedstock and the initial substrate concentration were examined
because of their decisive roles as operating variables and their
ease of manipulation. The steady state solutions were
calculated; the Hopf points were marked in Figure 5, with

Figure 4. Homology limit cycles adjoining the Hopf point.

(glucose) concentration of feedstock concentration, and


permeation area were chosen as bifurcation parameters, not
only for their decisive role as operating variables in this system
but also because of their ease of manipulation and control
during practical fermentations. To avoid the oscillatory
behaviors caused by the existence of the Hopf bifurcation,
which as stated decrease the fermentation yield, system
stability, and controllability, decline product quality, we put
forward a rational method to position the entire set of Hopf Figure 5. Steady state points and Hopf points.
points in the operation parameter panel. In this proposed
method, all Hopf singularity points were located and assembled the first Lyapunov coefficient labeled to imply that these Hopf
in the operation parameter panel as a “risky curve”. Thus, points are supercritical and will stimulate nearby periodic limit
operating parameters can be located away from the undesirable cycles. To observe the membrane-side dilution influence on the
singular points, unwitting steps across these points can be system by direct-viewing, results were combined within one
avoided, and appropriate optimized values of the parameters bifurcation map, as shown in Figure 5, under variable
adopted.


membrane side dilution rates.
In Figure 5, it can be seen that the membrane side dilution
PARAMETER INFLUENCES ON HOPF POINTS rate DMin changed from 0.1 to 0.9 h−1 at a step size of 0.2 h−1.
Hopf points in Z. mobilis fermentation result in orbital stable Under each given membrane side dilution rate, production
limit cycles and oscillation behaviors in the process, and it concentration is plotted against the feedstock dilution in the
follows from this, as mentioned, that the product conversion steady state solution. Note that the clusters of production
rate, system stability, controllability, and product quality are all concentration curves have generally delayed peaks with equal
adversely affected. changes in membrane side dilution. As may be expected, the
Here, four important operating parameters, including corresponding Hopf points were detected within the region of
dilution rate of feedstock, membrane side feedstock dilution low feedstock flow rate, where the negative first Lyapunov
rate, the initial substrate concentration, and the permeation coefficients indicate potential oscillation near the production
area, were investigated. Each case study was undertaken in the concentration peak. Another fact worth noticing is that the
following way. First, the changes in product concentration were product conversion rate peaks in the system decreased with
observed and calculated under various operating conditions and increasing membrane side dilution rate, which can be explained
the results shown in a single figure to confirm the existence of by the conversion inhibition effect brought by increasing
the Hopf points along with the steady state operation points in membrane side dilution rate. Fortunately, with the decrease in
the system. These two selected parameters were then changed concentration of product, the Hopf point becomes increasingly
over a range wide enough to find all Hopf points in the separated from the conversion peak, characterized by an
operating region formed orthogonally by these two parameters. increasing first Lyapunov coefficient at the same time.
As can be seen, every point in the figure was obtained under Details of this case study and its five Hopf points are listed in
fixed corresponding operating parameters. Among all steady Table 2. The same conclusion can be drawn after analyzing
state solutions of the system, a Hopf point intrinsically exists homologous state and operating variables.
with joint orbital stable limit cycles in practice, arising from the Figure 5 only describes the influence of a single parameter
oscillation. Furthermore, using the methodology proposed in (or two parameters to a limited extent) on the Hopf points.
this paper, not only can these undesirable points be avoided but However, it is more common for operating parameters to work
the distances between current operating conditions and together and change simultaneously. Thus, we examined
1650 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

Table 2. Hopf Points in the System Cs0 = 140 kg/m3, AM = results in Figure 7 indicate that the membrane permeation area
0.24 m2 has only a slight influence on product concentration as a
DMin Din Cp CpM first Lyapunov coefficient
function of feedstock dilution rate.
(h−1) (h−1) (kg/m3) (kg/m3) l1(10−4)
0.1 0.096 57.228 57.168 −8.968
0.3 0.395 53.154 52.988 −2.583
0.5 0.627 50.790 50.527 −1.825
0.7 0.832 48.995 48.641 −1.344
0.9 1.027 47.469 47.030 −1.073

another important parameter, the initial substrate concen-


tration. Specifically, while we fixed AM = 0.24 m2 and varied
DMin from 0.1 to 0.9 with a step-size of 0.2 h−1, we searched Din
and Cs0 to find the existence of Hopf points and then recorded
these latter two parameter values to plot the data in one figure.
Accordingly, all parameter values creating Hopf singularity
points over the range of fixed DMin and AM values are shown in
Figure 6.

