Anaerobic tapered fluidized bed reactor for starch wastewater treatment and modeling using multilayer perceptron neural network.
ABSTRACT Anaerobic treatability of synthetic sago wastewater was investigated in a laboratory anaerobic tapered fluidized bed reactor (ATFBR) with a mesoporous granular activated carbon (GAC) as a support material. The experimental protocol was defined to examine the effect of the maximum organic loading rate (OLR), hydraulic retention time (HRT), the efficiency of the reactor and to report on its steady-state performance. The reactor was subjected to a steady-state operation over a range of OLR up to 85.44 kg COD/(m3 x d). The COD removal efficiency was found to be 92% in the reactor while the biogas produced in the digester reached 25.38 m3/(m3 x d) of the reactor. With the increase of OLR from 83.7 kg COD/(m3 x d), the COD removal efficiency decreased. Also an artificial neural network (ANN) model using multilayer perceptron (MLP) has been developed for a system of two input variable and five output dependent variables. For the training of the input-output data, the experimental values obtained have been used. The output parameters predicted have been found to be much closer to the corresponding experimental ones and the model was validated for 30% of the untrained data. The mean square error (MSE) was found to be only 0.0146.
[show abstract] [hide abstract]
ABSTRACT: Anaerobic digestion is a multistep process involving the action of multiple microbes. In order to be able to design and operate anaerobic digestion systems efficiently, appropriate models need to be developed. Several Mathematical models have been introduced which suffer from lack of knowledge on constants, complexity and weak generalization. Novel techniques to provide correlation between the affecting factors and production criteria of reactors have been reported to be robust, simple and fast enough for control applications and on-line industrial implementations. In this paper, artificial neural networks (ANN), genetic algorithms (GA) and Fuzzy systems are reviewed. ANN models have been extensively used and gained a considerable attention among the researchers. However, integration of GA and Fuzzy systems looks extremely promising for the industrial fields in future. In addition, the advantageous and practical applications of these models for wastewater treatment are also fully discussed.
Journal of Environmental Sciences 19(2007) 1416–1423
Anaerobic tapered fluidized bed reactor for starch wastewater treatment and
modeling using multilayer perceptron neural network
RANGASAMY Parthiban1,∗, PVR Iyer1, GANESAN Sekaran2
1. Department of Chemical Engineering, Sri Venkateswara College of Engineering, Sriperumbudur 602 105, Tamilnadu, India.
2. Department of Environmental Technology, Central Leather Research Institute, Chennai 600 025, Tamilnadu, India
Received 6 March 2007; revised 21 May 2007; accepted 3 June 2007
Anaerobic treatability of synthetic sago wastewater was investigated in a laboratory anaerobic tapered fluidized bed reactor (ATFBR)
with a mesoporous granular activated carbon (GAC) as a support material. The experimental protocol was defined to examine the effect
of the maximum organic loading rate (OLR), hydraulic retention time (HRT), the efficiency of the reactor and to report on its steady-
state performance. The reactor was subjected to a steady-state operation over a range of OLR up to 85.44 kg COD/(m3·d). The COD
removal efficiency was found to be 92% in the reactor while the biogas produced in the digester reached 25.38 m3/(m3·d) of the reactor.
With the increase of OLR from 83.7 kg COD/(m3·d), the COD removal efficiency decreased. Also an artificial neural network (ANN)
model using multilayer perceptron (MLP) has been developed for a system of two input variable and five output dependent variables.
For the training of the input-output data, the experimental values obtained have been used. The output parameters predicted have been
found to be much closer to the corresponding experimental ones and the model was validated for 30% of the untrained data. The mean
square error (MSE) was found to be only 0.0146.
Key words: anaerobic digestion; tapered fluidized bed reactor; organic loading rate; biogas; mesoporous granular activated carbon;
modeling; artificial neural network
Water is one of the most important resources used by
mankind. Due to the increased industrial activities and
urbanization, availability of good quality water is dimin-
ishing day by day. Purification and recycling of industrial
waste waters have become imperative in view of reduced
availability and deteriorating water quality. The most com-
mon used systems for biological wastewater treatment are
the activated sludge system, biological-film system (trick-
ling filters) and anaerobic fluidized bed reactors (AFBR).
