Sei sulla pagina 1di 10

Process Biochemistry 46 (2011) 900909

Contents lists available at ScienceDirect


Process Biochemistry
j our nal homepage: www. el sevi er . com/ l ocat e/ pr ocbi o
Multi-objective cascade controller for an anaerobic digester
Carlos Garca-Diguez
a,
, Francisco Molina
b
, Enrique Roca
a
a
USC - University of Santiago de Compostela, Department of Chemical Engineering/School of Engineering, Rua Lope Gomez de Marzoa s/n, 15782 Santiago de Compostela, Spain
b
Faculty of Engineering, University of Antioquia, A.A. 1226, Medellin, Colombia
a r t i c l e i n f o
Article history:
Received 3 September 2010
Received in revised form 9 December 2010
Accepted 25 December 2010
Keywords:
Anaerobic digestion
Biogas
USBF reactor
Cascade control
Control systems
Optimisation
Wastewater treatment
a b s t r a c t
In this work, a new multi-objective control strategy based on the concentration of volatile fatty acids
(VFAs) in the efuent and the methane owrate (Qch
4
) has been proposed for an upowsludge bed-lter
(USBF) reactor, which is used in the anaerobic treatment of winery wastewater. The approach presented
here is novel due to the following reasons: (i) it considers two operational objectives, i.e., control of the
efuent quality and control of the maximum production rate of methane; (ii) it takes advantage of the
difference between the dynamics of the liquid and gas phases using variables from both phases. The
control system is based on a cascade control strategy with a reference signal for the methane ow rate.
The control systemcomputes the feed owrate for adjusting the organic load applied to the reactor. The
performance of the proposed control scheme is illustrated through numerical simulations and parameter
optimisation using the Anaerobic Digestion Model no. 1 (ADM1) with regards to inuent disturbances.
Moreover, the controller has been validated in the closed-loop control of a 1.15m
3
USBF reactor treating
wastewater containing ethanol, which emulates winery efuents under different operational scenarios:
restart-up, long duration organic overload, long duration organic underload and successive organic dis-
turbances. The control systemsupplied adequate control action in response to the different disturbances
tested, and it demonstrated high reliability in achieving the desired set-point.
2011 Elsevier Ltd. All rights reserved.
1. Introduction
Biochemical processes are difcult to control because micro-
organisms are highly sensitive to changes in environmental
variables and are unable to fully inuence the cells internal envi-
ronment by manipulating the external environment in which they
live. In general, biological systems are known to be highly variable
and difcult to measure, and no reliable biological law is available
for their measurement [1]. Some of the factors that contribute to
the difculty incontrolling anaerobic wastewater treatment (AWT)
systems are the following: (i) the nonlinear dynamic behaviour of
the models; (ii) the different levels of complexity presented by the
existing models and the unpredictable variation of the parameters,
that are partly inuenced by biomass adaptation; (iii) the unpre-
dictable load disturbances in the inlet streamdue to changes in the
inlet owrate and composition; and (iv) the lack of reliable sensors
to measure intracellular activities [2,3].
In recent years, there have been signicant advances in the
development of sensors, whichhaveledtotheavailabilityof on-line
sensors for measurement/monitoring of key biological variables
such as volatile fatty acids (VFAs), chemical oxygen demand (COD)
and total organic carbon (TOC) [48]. However, COD and TOC

Corresponding author. Tel.: +34 981 563100x16772.


E-mail address: carlos.garcia.dieguez@usc.es (C. Garca-Diguez).
analysers are expensive and are usually recognised as fragile mea-
surement devices [9]. Furthermore, their maintenance is costly
compared to VFA titrimetric sensors [7,8].
The aim of an anaerobic reactor controller is to maintain condi-
tions of stability in a bioreactor against possible changes in inuent
characteristics (ow rate or composition) and to attain adequate
efuent quality and maximise the methane productivity of the
AWT plant [10]. Controllers can be designed for the management
of the process under the following different operational scenar-
ios: organic overload, hydraulic overload, toxic effects of inhibitor
compounds, thermal shock and restart-ups [11].
Robust control of anaerobic processes is crucial for avoiding
possible instability due to disturbances. Consequently, important
research efforts have been focused on the development of different
feedback control strategies for AWT processes. Inthe last fewyears,
several control feedback structures have been developed to over-
come the difculty in controlling AWT processes. Researchers have
reported a simple on/off control algorithm that uses the alkalinity
consumption and the feed ow rate as the process state variable
and the manipulated variable, respectively [12]. A combination of
the on/off and neural networks algorithm to control bicarbonate
was proposed by Guwy et al. [13]. Other researchers have used
PID (proportionalintegralderivative) controllers for regulating
AWT processes [1417] because this type of controller can be eas-
ily implemented in wide variety of plants. However the tuning of
these controllers is based on heuristic rules. Anaerobic digestion is
1359-5113/$ see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.procbio.2010.12.015
C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909 901
a complex and nonlinear process. Therefore, traditional and linear
controllers (e.g., on/off, PID, etc.) may fail in certain ranges unless
they have been properly adapted and their parameters have been
precisely calibrated to handle specic situations.
Rule-based expert systems have also been reported for the
supervision and control of AWT systems [1822]. Most of the
recently published work about expert systems for AWT control
is based on fuzzy logic [2327]. Moreover, fuzzy structures have
been used to develop control systems from expertise of operators
and phenomenological information [23,2830] to derive gain-
scheduling schemes for tuning on-line control parameters and to
manage process uncertainties [10]. With this aim and due to the
initial difculty in modelling AWT systems, neural networks have
been applied to develop controllers using data from AWT plants
[13,31,32].
Adaptive controllers for AWT plants [28,3339] have been
developed to account for the nonlinearities and transient features
of the anaerobic digestion process. The drawback of such strate-
gies is that complete knowledge of the parameter structure of the
system is required, which can be difcult to obtain when dealing
with bioprocesses such as AWT. In addition, adaptive controllers
usually respond more aggressively against disturbances than lin-
ear controllers, and therefore, strong variations may occur in the
manipulated variable, which is usually the feed ow rate [16].
Interval controllers have been successfully applied to the AWT
of distillery vinasses to solve the problem of COD and VFA regula-
tion [40]. The use of this type of controller permits management of
the relatively large uncertainty in some key process variables. Nev-
ertheless, its performance is strongly dependent on the denition
of the uncertainty intervals. Another interesting approach of the
interval-based control application, which has not yet been experi-
mentally validated, is the regulation of the inuent COD acting on
the dilution rate [41].
A variant in the feedback control topology is cascade control,
whichconsists of twoor morefeedbackcontrol loops. Afewcascade
applications can be found in the literature. Alvarez-Ramirez et al.
[15] have shown a direct feedback control system using only PID
controllers in a cascade conguration. This systemallows the regu-
lationof ananaerobic digester working at a lowCODconcentration.
Liuet al. [10,42,43] developeda cascade controller embeddedinto a
fuzzy rule-based supervisory system. This approach demonstrated
the ability to achieve high productivity conditions for the produc-
tion of methane by the AWT reactor.
Control strategies have often been tested for different reactor
congurations, different scales and under different operating con-
ditions, which makes it difcult to compare their performance.
However, it has been recognised that a suitable combination of
direct and indirect feedback control provides the best control strat-
egy [15]. Direct feedback control acts at a regulatory level whereas
indirect feedback control acts at a supervisory level (e.g., fuzzy
logic). An ideal control system must full different characteristics
at the regulatory level for simultaneously ensuring the following
criteria: (a) high efuent quality; (b) maximum methane produc-
tion; (c) general system stability; and (d) applicability to different
types of wastewater [23].
In this work, a new robust multi-objective controller has been
developed and used for the control of an upow sludge bed-lter
(USBF) reactor. Methane ow rate has been used as the inner con-
trol loop variable and VFA has been chosen as the external control
loop variable. The nal control variable is the variation (increase
or decrease) in the feed ow rate. The proposed scheme has been
designed to deal with modelling errors, and its structure allows
the attenuation of disturbances in the COD concentration in the
inlet stream. Validation of the control system has been performed
using simulations with ADM1 (Anaerobic Digestion Model no. 1)
and closed-loop control of a pilot-scale USBF reactor. The approach
is novel as it considers the main operational criteria in the same
control structure. Moreover, this controller takes advantage of the
different dynamics of the liquid phase and gas phase through a
cascade structure with a reference signal.
2. Reactor, model description and problem statement
2.1. Anaerobic reactor
Experiments for the closed-loop validation of the cascade con-
troller were conducted in a USBF, pilot-scale reactor (Upow
Anaerobic Sludge BlanketUASB zone +Anaerobic FilterAF zone)
[26] with an approximate liquid volume of 1.15m
3
(Fig. 1). The
reactor temperature was controlled at 372

