Journal of Membrane Science 216 (2003) 55–65

Modelling of a membrane bioreactor system for

municipal wastewater treatment
T. Wintgens a,∗ , J. Rosen a , T. Melin a , C. Brepols b , K. Drensla b , N. Engelhardt b
a Department of Chemical Engineering, Institut für Verfahrenstechnik, Rheinisch-Westfälische Technische
Hochschule Aachen, Turmstrasse 46, 52056 Aachen, Germany
b Erftverband, Bergheim, Germany

Received 3 May 2002; received in revised form 14 January 2003; accepted 16 January 2003

Abstract
Within the presented study a model to describe the filtration performance of submerged capillary hollow fibre modules in
membrane bioreactor applications for wastewater treatment was developed. The model was implemented in a software platform
for simulation and calibrated using operating data of the full-scale municipal wastewater treatment plant Rödingen, which is
operated by the Erft River Association (Erftverband) and equipped with activated sludge treatment and microfiltration units
for biomass retention. Mathematical expressions for filtration resistances like cake layer forming and fouling are presented and
combined with an activated sludge model (ASM) to describe the biological treatment processes. The model proved successful
in simulating the long-term decrease in permeability of the membranes and the final effluent quality in terms of standard
parameters.
© 2003 Elsevier Science B.V. All rights reserved.
Keywords: Membrane bioreactor; Wastewater treatment; Fouling model; Microfiltration; Simulation

1. Introduction due to membrane replacement costs and high-energy

demand for aeration [1].
Membrane bioreactor systems are increasingly Fouling phenomena on the membrane surface and
applied for municipal wastewater treatment, with within the pores reduce the long-term stability of flux
submerged membrane units featuring rather low performance. Permeate back flushing and chemical
trans-membrane pressure (TMP) differences and cleaning are standard procedures applied to minimise
feed-sided air pulsing predominantly being employed. these effects and stabilise overall permeability of the
The combination of activated sludge units and mem- membrane systems, but result in losses of net filtra-
brane filtration for biomass retention generally results tion efficiency and possible damage the membrane by
in high effluent qualities and compact plant configu- cleaning agents. Neither the evolution of membrane
rations. While investment costs are already lower than permeability under certain operating conditions nor
for conventional wastewater treatment plants with the effect of cleaning measures can nowadays be pre-
secondary clarification, operating costs are still higher dicted. These uncertainties cause considerable diffi-
culties in plant layout, design and operation.
∗ Corresponding author. Tel.: +49-241-80-96-233; For many industrial processes as well as for conven-
fax: +49-241-80-92-252. tional wastewater treatment models are used to sim-
E-mail address: wintgens@ivt.rwth-aachen.de (T. Wintgens). ulate process performance and to derive optimisation

0376-7388/03/$ – see front matter © 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0376-7388(03)00046-2
56 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65

approaches based on the output of simulation studies.

Nomenclature No comprehensive models for membrane bioreactor
cb bulk concentration (g/m3 ) systems exist so far, which integrate interdependen-
cM concentration on membrane cies between biological processes and filtration per-
surface (g/m3 ) formance as well as mathematically describe the main
COD chemical oxygen demand (g/m3 ) flux-determining phenomena, occurring in submerged
dC characteristic diameter (m) membrane units for municipal wastewater treatment.
F trans-membrane flux (l/m2 h) Feed-sided air pulsing was described to be a main

flux-determining mechanism in submerged hollow
F dt temporal flux integral (m3 /m2 )
H0 , HP geometric heights (m) fibre applications [2]. The shear forces created by a
kC model parameter cake layer (m2 /kg) slug flow along the fibres were correlated to the mass
kF model parameter fouling (1/m) transfer coefficients at the membrane surface [3]. The
kP mass transfer coefficient (m3 /m2 s) mass transfer coefficient controls the concentration
phydro hydrostatic pressure (Pa) polarisation occurring in front of the membrane in
ppump suction pressure (Pa) all separation applications, as rejected components
pax axial pressure losses (Pa) accumulate at the membrane. In case of membrane
pTM trans-membrane pressure difference bioreactors, micro- or ultrafiltration processes are
(Pa) used to retain biomass in the system. Macromolecular
RC cake layer resistance (1/m) and particulate components of the activated sludge
RF fouling resistance (1/m) accumulate at the membrane leading to significant
RM membrane resistance (pure water flux) losses of driving force due to the resulting cake layer
(1/m) formation and colloidal fouling. Apart from these phe-
SF model parameter fouling saturation nomena on the membrane surface, pore-blocking may
(1/m) occur and contribute significantly to the long-term
SI inert dissolved organic matter decrease in permeability of the membrane [4].
The mentioned phenomena, determining filtration
(g/m3 )
performance in membrane bioreactors, will be math-
SNH ammonia–nitrogen
ematically described within this paper and an inte-
(NH4 + –N + NH3 –N) (g/m3 )
grated model will be presented. The model was used
SNOx nitrate–nitrogen and nitrite–nitrogen
to describe the evolution of membrane permeability at
(NO3 − –N + NO2 –N) (g/m3 )
a full-scale membrane bioreactor plant for municipal
SO2 dissolved oxygen, O2 (g/m3 )
wastewater treatment. Simulation results with respect
SS biodegradable dissolved organic
to biological parameters and flux performance will be
matter (g/m3 )
presented in comparison to the operational data.
T temperature (◦ C)
XA autotrophic organisms (g/m3 )
XH heterotrophic organisms (g/m3 ) 2. Model description
XI inert particulate organic matter
(g/m3 ) The integrated membrane bioreactor model de-
XS biodegradable particulate organic veloped consists of a compartment describing the
matter (g/m3 ) activated sludge processes to account for the biologi-
XSTO storage products (g/m3 ) cal treatment performance and an element to describe
the permeability of the filtration unit. Both elements
Greek letters were coupled within an integrated simulation envi-
␥ shear rate (1/s) ronment to account for interdependencies, mainly
η viscosity (Ns/m2 ) the membrane-initiated accumulation of biomass in
τW wall shear stress (N/m3 ) the system with consequences both for the metabolic
state of the micro-organisms and filtration resistance.
 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65 57