AFBRs were originally a chemical engineering tool used
to perform phase transformations, reactions, and diffu-
sions of various chemicals existing in solid, liquid, and
vapor phases. With the concept of maximum diffusion and
maximum chemical reaction within a minimum volume
in mind, AFBRs have been adapted to perform biological
wastewater treatment and are utilized in several process
configurations (Hickey et al., 1991; Barascud, 1992; Iza,
1991; Perez et al., 1997b, 1999, 2001; Seckler et al., 1996;
Souza et al., 2004).
The AFBRs containing immobilized micro organisms
are more efficient than other types of reactors. Due to the
large available surface area of the film, the time required
*Corresponding author. E-mail: firstname.lastname@example.org.
for treatment is drastically reduced as compared to that in
conventional processes such as trickling filters or activated
sludge processes. The volume of the reactor is extremely
small as the surface area of unit volume is much larges
than in any other biological reactor.
The superior performance of fluidized bed bioreactor
is: (1) high biomass concentration can be achieved due to
immobilization of cells onto or into the solid particles; (2)
the limit on the operating liquid flow rates imposed by the
microbial maximum specific growth rate as encountered
in the continuous stirred tank reactor system without solid
phase, is eliminated due to the decoupling of the residence
time of the liquid phase and of the microbial cells
The conventional fluidized-bed reactor has a severe
drawback of the application in its uniform cross-sectional
velocity. Wash-out of bio-particles occurs frequently if the
overcome this drawback, Scott and Hancher (1976) mod-
ified the geometrical circular fluidized bed configuration
(identical cross sectional area) along its length to a tapered
form. Hence, the incorporation of tapered portion in the
bottom portion of the AFBR is the most viable alternative.
By configuration it resembles an inverted truncated cone
rather than a constant cross-sectional column. Thus, there
is a gradual expansion from a relatively small cross-
No. 12Anaerobic tapered fluidized bed reactor for starch wastewater treatment and modeling······
sectional area of the entry to one that may be several times
larger. If the entry cross section is sufficiently small and
the expansion is gradual (an angle of few degrees), the
flow should be relatively stable throughout the reactor. As
there is a natural gradient of up flow velocity, it results in
the lowering of superficial velocity which allows a perfect
to stabilize the bed for a wider range operating conditions.
The gradual expansion of the column results in a stable
feed introduction without gross eddies and significant back
mixing. The slope of the reactor is very important, since it
can cause the particles to get stuck together and produce a
spouted bed, with a high turbulence and shear area close
to the bottom of the reactor. The COD removal efficiencies
for taper angles of 5◦and 10◦tended to be higher than at
a taper angle of 2.5◦. The experimental substrate removal
efficiencies for taper angles of 5◦and 10◦were nearly
the same. However, the efficiency in the removal of COD
in fluidized bed reactors is limited by the taper angle
(Boening and Larsen, 1982; Wu and Huang, 1995, 1996;
Huang and Wu, 1996).
This article reports and discusses a laboratory investi-
gation that evaluated the TAFBR as a means of treatment
of sago industrial wastewater in Salem District of South
India, which is processed from the tubers of tapioca. Pro-
cessing of tapioca requires 20000 to 30000 L of water per
ton of Sago; besides it produces equal quantity of wastew-
ater, which is highly organic, foul smelling and acidic
a tapioca processing factory discharges the wastewater
containing 11000–13500 mg O2/L in terms of COD, 4200–
7600 mg suspended solids/L and pH 4.5–5.0. Studies
by Mai et al. (2004) and Oanh et al. (2001) on large-
scale tapioca processing companies give similar tapioca
7000–41406 mg/L, a BOD5of 6200–23077 mg/L. Various
anaerobic technologies including conventional anaerobic
treatment (Sastry et al., 1964; Tongkasane, 1970; Saroja
and Sastry, 1972; Pescod and Thanh, 1977), high rate
anaerobic treatment such as anaerobic filter (Khageshan
and Govindan, 1995), Hybrid UASB (Rajesh Banu et al.,
2006) and Fluidized Bed (Saravanane et al., 2001) have
been attempted to treat Sago wastewater.
Computation through ANN is one of the recently grow-
ing areas of artificial intelligence. Neural networks are
promising due to their ability to learn highly non-linear
relationship. Wasserman (1989) defined ANN model as
a computing system made up of a number of simple,
highlyinterconnectednodes orprocessingelements, which
processes information by its dynamic system response to
Several applications of artificial neural networks for
modeling of nonlinear process system and subsequent
control have been reported in literature (Parthiban et al.,
2007; Bhat and McAvoy, 1990; Kumar and Roy, 2004). In
the present case, a software package for artificial neural
networking using Multilayer perceptrons has been devel-
oped and used using MATLAB 7 environment. MLPs are
layered feed forward networks typically trained with static
back propagation algorithm which is based on minimizing
the sum of squared errors between the desired and actual
outputs. These networks have found their way into count-
less applications and ability to perform non linear pattern
classification and function approximation.
