E3S Web of Conferences 141, 01007 (2020)

https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

Input/output Linearization Control Technique for Anaerobic

Digestion Reactor with Recirculation
Sura Srisuddee1, Malinee Sriariyanun1, Chanin Panjapornpon2, and Atthasit Tawai1,*
1Chemical and Process Engineering Program, The Siringhorn International Thai-German Graduate School of Engineering, King
Mongkut’s  University  of  Technology  North  Bangkok,  Bangkok,  Thailand
2Department of Chemical Engineering, Faculty of Engineering, Kasetsart University, Bangkok, Thailand

Abstract. Anaerobic digestion (AD) process has been generally applied in factories for wastewater
treatment and energy recovery. For the AD processes, the continuously stirred tank reactor (CSTR) with
recirculation are typically applied in industries for methane production from wastewater treatment. Since
the recycle stream affects the inlet concentration of the reactor, the control performance of traditional PI
and PID controllers used to manipulate control actuators may be limited. Additionally, the process control
loops related to biochemical reactions are generally employed the operational experiences without process
dynamic consideration. A control system based on input/output (I/O) linearization control technique for an
anaerobic digestion (AD) process was developed in this work. The control system applied the concept of
the I/O linearization technique, which followed dynamic behaviors of the reactor with a two-step
(acidogenesis-methanogenesis) kinetic model. The volatile fatty acid (VFA) concentration was regulated
by manipulation of the dilution rate to achieve the requested trajectories. Control performances of the
closed-loop system were investigated by a simulation under servo and regulatory problems. The
simulation results showed that the developed control scheme successfully forced the controlled output to
achieve the desired set points and handled the introduced control problems.

1 Introduction operational experiences without process dynamic

consideration. Moreover, the recirculation of the AD
Nowadays, wastewater is an important problem to be process can strongly affect the inlet concentration that
handled in industries. An alternative way to promote a need to be considered in the control system
greener production and solve this problem is anaerobic development. The advanced control systems such
digestion (AD) process. The AD process has been input/output (I/O) linearization technique have been
generally applied for wastewater treatment and energy applied for chemical processes to regulate the process
recovery because it has ability to transform the organic with nonlinear behaviors. The process models,
matter into biomethane [1]. The application of the AD containing the kinetic expressions, are required to
process allow us to obtain benefits, compared with formulate the model-based control systems. Since the
traditional aerobic treatment: low sludge production, aim of this work was to design a control strucuture by
little energy requirement, minimal safety requirement applying the process model, the governing equations of
[2]. Normally, the AD process deals with two main the process need to be investigated. An example of AD
biochemical reactions to produce biogas, acidogenesis process in the continuous stirred tank reactor (CSTR)
and methanogenesis. The organic substrate, which system was presented in [2]. The dynamic models
normally characterized by chemical oxygen demand composed of two main biochemical reactions with a
(COD), is consumed by acidogenic bacteria and consideration of the total alkalinity and the total
converted into volatile fatty acids (VFA). After that, inorganic carbon in water. The growth rate of acidogenic
VFA will be used as substrate of methanogenic bacteria bacteria was  presented  by  Monod  kinetics’  model, while
for the cell growth and biogas production. As mentioned that of methanogenic bacteria was presented by Haldane
that the AD process is a complex biochemical reaction kinetics’ model. There was a number of research works
process that presents the uncertain and nonlinear proposed the control system for the AD processes.
behaviors, process control systems that can handle the Zarrad et al. [3] proposed a linear quadratic (LQ) model
nonlinear dynamic behaviors are required. Therefore, a to regulate the output at desired setpoint by manipulation
good design, analysis and evaluation of efficiency  of the of activated sludge and compared the control
control systems are needed for the AD processes. performance with the PI controllers. Mu et al. [4]
However, the traditional PI (proportional-integral) or proposed a multivariable linear quadratic tracking (LQT)
PID (proportional-integral-derivative) controller applied approach by using by-pass and recycle stream to control
to the AD processes are generally employed the the COD concentration in an upflow anaerobic sludge

