You are on page 1of 9


discussions, stats, and author profiles for this publication at:

Regulation of the Organic Pollution Level in

Anaerobic Digesters by Using Off-Line COD

Article in Bioresource Technology May 2011

DOI: 10.1016/j.biortech.2011.05.053 Source: PubMed


14 88

5 authors, including:

Hugo Oscar Mndez-Acosta Juan Paulo Garca-Sandoval

University of Guadalajara University of Guadalajara


Vctor Gonzlez-lvarez Victor Alcaraz-Gonzlez

University of Guadalajara University of Guadalajara


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

Mexican Center for Innovation in Bioenergy View project

Waste2bioHy (FP7-PEOPLE-2012-Marie Curie action - IEF 326974) (Sustainable hydrogen production

from waste via two-stage bioconversion process: an eco-biotechnological approach. View project

All content following this page was uploaded by Hugo Oscar Mndez-Acosta on 21 April 2015.

The user has requested enhancement of the downloaded file.

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
Author's personal copy

Bioresource Technology 102 (2011) 76667672

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage:

Regulation of the organic pollution level in anaerobic digesters by using off-line

COD measurements
H.O. Mndez-Acosta , J.P. Garca-Sandoval, V. Gonzlez-lvarez, V. Alcaraz-Gonzlez, J.A. Juregui-Juregui
Departamento de Ingeniera Qumica, CUCEI-Universidad de Guadalajara, Blvd. M. Garca Barragn 1451, C.P. 44430, Guadalajara, Jal., Mexico

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

Article history: A sampled delayed scheme is proposed to regulate the organic pollution level in anaerobic digestion pro-
Received 4 March 2011 cesses by using off-line COD measurements. The proposed scheme is obtained by combining an error
Received in revised form 12 May 2011 feedback control with a steady state estimator to track constant references and attenuate process load
Accepted 18 May 2011
disturbances. The controller performance is tested experimentally for the treatment of tequila vinasses
Available online 27 May 2011
over a period of 68 days under different set-point values and several uncertain scenarios which include
badly known kinetic parameters and load disturbances. Experimental results show that the COD concen-
tration can be effectively regulated under the inuence of set-point changes and high load disturbances
Anaerobic digestion
Bioprocess optimization
by using only a daily off-line COD measurement, which makes the industrial application of the proposed
Sampled delayed control control scheme feasible.
Wastewater treatment 2011 Elsevier Ltd. All rights reserved.
Tequila vinasses

1. Introduction Over the past decade, the regulation of the organic pollution
level has been addressed by proposing different control algorithms
Anaerobic digestion (AD) has regained the interest of the waste- (Ahring and Angelidaki, 1997; Schogerl, 2001; Mndez-Acosta
water treatment scientic and industrial community to reduce the et al., 2004, 2005; Alcaraz-Gonzlez et al., 2005; Olsson et al.,
organic pollution level from industrial and municipal efuents be- 2005; Steyer et al., 2006). In fact, a summary of the advantages
cause of their low initial and operational costs, smaller space and drawbacks of several control schemes when applied to AD pro-
requirements, high organic removal efciency and low sludge pro- cesses has been recently reported by Steyer et al. (2006) who have
duction. In addition, AD has an extra energy benet through the shown that, most of these control laws have been based in contin-
production of biogas. However, its widespread application has uous control approaches that required continuous or, at least, fast
been limited because of the difculties involved in achieving the enough measurements (which can be approached to a continuous
efcient operation of these processes. This is why actual research measurement) in order to be implemented, a condition which is
aims not only to extend the application of AD, but also to optimize not always possible to satisfy in practice because of technological
and increase the robustness of these processes. Thus, the imple- or cost restrictions. The practical implementation of such control
mentation of appropriate, carefully designed and efcient control schemes is further compounded by the long time delays induced
strategies for AD processes has become a very important research by the sample processing (which include the off-line sampling rate,
area during the last years. sample preparation and laboratory analysis). Thus, a sampled de-
It is well known that, in order to guarantee the so-called oper- layed control scheme is devised here to cope with the aforemen-
ational stability (Hill et al., 1987) and to avoid the eventual break- tioned control problems, while taking into account all the system
down of AD processes, the organic matter in the liquid phase must information as it is available. The proposed control scheme is
be kept within a set of predetermined values, which depend on fac- designed and applied in the treatment of tequila vinasses to regu-
tors such as the reactor conguration and the characteristics of the late the organic pollution level measured as chemical oxygen de-
wastewater to be treated (Ahring and Angelidaki, 1997). However, mand (COD) by using the dilution rate as the control input. The
the complex nonlinear and nonstationary nature of the AD process, paper is organized as follows. First, the considered AD model used
the feed composition overloads, and the presence of toxic and in the controller design is briey described. Later, the control
inhibitory compounds enhance the control problems that are asso- scheme is proposed and tested experimentally in an up-ow
ciated to the regulation of the organic pollution level. xed-bed reactor (FBR) used for the treatment of tequila vinasses.
Then, the controller performance and its robustness are evaluated
Corresponding author. Tel.: +52 33 13785900; fax: +52 33 39425924. under different operating conditions and uncertain scenarios.
E-mail address: (H.O. Mndez-Acosta). Finally, some concluding remarks are given.