1 Materials and methods
1.1 Experimental set up
A schematic diagram of the experimental set up is
shown in Fig.1. The TAFBR consists essentially of conical
shaped acrylic column of 5◦taper angle with a total
volume of 7.8 L and the volume of tapered section is
1.5 L. The reactor column itself had a height 290 mm
with an increase in diameter from 46.6 mm at the base
to 91.5 mm at the top, and attached to this was an upper
settling section which was 1.073 m high and 91.5 mm
diameter. A static bed volume of 500 ml of mesoporous
GAC were used as a biomass carrier. The effluent was
recycled from the top to the bottom of the reactor using a
magnetic driven polypropylene centrifugal pump operated
at a constant rate enough to provide complete fluidization
conditions in the reactor. The settlement zone of the reactor
contained a conical gas liquid separator to allow venting
of the biogas produced. Sampling ports were provided
along the column length to obtain bed samples. Influent
was pumped in continuously at the bottom of the reactor
by means of a peristaltic pump initially for low flow rates
and for higher flow rates it was pumped from a magnetic
driven polypropylene centrifugal pump and effluent was
withdrawn from the top. The effluent discharge output port
Fig. 1 Schematic diagram of an anaerobic tapered fluidized bed reactor
and the gas liquid separator.
1418RANGASAMY Parthiban et al.Vol. 19
is located in the cylindrical section at a point below 55
mm from the top of the column. Biogas produced from
the reactor was collected by a 20-L displacement jar which
contains 10% NaOH solution initially when the gas flow
rate is less and when it exceeded a rate of 16 L/d it was
measured using a wet gas meter.
The wastewater used throughout the study was a syn-
therange 1100–7000 mg/Land wasanalyzed followingthe
standard methods (APHA etc., 2005). The characteristics
are presented in Table 1.
1.3 Support material
used as a growth support material because of its ability
to readily attach methanogenic bacteria. Further more,
some of the disadvantages of anaerobic treatment process
viz., the biological inhibition, and slow growth rate of
anaerobes; have been solved using the GAC as the support
media. The main characteristics of this carrier are given in
1.4 Start-up of the anaerobic fluidized-bed reactor
The reactor was initially filled for 1/3 the volume of
the reactor (500 ml) with mesopores GAC, and then 7 L
of supernatant liquid of the UASB reactor from Central
Leather Research Institute, Chennai, which treating the
municipal wastewater and the remaining volume with the
sago feed of 1100 mg/L. Then the reactor was operated in a
total recycle fashion with initial bed expansion maintained
at 30%. Sodium bicarbonate was added if necessary to
maintain the reactor pH in the range of 6.8–7.2. Diluted
synthetic sago was added to the reactor each day to
promote and sustain the growth of biofilm on the carbon
particles. After 45 d operation in this fashion there was
90%of COD removalwhichensures a complete adaptation
to the wastewater used.
1.5 Sampling and analysis
During the operation of the fluidized bed reactor, tem-
perature, pH for inlet and outlet waste water, the influent
and effluent COD concentrations, biogas production rate
and effluent total volatile fatty acids were monitored
daily. The total VFA composition was measured by a
Gas Chromatograph. The composition was found to be
acetic acid, butyric acid and propionic acids. All analytical
determinations were performed according to the standard
methods (APHA etc., 2005).
1.6 Reaction operation
The experimental protocol was designed to examine the
effect of the organicloading rate (OLR) on the efficiencyof
the tapered fluidized bed reactor (TFBR). The TFBR was
subjected to an operation of 535 d overa range of hydraulic
retention time (HRT) of 1.97–26.74 h. Initially it is oper-
ated for an OLR of 1 kg/(m3·d) and gradually increased
to 85.44 kg/(m3·d) with the optimum superficial velocity
(2.5 Umf) which gives the maximum COD removal. The
COD concentrations were varied in the range of 1100–
The start up period of the reactor was found to be 45
d. The attainment of the steady state was verified after
an initial period by checking whether the constant effluent
were the mean of the last measurements in each stage.