C using a single
on/off loop. The on-line measurement devices that were available
included feed and recycling electromagnetic ow-meters (ABB,
COPA-XE and Siemens, 7ME2531), input and output reactor pH
meters (Cole Parmer) and thermometers (Pt-100), a biogas ow-
meter (Brooks, 3240), an infrared gas analyser (Siemens, Ultramat
22P) for the measurement of CH
4
and CO concentrations in the gas
phase and an electrochemical hydrogen gas analyser (Sensotrans,
Sensotox 420). On-line TOC and total inorganic carbon (TIC) were
determined by catalyst combustion oxidation and non-dispersive
infrared (NDIR) CO
2
detection (Shimadzu, 4100). All data were
monitored on-line with the sensors and recorded at 15-min inter-
vals. Detailed descriptions of the equipment and data acquisition
system have been reported elsewhere [23,44].
During the experiments, the biomass inside the reactor showed
specic methanogenic activity of 0.660.18kg CODkgVSS
1
d
1
.
The average total biomass observed in the reactor was 18.61.5kg
VSS, corresponding to anaverage VSS concentrationof 16.91.4kg
VSS/m
3
[45]. These values corresponded to a maximum organic
loading rate of 12kg COD/m
3
d.
Synthetic wastewater, containing ethanol which emulates
winery efuents, was used in the experiments. The inuent com-
position consisted of dilute white wine, nutrients and sodium
bicarbonate. The wine was diluted in situ using a static mixer just
beforeit enteredthereactor toavoidpre-acidicationinthefeeding
tank. Nutrients and sodium bicarbonate were added to maintain a
COD/bicarbonate/N/P ratio of 1000/400/7/1, which is required for
biomass growth, and to maintain an adequate buffering capacity
inside the reactor [46].
A titrimetric AnaSense analyser [7,8] was used to determine
the following operational reactor parameters: VFA (volatile fatty
acids), bicarbonate, and partial and total alkalinity. After the
determination of these parameters, the IA/TA ratio (intermediate
alkalinity/total alkalinity ratio) was calculated (IA corresponds to
the difference between total and partial alkalinity).
2.2. Anaerobic Digestion Model no. 1 (ADM1)
ADM1 [48] is a complex model of the multi-step anaerobic
process transformations. This tool is adequate for predictions of
anaerobic wastewater treatment processes with sufcient accu-
racy for use in process development, optimisation, and control.
It is a standard benchmark for developing operational strategies
and evaluating controllers [16]. ADM1 incorporates processes such
as the hydrolysis of particulates, acidogenesis, acetogenesis and
methanogenesis, and it includes 26 dynamic state concentration
variables, 19 biochemical kinetic processes, 3 gasliquid transfer
kinetic processes, and 8 implicit algebraic variables per liquid ves-
sel. A modied version of the ADM1 toolkit was used in this study.
The modied version incorporates an extension to ethanol degra-
dation pathways through an additional group of ethanol degraders
and a new state variable for ethanol concentration, which is cali-
902 C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909
Fig. 1. USBF reactor and cascade control scheme. USBFupow sludge bed lter reactor; GFCgas phase controller; LFCliquid phase controller; GMgas measurements;
LMliquid measurements.
brated for a USBF reactor [49]. The extension for ethanol considers
thehydrogenandacetatepathways andaccounts for theVFAs path-
ways (propionate and butyrate) through stoichiometry. Thus, VFAs
are suitably predicted by the model, including any possible over-
load or transitions between steady states.
3. Controller design
3.1. Controller objectives
Previously, many authors have used the Haldane kinetic model
(Fig. 2) to describe the behaviour of anaerobic digesters [5052].
Haldane kinetic model also provides a simple method of explaining
controller objectives.
Haldane kinetics have an obvious equilibrium point (biomass
washout) andtwooperational regions, i.e., astableoperatingregion
and an unstable operating region (see Fig. 2), which have different
dynamic properties in terms of stability [50]. In practice, the risk
of destabilisation of an AWT process can be avoided by operating
in the stable region or far below the maximum reactor capacity
Fig. 2. Haldane kinetic model for an anaerobic digester. (- - - ) Methane ow rate
setpoint. () Operational point with the objective of fullling environmental regu-
lations.
(optimum methane ow rate production), i.e.,