Figure 7. Steady state points and Hopf points.

As can be seen in Figure 7, although the permeation areas


differed markedly, their impact on product concentration as a
function of feedstock dilution was small and the steady state
solution curves almost coincided. In a similar manner, the
calculated Hopf points were located very close to one another
and were almost constant. Thus, it can be concluded that the
permeation area was not a sensitive parameter for this system.
From Table 3, it can be seen that AM contributes little to Cp
and CpM, as shown in Figure 7.

Table 3. Hopf Points in the System with Cs0 = 140 kg/m3,


DMin = 0.5 h−1
AM Din Cp CpM first Lyapunov coefficient
Figure 6. All Hopf points in the parameter panel. (m2) (h−1) (kg/m3) (kg/m3) l1(10−4)
0.1 0.650 50.578 49.954 −1.564
0.3 0.624 50.817 50.607 −1.770
In Figure 6, the Hopf points can be seen to change in
0.5 0.620 50.861 50.734 −1.825
response to the variations in feedstock dilution rate and initial
0.7 0.618 50.879 50.788 −1.847
substrate concentration. When the feedstock dilution rate or
0.9 0.617 50.889 50.818 −1.827
the initial substrate concentration was low, the system displayed
a higher probability of generating Hopf points. Notably, even if
one parameter was altered only over a certain narrow range,
while the other was varied more aggressively, Hopf points were Furthermore, at a fixed value of permeation area, Hopf points
located, which reveals that there is no universal safe operating were fixed within the operating panel with respect to initial
region for every parameter. The position of the Hopf points, as substrate concentration and feedstock dilution rate, as shown in
expected, changed with dilution rate on the membrane side. It both Figures 7 and 8. Thus, in the analysis that follows, the
is demonstrated in Figure 6 that a greater membrane side permeation area was not considered because it is not a sensitive
dilution rate causes a heavier density of Hopf points within the variable in this system. In contrast, the dilution rate on both the
operating range as the feedstock dilution rate increases. feedstock and the membrane sides, together with the initial
Meanwhile, locations across the Hopf points are shown in substrate concentration in the feed were discussed in more
this figure by a dashed line, which marks the Hopf points under detail.
a constant substrate concentration of 140 kg/m3. Case 2: DMin and Cs0. In case 2, we focused on membrane
After examining D Min , the final parameter A M was side dilution and the initial substrate concentration in the
investigated because the diversity of membranes used in feedstock.
fermentation processes may cause different permeation areas Simulation was carried out under five given feedstock
of the membrane. Here, we selected AM from 0.1 to 0.9 at dilution rates and one membrane-side dilution rate. The
intervals of 0.2 m2, and fixed DMin = 0.5 h−1, to examine the product concentration changed with the initial feedstock
influence of membrane area on product concentration. The substrate concentration, as shown in Figure 9.
1651 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

Table 4. Hopf Points in the System AM = 0.24 m2, DMin = 0.5


h−1
Din Cs0 Cp CpM first Lyapunov coefficient
(h−1) (kg/m3) (kg/m3) (kg/m3) l1(10−4)
0.02 418.671 58.833 58.526 −3.048
0.05 250.692 58.201 57.897 −1.375
0.1 195.376 57.267 56.968 −1.070
0.3 155.680 54.330 54.047 −1.222
0.5 144.557 52.050 51.780 −1.574

Figure 8. All Hopf points in the parameter panel.

Figure 10. All Hopf points in the parameter panel.