1.7 Development of multilayer perceptron model
Multilayered perceptron is a supervised neural network
which consists of multiple layers of processing elements
(PE) connected in a feed forward fashion. It consists of
three types of layers: input, hidden, and output. It has
a one-directional flow of information, generally from the
input layer, through hidden layer, and then to the output
layer, which then provides the response of the network to
the input stimuli. In this type of network, there are gen-
erally three distinct types of neurons organized in layers.
The input layer contains as many neurons as the number of
input variables. The hidden neurons, which are contained
Table 1 Characteristics of the sago wastewater
Alkalinity (as CaCO3, mg/L))
Total solids (mg/L)
Total dissolved solids (mg/L)
Total suspended solids (mg/L)
Volatile suspended solids (mg/L)
Total phosphorus (as P, mg/L)
Kjeldahl nitrogen (as N, mg/L)
Table 2 Characteristics of granular activated carbon (GAC) support material
Parameter ValueParameter Value
Surface area (BET) (m2/g)
Micropore surface area (m2/g)
Mesopore surface area (m2/g)
Micropore volume (Vmicro) (cm3/g)
Mesopore volume (Vmeso) (cm3/g)
Total pore volume(Vtotal) (cm3/g)
Average pore diameter (µm)
Ash content (%)
Bulk density (g/ml)
No. 12 Anaerobic tapered fluidized bed reactor for starch wastewater treatment and modeling······
in one or more hidden layers, process the information and
encode the knowledge within the network. The hidden
layer receives, processes, and passes the input data, to the
output layer. The selection of the number of hidden layers
and the number of neurons within each affects the accuracy
and performance of the network. The output layer contains
the target output vector. The architecture of multilayer
perceptron is shown in Fig.2.
A weight coefficient is associated with each of the
connections between any two neurons inside the network.
Information processing at the neuron level is done by an
“activation function” that controls the output of each one.
Artificial neural networks (ANNs) train through adaptation
of their connection weights based on examples provided in
a training set. The training is performed iteratively until
the error between the computed and the real output over all
training patterns is minimized. Output errors are calculated
by comparing the desired output with the actual output.
Therefore, it is possible to calculate an error function that
is used to propagate the error back to the hidden layer
and to the input layer to modify the weights. This iterative
procedure is carried out until the error at the output layer is
reduced to a pre specified minimum or for a pre specified
number of epochs. A plot of epoch vs MSE of learning
curve is shown in Fig.3. Error correction learning works
in the following way: from the system response at PE i at
iteration n, yi(n), and the desired response di(n)for a given
Fig. 2 Architecture of multilayer perceptron neural network.
Fig. 3 Learning curve. T stands for “Training” and CV stands for “Cross
input pattern an instantaneous error ei(n)is defined by:
ei(n) = di(n) − yi(n)
Using the theory of gradient descent learning, each
weight in the network can be adapted by correcting the
present value of the weight with a term that is proportional
to the present input and error at the weight, i.e.,
wij(n + 1)= wij+ ηδi(n)xj(n)
The local error δi(n)can be directly computed from ei(n)
at the output PE or can be computed as a weighted sum of
errors at the internal PEs. The constant η is called the step
size. The back-propagation algorithm is most commonly
used for training multilayer perceptron (MLP) and is based
on minimizing the sum of squared errors between the
desired and actual outputs (Bose and Liang, 1998; Fu,
1.8 Multilayer perception algorithm
The algorithm for the multilayer perception is shown
below. It requires the units to have thresholding non linear
functions that are continuously differentiable, i.e., smooth
everywhere. A sigmoid function f(net) = 1/(1 + e−knet),
is used, since it has a simple derivative. All training and
testing data were normalized.
Initialize weights and thresholds: set all weights and
thresholds to small random variable.
Present input and desired output:
Present input: XP = x0, x1, x2,............., xn−1
Target output: TP = t0,t1,.........,tn−1
where, n is the number of input nodes and m is the number
of output nodes. Set w0to be −θ , the bias, and x0to be
always 1. For pattern association, Xpand TPrepresent the
patterns to be associated. For classification, TPis set to
zero except for one element set to 1 that corresponds to the
class the Xpis in.
Calculation of actual output: each layer calculates
layer. The final layer outputs values opjand passes that as
input to the next layer.
Adapt weights: start from the control layer, and work
wij(t + 1) = wij(t) + ηδpjopj
where, wij(t) represents the weights from node i to node
j at time t, η is a gain term, and δpjis an error term for
pattern p on node j.