*
Corresponding author: [email protected]
© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution
License 4.0 (http://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 141, 01007 (2020) https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

blanket (UASB) reactor. The controlled complex reaction  X 10 

of AD process, which described by ODE system, was the  0   0 
   0   0
challenging point in this work. Recently, the analytical  0    X2 
model predictive control (AMPC) based on the I/O  DZ in   0   0
F   ,Q
  ξ in
and Z 
linearization technique was applied for a UASB reactor S 0 
described by an PDE-ODE model [5,6].  DS1in   0 
 DS   0   10 
This work proposed an alternative control method for  2in     S2 
the AD process with a recycle stream. To formulate a  DCin   qC ( )   0
control structure, the I/O linearization technique was C 
applied to regulate the process variable at the desired
where ξ denotes the vector of state variables, ξ in is the
setpoints. The main objective was to control the VFA
concentration by manipulation of the dilution rate to the vector of variables of inlet stream, D and Q are the
achieve the desired target. An integral action was dilution rate and the gas flow rate of the reactor. F
combined as a part of the control system to eliminate the denotes the feed of reactor, K is a vector of the yield for
mismatch of process and model. This developed control substrate consumption or product formation. r  ξ  is the
system was then applied to the wastewater treatment
reaction rate of the reactor, y   y1 ,..., ym  denote the
T
process for the methane production. In section 2,
preliminaries of the mathematic models and I/O vector of controlled outputs, t  0,  is the time and h
linearization control definition were explained in this is a nonlinear function.
section. The description and mathematic modelling of
AD process were presented in section 3. The formulation The model in (1) was later used to investigate the
of the control system was presented in section 4. The process responses and to formulate a control strategy for
simulation results, which were evaluated with servo and the CSTR reactor with recirculation.
regulatory test were shown in section 5. And the last
section was conclusion that will be shown in section 6.
2.2 Input/output linearization control definition

2 Prilimnaries Consider the nonlinear system as the following form:

ξ f (ξ,
u), y h(ξ) (2)
2.1 Problem formulation

The system in equation (1) expressed the mathematical where u denotes the vector of manipulated yariables, y
model of a CSTR with recirculation for the AD process. denotes the vector of output variables and f is vector of a
The state variables were shown in the metrix as follow : nonlinear function. The relative order of the nonlinear
system in equation (2) is denoted by r1,…,  rm-1, rm where
ξ ri is the smallest integer [7] that  d ri yi / dt ri  / u  0
F  Dξ  Kr  ξ   Q
t (1)
as expressed by the following equation :
y  h ξ 
yi  h(ξ )
with the initial conditions
dyi  h ξ 
 h (ξ )
1
 
 (0)  ξ in
0
dt  ξ t 
d 2 yi  h1 ξ 
where    h (ξ )
2

 ξ t 
2
dt
(3)
 X1  10  
   
X2   0  d r 1 yi  h r  2 ξ 
 h (ξ )
r 1
Z    (ξ ) X 1   0   
ξ)  1
ξ   ,r ( dt r 1   ξ t 
  ,K
  
 S1    2 (ξ ) X 2    k1 0  d r yi  h r 1 ξ 
S   k  k     h (ξ, u )
r

 2   2 3
 dt r   ξ t 
C   k4 k5 

 D00000  The closed-loop responses of the process state

0D0000  variables were requested as the following linear form:
 
00D000 
D , (i D  1)ri 
νi (4)
000D00 
0000D0  where D is the differential operator (i.e. D=d/dt), ν i is
 
00000D  the compensated setpoint vectors and i is the tuning

2
E3S Web of Conferences 141, 01007 (2020) https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

parameter that can adjust the speed of the output governing equations from material balances that applied
response. the process model of [2] was shown as follow:

In order to formulate the controller, equation (3) and dX 1

 D  ( X 10  X 1 )  Y1  1 ( S1 )  X 1
(4) are applied. The manipulated input u can be written dt
in the compact form as: dX 2
 D  ( X 20  X 2 )  Y2  2 ( S 2 )  X 2
dt
u(t )   ( ξ , ν) (5)
dS1
D  (S10  S1 )  1 ( S1 )  X 1
dt
where ξ denotes the vector of the measured state dS 2
variable.  D  (S02  S2 )  c  1 ( S1 )  X 1  2 ( S2 )  X 2 (6)
dt
dZ
D  ( Z 0  Z )
3 Description and mathematic modeling dt
of anaerobic digestion process dC
 D  (C0  C )  q c ( S 2 , Z , C )  Y3  1 ( S1 )  X 1
dt
The biological reactions proposed by Bernard et al. [1]
 Y4  2 ( S 2 )  X 2
for macroscopic system of the anaerobic process can be
used to explain the general biological pathways. The
where X1, X2, S1, S2 are acidogenic bacteria,
two main pathways, acidogenesis and methanogenesis,
methanogenic bacteria, organic substrate and VFAs,
in the homogeneous continuous stirred tank reactor
respectively. The total alkalinity is shown by Z and the
(CSTR) can be shown as follow:
total inorganic carbon in water which consist of
dissolved carbon dioxide, bicarbonate and carbonate is
• Acidogenesis reaction with reaction rate r1=µ 1X1 shown by C.

k1S1 
1r
 X1  k2 S2  k4CO2 X 01 , X 02 , S 01 , S02 , C0 and Z0 are inlet concentrations
which are affected by the recycle stream and D (D=F/V)
• Methanogenesis reaction with reaction rate r2=µ 2X2 is dilution rate, The Monod kinetics for growths of
acidogenic bacteria (µ 1) and Haldane kinetics for
r
k3 S2 
2
 X 2  k5CO2  k6CH 4 methanogenesis bacteria (µ 2) are shown as follow:
1max  S1
where S1 represents the organic substrate characterized 1 ( S1 ) 
by COD. S2 is shown the total concentration of VFA. µ 1 Y1  ( K S 1  S1 )
and µ 2 are shown the specific growth rates of 2 max  S2 (7)
2 ( S2 ) 
acidogenesis and methanogenesis bacteria, respectively. S2
Y2  ( K S 2  S2  2
)
KI
A schematic of anaerobic digestion process with a
recycle stream was shown in Fig. 1 It was assumed that where µ 1max and µ 2max show the maximum growth rate of
the AD process containd 2 reaction steps and the bacteria the two groups of bacteria, Ks1 is the saturation constant
can be divided into two different groups with related with the organic substrate, and Ks2 and Ki2 are the
homogeneous characteristics. saturation and inhibition constant related with the VFAs,
respectively. The model parameters were listed in Table
1.
Table 1. Parameters of the AD process

Symbol Parameter Value

µ 1max Maximum acidogenic biomass 4.2 d-1
growth rate
Fig. 1. One stage of anaerobic digestion process µ 2max Maximum methanogenic 0.36 d-1
biomass growth rate
For the acidogenesis, the organic substrate (S1), Ks1 Half-saturation constant for S1 0.0023 g/L
characterized by COD, was consumed by acidogenic Ks2 Half-saturation constant for S2 2.3 mmol/L
bacteria (X1) to formulate the VFAs (S2) and carbon KI2 Inhibition constant for S2 66.67 mmol/L
dioxide as the products. For the methanogenesis, the y1, y2, Yield coefficients 0.11,0.00236,
y3, y4 50.6,343.6
methanogenic bacteria (X2) utilized the VFAs as the
c Stoichiometric coefficient of 16.95 mmol/g
substrates to produce methane and carbon dioxide gases. the conversion of S1 to S2
Introduction of the AD process in the CSTR system, the

