ELSEVIER

Chemical

Engineering

Journal 65 (1997)

Chemical
Engineering
Journal

133-143

A generalized approach for dynamic modelling and simulation of

biofilters: application to waste-water denitrification
J&Gme Jacob a, Jean-Marc Le Lann a,*, Hew2 Pingaud a, Bernard Capdeville b
a LEAPIAFP:
Groupe dAnalyse Fonctionnelle
des ProcPdtfs, INPT/ENSIGC,
I8 Chemin de la loge. 31078 Toulouse Cedex, France
h INSATIURTB:
Unitt de Recherche Traitement Biologique,
Complexe scientijique de Rangueil, 31077 Toulouse Cedex, France
Received

23 September

1996; accepted

15 November

1996

Abstract
This contribution is devoted to the use of dynamic simulation as a support tool to improve the knowledge of biofilters in the general context
of waste-water treatment.
After a brief overview of current waste-water treatment techniques, a general approach for modelling and simulation of biofilters in transient
states is presented. First of all, a comprehensive description of the development of the model, highlighting specific characteristics such as the
active/deactivated biomass approach is presented; discussion of the different phenomena involved is then pursued. The resulting set of Partial
Differential Algebraic Equations (PDAE), with the need for state event detection for filter clogging, is then solved using an extension of the
Differential-Algebraic
philosophy coupled with the method of lines.
Finally, the validity of the general developed model, as well as the reliability of the numerical approach is illustrated through different case
studies for denitritication purposes, with plant data comparison.
It demonstrates the utility of dynamic simulation to speed up the development of the model phase and to improve basic knowledge in a
new emerging area. 0 1997 Elsevier Science S.A.
Keywords: Dynamic simulation and modelling; Wastewater treatment; Denitrification; Biotilters; Partial differential algebraic equations (PDAJZ)

1. Introduction
Waste-water
treatment (and, more generally, environment
protection) is a topic that is currently receiving substantial
public attention. It consistsof a complex sequenceof interdependentbiological, physical and chemical processessubject to time-varying hydraulic and organic load conditions.
In this context, dynamic simulation based on rigorous
mathematicalmodelsis a powerful tool by which the designers of new biological waste-water treatment units can investigate the performance of complex systemsunder different
conditions. The useof general dynamic modelscan assistin
improving both basic knowledge at the outset and future
operability. Also it may help to develop control strategies,
start-up and shutdown proceduresand operator training in
the wider context of operational water-quality management.
The literature clearly showsthat pressureof different regulatory acts (strict limits on BOD, COD and nitrogen discharges) and the complexity of waste-water plants have
provided incentives to use processsimulation methodology
* Corresponding

author.

1385-8947/97/$17.00
PIIS1385-8947(96)03185-3

0 1997 Elsevier

Science S.A. All rights reserved

[ 11. Among the different case studies reported, it may be

noticed that most are devoted to the simulation of activatedsludgewaste-watertreatment plants [ 2-51 whereasfew contributions in the literature have involved biofilters [6].
However, even in the field of activated-sludge processes,
most of the specialistsare convinced that model improvement, model evaluation and better calibration
are necessary
as well as improvements in mathematical tools and tech-

niques,especially for processoperationandfull-scale dynamics applications.

As proof of this situation,

arecent motivation

at European level has been stimulated by the COST 682

project entitled: Optimizing the designand operation of biological waste-water treatment plants through the useof computer programs

based on dynamic

modelling

of the process.

This paper reports the systematic useof a model basedon

dynamic simulationfor water plant involving submergedbiofilters. This classof unit operationsis often used for denitrification, nitrification
and carbon
removal
(Biocarbon
process, Biofor process... ) . The main advantagesare compactnessand the high capacity for removal of carbon, nitrogen
and phosphorus.Furthermore, they are well integrated into
the environment (no noise, no smell and no aerosols), there

134

.I. Jacob et al. /Chemical

Engineering

is no need for secondary clarification except for certain aerobic applications, there is no need for liquid circulation for
nitrogen removal, and energy consumption along with sludge
production are minimum. One of the main difficulties, from
an operational point of view, is due to the cyclic mode of
operation and in bed clogging which needs to be anticipated.
With these objectives in mind, the proposed generalized
model has been developed for submerged biofilters.