Output units: δpj = kopj(1 − opj) (tpj− opj)
Hidden units: δpj = kopj(1 − opj)?
where, the sum is over the k nodes in the layer above node
The stopping condition may be weight change, number
of epochs, and so on.
and passes that as input to the next
1420RANGASAMY Parthiban et al.Vol. 19
2 Results and discussion
2.1 Influence of OLR on COD removal and the biogas
The loading regime and COD removal of the reactor
during the entire operation of 535 d are presented in the
Fig.4a. The reactor was started with an initial OLR of 1
kg/(m3·d). Low OLR was preferred to prevent the washout
of inoculated biomass (Hickely et al., 1991). Initially, the
study was done ranging from a flow rate of 7 to 16 L/d
with various influent concentrations 1100 to 7000 mg/L till
the end of 124 d (OLR of 14.4 kg/(m3·d)). From day 125
onwards the influent COD concentration was kept constant
and the flow rate was increased gradually by varying the
HRT. The increment between two successive OLR was 1
kg/(m3·d) and kept constant for 3–5 d for the stabilization
of the reactor. It was observed from Fig.4a that the COD
removal was increased with time of operation. This was
in general agreement with investigations of Chen et al.
(1988) and Saravanane et al. (2001) on continuous flu-
idized bed reactor fed with corn starch wastewater, and
Sreekrishnan et al. (1991) on the synthetic glucose as
substrates. The start up phase was found to be 48 d. When
the OLR was increased in a stepped manner upto 85.44
kg/(m3·d) over a period of 535 d, the biogas production
also increased gradually reaching a maximum of 200 L/d
which corresponds to 25.36 m3/(m3·d) for an OLR of 83.7
kg COD/(m3·d) (Fig.4b), resulting in the COD reduction
(Fig.4a). The production of biogas after the start up period
for an OLR of 1 kg/(m3·d) was found to be 0.3125
m3/(m3·d) of reactor. When ever there is a change in OLR
the system gets disturbed resulting in a reduction of the
COD removal efficiency. When the reactor is operated at
a particular OLR for 2–3 d the removal efficiency reaches
the maximum. Upto about 0.330–0.345 L of methane were
produced per gram COD removed which compares with a
theoretical quantity of 0.35 L of methane when the starting
compound is glucose (Lawrence and McCarty, 1969). It is
Fig. 4 OLR pattern and COD removal (a) and biogas formation (b) during
the entire operation of the reactor.
evident from Fig.4a that the COD removal was maximum
at 92% for an OLR up to 83.7 kg/(m3·d) and beyond 83.7
kg/(m3·d) the COD removal and the biogas production
2.2 Variation of volatile fatty acids
Figure 5 presents the COD removal and VFA con-
sumption pattern during startup period. The COD removal
increased with time; this is in conformity with the findings
of Saravanane et al. (2001) during the treatment of Sago
wastewater. The total VFAs increased from the day 1
to 18. Higher levels of VFA in the wastewaters during
the initial phases of operation indicate the prevalence of
acid fermentation (Van Hanndel and Lettinga, 1994). Then
there is a rapid decrease in the total VFAs after day 18
indicating healthy anaerobic environment and satisfactory
methanogenic activity. The overall performance of the
reactor during the start-up was more than satisfactory.
During the stable operational phase of the reactor, the total
VFA levels in the medium varied from 100 to 210 mg/L.
VFA started building up in the wastewater as the digestion
proceeded and the maximum concentration of 345 mg/L
was recorded at an OLR of 85.44 kg COD/(m3·d). VFA
has been recognized as one of the important intermediates
during the anaerobic digestion (Wang et al., 1999; Ahring
and Angelidaki, 1997; Rajesh Banu, 2006), and is consid-
ered a central parameter for anaerobic treatment (Ahring
and Angelidaki, 1995; Pind et al., 1999, 2002). The
impact of VFA accumulation was reflected in the marked
decrease of COD removal from 92%–85% when the OLR
was increased to 83.7 kg COD/(m3·d) as evident from
Fang et al. (1994) have reported that VFA accumulated
concentration is responsible for souring of the anaerobic
2.3 Effect of mesoporous granular activated carbon
The GAC not only functions as a media for bacterial
Fig. 5 Influence of OLR on COD removal and VFA consumption during
the startup of the reactor.