K
I
K
S
(K
I
is the
inhibition kinetic constant and K
S
is half saturation constant) for
the Haldane kinetic model [10].
Normally, AWT plants attempt to full environmental regula-
tions. Therefore, the process requires a controller, which keeps the
systemstable at a xed set-point. However, when AWT plants seek
tomaximise methane production, anoptimal or suboptimal control
strategy must be proposed. In practice, the risk of system overload
can be reduced by operating with a security margin that is below
the maximum reactor capacity. In any case, the main objective of
any controller for AWT plants is to keep the system in the stable
region [53].
As a consequence, the development of a controller that is able
to integrate both objectives, i.e., the emission level set-point and
the maximum methane ow rate, and is able to robustly regulate
the process in the same single structure is of great interest to such
wastewater treatment systems. Different control structures could
be considered for developing the controller. However, a cascade
framework has been used in our study.
3.2. Selection of control variables
The rst step during the development of a control system is
the selection of a group of process variables, which can provide
information about the metabolic state of the process. Different
combinations of variables werestudiedtoestablishthemost appro-
priate combination for maximumcontrol of regulation in a cascade
structure. In this sense, when two of the variables for state iden-
tication in the anaerobic process are considered with winery
wastewater, several combinations of variable pairs are able to pro-
vide accomplish a complete classication of states [54]. A similar
study [55] with other types of wastewaters has showed that an
appropriate combination of variables in the gas and the liquid
phases can be used to develop a monitoring, diagnosis and control
(MD&C) system. Furthermore, variables in the liquid phase present
higher response times than gas phase variables. Therefore, accord-
ing to this criterion an inner control loop with a secondary variable
in the liquid phase does not seemappropriate. Besides, the use of a
cascade control structure has been proposed to reduce the adverse
C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909 903
Table 1
Evaluation criteria for inner-loop variables suitability. (Yyes; Nno or not
appreciable; N/Anot applicable; Aadvanced instrumentation; Oin occasions;
Ssimilar).
Criterion Qch
4
%CH
4
Qgas %CO
2
CO H
2
Secondary variable is available A A Y O A A
Indicates a key disturbance Y O Y N N Y
Causal relationship with the
controlled variable
Y N Y N N N
Secondary dynamic faster than
primary
Y Y Y Y Y Y
effects of measurement delays andexploit thefaster responsetimes
of gas phase variables (e.g., biogas ow rate or methane ow rate)
compared to liquid phase variables.
Methane owrate (Qch
4
) was chosen as the inner-loop variable
after accounting for the technical, economical and dynamic process
response criteria. A combination of this variable with the VFA con-
centration allows the detection of any potential imbalance in the
AWT process.
The selected inner-loop variable determines the performance
of cascade structures because the cancellation of the error in the
inner-loop affects the cascade efciency. A suitable internal vari-
able must full a series of criteria or requirements such as those
listed above for the considered variables. These criteria establish
a ranking amongst the variables under consideration in terms of
their adequacy for the development of a cascade controller for an
AWT plant. The methane ow rate (Qch
4
) and the biogas ow rate
are considered the best inner-loop variables, as shown in Table 1.
A feature of the cascade control structure is that it can be
switched to operate as a single loop that seeks a suboptimal value
of Qch
4
. This is easily achieved by turning off the external control
loop and by providing an external reference trajectory to the slave
loop.
3.3. Methane productivity inner control loop
The scheme for the inner control loop corresponding to the
set-point of methane productivity is illustrated in Fig. 3. In this
case, the controller works as an auto-setting control system with a
methane reference signal (Qch
4ref
). Note that Qch
4ref
provides the
set-point of the inner controller, and it is changed to achieve opti-
mum methane productivity. In other words, the inner controller
was designed to maintain the process near the operational point
corresponding to maximum methane production (Qch
4max
) when
the output loop was disconnected, and thereby push the system
to higher organic loading rate (OLR), which avoided working in a
dangerous and unstable operational zone (see Fig. 2).
Theinner controller was implementedas aPIDcontroller. There-
fore, the equation of this controller can be expressed as follows:
Q
in
(t) = Q
in
(t 1) +K
Ps
eQch
4
(t) +K
Is

eQch
4
(t)dt
+K
Ds
d(eQch
4
(t))
dt
(1)
Fig. 3. Block diagram of the methane productivity controller (inner-loop).
where Q(t) represents the current reactor feed owrate or the con-
troller output, Q
in
(t 1) is the initial value of the feed owrate, K
Ps
is the proportional gain, K
Is
is the integral gain, K
Ds
is the deriva-
tive gain, and eQch
4
is the error between the actual value of Qch
4
and the reference signal (i.e., eQch
4
(t) =Qch
4ref
(t) Qch
4
(t)). The
problem of controlling the output of a system to track a prescribed
reference in the presence of model uncertainties and input distur-
bances is of great interest in the control of bioreactors.
The methane owrate was measured at intervals of 5s and was
ltered using a moving window of 15min. Identical time intervals
were considered for modifying the action of the controller. These
time intervals are considered appropriate when the scale of indus-
trial anaerobic bioreactors and the slow dynamics of this process
are taken into account.
3.4. Tracking methane reference
In anaerobic digestion, if one is primarily interested in the
amount of methane generated, the total methane production dur-
ing the transition period between two steady states can be used as
anappropriate measure of the systems performance, whichinturn
can be maximised (see Eq. (2) optimal control approach).
J(Q
in
(t)) =

tf
0
Qch
4
(t) dt (2)
However, optimal control is a very sensitive technique for the
proposed model. It requires complete knowledge of the process
model, including an analytical expression for each specic rate in
the system. In biotechnology, particularly in the eld of anaerobic
digestion, this assumption is never fullled in practice; an opti-
mal prole is generally calculated using a model that describes the
process correctly from a qualitative viewpoint. Therefore, it is very
useful to construct suboptimal strategies that do not suffer from
the above difculties.
Based on a two-step anaerobic model structure provided by
Bernard et al. [51], it is possible to express the methane produc-
tion rate (q
M
) as a function of VFAconcentration and methanogenic
biomass concentration.
q
M
= k
M

max
VFA
VFA +K
s
+(VFA
2
/K
I
)
X
met
(3)
where k
M
represents the yield coefcient for methane production
determined from previous experimental data calibration,
max
is
the maximumspecic growth rate for methanogenic bacteria, X
met
is the methanogenic biomass concentration, and K
s
and K
I
are the
kinetic parameters for Haldanes kinetic expression(half saturation
constant and inhibition constant, respectively).
Consider that the nonlinear adaptive parameter
1
is dened
according to Eq. (4):

1
=
q
M
VFA

K
s
+VFA +

VFA
2
K
I

(4)
where q
M
is expressed in kgmol/m
3
d and therefore needs to be
converted into units of m
3
/d through the ideal gas law Eq. (5),
assuming a constant temperature (T) of 37

C and a pressure equiv-


alent to 1atm. The same temperature and pressure conditions are
used in the remaining equations in this manuscript.
Qch
4
= q
M

RT
P
(5)
where VFA* is the concentration of the volatile fatty acids at the
optimum methane ow rate production (see Eq. (6) and Fig. 2):
VFA

K
I
K
S
(6)
904 C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909
Then, the suboptimal methane ow rate can be computed from
Eq. (7) assuming a 10% margin of security beyond the beginning of
the unstable zone (see Fig. 2). Eq. (7) provides the reference sig-
nal, in this case the set-point, for the rst control objective (i.e., to
maximise methane production).
Qch