2. Modelling

of the biofilter

2.1. Basis for biojilter modelling

A biological filter may be described as a three-phase
packed column. The solid phase consists of two components:
carrier particles (packing) and biofilm (biomass) consisting
of water, microorganisms and saccharidic polymers. The gaseous phase consists of air (for aerobic reactions) and products
of the reaction (carbon dioxide and, in the case of denitrification, gaseous nitrogen).
Generally, two main parts can be distinguished in a model
of a packed biofilter: 0 a reactor model (hydrodynamic
aspect) ; and l a biofilm model with substrate uptake, product
formation and biomass development.
In our particular case, another part has to be added for deep
filtration as the suspended particles present in the waste water
(and their retention in the filter) are taken into account.
For hydrodynamic aspects, the gaseous phase is not taken
into account because the gaseous hold-up is markedly lower
than the liquid hold-up.
The biofilter should, rigorously, be described as a series of
completely mixed tanks or as a plug-flow reactor with axial
dispersion model [ 71. Experimental hydrodynamic studies
on a biological filter showed that the axial dispersion may be
neglected especially for anaerobic reactions: the filter was
described by a series of 16 to 60 (for anaerobic treatment)

Journal

65 (1997)

133-143

and 11 to 34 (for aerobic treatment) completely mixed tanks.

This number, which depends on the experimental conditions
(aerobic or anaerobic reactions, air/water flow-rate ratio,
degree of clogging of the filter.. ), is fairly high and we may
consider, like most authors [ 6,8], that the liquid is in plugflow.
In most biofilter models, stationary and uniform flow is
hypothesized. This hypothesis, which simplifies the equations and leads, sometimes, to analytical solutions, may be
applied only to very short treatment times. For longer treatment times (as in waste-water treatment), there is progressive
clogging of the filter due to biomass growth and retention of
suspended particles; it is taken into account in the model by
writing a space balance which expresses the progressive diminution of the porosity (volume fraction for the liquid).
For biofilm modelling, four steps are generally considered:
0 diffusion of substrates through the liquid film layer around
the biofilm to the liquid/biofilm interface; 0 diffusion of
substrates through the biofilm; 0 biological reaction; and l
biofilm detachment (which is not taken into account in the
model because it is not observed experimentally-the
liquid
flow is laminar).
In most biofilter models [ 8,9], diffusion phenomena are
taken into account and described by Ficks law. From a theoretical point of view, the diffusion is a phenomenon which
occurs perpendicularly to the liquid flow through a flat surface. In biological filtration, the biofilms are very irregular
and filamentous which accords poorly with the diffusional
theory hypothesis.
Furthermore, experimental studies on aerobic [ 10 ] and
anaerobic [ 111 biofilms show that the substrate removal reaction occurs only at the edge of the biofilm in a very thin part
of it-the active thickness is estimated at about 10 p,rn
whereas the biofilm may reach 500 pm. Thus, the reaction
may be considered as a surface reaction.
Fig. 1 shows the evolution of the amount of biomass
(active and total) and the biofilm aspect (front and profile

biarmsa
amaunt

total

Fig. I. Total and active biomass evolution.

biomass

active biomass

J. Jacob et al. /Chemical

Engineering

views) with time. At the beginning, the active biomass grows

at the edge of the particles. Then, when the support is fully
settled by microorganisms, the active biomass becomes constant while the total biomass keeps on growing; when reaction
occurs the microorganisms grow at the edge of the biofilm
and a new active layer is generated which deactivates the
previous one.
For these reasons, diffusion phenomena, as in some other
works [ 6,121, are not taken into account in the model.
2.2. Model of the biological reaction
Initially, a model is built considering the whole biomass
active in the filter. This approach is based on a contradiction
but it is interesting to try it in order to emphasize the falseness
of the hypothesis it is based on.
The biological model is adapted from the genera1 model
of the IAWPRC for activated sludge process [ 51; in this first
approach, microorganism growth is first-order with total biomass (concentration X,) with double substrate limitation
characterized by a double Monod-type kinetic law for the
electron donor D and the electron acceptor A. The general
form is:

and it may be applied to every kind of reaction which might

occur in a biofilter (Table 1) .
As in the first IAWPRC model, the stoichiometric coefficients are obtained by writing reduction-degree balances for
the biological reaction. For example, in denitrification:
1
1 - Y
7 organic substrate + nitrate -+ biomass
2.86YH
H
where YH is the heterotrophic biomass yield.
For l/Y, mol (COD) of electrons given by the organic
substrate, I mole (COD) of biomass is formed, the nitrates
using the rest are completely reduced to gaseous nitrogen.
The organic compound concentrations are given in units
of chemical oxygen demand (COD), likewise the balances
Table 1
Kinetics for the different

applications

Characteristics

Denitrilication
Reduction of nitrates to gaseous nitrogen

Anaerobic
Electron
Electron
Anaerobic
Electron
Electron
Aerobic
Electron
Electron
Aerobic
Electron
Electron

Denitritation
Reduction of nitrites to gaseous nitrogen

Carbon removal
Oxidation of organic

to nitrates

substrate

65 (1997)