0960-8524/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
Author's personal copy

H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672 7667

Z d
2. Methods
Ad0 eA0 d ; Bd0 eA0 k dkB0 ; C d0 C 0 ;
2.1. AD model Z 1ed
Bd1;0 eA0 k dkB0 ; Bd2;0 Bd0  Bd1;0 3
In this work, a simplied version of the mass-balance model
proposed by Bernard et al. (2001) is used in the controller design. where A0 @f =@xx;u;kxss ;uss ;0 ; B0 @f =@ux;u;kxss ;uss ;0 and C 0
This model has been widely used for monitoring and control pur- @h=@xx;kxss ;0 . Then, from the robust regulation theory (Isidori,
poses due to its simplicity and its capability to represent the 1995; Garca-Sandoval et al., 2007) and by assuming that the pairs
dynamics of various continuous AD bioreactors. Then, without loss    
  Ad0 Bd0
of generality, the dynamics of the total soluble COD concentration Ad0 ; Bd1;0 and Ad ; C d Ad , where Ad and
0 I
in continuous AD processes operating under isothermal conditions
C d C d0 0 are controllable and observable, respectively; a
can be described by the following equations:
local solution for the proposed delayed output discrete robust reg-
X_ lST  aDX ulation problem can be obtained by using the following discrete
1 controller
S_ T ST;in  ST D  k lST X  
bn j nj  Gd1 C d0 nj  ej 4
where X (g/L) denotes the microorganisms population involved in  
bf j fj  Gd2 C d0 nj  ej 5
the AD process, D (1/d) is the dilution rate, while ST and ST;in
(gCOD/L) are the total soluble COD concentration at the process nj1 Ad0 bx j  Bd0 bf j Bd1;0 uj Bd2;0 uj1 6
efuent and inuent, respectively. lST (1/d) represents the spe-
cic growth rate of the bacterial population given by a Monod type
fj1 bf j 7
uj K d b
n j bf j
function and k (gCOD/g X) is the yield coefcient related to the deg- j 1; 2; 3; . . . 8
radation rate of the total soluble COD. Finally, a represents the frac-
tion of microorganisms that is not xed on the bed, and therefore, where uj is the discrete input applied to the nonlinear system (2)
may be affected by the dilution effect. Thus, it is evident that by set- between jd < t 6 j 1d, nj and fj represent the regulation error
ting a = 1, Model (1) describes the dynamics of a Continuous Stirred and the steady state input estimators, while b
n j and b
f j are their up-
Tank Reactor (CSTR) where the biomass is completely suspended in date values when the regulation error ej is available. K d and
the liquid phase. Moreover, it has been shown that with 0 < a < 1,  T
Gd GTd1 GTd2 are such that the matrices Ad0 Bd0 K d and
Model (1) can be successfully used to describe the dynamics of  
uidized-bed and xed-bed bioreactors (Escudie et al., 2005). Ad  Gd C d Ad are Schur (i.e., all its eigenvalues lie within the uni-
tary circle). Thus, given the existence of the symmetrical matrices
2.2. Controller design Q > 0 and M > 0, a sufcient condition to guarantee the stability
of the proposed discrete controller (4)(8) is that the following
2.2.1. Basic concepts Linear Matrix Inequalities (LMIs) hold
First, let us consider the following nonlinear system !
T M WT0 Q W1
W 0Q W0  Q 2M 6 0; >0 9