From Figure 10, it can be seen that, once again, the entire
system has a turning point on each curve. The position of the
Hopf points, although always below the turning point, changes
markedly as the initial substrate concentration increases,
whereas the rates decreased above the turning points. The
dilution rate of feedstock influenced the curve shape to a large
extent, with larger dilution values giving lower initial substrate
Figure 9. Steady state points and Hopf points. concentrations at which Hopf points were encountered.
Case 3: Din and DMin. In this case, the dilution rate on both
sides of the membrane were considered. Calculation results are
depicted in Figure 11, showing a complete three-parameter
Product concentration increased gradually with the increase bifurcation diagram. When the initial substrate concentration
of initial substrate concentration of feedstock. And the turning and the feedstock dilution rate were both fixed, the product
points seem to serve as an upper limit of concentration no concentration decreased progressively with the membrane side
matter how the substrate concentration increases. This baffle dilution rate because the feedstock dilution rate had only a low
effect can be interpreted since at the begin in the fermentation value (0.02 h−1), which gave a sufficiently long resident time for
process, the Z. mobilis grow with the increase of initial the product concentration to achieve its maximum. With an
concentration of feedstock, but when the growth reaches the elevated membrane-side dilution rate, the ethanol was gradually
maximum, the produce capability has no potential to constantly removed through entrainment, resulting in a drop of the
increase. However, the dilution rate of feedstock also has an product concentration. This effect would be intensified by
impact on product concentration. It is worth noting that the further increases in the membrane-side dilution rate. Mean-
greater feedstock dilution rate is, the earlier turning points will while, when the initial substrate concentration in the feedstock
appear, with low concentration at high dilution rate. Hopf increased, the product could maintain maximum growth rate
points, shown in Figure 9, tighten therewith the maximum for a longer time, even at the same membrane-side dilution rate.
point. Corresponding values are listed in Table 4. All Hopf points are marked in Figure 11, and actual values are
Similarly, in Table 4, we find all the first Lyapunov listed in Table 5.
coefficients are negative, indicating the existence of stable Contrary to to the preceding cases, Table 5 clearly indicates
limit cycles. The location of Hopf singularity points in the that Hopf points are generated at the maximum growth rate,
parameter panel changed significantly with the initial substrate adjacent to the value of product concentration, as for the
concentration and the membrane side dilution rate, as shown in membrane-side product concentration. The Hopf points in the
Figure 10. dilution rate and membrane-side dilution rate panel are
1652 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

feedstock and membrane side, together with the initial substrate


concentration, are the most sensitive parameters contributing
to the product yield and product quality, and they significantly
influence the location of Hopf points.

■ RESULTS AND DISCUSSION


In our research work, four important operating parameters,
including dilution rate of feedstock, membrane side feedstock
dilution rate, the initial substrate concentration, and the
permeation area, were investigated, because of their decisive
roles as operating variables and their ease of manipulation. As
operating conditions changed, the steady state solution curve
and the location of Hopf point appear in very distinguishing
manners. Specifically, we found that
1. Hopf points, in the present fermentation dynamic
system, locate very close to the maximum conversion
point, and this fact highlights the importance of Hopf
point investigation.
Figure 11. Steady state points and Hopf points.
2. Increasing the membrane-side dilution rate will result in
delayed and decreased the product conversion rate peak
Table 5. Hopf Points in the System Din = 0.02 h−1, AM = 0.24 value, while the undesired Hopf point becomes
m2 increasingly separated from the desired conversion peak.
3. Hopf points have a high probability to appear in areas of
Cs0 DMin Cp CpM first Lyapunov coefficient either low feedstock dilution rate or low initial substrate
(kg/m3) (h−1) (kg/m3) (kg/m3) l1(10−3)
concentration; thus, there is no universal safe operating
150 0.049 58.701 58.671 −5.710 region for a single parameter.
200 0.133 58.731 58.648 −3.884 4. The membrane permeation area is not a sensitive
250 0.218 58.757 58.622 −2.598 parameter and only has a slight influence on product
300 0.302 58.781 58.594 −1.651 concentration. Almost coincident steady state solution
350 0.386 58.803 58.565 −0.949 curves and Hopf points were detected.
5. Baffle effect of product concentration and the maximum
illustrated in Figure 12. Hopf points exist at each given initial turning point were observed, while the position of Hopf
substrate concentration with various dilutions. points changes markedly as initial substrate concen-
tration increases.
6. The three-parameter bifurcation diagram showed that
when the initial substrate concentration is fixed and the
feedstock dilution rate is low, the product concentration
decreased progressively with the membrane side dilution
rate, but when the initial substrate concentration in the
feedstock increased, the product could maintain a
maximum growth rate for a longer time.
7. At low initial concentration, the Hopf points appear at
almost the same level of feedstock dilution rate, whereas
at higher initial concentrations, the Hopf points arise in a
region of relatively low dilution rate but high membrane
dilution.
8. Product conversion rate, system stability, controllability,
and product quality all depend heavily on these four
operating variables. The synergism between multivariate
operating parameters has been first discussed.
The emphasis of this article is on locating Hopf singularity
points in the process. As a result, with four major manipulate
Figure 12. All Hopf points in the parameter panel. parameters and their combinations, also within a large
operation condition parameters scopes, we have studied the
Hopf singularity points thoroughly in our research work. The
It can be observed from Figure 12 that at low initial results showed the universal existence of Hopf singularity
concentration, the Hopf points appear at almost the same level points in this process within a large operational space. Although
of feedstock dilution rate, whereas at higher initial concen- it seems there were many singularity points in the panel
trations, the Hopf points arise in a region of relatively low constructed by operational variable, suitable operation points
dilution rate but high membrane dilution. for designing process were even more in the panel. Because
From all of these analyses, we can now conclude that, among only the points located exactly in the critical curve were not
all operating parameters, the dilution rates on both the recommended, many more points away from the curve in the
1653 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655
Industrial & Engineering Chemistry Research Article