3
E3S Web of Conferences 141, 01007 (2020) https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

At the outlet of AD process, there were two mixed In this study, the control objective was to regulate
gases which were CO2 and CH4. The outlet molar of CH4 effluent VFA concentration   of the CSTR with recycle
flow rate (qm) can be calculated by   stream at the desired set point by manipulating the
qm  Ym 2 ( S2 ) X 2   (8) dilution rate (D). From the material balance equation, it
And the outlet molar of CO2 flow rate (qc) can be was shown that the dilution rate affected on both COD
calculated by (S1) and VFA (S2), which were the time-varying
parameters and related to the methane production. The
qc kL a[C  S2  Z  K H PC ]
 (9) control structure of the proposed control system was
shown in Fig. 3 The control system consisted of an I/O
where the partial pressure of carbon dioxide gas was linearizing controller and a compensator, which expressed
shown by Pc that can be calculated by as an integral action.

   2  4 K H PT (C  S 2  Z ) 4.1. Input/output linearizing controller

Pc  (10)
2K H
To develop the I/ O linearizing control, the definition of
with relative order of (3) was applied. The control output, the
VFA concentration, which was represented by S2 was
Ym considered as a lumped variable. The controlled output
 C  S2  Z  K H PT  2 ( S2 ) X 2 (11) can be shown in linearizing form as
Y2 k L a
 DS2  S2 
ν (12)
where Kh, Pt and kLa showed the  Henry’s  constant,  total  
pressure and liquid/gas transfer rate, respectively. A where  is a tuning parameter [8]. By an application of
block flow diagram of Fig. 2 presented the CSTR system
with the process model. the process model of equation (4) in (12), the equation of
the control system can be solved. The compact form for
the feedback control can be shown as
u (t )   ( S2 , ν,  ) (13)

This proposed controller was used to calculate the

control action for each time step.

4.2. Integral action

For this proposed controller, the integral action was

applied as a compensator and combined to the controller
system. The compensator was used to handle the
mismatch between the process and mathematical model.
The equation of the proposed integral action was shown
as

Fig. 2. A diagram of AD model in the CSTR system   ( y  ysp )


(14)
ν S2, sp  

The state variables of this model involved the
substrates and microorganism that directly affected on where ղ  is an integral action. The proposed I/ O
CH4 and CO2 production, while total alkalinity and linearizing controller of equation (12) was combined
inorganic carbon affected on the gas-liquid equilibrium. with the integral action for the control system
development.
4 Formulation of control system for the
AD process

Fig. 3. The control structure of the proposed control system.

4
E3S Web of Conferences 141, 01007 (2020) https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

5 Simulation results 5.2. Control performance

The control performance of the developed control
5.1. Closed-loop responses system was evaluated by introducing the servo and
regulatory problems. For the servo test, it was applied
In this work, the nonlinear dynamic model of the CSTR when the control output achieved the desired set point.
reactor was simulated to investigate the process
responses under the closed-loop system of the developed From the initial set point, which was set at 1 mmol /L,
control strategy. The computer software such MATLAB then it was adjusted to be 2.5 mmol/L. The dilution rate
was applied. To promote the VFA formation by the of the reactor was adapted by the proposed controlle to
acidogenic bacteria, a minimum dilution rate of the force the control output to the new set point (Fig. 6 and
reactor was set to be 0.15 d-1 at the beginning of the 7,). The results showed that the developed controller can
operation. Then the VFA (S2) was controlled by regulate the control output to follow the desired
manipulating the dilution rate (D) that contain a recycle conditions asymptotically. For the PI controller, it
stream with R=10%. The feed stream was mixed with the presented the aggressive adjustment of the manipulated
recycle stream before feeding to the reactor, the inlet input, which caused strong fluctuations of the output.
concentration was continously updated during the
process operation. The control performance of the
proposed control scheme was compared   to the PI
controller with Kp = 0.12, KI = 0.15. The initial inlet
concentration of the reactor are S10  20 g/L, S20  40
mmol/L, X10  0.1 g/L, X 20  0.58 g/L, Z0 = 62 mmol/L
and C0 = 65 mmol/L. The tuning parameters of the I/O
linearization based control system were   0.4 and
  0.15. The simulation results of the closed-loop
system of the developed control strategy were compared
to the traditional PI controller. Figures 4 and 5 showed Fig. 6. Closed-loop response of the effluent of VFA concentration
the closed-loop responses of the controlled output (S2) under servo test.
and the manipulated input variable (D), respectively.
The results showed that both control systems
successfully force the controlled output at the desired set
point. For the proposed control, the output achieved to
the desired setpoint faster and presented less oscillation
than PI controller. Although, the inlet concentration
affected by the recycle stream led to the fluctuation of
the output stream concentration. It can be seen from the
process responses that the control systems attempted to
manipulate the output around the required target.
Fig. 7 Dilution rate corresponding to the servo test.