4ref
= 0.9
1

K
I

K
I
+2

K
s

R T
P
(7)
A disadvantage of this strategy is the dependency of kinetic
parameters on the control law, which can be solved by developing
a mechanism that changes the parameters of the control law with
time to account for biomass changes (i.e., biomass adaptation). An
option to the adaptation mechanism may be to approximate the
stable zone in Haldanes model (see Fig. 2) with a straight line (Eq.
(8)).
q
M
= k
met
VFA (8)
Evidently, if the slope of this line (methanogenic adaptation
slope k
met
) increases with time, we can conclude that biomass
adaptation has occurred. On the other hand, a decrease in the slope
of the methanogenic adaptationimplies a loss of biomass, the death
of biomass, the entry of a toxic substance into the reactor or any
other operational circumstance that diminishes biomass activity.
In both cases the parameters need to be varied as per the mea-
surements. Thus, the adaptive mechanism consists of frequently
checkingthe methanogenic adaptationslope andcomparingit with
previous values. A switching mechanism (Eq. (9)) has been pro-
posed in this study. If the absolute value of the difference between
the methanogenic adaptation slope and its average is greater than
0.5, then the parameters are modied to account for biomass
adaptation. However, if the difference between the methanogenic
adaptation slopes is small (i.e., less than 0.5), then the parame-
ters are allowed to remain at their previous values. The values are
checkedwiththe average k
met2
values after 12hof operation, which
is sufcient considering the duplication time for methanogenic
bacteria (approximately 7 days). k
met
is initially established from
Haldanes model (Fig. 2) and then it is progressively updated using
Eq. (10).
if |

k
met 1


k
met 2
| > 0.5 then modify parameters to account
for biomass adaptation
if |

k
met 1


k
met 2
| > 0.5 then maintain parameter values
(9)
If biomass adaptation is detected, then measurement of both
parameters, the methane ow rate and the VFA concentration, is
necessary for computing a new slope as shown in Eq. (10).
k
met
=
Qch
4t
VFA
t

P
R T
(10)
Finally, with the newslope of methanogenic adaptation, assum-
ing that biomass adaptation occurs only by increasing substrate
afnity (i.e., K
s
decreases), then K
s
can be adapted as shown in Eq.
(11). It is very important to ensure that K
s
remains positive (i.e., if
K
s
does not remain positive, then it must retain its previous value).
K
s
=

1
K
1
K
met
k
met
(11)
While controlling the concentration of organic matter in the
efuent (i.e., second controller objective), the methane reference
signal is established as a function of VFA, methanogenic biomass
concentration (X
met
) and some model parameters (see Eq. (12)).
Qch
4ref
= k
M

max

VFA
VFA +K
s
+(VFA
2
/K
I
)


X
met

R T
P
(12)
A simple mass balance observer (Eq. (15)) can be derived from
the mass balance of the methanogenic biomass (Eq. (13)) and the
methaneproductivityequation(Eq. (14)). This observer permits the
estimation of the methanogenic biomass concentration (X
met
). It is
easy to demonstrate that the observer converges asymptotically
when D>0, which is always true except when D is equal to zero
(i.e., the reactor is shut down) [56].

X
met
= r
M
D

X
met
(13)
Qch
4
= k
M
r
M

R T
P
(14)

X
met
=
Qch
4
k
M

P
R T
D

X
met
(15)
where Dis the dilutionrate computedby dividing the feedowrate
value by the volume of the reactor and describes the deviation of
a completely stirred tank reactor (CSTR) from the ideal behaviour,
which allows the simplication of the solid retention model. The
value of is 0.005, and it is estimated using the average of the solid
retention time (SRT) and the average of the dilution rate (Eq. (16)).
These values were estimated from experimental data of the USBF
reactor.
=
1
SRTD
(16)
3.5. VFA control external control loop
A bioreactor with a single loop performs sufciently well when
the dynamics is fast, dead time is small and the disturbances are
small and slow, as in the case of the methane productivity con-
troller. However, a single loop for controlling VFA is insufcient
because it lacks the characteristics described above. In this case, a
cascade control structure may be considered as a good alternative.
The cascade structure proposed for this task is shown in Fig. 4.
The feed ow rate (Q
in
) is constrained by a saturation func-
tion (according to the physical restrictions on the manipulated
variables). In practice, the ow rate that is physically applied to
the anaerobic reactor should be positive and must have an upper
bound. The minimum feed ow rate (Q) was zero, and the max-
imum feed ow rate (Q) was established using the minimum
hydraulic retentiontime (HRT) criterionof 9hfor preventing unde-
sired washout of biomass. Under this criterion, the operational
limits of the pumps must be respected.
sat(Q
in
) =

Q, Q
in
Q
Q
in
, Q < Q
in
< Q
Q, Q
in
Q
(17)
In addition, variation in the feed ow rate (Q
in
) was limited
to 0.012m
3
/d for each 15min interval to avoid extreme behaviour
such as oscillations in operational zones that were further away
fromthe desired operating conditions. Due to the integral action of
the inner controller, the saturationof the control input couldpoten-
tially induce undesirable phenomena such as reset windup, which
could lead to large overshoots and high settling times. Therefore,
an anti-reset windup scheme was added to the control structure
by introducing static gain feedback of the difference between the
computed control signal and the actual saturated control action.
The master controller was also implemented as a PID controller
with the following equation:
Qch
4sp
(t) = K
Pm
eVFA(t) +K
Im