panel are feasible to the process design. Also, in practice, the Notes
operation condition parameter space was limited into a smaller The authors declare no competing financial interest.


operational space, which would meets complex constraints; the
operation variables and their combinations also will be reduced, ACKNOWLEDGMENTS
so there would be less Hopf singularity points, even no Hopf
The authors gratefully acknowledge support from China
singularity points would be encountered in some situations.
Postdoctoral Science Foundation (No. 20100480282), China
In general situations at process design phase, it is a better
Postdoctoral Science special Foundation (No. 2012T50099),
method to determine the operation ranges first, after that, to and the National Basic Research Programme (No.
use the method in this article to detect whether there are Hopf 2012CB720500).


singularity points or not. Most of the time, there would be
seldom Hopf singularity points within the small feasible
NOMENCLATURE
operational range that meets all the complicated practical
constraints. Even in the worst situation, there was a Hopf Ci Concentration of component i (kg/m3)
singularity curve in the operational space; these points are ri Production rate of component i (kg/m3 h)
feasible to process design except that these operation points Cs0 Inlet substrate concentration (kg/m3)
just locate in the critical curve; for insurance, these points very DMin Inlet membrane dilution rate, inlet flow rate/membrane
close to the critical curve also can be excluded in process volume (h−1)
design. DMout Output membrane dilution rate, output flow rate/
membrane volume (h−1)

■ CONCLUSION
In this article, a typical fermentation process using Z. mobilis to
Din Inlet fermentor dilution rate, inlet flow rate/for monitor
volume (h−1)
Dout Output fermentor dilution rate, output flow rate/
produce ethanol was investigated. Although it is a more fermentor volume (h−1)
promising microorganism than yeast for the industrial AM Area of membrane (m2)
production of ethanol, continuous fermentation of Z. mobilis XS Substrate conversion
undergoes oscillatory behavior in which biomass, product, and KS Monod constant (kg/m3)
substrate cycle under certain fermentation conditions. There- k1 Empirical constant (h−1)
fore, it is important to search for a reasonable approach that k2 Empirical constant (m3/kg h)
helps to avoid the oscillatory behaviors caused by Hopf k3 Empirical constant (m6/kg2 h)
bifurcations, which decrease fermentation yield, system stability ms Maintenance factor based on substrate (kg/kg·h)
and controllability, and decline product quality. This article mp Maintenance factor based on product (kg/kg·h)
rigorously analyzed the Hopf point variation under carefully Ysx Yield factor based on substrate (kg/kg)
designed operating conditions and discussed its influence on Ypx Yield factor based on product (kg/kg)
product yield and quality. The article primarily focused on the VM Membrane volume (m3)
operating conditions and product yield rather than complicated VF Fermentor volume (m3)
theoretical bifurcations and chaotic behavior. A numerical Pm Membrane permeability (m/h)
analysis method for locating the Hopf singularity point and Greek Symbols
limit cycles was introduced and applied to the Z. mobilis ρ Ethanol density (kg/m3)
fermentation process for producing ethanol. In our research, μ Specific growth rate (h−1)
the entire set of Hopf points have been solved in the operating
Subscripts i
parameter panel. Furthermore, to avoid tedious explicit
computation of all the eigenvalues of a matrix, the current s Substrate (glucose) inside the fermentor
work used a method based on the bialternate product to s0 Influent glucose to the fermentor
compute the test function to determine Hopf point. It is the p Product (ethanol) inside the fermentor
proposed method makes it possible for us to first locate and p0 Influent ethanol to the fermentor
e Internal key component inside the fermentor
assemble these Hopf points as a “risk curve”, with reduced
e0 Influent internal key component to the fermentor
computational work and better accuracy. The calculated
x Biomass (microorganisms) inside the fermentor
distribution of Hopf points can enable all operating parameters
x0 Influent biomass to the fermentor
to be located away from undesirable singular points and
pM Product (ethanol) inside the membrane
avoidance of unwitting steps across these points, with
pM0 Influent ethanol to the membrane