2.1. Activated sludge model particular components involved in the decomposition

of carbon and nitrogen compounds in wastewater. The
The activated sludge model 3 (ASM3) was fre- biochemical processes undergone by the ASM3 com-
quently used as a basis for the description of pro- ponents are visualized in Fig. 1. While autotrophic or-
cesses in the biological stage. Like the first version ganisms use inorganic substances as the main energy
of the activated sludge model (ASM1), the ASM3 has source, heterotrophic organisms utilise reduced carbon
been designed as a tool for description and simulation sources for growth and endogenous respiration.
of aerobic and anoxic stages in municipal wastewater The reactions described by ASM3 represent a
treatment [5]. system of coupled ordinary differential equations,
Unlike in ASM1, the endogenous respiration con- which can be numerically solved for a given set
cept is used, introducing storage substances as major of initial concentrations in the bioreactor. Hence, a
components. They were used in ASM2 for the first dynamic calculation of the time-dependent concen-
time and are specifically important for cell metabolism tration schemes for the individual components can be
in conditions of low organic loads, which are typically accomplished.
encountered in the operation of membrane bioreactors
for municipal wastewater treatment in order to min- 2.2. Modelling of the filtration process in
imise excess sludge production. Extra-cellular poly- membrane bioreactors
meric substances (EPS), produced by micro-organisms
in biological wastewater treatment systems, exert a In membrane bioreactor configurations for mu-
critical influence on the flux-rate achieved in mem- nicipal wastewater treatment micro- or ultrafiltration
brane filtration of activated sludge suspensions and membranes are used to retain the biomass in the sys-
are not yet included in the description of biotransfor- tem, leading to a considerably high total suspended
mation processes. Their generation is supposed to be solid concentration in the bioreactors. The filtration
strongly dependent on the metabolic stage, which cor- performance, namely, the permeability and its evo-
relates with the organic loading rate of the biological lution of long periods of time, is the main focus
treatment system [6]. of interest due to its crucial importance for system
The ASM3 in its basic version considers 12 micro- reliability and economic feasibility [1]. The under-
bial transformation processes with 13 dissolved and standing and quantitative description of performance

Fig. 1. Metabolic pathways according to activated sludge model 3 [1].

58 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65

Fig. 2. Phenomena limiting flux performance in membrane filtration.

determining phenomena is supposed to be a presup- permeate is withdrawn through submerged membrane