xt f xt; ut; k WT1 Q W0 M  WT1 Q W1
et hxt  s; r; k
where x 2 Rn and u 2 R are the state and input process variables, 0 1
respectively. k 2 Rp denotes an uncertain parameter vector which Ad0 Bd1;0 K d Bd1;0 K d Bd1;0
may take values in a neighborhood } 2 Rp of the system nominal W0 @ 0 Ad0  Gd1 C d0 Ad0 Bd0 Gd1 C d0 Ad0 A
values. r represents a constant reference signal, while e describes 0 Gd2 C d0 Ad0 1 Gd2 C d0 Ad0
the delayed regulation error given by the difference between the
0 1
system output and the reference signal, with s as a constant delay. Bd2;0 K d Bd2;0 K d Bd2;0
It is also considered that the output signal is acquired with a sam- B C
W1 @ 0 0 0 A
pling period, d, which may be bigger than the acquisition delay (i.e.,
0 0 0
s ed for 0 < e < 1). Then, the delayed output discrete robust reg-
ulation problem can be stated as follows. Find, if possible, a feed-
Eqs. (4) and (5) render the state controller updates at each sampling
back dynamic discrete controller with sampling period d, such
instant used to generate the control input (8), while by using (6)
that, for all admissible parameters values e and k around the system
and (7), it is possible to predict the next states of the controller. This
nominal values, the following conditions are met:
updating procedure takes advantage of the measurements as they
are readily available. The solution and existence of the LMIs (9)
Stability. The solution of the closed-loop system, without distur-
guarantee the system stability regardless of the output delay. Then,
bances but with parametric variations at the sampling
n 2 Rn and its update b n, represent an observer for x, while f 2 R and
instant goes asymptotically to zero.
its update b f are used to calculate the required steady-state input. A
Regulation. For each initial condition of both the process and the more detailed discussion about the results summarized in this
controller in a neighborhood of the origin, the solu- section can be found in the Appendix.
tion of the closed-loop system under the inuence
of disturbances and parametric variations must 2.2.2. Sampled delayed COD control
guarantee limt!1 et 0. By following the previously described ideas, a robust sampled
delayed control can be designed to regulate the organic pollution
Now, let us consider the following matrices Ad0 ; Bd0 and C d0 , level in AD processes by using off-line COD measurements. Thus,
which are the discretized nominal matrices of the linear approxi- from model (1), the linear matrices around the process nominal
mation of the nonlinear system (2), i.e., values are given by
Author's personal copy

7668 H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672

lST;r 1 k0 H k  0
 1 described in the following section. A remotely controllable peristal-
A0  ; B0 ST;in  ST;r 0 ; tic pump was connected to the dilution tank to ensure the desired
a H 0 1=k
C0 1 0 inuent ow rate. Fresh substrate was mixed with the recycled
  liquid just before entering the reactor in order to ensure homoge-
where H ST;in  ST;r k0KlS S2 and lST;r lm ST;r =K S ST;r . neous conditions. The process efuent was collected by overow
m T;r