133-143

135

are written in terms of COD, and hence a conversion term

(here 2.86) is used to transform to COD the nitrogen compound concentrations (given as nitrogen).
In the IAWPRC mode1 for activated sludge, the decay of
the microorganisms is taken into account (as hydrolysis and
ammonification) because, in these types of process, the biomass is not fixed and the solid residence time is long (approximately 3-15 days). The biofilter, on the other hand, has to
be washed frequently to prevent clogging, that induces short
solid residence time (approximately l-3 days) and the biomass decay may be neglected. Likewise, after experimental
observation, it has been decided to neglect biofilm detachment during the treatment phase.
As the first approach is based on a contradiction (we consider the whole biomass active although it has been shown
that only microorganisms present at the edge of the biofilm
are working in the biological reaction), we use a more realistic approach based on the concept of active/deactivated
biomass (Ma/M,). The principle is described below [ 10,l l] .
l The active microorganisms
are responsible for substrate
degradation and are characterized by a specific growth rate
PO:

0 The deactivated microorganisms do not play any role in

substrate degradation (even though they might maintain their
enzymic activity) but continue to play arole infilterclogging.
l The total biomass is defined as the sum of the two types
of biomass X, =X, +X,.
An intrinsic kinetic (without considering deactivation of
the active biomass) is defined:

As shown in Fig. 1, when the support is fully populated by

microorganisms, part of the active biomass (at the edge of
the biofilm) keeps on growing while another part has lost its
activity, which induces an accumulation of deactivated
biomass:

of biofiltration

Reaction

Nitrification
Oxidation
of ammonia

Journal

work of heterotrophic
biomass
donor: organic substrate
acceptor: nitrates
work of heterotrophic
biomass
donor: organic substrate
acceptor: nitrites
work of autotrophic biomass
donor: ammonia
acceptor: oxygen
work of heterotrophic
biomass
donor: organic substrate
acceptor: oxygen

Kinetics

136

J. Jacob et al. / Chemicul

Engineering

Journal

65 (I 997) 133-143

k=@(c)

= rd = K,nr,
where I is the concentration of inhibiting products, proportional to the active amount of biomass
I=a&-rd=K,c&,=K&,

Different forms of the function F(a)

were tried
(F(a) = 1- (T/u~, F(a) = 1- ((~/a,)*, where vM is the
maximum retention when all the sites are occupied and the
retention probability is equal to zero) but asnone gave significant results, we kept the first option.

So, the real active biomass accumulation is:

2.4. Resulting setof equationsfor the model

When the steady state is reached:

=OandX,=X,,,*K,=a

A material balance is written for each component on an

elementary volume, for a liquid in plug flow inside the filter:accumulation = input - output * reaction

X Amax

Thus, the real growth kinetics for the active and total biomass are:
~=~A(l-$--)

2.3. Model for the physical reaction (deep filtration)

Unlike other biofilter models, we take into account the
suspendedparticles in the influent and their retention in the
filter. Two types of deep filtration modelsexist:
l macroscopic models with material balances and a
kinetic equation introducing a hold-up probability [ 131; and
0 microscopic models which describe the biofilter as an
aggregationof singlecollectors with specific geometry, introducing the suspended-particlediameter [ 141.
A granulometric analysisof the wastewater showsa range
of particle size (with two sensiblepeaksaround 10 pm and
50 pm); thus, it seems better to use the macroscopic
approach.
The physical filtration reaction may be written:suspended
particles (M,) + retained particles (M,)
The retention kinetic is:
axMr-- Q@L,,
atn
where k is the filtration coefficient, correspondingto the probability density of particle capture.
To a first approximation, this coefficient may beconsidered
as a constant; this is the choice used in the model. But, in
fact, it is logical to think that the filtration efficiency decreases
as soon as the retention sites are occupied by the captured
particles.
The filtration coefficient is defined as a function of the
retention in the filter (volume fraction occupied by captured
particles) [ 151;

where v is the stoichiometric coefficient for the component

involved in reaction r.
Thesematerial balancesare combined with the spacebalance-the variations of the void fraction (or porosity E) are
due to the growth of total biomassand the retention of suspendedparticles:

2.4.1. Basic assumptionsof the model

l The liquid is in plug flow. l The liquid flow-rate is
supposedto be constant insidethe filter but variations in feed
flow-rate are taken into account. 0 Suspendedparticles in the
influent are taken into account. l Filter clogging is taken into
account. l Diffusion phenomenaare not taken into account.
0 The gaseousphaseis not taken into account. l The initial
amountof biomassis uniform in the filter. 0 Decay of biomass
is neglected. 0 Detachment of biofilm and retained particles
is neglected (laminar flow). l Temperature and pH (close
to 7) are constant in the filter.
For the M,/M, approach, a system of eight partial differential equations, with adequateinitial and boundary conditions, is obtained (Table 2). For floating bed biofilters, an
algebraic equation may be addedto representthe bed rising
in the filter. For aerobic reactions (carbon removal, nitrification), a term for oxygen transfer in the liquid phase,k,a
(S,* - S,), has to be added in the oxygen material balance
(S,* is the concentration in equilibrium with the gaseous
phase,the maximum solubility of oxygen in the water, kLa is
the transfer coefficient of the oxygen in the liquid).
The pressuredrop in the filter is calculated by meansof
the Blake-Kozeny correlation [ 161:
AP=Az

15O/LU,( 1- l )2
02$
P

where p is the dynamic viscosity, D, the particle diameter

and V,,, the superficial velocity of the liquid.
If thereactor isdivided into n elementaryvolumes ofheight
AZ, the total pressuredrop is:

.I. Jacob et al. /Chemical

Table 2
System of equations

for Ma/Md

Engineering

Journal

65 (1997)

137

133-143

approach

Component

Balance

Initial and boundary

Total biomass

f$!=

x, (r=O)

Active

biomass

Retained

particles

Electron

donor

Electron

acceptor

h&-&--&--*

?=4&lkb~tl-2)
Ww,
Qku,
_=ar
R
a(e&,)
-= at
-g$-

X,(t=O)=X,,
XMr

(t=O) =o

Vuk&&J(*&

S,(t=O)

=S,Az=O)

=SD~,p,)

-g-

Wmax(&&-$jY*

.s,(t=O) =S,(z=O)

=SA,,p,)

W,.,,(

Sdt=O)

=&d(z=O)

=s,,,pr,

Xbf,(t=O)

=xw,(z=O)

(or nitrates in case of denitrification)

a(eN)
-=

-X2-

Suspended

particles

a(ad
-=

---_-Q ax,,

at

Qusw,

Porosity

e(t=O)

APT= t AP,
I=1

3. Numerical method and implementation aspects

The equationsdescribing filter behaviour lead to a partial
differential algebraic system (PDAE) which may be written
in vectorial form with adequate initial and boundary
conditions:

s(t=O)

=sg,

s(z=O)

=s,p,,

=x,,

a( ES*)
-= at

Ammonia

ms, r$=f(s. r, u,p>+g(s, t, &&$

conditions

(1)

where s is the vector of the statevariables, u the vector of the

control variables andp the vector of the input variables.
Different numerical methods (orthogonal collocation,
method of lines.. .) may be usedto solve this kind of system.
Bearing in mind generalization, a numerical strategy issued
from the solving of mixed differential andalgebraicequations
system (DAE) , coupled with the method of lines with fixedgrid discretization hasbeen implemented.
The numerical method of lines (NMOL) [ 171 may be
consideredasone of the most powerful tools for the solution
of time-dependent coupled ODE/PDE systems.The attraction of this method is that the complex system of coupled
ordinary and partial differential equationsarising in mathematical modelling can be solved by using the sophisticated
tools which have beendevelopedfor initial valuedifferentialalgebraic equations. Additionally, the proposed extended
DAE approach to PDAE allows global resolution with no
discrimination between the variables of the system as it is
given by the model designerand the useof robustandreliable
implicit numerical integrators.

=x.M,,,w,>

= co

The principle of the method of lines is to discretize the

spatial variable (z) into n - 1 pieces (n is the discretization
order). Each variable is transformed into n variables corresponding to its value at each of the n positions in the filter.
The spatial derivatives are approximated using finite differences.After preliminary numerical studies,an approximation
formula with five points hasbeen chosen. The general term
is
as, 2s,~~-16s,~~,+16s,+,-2s,+~
az=
Az.4!

i=3 to i=n-2

while special formulae are used for the upper and lower
boundaries:

!!L
az -

- 5Os,+ 96s, - 72s, + 32.~ - 6s,

Az.4!

as, - 6s, - 2Os,+ 36s, - 12sq+ 2s,

az=
Az.4!
%I

- I _
aZ

-2s,-,+12s,-,-36s,-2+20s,-,+6s,
Az.3!

as,-6s,p,-32s,-,+72s,-,-96s,-,+SOs,
Az.4!
az Thus, the systemis transformed into a system of 8n ordinary (or algebraic) differential equations which are solved
by Gears multistep multi-order implicit method basedon a
predictor corrector scheme[ 18-201.
At any time step, a set of variables is computed using
extrapolation polynomials basedon the values of the variables and their derivatives at previous steps (prediction
phase). The number of previous stepsusedis the order of the
method. The correction phaseconsistsin correcting the predicted valuesto reachthe solution, by usinganapproximation
of the derivative

138

J. Jacob ef al. /Chemical

Engineering

Journal

65 (1997)

133-143

as i
where h is the time step, p a parameterdependingof the order
and 9 a function of the variable values at previous steps.
Thus, the systemdescribedby Eq. ( 1) becomesa non-linear
algebraic equationssystem:

This is then solved by a Newton-Raphson method. The

values used for the initialization came from the prediction
phase:
sk+l=sk+&k
where 6skis the solution of the linear system:
hpJI-$(s*-9)

-Dk

6sk= -g(s,

t, uk,p)

where J is the jacobian matrix of the elementsaf;/as,.