eVFA(t)dt
+K
Dm
d(eVFA(t))
dt
(18)
C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909 905
Fig. 4. Block diagram of the cascade controller. Note that the reference signal Qch
4
modies the set-point to the inner controller in the cascade scheme.
where eVFA can be calculated through the following expression:
eVFA(t) = spVFA(t) VFA(t) (19)
Theinner loopcontroller equationis thesameas that usedinsin-
gle loop control (i.e., to maximise methane production). However,
the error in measurement/calculation of the methane ow rate
(eQch
4
) is modied by an appropriate reference signal (Qch
4ref
):
eQch
4
(t) = Qch
4sp
(t) +Qch
4ref
(t) Qch
4
(t) (20)
When the second control objective is pursued (i.e., VFA regula-
tion), then the Qch
4ref
value allows to modify the error (Eq. (20))
in the inner loop eQch
4
, because it is necessary to improve the
behaviour of the control system when the process is near to the
VFA set-point. On the contrary, whether Qch
4ref
was not included
aninverse behaviour couldtake place whenthe error onVFA(eVFA)
was small. Moreover, Qch
4ref
accelerates the controller response
when eVFA is high. Therefore, it helps to reduce restart-up time
andconvenientlydecreasingthefeedowrateduringorganic over-
loads.
Due to the inherent uncertainty in the VFA measurement, a
mechanismavoiding the excessive oscillations inthe control action
around to the desired set-point was incorporated. This mechanism
consists of introducing an uncertainty band equivalent to 5% of the
set-point value.
To obtain controller settings, the system was tuned through
parameter optimisation. These parameters were computed using
ADM1, and an integrated square error (ISE) was used as the crite-
rion for optimisation. The parameters (see Table 2) were found to
be very similar for the range of concentrations studied, i.e., from
7 to 16kg COD/m
3
for COD
in
, which is a characteristic of winery
wastewater [51]. Observe that derivative and integral values of the
PID controllers were negligible.
4. Validation
Taking into account the performance of the reactor at equilib-
rium, any possible imbalance in the process can be attributed to the
characteristic changes of the inuent (e.g., a pronounced change in
its organic matter content knownas organic overload or underload,
variation in pH, variation in feed ow rate and presence of toxic
substances), sudden changes in the operational environment (e.g.,
excessive temperature uctuations) and excessive loss of microbial
biomass.
Therefore, to validate controller performance, numerical sim-
ulations were carried out with the ADM1 virtual plant, and pilot
plant experiments were performed under three different operat-
ing conditions: organic overload, organic underload and restart-up
of the plant after a short periodof rest (23weeks) for the anaerobic
digester described previously.
4.1. Validation using ADM1
The cascade control system that was developed in this study
was tested using data generated by the ADM1 simulation. Three
different operational scenarios were studied: automatic restart-up,
organic overloadandorganic underload. Thesimulations areshown
in Fig. 5. The initial condition (i.e., the initial state of the reactor)
for the simulations was obtained for a medium value of OLR (8kg
COD/m
3
d) over a large period of time to ensure that a steady state
was reached. The inuent COD concentration used in the simula-
tions was 812kg COD/m
3
. This is a typical concentration range for
industrial winery wastewater [51,57]. The inlet inorganic carbon
Fig. 5. Validations by means of simulations withADM1. (a) Restart, (b) organic over-
load, (c) organic underload. VFAvolatile fatty acids concentration; eVFAexternal
controller volatile fatty acids error; Q
in
feed owrate; and eQch
4
inner controller
methane ow rate error.
906 C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909
Table 2
Parameters of the control system.
Parameters Inner controller Parameters External controller
KPs 0.0172 (m
3
/m
3
CH
4
) KPm 0.38 (m
3
CH
4
L/mgAc d)
KIs 7.210
7
(m
3
/m
3
CH
4
d) KIm 0
KDs 0.0006 (m
3
d/m
3
CH
4
) KDm 0
concentration, inlet inorganic nitrogen concentration, cations and
anions concentrations inthe ADM1were estimatedconsidering the
same bicarbonate and nutrients ratio that was used in the pilot
plant experiments (COD/bicarbonate/N/P ratio of 1000/400/7/1)
[58].
AWT plants are often restarted after a short period of rest in
certain industrial applications such as the treatment of winery and
distillery wastewaters where wine productionis seasonal [16]. This
is also true for the treatment of wastewater from other agro-food
industries where the collection of the agricultural products is a sea-
sonal activity. Automatic control systems can help minimise the
time required to restart-up the AWT plants. The cascade controller
(see Fig. 5a) restart-up the AWT plant in 3 days with a very small
initial feed ow rate (0.12m
3
/d) and a COD
in
of 10kg COD/m
3
.
However, the desired set-point is achieved only after two days.
As shown in Fig. 5b (organic overload), the controller was able
to maintain the process in a stable condition after a severe increase
of COD
in
from 10kg COD/m
3
to 15kg COD/m
3
, which corresponds
to a 50% increase in the organic load. Without automatic control,
this disturbance would cause the treatment systemto fail, because
it exceeds the treatment capacity of the reactor. When it occurs,
VFAs accumulation produces stronger inhibitory effects [5]. Sim-
ilarly, the controller responded to a 30% decrease of the organic
load by increasing the feed ow rate, thereby accurately handling
the organic underload. In this case, a minimumvariation of the VFA
concentration would be detected. In all the simulations, the pHwas
maintained close to a constant neutral value. Also, the IA/TA ratio
was lower than 0.2 (results not shown here), which is considered
to be adequately lower than the recommended limit of 0.3 [59].
These simulations demonstrate that the anaerobic digestion
process is accurately regulated through cascade control, which can
be designed and implemented with minimal effort.
Afewother researchers have successfully used cascaded cong-
urations for controlling anaerobic digesters. Alvarez-Ramirez et al.
[15] have shown that direct feedback control using linear con-
trollers in a cascaded conguration has potential application in the
optimal operation of AWT plants. However, these authors set their
control variables as VFA and COD, which belong to the liquid phase
and would delay the response of the controller system. Another
interesting application using the cascade control of pH and the
ow rate of biogas was developed by Liu et al. [10,42,43]. In these
studies, good control performances were achieved in a closed-loop
conguration during operation of anaerobic reactors at lab. How-
ever, the use of pHas aninner loopvariable requires precise control
over the addition of the source of alkalinity, especially at industrial
scales to attain a fast response. Moreover, this controller requires a
sensitive pH-meter, which is not feasible at industrial scales. Such
controllers are only designed to achieve maximum methane pro-
ductivity. Therefore, they are unable to regulate the concentration
of organic matter in the reactor efuent.
4.2. Closed-loop validation at pilot scale
The cascade controller was experimentally validated using a
USBF pilot scale reactor working in a closed loop. To evaluate
its behaviour, the control system was tested under both restart-
up conditions and different disturbances (i.e., long duration of
organic underload, long duration of organic overload and consecu-
tive overload and underload). The controller was roughly tuned in
simulation using the modied ADM1 incorporating ethanol degra-
dation. The same parameters were used in both simulations and
during the different experimental tests. However, it is recom-
mendedtoperiodicallyrecalibratethecontroller parameters before
using it, because important changes on biomass concentration and
activity may occur after long periods of operation that may affect
the controller performance. It is also very important to take into
account the type of wastewater to be treated. This control approach
has been validated with winery wastewater, but it should work
with other similar types of wastewater.
4.3. Restart-up operation
As previously mentioned, the restart-up of AWT plants is of
interest to the treatment of winery wastewater because wine pro-
duction is a seasonal activity, and digesters must often be restarted
[16,57,60].
Two restart-up experiments were performed in this study. The
results presentedinFig. 6correspondtotwodifferent VFAset-point
(spVFA) values. Fig. 6 shows a restart-up situation where the input
COD (COD
in
) concentration was initially xed at 8kg COD/m
3
for
a set-point value of 120mgAc/L. In this rst experiment, the con-
troller only neededthree days torestart-upthe process after a short
stop of 3 weeks, and produced 2.7m
3
/d of high quality methane
(in agreement with the stoichiometry for this residue) containing
close to 71% of methane. The nal OLR achieved was nearly 8.35kg
COD/m
3
d and the pH was close to the neutral value.
The objective of the second restart-up experiment was to test
the robustness of the controller design using a different VFA set-
point (550mgAc/L) while retaining the controller parameters that
were established for the previous experiment. Fig. 7 shows the
response of the reactor during restart-up after a short stop of 1
month with an approximate COD
in
concentration of 10kg COD/m
3
.
The main criterion used for these stops was the seasonal behaviour
in the winery industry [11].
Note that at time around 1.8 and 3.4 days a rapid change on
VFA had occurred. In these cases the calibration frequency of the
AnaSense

was of 4 days, lesser than in the rest of the experiment.