Then, depending on the set-point value, matrices K d and Gd are cal- in a receiving vessel. The digester temperature was regulated at
culated in order to guarantee that matrices Ad0 Bd0 K d and 35  1  C by using an immersion circulator and water as heat trans-
  fer liquid which was conducted through the digester jacket.
Ad  Gd C d Ad are Schur and that the LMIs dened in (9) are
fullled. 2.3.3. On-line measurements
The digester was fully instrumented and automated, allowing
2.3. Experimental set-up the on-line measurement of variables such as pH, temperature,
pressure and the biogas and wastewater ow rates. A National
2.3.1. On the inuent characteristics Instruments cRIO9004 device equipped with analogical and digi-
Experiments were performed by using raw tequila vinasses as tal cards was used in the acquisition, treatment and storage of the
substrate. Tequila vinasses were collected from a cooling reservoir data. This device also included the appropriate ports and capabili-
in a tequila factory located at La Laja-Jalisco, Mexico. It is impor- ties that allow the remote monitoring and control of the process
tant to remark that around 95% of the total COD is soluble in this from an internet connection. The programming of this device
type of efuents. Additional characteristics of the tequila vinasses was carried out by using the LabVIEW 8.2 software.
used in this work are listed in Table 1.
2.3.4. Off-line measurements
Additionally to the previously described on-line measurements,
2.3.2. The AD process off-line measurements were also performed during this work. The
The schematic layout of the tequila vinasses treatment used in COD was determined by the closed reux colorimetric method by
this work is shown in Fig. 1. The AD process was carried out in an using the HACH digester DBR200 and the spectrophotometer
up-ow xed-bed digester, which was made of Polyvinyl Chloride DR2800. It is important to remark that the COD determined during
(PVC) with an effective volume of 2.8 L. The support used consists the experiments corresponds to the soluble COD because: (a) most
of three PVC tubes (30 cm length, 5 cm diameter) with an inner of the COD present in Tequila vinasses is soluble and (b) a possible
cross shaped plates that increase the surface area available for bio- contribution of washed out biomass in the COD measurement is
lm attachment. The rst part of the process is conformed by a 6 L eliminated. Partial (PA), total (TA) and intermediate (IA) alkalini-
dilution tank, where known volumes of tap water and raw vinasses ties as well as bicarbonate (B) were also determined according to
can be mixed in different proportions allowing the manipulation of Ripley et al. (1986). Volatile Fatty Acids (VFA) were measured with
the inuent COD concentration. At this point, the vinasses pH was a HPLC Watters 600 using a 2487-Absorbance detector with a wave
regulated between 6.5 and 7.0 by adding a NaOH solution through length of 204 nm and the Alltech colon OA-1000 which is special
an offon control scheme, which is computed by the control device for organic acids. HPLC operating conditions were the following:
mobile phase, sulphuric acid 0.01 N; mobile phase ow rate,
0.8 mL/min and column temperature, 60 C. These operating condi-
Table 1
tions allow the detection of acetic, propionic, butyric, isobutyric
Characteristics of the industrial tequila vinasses.
and valeric acids in a concentration range from 0 to 25 g/L by each
Component Value (g/L) acid. Finally, the biogas composition was determined by gas chro-
pH, 25 C 35 matography using a Perkin Elmer AutoSystem XL with a TCD detec-
Biochemical oxygen demand (BOD) 1324 tor and an Alltech packed colon Hayesep D 100/120. The operating
Total chemical oxygen demand (COD) 2850
conditions used by the chromatograph were the following: carrier
Total suspended solids (TSS) 12
Volatile suspended solids (VSS) 9.8 gas, nitrogen; pressure of 386 kPa; temperature of the injector,
Total Volatile Fatty Acids CH3 COOH 2.53.4 120 C; temperature of the detector, 100 C and oven temperature,
60 C. The samples used to determine the inuent composition

Fig. 1. Schematic view of the AD process used in the treatment of tequila vinasses.
Author's personal copy

H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672 7669

were taken from the dilution tank, while the efuent samples were
withdrawn from the recycling loop.