This systemmay be written
Mk&=

Fig. 2. The dynamic operator.

schemewhich is a fixed-grid approachandmay be overcome

by applying a variable grid strategy. However, in our case,
this was not a necessityasthere were no sharpspaceprofiles.

-g(s, t)

whereMk, called the dynamic operator, is the dynamic equivalent of the jacobian matrix in steady-state.
When the corrector converges, numerical tests are
performed to validate the integration step with a tolerance
given by the user. Thesetestsactivate a stepandorder change
strategy to provide a good result quality for a lower cost.
As seenpreviously, at any time step a non-linear system
in the corrector loop hasto be solved. For a rapid convergence
and becausethere is a good initialization with the predictor,
the Newton-Raphson method is chosen,involving analytical
derivatives in the dynamic operator.
The eight equations of the model are arranged for each
discretization point which leadsto a block structure for the
dynamic operator (Fig. 2). Each block is an 8 X 8 matrix
corresponding to the model equation at one discretization
point. These equations are arranged to bring the non-zero
terms closer to the diagonal and to preferably fill the upper
triangular part of the block. The extra diagonal blocks come
from the spatial five-point derivative approximation. The
operator, partially hollow, is an n-diagonal block matrix and
computation time, reliability and efficiency is gainedby treating it like a banded matrix: it is transformed in a smaller
rectangular matrix for which the pivot researchin the Gaussianelimination is done vertically.
As the biofilter hasa cyclic operating mode and is cleaned
when the pressuredrop reachesanupperlimit ( AP = A P,,,) ,
an automatic procedureto detect stateevents hasbeenimplemented [ 211. In the case of time events, input disturbance
discontinuities arerepresentedby usinga polynomial approximation to simulate stepor ramp functions.
The numerical strategy has appearedto be very efficient
and reliable. The main interest in the method of lines leads
to the fact that all the development involved in DAE systems
may be re-used.One of its drawbacksis in the discretization

4. Experimental and dynamic simulation studies

4.1. Pilot plant description
Even if the type of applications (denitrification, denitritation, waste water, drinking water.. . ) and operational conditions (up-flow, down-flow...) used are different, all of the
experimentspresentedin this paperwere performed with the
samepilot biofilter, which wasdesignedand performed asa
multipurposeflexible biofilter.
The biofilter diameterwas6.8 cm andpackedheight 2.1 m;
samplepoints were placed every 30 cm and it was packed
with pouzzolane (a very porous volcanic rock). The reactor
configuration is presentedfor the caseof denitrification of a
drinking water (up-flow for a synthetic water using ethanol
as the organic carbon source) in Fig. 3.
Table 3 gives the values of kinetic, stoichiometric and filtration parametersusedin the following applications.
4.2. Application to the denitrification of waste water
The experimental studies were concentrated on a downflow denitrification filter andtwo biological approacheswere
tested.
4.2.1. Biomassfully active
Fig. 4 showsthat, unlike the experimental results, the simulated nitrate levels run out too quickly in the filter (other
simulationsfailed with lower valuesof the maximum specific
growth rate). This is due to the exponential growth of the
biomass (first order) which leads to faster filter clogging;
even with a very low value for the initial amount of biomass
(this is the case in the figures), it is impossibleto simulate

J. Jacob et al. /Chemcal

Engineering

Journal

65 (1997)

133-143

i39

nitrates tank

Fig. 3. Experimental pilot plant for drinking water denitrification.