Fig. 6 Influence of OLR on COD removal and VFA accumulation in the
No. 12Anaerobic tapered fluidized bed reactor for starch wastewater treatment and modeling······
attachment, but can also work as an adsorbent. Com-
bination of activated carbon with the biological process
provides a possibility of grading up the function of the bi-
ological process. Adsorption and desorption are distinctive
characteristics of activated carbon. Its adsorption capacity
protects the attached biofilm from shock organic loading
(Fox et al., 1990). It is also ideal for rapid biomass colo-
nization and provides shelter from shear forces (Characklis
et al., 1982; Suidan and Nakhla, 1987). These functions
can alter the concentrations of substrate in bulk water and
at the surface of the biofilm. It has been proved that the ap-
plication of activated carbon to anaerobic process enables
better degradation of inhibitory wastewater (Hanaki et al.,
We have used the mesoporous GAC as a carrier in
the reactor while most of the people have used only mi-
croporous and macroporous GAC. While in microporous
the adsorption is going to be a slow process due to its
fine size of the pores, whereas in mesoporous the surface
Fig. 7 Effect of mesoporous carbon as a carrier in the reactor.
area is greater than macrporous for the same volume.
This is the main advantages of mesoporous GAC over the
To study the effect of the GAC, the reactor is operated
with a bed expansion in such a manner that the sludge
does not get washed away when it is operated without
carbon. Apart from the advantages cited above, the GAC
is also used as a biogas (methane) storage (Najibi et al.,
2007). The methane gas formed gets stored in the GAC
which reduces the apparent density of the carbon and so
it rises to the surface of the bed where the methane gas
gets released. The GAC particles come down due to the
normal density. This transport of carbon particles creates
a secondary swirling motion over and above the fluidized
motion. The biofilm attached to the carbon also follows the
secondary motion mixing thoroughly with the waste water.
This enhances the efficiency of the reactor. The effect of
GAC has been clearly shown in the Fig.7. The difference
in efficiency of 13% can be attributed to the mesoporous
carbon and the tapered geometry of the fluidized reactor.
2.4 Validation of the MLP Model developed
Actual validation of an already trained ANN requires
testing the network performance on an exclusive set of
data, called testing data, which is composed of data that
was never presented to the network before. Cross valida-
tion computes the error in a test set at the same time that
the network is being trained with the training set and its
performance is shown in Fig.8a. If the error obtained in
Fig. 8 Cross validation plot (a) and output vs. desired plot (b).
1422RANGASAMY Parthiban et al.Vol. 19
both training and testing phases is satisfactory, the NN is
considered adequately developed and thus can be used for
MLP is trained with the 2 input (OLR, pH) and 5 output
parameters (biogas formation and the effluents of COD,
VFA, alkalinity and pH). The output versus desired plot
is shown in Fig.8b. The MSE is 0.0146, which is quite
The experimental data obtained in this investigation
clearly indicate that the anaerobic tapered fluidized-bed
system could remove 92% of feed COD at an OLR as
high as 83.7 kg COD/(m3·d). The average biogas yield
was found to be 25.38 m3/(m3·d) of reactor and the
maximum rate of generation was found to be 198 L/d.
Previously reported efficiency is 82%–85% of COD re-
moval for the same treatment of synthetic sago wastewater
using a columnar AFBR. The anaerobic tapered fluidized
bed reactor (ATFBR) could give a higher efficiency of
minimum 5%–7% more than that of columnar AFBR.
Also the reactor can be operated for an OLR upto 83.7
kg COD/(m3·d) compared to 60.5 kg COD/(m3·d) for
the columnar AFBR. The high COD removal-efficiency
as well as the high methane-production rate observed in
this investigation indicates that the process intensification
achieved in the ATFBR is very impressive. All the biomass
in the anaerobic reactor is retained in biofilm grown on the
fluidized particles; the possibility of biomass wash-out at
higher hydraulic and/or organic loadings is dramatically
reduced due to the tapering effect in the column. Further
the efficiency is also increased due to the mesoporous
GAC. The results are significant, especially in the context
of wastewater treatment in tropical developing countries,
where reactor design with low HRT and high OLR would
be a technologically viable and economically affordable
option. MLP model is trained with the experimental values
obtained. The model identifies the nonlinear relationship
between the various experimental parameters. We have
validated our technique by replicative testing and had
testing error of 0.0146 which is very much encouraging
for further research in this area. Further no such work on
modeling using MLP in the area of waste water treatment
has not been reported in the literatures
The authors wish to express their gratitude to the man-
agement of Sri Venkateswara College of Engineering for
providing the financial support and necessary infrastruc-
ture facilities to carry out this research work.
Ahring B K, Angelidaki I, 1995. Volatile fatty acids as indica-
tors of process imbalance in anaerobic digestion[J]. Appl
Microbiol Biot, 43: 559–565.