2.3.5. Controller implementation

The proposed control scheme was tested during 68 days in the
previously described AD process used for the treatment of Tequila
vinasses. Such a scheme was implemented by using a sampling
rate of 1 day and a sampling processing delay of 5 h which is the
time needed to determine the COD concentration of the sample
by the HACH methodology (i.e., d = 1 day and s = 5 h). The inlet
ow rate was used as manipulated variable q = DV, where
V is the effective bioreactor volume. A remotely controllable
peristaltic pump was connected to the dilution tank to ensure
the desired inuent ow rate, which was constrained in order to
avoid undesired effects such as the washout phenomena, i.e.,
q 2 14:5; 151:7 mL=h .
In order to show the controller response in the face of paramet-
ric uncertainties, the nominal values used during the experimental
run were those reported by Alcaraz-Gonzlez et al. (2005) for the
anaerobic treatment of red wine distillery vinasses: lm
1 0
1:30 d , K S 3:72 gCOD=L; k 86:2 gCOD=gX and a = 0.3. The
performance and robustness of the proposed control scheme were
evaluated in the face of different set-point changes and load distur-
bances. Thus, three set-point changes between 1.5 and 2.5 gCOD/L
were induced at time t = 0, 15 and 45 days.
On the other hand, since the composition of the tequila vinasses
Fig. 2. (a) Inuent soluble COD concentration. (b) Inlet ow rate computed by the
can change drastically from company to company and even from control scheme. (c) Response of the efuent soluble COD concentration under the
batch to batch in the same company, different uncertain load inuence of the sampled delayed control.
disturbances were induced by choosing a nominal inuent COD
concentration ST;r 30 gCOD=L, while diluted vinasses (50% vin-
asses50% water) were fed to the digester during the rst 30 days
practical approach because besides the considerable sampling
resulting in an inuent COD concentration around 20 gCOD/L.
time, it allows the use of delayed off-line measurements. Even
Finally, at day 30, additionally to the load disturbances induced
when the mathematical stability of this control law has been for-
by the composition change of the vinasses from batch to batch, a
mally developed in Section 2.2 and Appendix A, the excellent
bigger load disturbance was induced by feeding raw vinasses
results shown in Fig. 2c can be also explained under a practical
reaching an inuent COD concentration close to 35 gCOD/L.
point of view by the fact that the time scale for COD determination
is much less than the time for the process development in the AD.
3. Results and discussion Hence, there will be enough time for automatic feed-back (or even
manually by an operator) which makes this proposed controller
The dynamic response of the efuent COD concentration under suitable for small and medium size industries that often lack on-
the inuence of the sampled delayed control is depicted in Fig. 2c. line and expensive laboratory equipment.
Clearly, the proposed control scheme was able to regulate the COD Recent studies have highlighted the importance of preserving
concentration around the desired set-point values. Notice the the operational stability in AD processes (Mndez-Acosta et al.,
excellent performance of the proposed control scheme in both 2010). One of the most well known and accepted practical stability
cases, the set-point tracking and the load disturbance rejection, criteria for AD processes is that proposed by Ripley et al. (1986),
with a quite fast controller response regardless of the daily sam- who found that the successful digester operation occurs when
pling rate and the lack of the exact knowledge of the process nom- the ratio IA/PA, known as alkalinity factor, is less than 0:35 mEq=L .
inal values (i.e., parametric uncertainties). It is worth noticing that Fig. 3b depicts the behavior of the alkalinity factor during the
the controller needed just one sampling period to drive the process whole 68 day run. As seen, such alkalinity factor was kept below
close to the new reference after a set-point change was introduced the operational stability limit (see dotted-line) during most of
during the experimental run and just two sampling periods in or- the experimental run due to the control effort of the proposed
der to reject the load disturbance induced at day 30 by feeding raw scheme. The excellent features of the control scheme were fully
vinasses. tested when pure vinasses were fed to the digester at day 30. This
On the other hand, Fig. 2b illustrates the behavior of the manip- load disturbance drove the system towards instability for a rather
ulated variable, the inlet ow rate q, which never saturated neither short period of time but was corrected by the controller after only
when a set-point change was induced in the experimental run nor three sampling periods.
by the load disturbance caused by feeding pure vinasses to the Another important variable that is closely linked to the opera-
reactor. Although the control law was recalculated on a daily basis, tional stability of AD processes is the VFA concentration. Fig. 3a
the behavior of the inuent ow rate was quite smooth. This shows the response of the total VFA concentration measured as
behavior is a desirable feature from a practical point of view, in or- acetic acid (HAc). As expected, once the efuent COD concentration
der to guarantee safe operating conditions and increase the life- was regulated, the total VFA concentration was also regulated
time of the feeding pump. Furthermore, by taking into account since the total efuent COD concentration is mainly composed by
the technological and cost restrictions that may be involved for the VFA produced during the AD process and other than VFA solu-
having on-line COD measurements, the use of this discrete control ble COD. Because of the control action, the total VFA concentration
law for regulating the COD concentration represents a useful and was maintained at low values during the experimental run
Author's personal copy