For achieving the set point, the controller

adjusted the dilution rate to balance the substrate for
the microorganisms. The results of regulatory tests
showed that the proposed control system had a higher
capability in handling the regulatory problem compared to
the PI controller (Fig. 8 and 9). It was clear that the
developed control scheme, which applied the
biochemical reaction model to formulate the controller
Fig. 4. Effluent of VFA concentration response of the closed-loop based on the I/O linearization technique with a
system. compensator, successfully forced the process to the
desired set points and stabilize the system under the
condition with disturbances.

Fig. 5. Dilution rate corresponding to the closed-loop system of

Fig. 4. Fig. 8. Closed-loop response of the effluent of VFA concentration
under regulatory test.

*
Corresponding author: [email protected]

5
E3S Web of Conferences 141, 01007 (2020) https://doi.org/10.1051/e3sconf/202014101007
RI²C 2019

the condition with disturbances. In addition, the

requirement of a few tuning parameters allow us to
easily implement the proposed control strategy to the AD
processes.
This research was funded by King Mongkut's University of
Technology North Bangkok. Contract no. KMUTNB-62-
KNOW-10

References
1. O. Bernard, Z. Hadj-Sadok, D. Dochain, A.
Fig. 9. Dilution rate corresponding to the regulatory test Genovesi, J.P. Steyer, Biotechnol. Bioeng. 75, 424-
438 (2001)
6 Conclusion 2. Z. Duan, MNC. Bournazou, C. Kravaris, Chem Eng
J. 327, 1102-1116 (2017)
The development of a control system based on the 3. W. Zarrad, J. Harmand, M.Devisscher, J.P. Steyer,
feedback I/O linearization technique for an AD process
Control Eng. Pract. 12, 323-333 (2004)
with recirculation was proposed in this study. To handle
the nonlinear behaviors of the reactor with biochemical 4. S. Mu, Y. Zeng, P.Wu, J. Chem. Technol.
reactions and the effects of the recycle stream, the Biotechnol. 83, 892-903 (2008)
controller applied the kinetic models with an updated 5. A. Tawai, C. Panjapornpon, M. Sriariyanun, K.
mixed stream concentration. Since methane production Cheenkachorn, IFAC-PapersOnLine. 51(28), 109-
was related to the VFA concentration, the control 114 (2018)
objective was focused on the regulation of the effluent 6. A. Tawai, C. Panjapornpon, P. Dittanet, Ind. Eng.
VFA concentration at the desired setpoint by Chem. Res. 55, 10121-10131 (2016)
manipulating the dilution rate. The controller was
7. P. Sanposh, W. Leenanithikul, S.
combined with an integral action to provide the
Phoojaruenchanachai, C. Panjapornpon, ECTI-CON
compensation of the process/model mismatch, and then
2008. 2, 613-616 (2008)
the developed control scheme was applied to the process.
Simulation results showed that the developed control 8. W. Chonwattana, C. Panjapornpon, A. Tawai,
system successfully forced the process to the desired Aided Chem. Eng. 108, 372-381 (2018)
targets and had a capability to stabilize the system under