Ahring B K, Angelidaki I, 1997. Monitoring and controlling the
biogas process[C]. Proceeding at 8th International Confer-
ence on Anaerobic Digestion. Sendai, Japan. 1: 40–45.
APHA, AWWA, WEF, 2005. Standard Methods for the examina-
tion of water and wastewater[M]. 21st ed. Washington, DC:
American Public Health Association.
Barascud M C, 1992. COD removal in a closed water circuit of
a paper mill by an anaerobic fluidized bed reactor[J]. Water
Sci Technol, 26: 445–454.
Bhat N, McAvoy T J, 1990. Use of neural nets for dynamic mod-
eling and control of chemical process systems[J]. Comput
Chem Eng, 14: 573–580.
Boening P H, Larsen V F, 1982. Anaerobic fluidized bed whey
treatment[J]. Biotechnol Bioeng, 24: 2539–2556.
Bose N K, Liang P, 1998. Neural network fundamentals with
graphs, algorithms, and applications[M]. New Delhi: Tata
McGraw-Hill Publishing Company Limited.
Characklis W G, Trulear M G, Bryers J D, 1982. Dynamics of
biofilm processes: methods[J]. Water Res, 16: 1207–1216.
Chen S J, Li C T, Shieh W K, 1988. Anaerobic fluidized bed
treatment of an industrial wastewaterJ]. J Water Pollut Con
F, 60: 1826–1832.
Fang H H P, Liu G H, Zhu J F et al., 1994. Treatment of brewery
effluent by UASB process[J]. J Environ Eng, 116: 454–460.
Fox P F, Suidan M T, Bandy J T, 1990. A comparison of media
types in acetate fed expanded-bed anaerobic reactors[J].
Water Res, 24: 827–835.
Fu L M, 1994. Neural networks and computer intelligence[M].
Singapore: McGraw Hill Inc.
Hanaki K, Saito T, Matsuo T, 1997. Anaerobic treatment utilizing
the function of activated carbon[J]. Water Sci Technol,
Hickey R F, Wu W M, Veiga M C et al., 1991. Start-up, operation,
monitoring and control of high-rate anaerobic treatment
systems[J]. Water Sci Technol, 24(8): 207–255.
Hien P G, Oanh L T K, Viet N T et al., 1999. Closed wastewater
system in the tapioca industry in Vietnam[J]. Water Sci
Technol, 39(5): 89–96.
Huang J S, Wu C S, 1996. Specific energy dissipation rate for
fluidized-bed bioreactors[J]. Biotechnol Bioeng, 50: 643–
Iza J, 1991. Fluidized bed bioreactors for anaerobic wastewater
treatment[J]. Water Sci Technol, 24(8): 109–132.
Kennedy L J, Vijaya J J, Sekaran G, 2004. Effect of two-
stage process on the preparation and characterization of
porous carbon composite from rice husk by phosphoric acid
activation[J]. Ind Eng Chem Res, 43: 1832–1838.
Kennedy L J, Vijaya J J, Sekaran G et al., 2007a. Equilibrium,
kinetic and thermodynamic studies on the adsorption of
m-cresol onto micro- and mesoporous carbon[J] J Hazard
Kennedy L J, Vijaya J J, Kayalvizhi K et al., 2007b. Adsorp-
tion of phenol from aqueous solutions using mesoporous
carbon prepared by two stage process[J] Chem Eng J, DOI
Khageshan P, Govindan V S, 1995. Anaerobic filter for treatment
of Sago wastewater[C]. Proceedings of 4th National Sym-
posium on Environment, Chennai, India. 248–252.
Kumar A, Roy K, 2004. Artificial neural network-based predic-
tion of bed expansion ratio in gas-solid fluidized beds with
disk and blade promoters[J]. J Insti of Engrs (India), 85:
Lawrence A W, McCarty P L, 1969. Kinetics of methane fermen-
tation in anaerobic treatment[J]. J Water Pollut Con F, 41:
Mai H N P, Duong H T, Trang T T T et al., 2004: UASB treatment
No. 12Anaerobic tapered fluidized bed reactor for starch wastewater treatment and modeling······
of tapioca processing wastewater in South Vietnam[C]. In
the Proceedings of the 10th World Congress: Anaerobic
Digestion 2004, the International Conference, Canada. 4:
Najibi H, Chapoy A, Tohidi B, 2007. Methane/natural gas storage
and delivered capacity for activated carbons in dry and wet
conditions[J]. Fuel, doi:10.1016/j.fuel.2007.03.044.