Thus, it is recommended to recalibrate the sensor with at a higher
frequency (e.g., each 2 days) in order to avoid this drawback.
Fig. 6. Experimental closed-loop control validation of the control system during a
restart of the USBF reactor (spVFA 120mgAc/L).
C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909 907
Fig. 7. Experimental closed-loop control validation of the control system during a
restart of the USBF reactor (spVFA 550mgAc/L).
4.4. Organic underload and overload
The robustness of the controller was also tested with distur-
bances in COD
in
. In this test, the inuence on the process variables
of a change in the COD
in
value was investigated over a long period
of time. Inthis case, the objective was to analyse the response of the
control system when faced with a sudden underload or a sudden
overload during the AWT process. The underload condition was
tested by diluting the inuent to 50% of its initial concentration
(underload of 50%) over 4 days and then returning it to its initial
concentration (8kg COD/m
3
).
As shown in Fig. 8, under conditions of underload, the controller
adjusted the applied load by increasing the feed owrate (Q
in
). This
disturbance did not induce noticeable changes in the output VFA
concentration. The appliedunderloadresultedina suddendecrease
of the methane ow rate and only a slight decrease in VFA con-
centration. This behaviour can be attributed to the faster response
of the VFA concentration (liquid phase variable) compared to the
methaneowrate(gas phasevariable). Theperformanceof thecon-
troller was found to be similar to the behaviour predicted by the
modiedADM1simulations (seeFig. 5c). The disturbance was over-
come in approximately one day, and the process remained stable
until day 13.5, when an organic overload was applied.
When the inlet COD concentration returned to 8kg COD/m
3
,
at rst, the methane ow rate and VFA concentration increased
considerably. However, in this case, the overload corresponds to a
COD concentration increase of 100%. Thus, the controller response
to this overload condition was simply to decrease the input ow
Fig. 8. Experimental closed-loop control validation of the control system during a
long duration organic underload and overload (spVFA 120mgAc/L).
Fig. 9. Experimental closed-loop control validation of the control system during a
long duration organic underload and overload (spVFA 550mgAc/L).
rate back to the initial value before the overload was applied. Fig. 8
also shows the delay in the response time of VFA during overload.
Although the controller response was appropriated in each case
the set point error could be improved using a better adaptive
mechanism of the controller parameters. In this sense, the use of
nonlinear control techniques could be of interest due to the non-
linear nature of the anaerobic wastewater treatment process.
Fig. 9 shows another similar experiment that was performed
with a higher VFA set-point (spVFA 550mgAc/L). First, the concen-
trationof organic matter was decreasedby50%, andthenanorganic
overload corresponding to a 100% COD
in
concentration increase
was applied. Withthe helpof the control system, the process recov-
ered from the instability induced by both disturbances. In this
case, the VFA concentration decreased during organic underload
and increased signicantly during overload, because the treatment
systemwas close to its maximummethanogenic capacity. Manage-
ment of overloads is riskier than underload management, because
the increase of VFA induced during an organic overload can drive
the treatment process towards an unstable operating zone, as can
be seen in Fig. 2. During an underload, however, only a part of the
treatment capacity is wasted. Nevertheless, the controller avoids
acidication, and thereby, process failure. The maximum capacity
was estimated for an OLR of 12kg COD/m
3
d. This can be observed
by comparing the methane ow rates during organic overload in
both experiments (Figs. 8 and 9). The effect of the disturbances
remained for some time even after the perturbations had been
removed.
In general during long-term COD changes, the controller was
able to maintain the stability of the process for the complete
duration of both experiments. Furthermore, other relevant pro-
cess variables such as pH and alkalinity ratio (IA/TA) (data not
shown) were maintained within acceptable limits. The measured
pHwas near neutral with a slight decrease during overload. On the
other hand, the IA/TA ratio remained lower than 0.3, which is the
recommended value [59]. This value was only slightly surpassed
during organic overload in the second experiment. The addition of
alkalinity must be regulated with a simple control system. Several
promising controllers showninthe literature [61] couldpotentially
help optimise the addition of alkalinity. Other interesting issue
about this type of process is the entrance of a toxic in the feed-
ing line. But, there exist a great variety of toxic compounds with
a different grade of affectation on anaerobic biomass depending of
the toxic compound nature and its concentration in the wastew-
ater. In the experimental tests showed here the wastewater used
was synthetic andtherefore the effect of a toxic must be established
in a further research. However, if the effect of a toxic on anaerobic
treatment process is known, it is possible to establish howthe con-
908 C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909
Fig. 10. Experimental closed-loop control validation of the control system during
consecutive variations of COD
in
(spVFA 120mgAc/L). (a) Inuent chemical oxygen
demand (COD
in
), organic loading rate (OLR). (b) Methane owrate (Qch
4
), feed ow
rate (Q
in
), volatile fatty acids concentration (VFA).
troller would act. If the controller is unable to manage the toxicant
effects will be necessary to introduce some changes in its structure
or use a supervisory control.
4.5. Successive variations
The practical application of the control law was also demon-
strated in an additional experiment consisting of successive COD
changes without permitting the process to reach stable or steady-
state conditions. The variations in OLR, COD
in
, methane ow rate,
inuent COD, feed ow rate (i.e., the control action), spVFA and
VFA are shown in Fig. 10. The remaining variables are not pre-
sented because the key aspects of the control action are sufciently
demonstrated with these variables.
In this experiment, two successive organic overloads were
applied to the reactor. Initially, the inlet COD concentration was
increased from 6 to 8kg COD/m
3
, and then another increase from
8 to 12kg COD/m
3
was induced. Then, an underload was induced,
decreasingCOD
in
to8kgCOD/m
3
. Fig. 10depicts thevariationof the
manipulated variable (Q
in
) with time, such that the control strat-
egy adjusted the loading rate in accordance with changes in the
concentration of the inuent.
The increasing availability of reliable on-line sensors, like
AnaSense