7670 H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672

(a) (b)

(c) (d)

Fig. 3. Response of the (a) total VFA concentration, (b) alkalinity factor, (c) biogas production and (d) biogas composition under the effect of the proposed control scheme.

< 800 mgHAc=L preserving the AD operational stability. Note, in (1.5 gCOD/L). This result may be explained by the fact that the ap-
particular, at day 30 also when pure vinasses were fed to the plied organic loading rate (OLR) was higher during the rst set-
digester. As a result of this drastic load change, the VFA concentra- point (17.2 gCOD/L d) than that used in the third set-point
tion exhibited a sudden increase that affected the process stability. (12.9 gCOD/L d). Finally, the biogas composition is depicted in
Nevertheless, after two sampling periods, the control scheme was Fig. 3d. As seen, the methane composition was kept around 70%
able to reject this disturbance driving the total VFA concentration during the whole controller implementation but dropped to a low-
back to the previous safe conditions. These results also indicate er methane percentage when pure vinasses were fed to the diges-
the excellent adaptation of the microbial consortium in severe ter at day 30. However, the proposed controller was able to drive
and hard conditions and particularly during and after the strong the methane composition back to the previous conditions.
load disturbance at day 30, when the digester was fed with pure
4. Conclusions
Here, it is important to remark that although the proposed con-
troller was originally devised to regulate the COD concentration, it
A sampled delayed control scheme for the regulation of the
was capable to maintain the VFA concentration under stable condi-
organic pollution level in AD processes was proposed and experi-
tions preserving the so-called operational stability of the AD pro-
mentally evaluated in a FBR used for the treatment of tequila vin-
cess. However, the variable composition of the inuent vinasses
asses. The controller yields robustness in the face of parametric
and the complex nature of the AD process call for a detailed anal-
uncertainties, load disturbances, and variable set-points by using
ysis of the control problem prior to the design and development of
only a daily off-line COD measurement. The proposed scheme is
the proper controller. For instance, even under stable COD concen-
particularly encouraging to scale it up to industrial applications
tration conditions, one should pay attention to the excessive accu-
because: (a) its simple structure is easy to implement, (b) expen-
mulation of VFA which causes the biomass wash-out and the
sive on-line sensors are not required and (c) neither the exact
eventual breakdown of the bioreactor. One should expect further
knowledge of the inuent composition nor the process kinetics
process failures if the control design does not take into account
are also required.
the effect of inuent toxic and/or inhibitory compounds (even in
small quantities) on the process stability. Therefore, the proposal
and implementation of control schemes based in the here pro-
posed methodology but focused on the process stability will be
This work was supported by Projects 25927/CONACyT-CB/
an interesting case for future works (Mndez-Acosta et al., 2010;
J50282-Y, 116655/CONACyT-FORDECyT and PROMEP/103.5/08/
Ward et al., 2011).
To complete the picture, Fig. 3c shows the biogas produced un-
der the effect of the proposed robust control approach. It is well
known that the biogas production is a key variable to determine Appendix A. Mathematical results useful in the controller
the AD efciency and can be related to the COD and VFA degrada- design
tion. Notice that the biogas production was lower during the third
set-point change compared to that reached during the rst set- Let us consider the existence of mappings xss pr; l, and
point even when the efuent COD was regulated at the same value uss cr; l which solve the equations
Author's personal copy