appropriate and optimized values. Furthermore, the proposed
method can be applied to other fermentation processes and
improved models. Design and control strategies can be REFERENCES
developed at a very early stage, due to the a priori identification (1) Abashar, M. E. E.; Elnashaie, S. S. E. H. Multistablity, bistability,
of potential Hopf points, and this will contribute to the and bubbles phenomena in a periodically forced ethanol fermentor.
achievement of high productivity, product quality, and stability Chem. Eng. Sci. 2011, 66 (23), 6146−6158.
(2) Sridhar, L. N. Elimination of oscillations in fermentation
in dynamic processes. processes. AIChE J. 2011, 57 (9), 2397−2405.

■ AUTHOR INFORMATION
Corresponding Author
(3) Paz Astudillo, I. C.; Cardona Alzate, C. A. Importance of stability
study of continuous systems for ethanol production. J. Biotechnol.
2011, 151 (1), 43−55.
(4) Shen, Y.; Zhao, X. Q.; Ge, X. M.; Bai, F. W. Metabolic flux and
*Tel.: +86 10 62784572. Fax: +86 10 62770304. E-mail: cell cycle analysis indicating new mechanism underlying process
dcecbz@tsinghua.edu.cn. oscillation in continuous ethanol fermentation with Saccharomyces

1654 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655


Industrial & Engineering Chemistry Research Article

cerevisiae under VHG conditions. Biotechnol. Adv. 2009, 27 (6), 1118− http://www.computeralgebra.nl/systemsoverview/special/diffeqns/
1123. content/gcbody.html.
(5) Bai, F. W.; Ge, X. M.; Anderson, W. A.; Moo-Young, M. (24) Dhooge, A.; Govaerts, W.; Kuznetsov, Y. A. MATCONT: A
Parameter oscillation attenuation and mechanism exploration for MATLAB package for numerical bifurcation analysis of ODEs. ACM
continuous VHG ethanol fermentation. Biotechnol. Bioeng. 2009, 102 Transactions on Mathematical Software (TOMS) 2003, 29 (2), 141−
(1), 113−121. 164.
(6) Bai, F. W.; Anderson, W. A.; Moo-Young, M. Ethanol (25) Dhooge, A.; Govaerts, W.; Kuznetsov, Y. A. MATCONT: A
fermentation technologies from sugar and starch feedstocks. Biotechnol. Matlab package for numerical bifurcation analysis of ODEs. ACM
Adv. 2008, 26 (1), 89−105. SIGSAM Bull. 2004, 38 (1), 21−22.
(7) Sprenger, G. A. Carbohydrate metabolism in Zymomonas mobilis: (26) Back, A.; Guckenheimer, J.; Myers, M.; Wicklin, F.; Worfolk, P.
A catabolic highway with some scenic routes. FEMS Microbiol. Lett. DsTool: Computer assisted exploration of dynamical systems. Notices
1996, 145 (3), 301−307. Amer. Math. Soc 1992, 39 (4), 303−309.
(8) Li, C. C. Mathematical models of ethanol inhibition effects during (27) Clewley, R. H.; Sherwood, W. E.; LaMar, M. D.; Guckenheimer,
alcohol fermentation. Nonlinear Anal.: Theory, Methods, Appl. 2009, 71 J. M. PyDSTool, A Software Environment for Dynamical Systems
(12), 1608−1619. Modeling; 2007. http://pydstool.sourceforge.net (accessed 2012-5-9).
(9) Diehl, F. C.; Trierweiler, J. O. Control strategy for a Zymomonas (28) Ermentrout, B. Simulating, Analyzing, and Animating Dynamical
mobilis bioreactor used in ethanol production. In Computer Aided Systems: A Guide to XPPAUT for Researchers and Students; Society for
Chemical Engineering; de Brito Alves, R. M.; Oller do Nascimento, C. Industrial Mathematics: Philadelphia, PA, 2002; Vol. 14.
A.; Biscaia, E. C., Eds.; Elsevier: Amsterdam, 2009; Vol. 27, pp 1605− (29) Wang, H. Z.; Chen, B. Z.; He, X. R.; Zhao, J. S.; Qiu, T.