, andtheincreasingof theavailableknowledgeof biolog-


ical phenomena involved in anaerobic digestion processes allowed
the application of a cascade structure to propose a controller able
to consider together the quality of the efuent (VFA regulation)
and the maximisation of the methane production objectives. VFA
measure could be changed by other variable related to organic
matter content in the wastewater, as for example the chemical
oxygen demand although with a high cost. In this respect, the con-
troller system presented in this paper had advantages over other
controllers, and it was widely validated by simulations and on a
pilot-scale USBF reactor of 1.15m
3
. In addition, the relative low
cost of the measurement devices required, make this controller
very interesting for industrial application.
5. Conclusions
A new multi-objective cascade control system was developed
for the regulation of anaerobic wastewater treatment. It was vali-
dated using simulations performed with a virtual ADM1 plant in a
closed-loop anaerobic reactor. The controller demonstrated excel-
lent performance even under conditions of severe disturbance by
providing an adequate control action to achieve the desired set-
point and manage sudden changes in the inuent concentration. It
was also able to restart-up the AWT plant in a short period of time.
The main drawback of the proposed cascade scheme is that an
excessively large VFAset-point canintroduce additional nonlinear-
ities in the system due to inhibition effects. However, in this case,
the appropriate control action can be employed using an accurate
alkalinity control system. Both control modes (cascade and sin-
gle loop) are compatible with promising alkalinity monitoring and
control systems that are currently available, such as the controller
developed by [61], which permits the control of carbon dioxide
concentration in the gas phase (i.e., biogas quality).
From a practical point of view, the resulting control scheme
presents the following advantages: (i) tuning of the controller is
easytoachieveduetoits linear structureandthefact that onlyafew
parameters must beconsidered(i.e., PIDslaveparameters, PIDmas-
ter parameters, anda fewkinetic parameters); (ii) therequirements
for practical implementation are easily satised; (iii) it requires the
least amount of prior knowledge of the system, whichincreases the
range of situations to which it can be applied; and (iv) the relatively
low cost of the measurement devices required. As a consequence,
the resulting scheme can be potentially used in actual industrial
applications.
Further closed-loop tests of the inuence of temperature may
be of interest to ensure the robustness of the controller in an indus-
trial environment. However, in this case a supervising system may
sense a failure in the temperature control system and deactivate
the cascaded controller as a preventive measure.
Acknowledgments
The authors wish to acknowledge the Spanish National R&D
Program and European Regional Development Fund (ERDF) for the
Project ANACOMCTQ2004-07811-C02-01, andthe SpanishFPI pro-
gram for the grant BES-2005-10878.
References
[1] Yamuna K, Ramachandra V. Control of fermenters: a review. Bioprocess Eng
1999;21:7788.
[2] Olsson G. Instrumentation, control and automation in the water
industrystate-of-the-art and new challenges CD. Water Sci Technol
2006;53:116.
[3] Mendez-Acosta HO, Campos-Delgado DU, Femat R, Gonzalez-Alvarez V. A
robust feedforward/feedback control for an anaerobic digester. Comput Chem
Eng 2005;29:161323.
[4] Steyer JP, Bouvier JC, Conte T, Gras P, Harmand J, Delgenes JP. On-line mea-
surements of COD, TOC, VFA, total and partial alkalinity in anaerobic digestion
processes using infra-red spectrometry. Water Sci Technol 2002;45:1338.
[5] Olsson G, Martinus K, Zhiguo Y, Jensen AL, Steyer JP. Instrumentation control
and automation in wastewater systems. London: IWA Publishing; 2005.
[6] Spanjers H, van Lier J. Instrumentation in anaerobic treatment-research and
practice. Water Sci Technol 2006;53:6376.
[7] Molina F, Ruiz-Filippi G, Garcia C, Lema JM, Roca E. Validation at pilot scale of a
newsensor for on-line analysis of volatile fatty acids andalkalinity inanaerobic
wastewater treatment plants. Environ Eng Sci 2009;26:6419.
C. Garca-Diguez et al. / Process Biochemistry 46 (2011) 900909 909
[8] Ruiz-Filippi G, Molina F, Steyer JP, VanrolleghemP, Zaher U, Roca E, Lema JM. In:
KimC, Park TJ, Ko JH, editors. 9th International Conference on Instrumentation,
Control &Automation in Water &Wastewater Treatment &Transport Systems.
Busan: IWA Publishing; 2006. pp. 1338.
[9] Steyer JP, Bouvier JC, Conte T, Gras P, Sousbie P. Evaluationof a four year experi-
ence with a fully instrumented anaerobic digestion process. Water Sci Technol
2002;45:495502.
[10] Liu J, Olsson G, Mattiasson B. Monitoring and control of an anaerobic upow
xed-bedreactor for high-loading-rateoperationandrejectionof disturbances.
Biotechnol Bioeng 2004;87:4353.
[11] Garcia C, Molina F, Carrasco EF, Roca E. Control of restart-up of anaerobic USBF
reactors after short stops. Ind Eng Chem Res 2009.
[12] Denac M, Lee PL, Newell RB, GreeneldPF. Automatic-control of efuent quality
from a high-rate anaerobic treatment system. Water Res 1990;24:5836.
[13] Guwy AJ, Hawkes FR, Wilcox SJ, Hawkes DL. Neural network and on-off con-
trol of bicarbonate alkalinity in a uidised-bed anaerobic digester. Water Res
1997;31:201925.
[14] von Sachs J, Meyer U, Rys P, Feitkenhauer H. New approach to control the
methanogenic reactor of a two-phase anaerobic digestion system. Water Res
2003;37:97382.
[15] Alvarez-Ramirez J, Meraz M, Monroy O, Velasco A. Feedback control design for
an anaerobic digestion process. J Chem Technol Biot 2002;77:72534.
[16] Batstone DJ, Steyer JP. Use of modelling to evaluate best control practice for
winery-type wastewaters. Water Sci Technol 2007;56:14752.
[17] Strik DPBT, Domnanovich AM, Holubar P. A pH-based control of ammonia in
biogas during anaerobic digestion of articial pig manure and maize silage.
Process Biochem 2006;41:12358.
[18] Chynoweth DP, Svoronos SA, Lyberatos G, Harman JL, Pullammanappallil P,
Owens JM, Peck MJ. Real-time expert-system control of anaerobic-digestion.
Water Sci Technol 1994;30:219.
[19] Pu nal A, Roca E, Lema JM. An expert system for monitoring and diagnosis of
anaerobic wastewater treatment plant. Water Res 2002;36:265666.
[20] Pu nal A, Lorenzo A, Roca E, Hernandez C, Lema JM. Advanced monitoring of an
anaerobic pilot plant treating high strength wastewaters. Water Sci Technol
1999;40:23744.
[21] Moletta R, Escofer Y, Ehlinger F, Coudert J, Leyris J. On line automatic control
system for monitoring an anaerobic uidized bed reactor: response to organic