the filter behaviour during all the treatment. Hencethe kinetic

model is not correct and the development the active/deactivated biomassconcept waschosen
4.2.2. Active/deactivated concept
It is now possibleto simulatebiofilter behaviour during its
whole lifetime. Good fits are obtained, with different liquid
flow-rates, for the outlet concentration of suspendedparticles
which instantaneouslyreachesa constant value (Figs. 5 and
6). Thus, the filtration part of the model may be validated.
As expected, a steady-state is observed in Fig. 7 with
constantoutlet concentrations (the simulationis stoppedwith
the state event detection: AP= AP-)
and the nitrates run
out only in the secondhalf of the filter (Fig. 8). Thus, from
a qualitative point of view, good resultsare obtained for the
biological part of the model. Unfortunately, theseexperimental denitrification resultscannot be usedfor quantitative studies becausethey were obtained with waste water coming
directly from a treatment plant with high uncontrolled variations in input concentrations. That is the reasonwhy it has
been decided to test the model for different applications on
synthetic waters, with better control of input variables.
4.3. Application to the denitritation of a synthetic water
The experiments were achieved on a synthetic water (with
sodium acetate as carbon source) which allows a constant

feed concentrationto be obtained.The useof a synthetic water

enablesseparatestudy of the biological and filtration aspects
and, in this case,for the validation of the biological model as
the experimentswereconducted without suspendedparticles.
Once again the filter (with liquid in down flow) was
packed with a very porous material (pouzzolane) which
retains a large amountof residualactive microorganismsafter
washing. Thus, the biofilter reaches steady-state nearly
instantaneously with constant outlet concentrations; this is
why only figures with space concentration profiles are
presented.
Good agreement was obtained between simulated and
experimental data for concentrations of nitrites (Figs. 9 and
11) and carbon (Figs. 10and 12)) with different liquid flowrates, during the whole treatment time. Thus, this approach
appearsmuch more realistic than the previous one and the
biological part of the model may be validated.
4.4. Application to the denitrification of a synthetic
drinking water
For drinking water treatment, very drastic constraintshave
to be observed concerning oxidized nitrogen ( [NO;] +
[NO; ] < 5.6 mg N L-l). Thus, both nitrite and nitrate concentrations have been measured;this is the reasonthey have

140

.I. Jacob et al. /Chemical

Table 3
Kinetic, stoichiometric

and filtration

parameter

Engineering

values used for different

g - DC0

substrate)

65 (1997)

of waste water

Denitritation

Denitrification

0.67a
4b

0.4a
5

0.3a
-

480
40
5

I OOOd
lob
o.2b
0.6*

a Experimental
values.
Values given with the IAWPRC model for waste water
Fixed values obtained by a synthesis of the appropriated
d Values obtained by a quick trial/error
strategy

of drinking

water

5
7.5
740d
40
I
3
-

[ 51.
literature

10

40

133-143

applications

Denitrification
Yu ( / (g DC0 biomass
fiH C/j-)
pLI CL-)
CLZ C/j-)
X hax(~(mgDCOL-))
Ks (/(mgCDOL-))
ILo3 (/(mgNLm))
ho2 (/(mg
N L-l))
k (/m-l)

Journal

Q=lZUh

Q=ZOlih
30
--

k20h

(sim.)

k30h

(sim.)

k40h

(sim.)

.- t=50h

(sim.)

-- l=60h

(sim.)

k146h

(exp.)

t1

20

40

60

60

time
0

30

60
vertical

90

120

position

Fig. 4. Waste water denitrification:

filter (mg N-NO1 L-l).

150

180

Fig. 6. Waste water denitritication:

cles (mgL-).

210

(cm)

concentration

profiles

for nitrates in the

100

Q=Blh

60

40

sim.

120

outlet concentration

140

of suspended parti-

60

60

100

(h)

Q=12Uh

60
:
40

exp.

CARBON OmLET

CONCENTRATtON

(mgDCOn)

04
0

40

60

120
time

Fig. 5. Waste water denitrification:

cles (mgL-).

160

NITRATES OUTLET CONCENTRATtON

200

20

40

(h)

outlet concentration

60

60
time

of suspended

parti-

Fig. 7. Waste water denitrification

and nitrate.

(Ma/M,):

100

outlet concentrations

The denitritation reaction (2) is written as:

(I)
(2)
+ NO, + N2

In terms of degrees of reduction, the denitratation reaction

( 1) may be written:
NO;
with a corresponding

1 - Yn
-+ biomass + 1.71YNoz
growth rate:

with

I
140

(h)

been dissociated in the model. The biological reaction

(NO; + N2) is represented schematically in two consecutive
reactions, denitratation ( 1) and denitritation (2) :
NO,

(mgNNo3)
120

of carbon

J. Jacob et al. /Chemical

Engineering

100

~-

Q=lZVh

Journal

60
60

k40h
-

133-143

141

-,=,otl

--- ,=ZOh

65 (1997)

Q=16Vh
8

8
8

t=60h

--

t=QOh=120h
k146

(exp.)

40 i

simulation
0

t=Zlh

(exp.)

t=70h

(exp.)

20

f
0

30

60

0--

90

wrtcal

120

posttm

150

160

210

30

60

(cm)

Fig. 8. Waste water denitrification

nitrates in the filter (mg N-NO, L-l)

90

vertical

(MJM,):

concentration

profiles

for

Fig. 12. Denitritation:

(mg DC0 L-).