1610. Numerical analysis tool for obtaining steady-state solutions and
(10) Ghommidh, C.; Vaija, J.; Bolarinwa, S.; Navarro, J. M. analyzing their stability characteristics for nonlinear dynamic systems.
Oscillatory behavior of Zymomonas in continuous cultures: A simple J. Chem. Eng. Jpn. 2010, 43 (4), 394−400.
stochastic model. Biotechnol. Lett. 1989, 11 (9), 659−664. (30) Abashar, M.; Elnashaie, S. Dynamic and chaotic behavior of
(11) Jöbses, I.; Egberts, G.; Luyben, K.; Roels, J. Fermentation periodically forced fermentors for bioethanol production. Chem. Eng.
kinetics of Zymomonas mobilis at high ethanol concentrations: Sci. 2010, 65 (16), 4894−4905.
Oscillations in continuous cultures. Biotechnol. Bioeng. 1986, 28 (6),
868−877.
(12) Bai, F. W.; Chen, L. J.; Anderson, W. A.; Moo-Young, M.
Parameter oscillations in a very high gravity medium continuous
ethanol fermentation and their attenuation on a multistage packed
column bioreactor system. Biotechnol. Bioeng. 2004, 88 (5), 558−566.
(13) McLellan, P. J.; Daugulis, A. J.; Li, J. H. The incidence of
oscillatory behavior in the continuous fermentation of Zymomonas
mobilis. Biotechnol. Prog. 1999, 15 (4), 667−680.
(14) Daugulis, A. J.; McLellan, P. J.; Li, J. H. Experimental
investigation and modeling of oscillatory behavior in the continuous
culture of Zymomonas mobilis. Biotechnol. Bioeng. 1997, 56 (1), 99−
105.
(15) Jarzebski, A. B. Modeling of oscillatory behavior in continuous
ethanol fermentation. Biotechnol. Lett. 1992, 14 (2), 137−142.
(16) Garhyan, P.; Elnashaie, S. Static/dynamic bifurcation and
chaotic behavior of an ethanol fermentor. Ind. Eng. Chem. Res. 2004,
43 (5), 1260−1273.
(17) Garhyan, P.; Elnashaie, S. S. E. H.; Al-Haddad, S. M.; Ibrahim,
G.; Elshishini, S. S. Exploration and exploitation of bifurcation/chaotic
behavior of a continuous fermentor for the production of ethanol.
Chem. Eng. Sci. 2003, 58 (8), 1479−1496.
(18) Garhyan, P.; Elnashaie, S. S. E. H. Experimental investigation
and confirmation of static/dynamic bifurcation behavior in a
continuous ethanol fermentor. practical relevance of bifurcation and
the contribution of Harmon Ray. Ind. Eng. Chem. Res. 2004, 44 (8),
2525−2531.
(19) Mahecha-Botero, A.; Garhyan, P.; Elnashaie, S. Non-linear
characteristics of a membrane fermentor for ethanol production and
their implications. Nonlinear Anal.: Real World Appl. 2006, 7 (3), 432−
457.
(20) Doedel, E. J. AUTO: A program for the automatic bifurcation
analysis of autonomous systems. Congr. Numer. 1981, 30, 265−284.
(21) Doedel, E.; Kernevez, J. P. AUTO: Software for continuation
and bifurcation problems in ordinary differential equations, including
the AUTO 86 user manual. In Applied Mathematics; California
Institute of Technology: Pasadena, CA, 1986.
(22) Doedel, E.; Champneys, A.; Fairgrieve, T.; Kuznetsov, Y.;
Sandstede, B.; Wang, X. AUTO 97: Continuation and Bifurcation
Software for Ordinary Differential Equations, User’s Manual; Center for
Research on Parallel Computing, California Institute of Technology:
Pasadena, CA, 1997.
(23) Kuznetsov, Y.; Levitin, V. CONTENT: Dynamical System
Software; CAIN Europe: The Netherlands, 1997; available online:

1655 dx.doi.org/10.1021/ie3013049 | Ind. Eng. Chem. Res. 2013, 52, 1645−1655

You might also like