overload. Water Sci Technol 1994;30:1120.
[22] Ehlinger F, Escofer Y, Couderc JP, Leyris JP, Moletta R. Development of an
automatic-control system for monitoring an anaerobic uidized-bed. Water
Sci Technol 1994;29:28995.
[23] Garcia C, Molina F, Roca E, Lema JM. Fuzzy-basedcontrol of ananaerobic reactor
treating wastewaters containing ethanol and carbohydrates. Ind Eng ChemRes
2007;46:670715.
[24] Carrasco EF, Rodriguez J, Pu nal A, Lema JM. Rule based diagnosis and supervi-
sion of a pilot plant wastewater treatment plant using fuzzy logic techniques.
Expert Syst Appl 2002;22:1120.
[25] Pu nal A, Rodriguez J, Carrasco EF, Roca E, Lema JM. Expert system for the on-
line diagnosis of anaerobic wastewater treatment plants. Water Sci Technol
2002;45:195200.
[26] Pu nal A, Rodriguez J, Carrasco EF, Roca E, Lema JM. Advanced monitoring and
control of anaerobic wastewater treatment plants: diagnosis and supervision
by a fuzzy-based expert system. Water Sci Technol 2001;43:1918.
[27] Sousa R, Almeida P. Design of a fuzzy system for the control of a biochemical
reactor in fed batch culture. Process Biochem 2001;37:4619.
[28] Bernard O, Polit M, Hadj-Sadok Z, Pengov M, Dochain D, Estaben M, Labat P.
Advanced monitoring and control of anaerobic wastewater treatment plants:
software sensor and controllers for an anaerobic digester. Water Sci Technol
2001;43:17582.
[29] Muller A, Marsili-Libelli S, Aivasidis A, Lloyd T, Kroner S, Wandrey C.
Fuzzy control of disturbances in a wastewater treatment process. Water Res
1997;31:315767.
[30] Pu nal A, Palazzotto JC, Bouvier JC, Conte T, Steyer JP. Automatic control of
volatile fatty acids in anaerobic digestion using a fuzzy logic based approach.
Water Sci Technol 2003;48:10310.
[31] Wilcox SJ, Hawkes DL, Hawkes FR, Guwy AJ. A neural-network based
on bicarbonate monitoring, to control anaerobic-digestion. Water Res
1995;29:146570.
[32] Holubar P, Zani L, Hager M, Froschl W, Radak Z, Braun R. Advanced controlling
of anaerobic digestion by means of hierarchical neural networks. Water Res
2002;36:25828.
[33] Renard P, Vanbreusegem V, Nguyen MT, Naveau H, Nyns EJ. Implementa-
tion of an adaptive controller for the startup and steady-state running of a
biomethanationprocess operatedintheCstr mode. Biotechnol Bioeng1991;38:
80512.
[34] Dochain D, Perrier M, Pauss A. Adaptive control of the hydrogen concentration
in anaerobic digestion. Ind Eng Chem Res 1991;30:12936.
[35] Dochain D, Perrier M. Control Design for nonlinear wastewater treatment pro-
cess. Water Sci Technol 1993;28:28393.
[36] Seok J. Hybrid adaptive optimal control of anaerobic uidized bed bioreactor
for the de-icing waste treatment. J Biotechnol 2003;102:16575.
[37] Mailleret L, Bernard O, Steyer JP. Nonlinear adaptive control for bioreactors
with unknown kinetics. Automatica 2004;40:137985.
[38] Mailleret L, Bernard O, Steyer JP. Robust regulation of anaerobic digestion pro-
cesses. Water Sci Technol 2003;48:8794.
[39] Monroy O, Alvarez-Ramirez J, Cuervo F, Femat R. An adaptive strategy to
control anaerobic digesters for wastewater treatment. Ind Eng Chem Res
1996;35:34426.
[40] Alcaraz-Gonzalez V, Harmand J, Rapaport A, Steyer JP, Gonzalez-Alvarez V,
Pelayo-Ortiz Q. Robust interval-based regulation for anaerobic digestion pro-
cesses. Water Sci Technol 2005;52:44956.
[41] Grognard F, Bernard O. Stability analysis of a wastewater treatment plant with
saturated control. Water Sci Technol 2006;53:14957.
[42] Liu J, Olsson G, Mattiasson B. Control of an anaerobic reactor towards maximun
biogas production. Water Sci Technol 2004;50:18998.
[43] Liu J, Olsson G, Mattiasson B. Extremum-seeking with variable gain control for
intensifying biogas production in anaerobic fermentation. Water Sci Technol
2006;53:3544.
[44] Molina F, Garcia C, Roca E, Ruiz-Filippi G, Lema JM. Winery efuent treatment
at an anaerobic hybrid USBFpilot plant under normal and abnormal operation.
Water Sci Technol 2007;56:2531.
[45] Molina F, Garcia C, Roca E, Lema JM. Characterization of anaerobic gran-
ular sludge developed in UASB reactors that treat ethanol, carbohydrates
and hydrolyzed protein based wastewaters. Water Sci Technol 2008;57:
83742.
[46] Speece R. Anaerobic biotechnology for industrial wastewaters. Nashville, Ten-
nessee: Archae Press; 1996.
[48] Batstone DJ, Keller J, Angelidaki I, Kalyuzhny S, Pavlostathis SG, Rozzi A, Sanders
W, Siegrist H, Vavilin VA. Anaerobic Digestion Model No. 1 (ADM1) vol. Report
No. 13. London: IWA Publishing; 2002.
[49] Ruiz-Filippi G, Rodriguez J, Roca E, Lema JM. Modication of the IWA-
ADM1 for application to anaerobic treatment of ethanolic wastewater
from wine factories. In: Guiot S, Pavlostahis SG, van Lier JB, editors. 10th
IWA Congress on Anaerobic Digestion. Montreal: IWA Publishing; 2005.
pp. 13414.
[50] Bastin G, Dochain D. On-line estimation and adaptative control of bioreactors.
Amsterdam: Elsevier Science Publisher B.V.; 1990.
[51] Bernard O, Hadj-Sadok Z, Dochain D, Genovesi A, Steyer JP. Dynamical model
development and parameter identication for an anaerobic wastewater treat-
ment process. Biotechnol Bioeng 2001;75:42438.
[52] Gavala HN, Angelidaki I, Ahring BK. Kinetics and modeling of anaerobic diges-
tion process. Adv Biochem Eng Biotechnol 2003;81:5793.
[53] Steyer JP, Bernard O, Batstone D, Angelidaki I. Lessons learnt from 15 years of
ICA in anaerobic digesters. Water Sci Technol 2006;53:2533.
[54] CastellanoM, Ruiz-Filippi G, Gonzalez W, RocaE, LemaJM. Selectionof variables
using factorial discriminant analysis for the state identication of an anaerobic
UASBUAF hybrid pilot plant, fed with winery efuents. Water Sci Technol
2007;56:13945.
[55] Molina F, Castellano M, Garcia C, Roca E, Lema JM. Selection of variables for on-
line monitoring, diagnosis, and control of anaerobic digestion processes. Water
Sci Technol 2009;60:61522.
[56] Bernard O, Zakaria H, Dochain D. Software sensors to monitor the dynamics
of microbial communities: application to anaerobic digestion. Acta Biotheor
2000;48:197205.
[57] Moletta R. Winery and distillery wastewater treatment by anaerobic digestion.
Water Sci Technol 2005;51:13744.
[58] Kleerebezem R, van Loosdrecht MCM. Waste characterization for imple-
mentabon in ADM1. Water Sci Technol 2006;54:16774.
[59] Ripley LE, Boyle WC, Converse JC. Improved alkalimetric monitoring for
anaerobic-digestion of high-strength wastes. J Water Pollut Control Federation
1986;58:40611.
[60] Wolmarans B, DelHomme Ch. Start-up of a UASB efuent treatment plant on
distillery wastewater. Water SA 2002;28:638.
[61] Hess J. Modlisation de la qualit du biogaz par un fermenteur mthanogne
et stratgie de rgulation en vue de sa valorisation. University of Nice (France);
Ph.D thesis; 2007.

Potrebbero piacerti anche