25

25

120

position

150

180

210

(cm)

concentration

profiles

for

carbon

in the filter

Q=14l/h

7
-

10

,= 71 h

07

sim.

nitrites

sim.

nitrates

exp.

nitrites

exp.

P
0

30

60

90

verhcat

Fig. 9. Denitritation:
NO,L-).
250

nitrates

120

position

150

concentration

160

210

30

(cm)

60

90

vertical

profiles

for nitrites

in the filter (mg N-

120

posatmn

150

180

210

(cm)

Fig. 13. Drinking water denitrification:

and nitrites in the filter (mg N L- I).

concentrations

profiles

for nitrates

200

Q=Bl.h

.
l

t=71

30

60

90

vert~~l

Fig. 10. Denitritation:

NO,L-).

120

postlion

concentration

150

160

o-0

210

exp.

90
position

120
z

Fig. 14. Drinking

water denitritication:
the filter (mg DC0 L-l).

150

160

210

(cm1

concentration

profiles

for carbon in

Good agreement between experimental and simulated data

were obtained for nitrate and nitrite levels (Fig. 13). The
retention of nitrites in the filter can be observed as their
concentration gets close to zero when there are no more
nitrates.
The results for carbon are not as good (Fig. 14) but it must
be noted that all these simulations were performed without
real estimation of the kinetic and stoichiometric parameters.
A better fit between experiments and simulations could be
obtained by using a rigorous parameter estimation procedure
for the determination of very sensitive parameters such as
maximum growth rates or maximum active biomass concentrations in the filter.

------I
-

simulation

t=Zlh

(exp.)

k70h

(exp.)

30

60
vertical

Fig. 11. Denitritation:

(mgDCOL-).

The
nearly
yields
The
(using
ilow.

60
vertical

carbona

30

(cm)

profiles for carbon in the filter (mg N-

:I;

sim.

01
0

carbone

90
position

120
z

concentration

150

180

210

(cm)

profiles

for

nitrites

in the filter

same bacterial species work in the two reactions, in

the same environment, so we may assume that the two
(Yn) are equal.
experiments were performed with synthetic water
ethanol as the organic carbon source) in the upstream

5. Conclusion
A general model has been developed for the dynamic simulation of waste-water treatment units by submerged biofil-

142

J. Jacob et al. /Chemical

Engineering

ters. When used for carbon removal, denitrification or

nitrification, this model allows for the investigation of the
response of this type of unit to load and flow-rate perturbation.
It will help designers to predict overall behaviour (removal
efficiency and filter clogging) with control strategies in mind.
The accuracy of the model and the efficiency of the implemented numerical method (even in hard integration conditions such as biofilter start-up) have been illustrated in the
cases of denitritation and denitrification with comparison versus pilot plant data.
Further work needs to be performed especially in the evaluation of the pressure drop in the filter. The model should
also be improved by taking into account the influence of
temperature on the biological kinetics. The development of a
specially adapted method for parameter estimation PDAE
system, as well as the opportunity to introduce a moving grid
in the method of lines should be investigated.
This work is dedicated to the memory of Professor B.
Capdeville.

6. Nomenclature
dcell
dP
D
DP

k
K4
KD
M
n
N
P

Q
r

cell density in the biofilm (g (COD) rn- 3,

particle density in the particle deposit (g m 3,
matrix of the coefficients associatedwith the
dynamic terms as/& (-)
packing particle diameter (m)
jacobian matrix (-)
filtration coefficient (m- )
electron acceptor half-saturation coefficient
k mm3>
electron donor half-saturation coefficient
g (COD) me3
dynamic operator matrix (-)
number of equationsof the PDAE system (-)
number of discretization points (-)
vector of input variables (-)
liquid flow rate ( m3h - )
reaction rate (biological or physical)
&Ei L,,iie variables (-)
electron acceptor concentration in liquid phase
(g me31
electron donor concentration in liquid phase
(g (COD) me3)
time (h)
vector of control variables (-)
superficial velocity (m h- )
active biomassconcentration in the filter
(g (COD) mp3)
total biomassconcentration in the filter
(g (COD) mm3)
deactivated biomassconcentration in the filter
g (COD) mm3)

Journal

XMl
XMr
Y
2
bP
E
I-L
PO
Pmar

rl
CT

65 (1997)

133-143

suspendedsolid concentration in the liquid phase

gm-3
retained particles concentration in the filter g mm3
heterotrophic biomassyield (-)
vertical position in the filter (m)
pressuredrop in the filter (Pa)
porosity (-)
dynamic viscosity of the liquid (kg m- s~ )
specific growth rate for biomass(h- )
maximum specific growth rate for heterotrophic or
autotrophic biomass( pH or pA) (h- )
stoichiometric coefficient (-)
cross-sectionalareaof the filter ( m2)
retention in the filter (-)

Acknowledgements
This work is part of J. JacobsPh.D. Thesis.We gratefully
acknowledgethe French Ministry of Researchand Technology (MRT) and the Midi-Pyr6ntes Region (CCRRDT) for
their financial support aswell asthe SAUR (BOUYGUES)
Company for support of laboratory research.