H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672 7671

0 f pr; k; cr; l; l 10a where

0 hpr; k; r; l 10b Z Z
ds d

for all admissible values of l  }, with p0; l 0 and c0; l 0, Bd0;1 eA0 k B0 dk and Bd0;2 eA0 k B0 dk
0 ds
both dened in a neighborhood of the origin of r; l 0; 0. Notice
that pr; k denes the state vector for the zero output submanifold Thus, the discrete linear approximation of system (11)-(13 ) using
of system (2), while cr; l is the necessary input to make invariant input (8) is the following
this submanifold. Notice that this input depends only on the 
xd k 1 Ad0 xd k  Bd0 zd k Bd0;1 K d nd k
reference, r, and on the uncertain parameter vector; therefore, it    18
fd k Bd0;2 K d nd k  1 fd k  1
is constant and can be generated from the dynamic system
z_ 0; cr; l z. By dening
Now, if one considers the deviation of the steady state !  
xt xt  pr; k, by using the central manifold theory one may nd k xd k  nd k
#d k
nd a suitable mathematical model of system (2) which is given by fd k zd k  fd k
xt A0 xt  B0 zt B0 ut f2 xt; k 11 since x xd k and z
d k d k zd k, the dynamics of #d is given by
z_ t 0 12  
et C 0 xt  s h2 xt  s; r; k 13 #d k I  Gd C d #d k 19

where #d k 1 Ad #d k 20
@f @f @h where Ad and C d are dened in (3). Then, by taking (19) to the next
A0 ; B0 and C 0 sampling instant and by substituting (20), the following expression
@x 0;0 @u 0;0 @x 0
is obtained
while f2 and h2 contain the second or higher order terms, which  
vanish at the origin with their rst order derivatives. z 2 R repre- #d k 1 I  Gd C d Ad #d k 21
sents an immersion of the steady state input, cr; l. As can be seen,  
(11)(13) is an extended representation of system (2) since the Therefore, if Gd renders Schur the matrix Ad  Gd C d Ad , then

immersion (12), which renders the steady state input cr; k, was limk!1 #
d k 0, which implies that nd k and fd k of the proposed

also incorporated in the model. controller converge to xd k and zd k, respectively.

Let us recall that the discrete output is sampled with a period d To verify that the stability condition is fullled with the pro-

at time t kd, for k 0; 1; 2; . . ., but is available after a delay s ed posed controller, let us dene #
d k col xd k; nd k; fd k . Thus,
at time tk k ed, i.e., from (18) and (19) it holds that

et k hxkd; r; l : ed k #d k 1 W0 #d k W1 #d k  1 22

then at time t k , it is possible to estimate the values of xkd where W0 and W1 are described in Section 2.2.1. Notice
 that K d and 
xt k  s and zkd zt k  s by using ed k in a discrete observer. Gd render Schur the matrices Ad0 Bd1;0 K d and Ad  Gd C d Ad ,
The discretized version of the linear approximation of (11)-(13) respectively. Then, W0 is also Schur and by dening the Lyapunov
from one sampling instant to the next (i.e., from t kd to function
t k 1d) is
V #T T
d kP#d k #d k  1Q #d k  1
Z d
xd k 1 Ad0 xd k  Bd0 zd k eA0 k B0 ukd d  kdk 14 the increment of V is given by
d k 1P#d k 1  #d kP  Q#d k
zd k 1 zd k 15
d k  1Q #d k  1
ed k C 0 xd k 16 T T
#T T
d kW0 P W0  P Q#d k #d k  1W1 P W0 #d k
where xd k xkd, zd k zkd, T T
#T T
d kW0 P W1 #d k  1 #d k  1W1 P W1 #d k  1
Z d
Ad0 eA0 d ; Ud eUd and Bd0 eA0 k B0 dk 17 Thus, if the rst LMI in (9) holds, the following inequality also holds
0 !T
#d k
Notice that in order to calculate the integral in (14), it is necessary DV  
to know the behavior of the input from t kd to t k 1d. Thus, #d k  1  #d k
! !
if the controller (4)(8) is used, it is evident that (4) and (5) perform Q WT0 PW1 #d k
the updating values of nd and fd at time t k k ed when the out-
WT1 PW0 Q  WT1 PW1 #d k  1  #d k
put is available. Therefore, for kd  t < k 1d the input (8) is
equal to Furthermore, if the second LMI of (9) is satised, it can be guaran-
( teed that DV < 0. As a consequence, system (22) is asymptotically
K d nd k  1 fd k  1 if kd  t < kd s stable, guaranteeing that the proposed discrete controller (4)(8)
K d nd k fd k if kd s  t < k 1d stabilizes the nonlinear system (2) in a neighborhood of the origin.