Oanh L T K, de Jong K, Mai H N P et al., 2001. Removing sus-
anaerobic filter[C]. International Conference: Industry and
Environment in Vietnam.
Parthiban R, Iyer P V R, Sekaran G, 2007. ANFIS modeling
for an anaerobic tapered fluidized bed reactor for industrial
wastewater treatment[J]. J Syst Sci Eng, 16. (in press)
Perez M, Romero L I, Sales D, 1997. Steady state anaerobic ther-
mophilic degradation of distillery wastewater in fluidized
bed bioreactors[J]. Biotechnol Progr, 13(1): 33–38.
Perez M, Romero L I, Sales D, 1999. Anaerobic thermophilic
fluidized bed treatment of industrial Wastewater: effect of
F:M relationship[J]. Chemosphere, 38(14): 3443–3461.
Perez M, Romero L I, Sales D, 2001. Organic matter degradation
kinetics in an anaerobic thermophilic fluidized bed bioreac-
tor[J]. Anaerobe, 7: 25–35.
Pescod M B, Thanh N C, 1977. Treatment alternatives for
wastewaters from the tapioca starch industry[J]. Water Sci
Technol, 9(3): 563–574.
Pind E P, Angelidaki I, Ahring B K, 1999. The use of VFA mea-
manure[C]. 2nd International Symposium on Anaerobic
Digestion of Solid Waste, Barcelona. 41–44.
Pind E P, Angelidaki I, Ahring B K, 2002. Dynamics of the anaer-
obic process: Effects of volatile fatty acids[J]. Biotechnol
Bioeng, 82(7): 791–801.
Rajesh B, Kaliappan J S, Beck D, 2006. High rate anaerobic
treatment of Sago wastewater using HUASB with PUF as
carrier[J]. Environ Sci Technol, 3(1): 69–77.
Saravanane R, Murthy D V S, Krishnaiah K, 2001. Anaerobic
treatment and biogas recovery for Sago wastewater man-
Saroja K, Sastry C A, 1972. Report on treatment of Sago
wastes[R]. National Environmental Engineering Research
Sastry C A, Mohan G J, 1963. Anaerobic digestion of industrial
wastes[J]. Indian J Environ Health, 5(3): 20–25.
Sastry C A, Murahari R, Saroja K, 1964. Studies on the treatment
of Sago mill wastewater[C]. Seminar on Environmental
Pollution, Kerala, India.
Scott C D, Hancher C W, 1976. Use of a tapered fluidized bed as
a continuous bioreactor[J]. Biotechnol Bioeng, 18: 1393–
Seckler M M, van Leeuwen M L J, Bruinsma O S L et al.,
1996. Phospate removalin a fluidized bed process optimiza-
tion[J]. Water Res, 30(7): 1589–1596.
Souza R R, Bresolin I T L, Bioni T L et al., 2004. The perfor-
mance of a three-phase fluidized bed reactor in treatment
of wastewater with high organic load[J]. Braz J Chem Eng,
Sreekrishnan T R, Ramachandran K B, Ghosh P, 1991. Effect of
an anaerobic fluidized bed reactor[J]. Biotechnol Bioeng,
Suidan M T, Nakhla G F, 1987. Anaerobic fluidized-bed treat-
ment of hazardous wastes[C]. Conference on Reducing
Risks from the Environment through Biotechnology, Seat-
tle, WA, July 19–22, 1987.
Tongkasane C, 1970. Anaerobic treatment of tapioca starch
wastewater[J]. Wat Wast Treat, 13: 392–398.
Van Haandel C, Lettinga G, 1994. Anaerobic sewage treatment:
A practical guide for regions with a hot climate[M]. UK:
Wang Q H, Kuninobu M, Ogawa H I et al., 1999. Degradation of
volatile fatty acids in highly efficient anaerobic digestion[J].
Biomass Bioenerg, 16: 407–416.
Wasserman P D, 1989. Neutral computing: Theory and prac-
tice[M]. New York: Van Nastrand Reinhold.
Wu C S, Huang J S, 1995. Performance enhancement with
tapered anaerobic fluidized-bed bioreactors[J]. J Chem
Technol Biot, 63: 353–360.
Wu C S, Huang J S, 1996. Bioparticle characteristics of tapered
anaerobic fluidized-bed bioreactors[J]. Water Res, 30(1):