References
[ 1) I. Farag, I. Gosling, R. Field and C.C. Chen, Simulating waste water
treatment processes, The Chemical Engineer, (Sept. 1990) 31-38.
[2] G.G. Patry and 1. Takacs, GPS-X: A waste water treatment plant
simulator, Proc. Mathmod
Vienna (IMACS).
Vienna, Austria, 1994,
pp. 456-459.
[3] M.N. Pons, 0. Potier, N. Roche, F. Colin and C. Prost, Simulation of
municipal waste water treatment plants by activated sludge, Camp.
Chem. Eng., 17(suppl.)
(1993) 227-232.
[4] P. Lessard and B. Beck, Simulation
dynamique
dune usine de
traitement des eaux US&S, Tribune de leau, no 533/5 ( 1991) 35-5 1.
[ 51 M. Henze, C.P. Leslie Grady, W. Guyer, G.V.R. Marais and T. Matsua
(IAWPRC
task group), A general model for single sludge waste water
treatment systems, Water Res., 21(5) ( 1987) 505-515.
[6] P. Le Cloirec, G. Martin, B. Ben Barka and A.Y. Le Roux, Un mod&le
mathkmatique
de la dt?nitrification
sur filtres soufre-carbonate
de
calcium, Chem. Eng. J., 31 (1985) 9-18.
[7] S.E. Jorgensen and M.J. Gromiec. Mathematical
models in waste water
treatment,
in Developments
in Environmental
Modeling,
Elsevier,
Amsterdam,
1985, pp. 473-524.
[ 81 G.E. Speitel, K. Dovantzis andF.A. Di Giano, Mathematical
modelling
of biogeneration
in G.A.C. column, J. Environ. Eng., I13 ( 1987) 3248.
[ 91 K.J. Williamson
and P.L. MC Carty, Verification
studies of the biofilm
model for bacterial substrate utilisation, J. Water Pollut. Control Fed.,
48 (1976) 281-296.
[lo] K.M. NGuyen,
Description
et modelisation
des films biologiques
aerobies, Ph.D. Thesis, INSAT (no 96), France, 1989.
[ 1 I] R. Belkhadir,
Etude fondamentale
des biomasses fixees. Description
et modklisation
des biofilms biologiques
anaerobies, Ph.D. Thesis,
INSAT (n 18), France, 1986.
[ 121 S. Elmaleh, H. Labaquereand
R. Ben Aim, Biological filtrationthrough
a packed column. Water Res., 12 ( 1978) 4146.
[ 131 K.J. Ives and A. Gur, Recherche sur loptimisation
de la filtration,
Centre Beige dEtude etde Documentation
des Eaux, 333-334 (1971)
377-384.

J. Jacob et al. /Chemical

Engineer *ing Journal

[ 141 CR. OMelia and W. Ali. The role of retained particles in deep bed
filtration, Prog. Water Technol., 10 (1978) 167-182.
[ 151 J.P. Herzig and P. Le Goff, Le calcul previsionnel
de la filtration a
travers un lit Bpais, Chimie et Industrie G&e Chimique, 104 ( 197 1)
2337-2346.
[ 161 R.B. Bird, Transport Phenomena, Wiley international
Edition, 1960.
[ 171 W.E. Schiesser, An Introduction
to the Numerical
Method of Lines
Integration
of Partial Differential
Equations
(Differential
Systems
Simulator,
Version
2). Lehigh
Universities
and Naval
Air
Development
Center, 1977.

65 (1997)

133-143

143

[ 181 C. Gear, The automatic integration of ordinary differential

equations,
Commun. ACM, 14(3) (1971) 176-179.
[ 191 A.C. Hindmarsh, LSODE and LSODI. two new initial value ordinary
differential
equation solvers, ACM SIGNUM
Newsletter,
I5 (1980)
lo-l 1.
[20] L. Petzold, DASSL: differential
algebraic system solver, Technical
Report, Sandia National Laboratories,
Livermore,
California.
1983.
[ 211 J. Albet. Simulation rigoumuse de colonnes de distillation discontinue
a sequences opkatoires
multiples, Ph.D. Thesis. INPToulouse,
France.
1992.