because at time t < t k the discrete controller state has not been up- References
dated. Replacing this input, the integral in (14) becomes
Alcaraz-Gonzlez, V., Harmand, J., Rapaport, A., Steyer, J.P., Gonzlez-lvarez, V.,
Z d Pelayo-Ortiz, C., 2005. Robust interval-based regulation for anaerobic digestion
A0 k
e B0 ukd d  kdk Bd0;1 K d nd k Bd0;1 fd k processes. Water Sci. Technol. 52, 449456.
0 Ahring, B., Angelidaki, I., 1997. Monitoring and controlling the biogas process. In:
Bd0;2 K d nd k  1 Bd0;2 fd k  1 Proceedings of the 8th International Conference on Anaerobic Digestion, vol. 1,
pp. 4049.
Author's personal copy

7672 H.O. Mndez-Acosta et al. / Bioresource Technology 102 (2011) 76667672

Bernard, O., Hadj-Sadok, Z., Dochain, D., Genovesi, A., Steyer, J.P., 2001. Dynamical Mndez-Acosta, H.O., Palacios-Ruiz, B., Alcaraz-Gonzlez, V., Gonzlez-lvarez, V.,
model development and parameter identication for anaerobic wastewater 2010. A robust control scheme to improve the stability of anaerobic digestion
treatment process. Biotechnol. Bioeng. 75, 424438. processes. J. Process Control 20, 375383.
Garca-Sandoval, J.P., Castillo-Toledo, B., Gonzlez-lvarez, V., 2007. Control of a Olsson, G., Nielsen, M.K., Yuan, Z., Lynggaard-Jensen, A., Steyer, J.P., 2005.
bioreactor with sampled delayed measurement. In: Proceedings of the 10th Instrumentation, Control and Automation in Wastewater Systems. IWA
International Symposium on Computer Applications in Biotechnology. IFAC, Scientic Technical Report, No. 15, IWA Publishing.
Cancn, Mxico, pp. 315320. Ripley, L.E., Boyle, W.C., Converse, J.C., 1986. Improved alkalimetric monitoring for
Escudie, R., Conte, T., Steyer, J.P., Delgenes, J.P., 2005. Hydrodynamic and biokinetic anaerobic digestion of high-strength wastes. J. Water Pollut. Control Federation
models of an anaerobic xed-bed reactor. Process Biochem. 40, 23112323. 58, 406411.
Isidori, A., 1995. Nonlinear Control Systems, third ed. Springer, London. Schogerl, K., 2001. Progress in monitoring, modeling and control of bioprocesses
Hill, D., Cobbs, S., Bolte, J., 1987. Using volatile fatty acid relationships to predict during the last 20 years. J. Biotechnol. 85, 149173.
anaerobic digester failure. Trans. ASAE 30, 496501. Steyer, J.P., Bernard, O., Batstone, D., Angelidaki, I., 2006. Lessons learnt from 15
Mndez-Acosta, H.O., Femat, R., Campos-Delgado, D., 2004. Improving the years of ICA in anaerobic digesters. Water Sci. Technol. 53, 2533.
performance on the chemical oxygen demand regulation in anaerobic Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.J., 2011. Evaluation of near infrared
digestion. Ind. Eng. Chem. Res 43, 95104. spectroscopy and software sensor methods for determination of total alkalinity
Mndez-Acosta, H.O., Campos-Delgado, D., Femat, R., Gonzlez-lvarez, V., 2005. A in anaerobic digesters. Bioresour. Technol. 102, 40834090.
robust feedforward/feedback control for an anaerobic digester. Comput. Chem.
Eng. 29, 16131623.

View publication stats