position for further process optimisation. modules using a suction pressure on the permeate-side
At the high fluxes encountered when using porous to increase the driving force. Hollow fibre modules
membranes as in ultra- and microfiltration concentra- with a high specific membrane surface in outside–in
tion polarisation on the feed side has a strong effect. filtration mode are frequently used in this configura-
A cake layer of the retained components is formed tion [11].
on the membrane surface causing filtration resistance, The hollow fibres with a nominal pore-diameter in
which is a critical factor for the overall performance the range of 0.1 ␮m are submerged in upright position
achieved in membrane filtration of activated sludge and exposed to a flow of relatively coarse air bubbles
suspensions. Components held back by the membrane from an aeration device situated under the module in
can cause scaling or colloidal fouling, defined as any order to minimize cake formation and clogging.
performance-limiting interaction between feed com- In contrast to the flow-regime in a tubular module,
ponents and the membrane material, on the membrane which can be described rather precisely, cross-flow fil-
surface and within the pores (Fig. 2). It has been shown tration in a submerged capillary hollow fibre module
that the presence of extra-cellular storage components is characterized by a rather complex flow-regime con-
increases the fouling potential of the feed suspension sisting of the two-phase air and activated sludge sus-
dramatically [7–10]. pension in a barely describable cross-section between
The formation of the concentration profile and thus and around the fibres.
the influence of the cake layer upon total filtration
resistance can be positively influenced by a suitable 2.4. Modelling approach for the time-dependent
flow-regime along the membrane. One measure to en- behaviour of total filtration resistance
hance mass transfer at the membrane is air pulsing on
the feed side, which results in a multiphase flow in- The two-phase flow between capillary hollow fi-
ducing enhanced shear forces to the membrane [3]. bres consisting of air and a suspension was described
as a slug flow-regime by Chang et al. [12]. For their
2.3. Semi-empirical modelling of a capillary experiments, conducted with a couple of hollow fibre
hollow fibre module membranes, they introduced a system of equations
consisting of mass-balances and empirical correla-
Due to their low specific energy consumption and tions which can be coherently solved in an ideal case
the economic advantages arising from that fact, low- (gas-content of the liquid-slug is zero) and for known
pressure processes are predominantly applied in mem- boundary conditions (air flow-rate, geometry). Previ-
brane filtration of activated sludge. In these processes, ous modelling approaches have explained the effect
 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65 59

of enhanced mass transfer induced by slug flow based submerged, the suction pressure p (Pa) applied by
on computational fluid dynamics. The highly turbu- the permeate suction pump, and the pressure losses
lent wake region of the gas slug is supposed to create caused by the permeate flow through the hollow
a peak in wall shear stress [13]. fibres.
In contrast, the approach introduced here accounts The effective trans-membrane pressure difference
for average hydrodynamic shear effects upon mass pTM (TMP) for a membrane element is composed
transfer and consequently concentration polarisation, of three pressure components as follows (Fig. 3):
while an accurate description of the geometric con-
straints is not possible in case of submerged capil- pTM = phydro + ppump − pax
lary hollow fibre systems without fixed flow channels
between the membranes. As the diffusivity of parti- Cake resistance RC is assumed to be dependent on
cles and macromolecules encountered in membrane a cake layer forming component at the membrane sur-
filtration of activated sludge is low, back-transport of face, where cM represents the corresponding concen-
species accumulated on the membrane surface to the tration at the membrane surface:
bulk flow can only be shear-induced. RC = kC cM
Formally, the model is based on the resistance-in-
series model, describing the area specific permeate The concentration cM at the membrane surface de-
flux F as the ratio of driving force pTM and a sum of pends on the concentration profile at the membrane
resistances, comprising the hydraulic resistance of the and is expressed in another relation, accounting for
clean membrane RM , a cake layer resistance RC , and a concentration polarisation effects:
fouling resistance RF , multiplied by the temperature-  
cM
dependent dynamic viscosity of the permeate ηP : F = kP ln
cb
pTM
F = This expression describes the deviation between
(RM + RC + RF )ηP
the bulk concentration cb and the concentration at the
The pressure gradient along the fibre is deter- membrane surface of cake layer forming components,
mined by the depth H (m) to which the module is which depends on the local mass transfer coefficient

Fig. 3. Submerged capillary hollow fibre modules and idealised trans-membrane pressure [15].
60 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65

kP and the permeate flux: 3. Operational data from a full-scale

wastewater treatment plant with membrane
τW dC bioreactor configuration
kP =
ηF
The wastewater treatment plant Rödingen is oper-
where τ W characterizes the mean wall shear stress ated by the Erft River Association in the western part
exerted by the two-phased medium of activated sludge of Germany and was the first full-scale municipal
and air upon the membrane surface, ηF the viscosity of wastewater treatment plant with a membrane biore-
the activated sludge and dC a characteristic diameter, actor configuration when it went into operation 1998
which is dependent on the particle size distribution in [16]. The plant is equipped with a 3 mm screening unit
the system and on the lower end determined by size of and a grit chamber for mechanical pre-treatment. The
particles rejected by the applied membrane. In order two bioreactors can perform alternating denitrification
to estimate the local mass the shear rate was assumed and nitrification phases (Fig. 4). An additional 0.5 mm
to be γ = 1/40 [14] and characteristic diameter dC = drum screen removes filamentous material from the
1 ␮m, respectively [15]. process, which otherwise may lead to a clogging of
The fouling resistance RF is assumed to be depen- the membrane filtration units. The biomass is retained
dent on the total permeate volume produced in a fil- in the system by submerged Zee-Weed® microfil-
tration interval under consideration, e.g. between two tration units [14], with a nominal pore-size 0.1 ␮m.
chemical cleanings: There are two filtration units to handle peak hydraulic

t  loads.
RF = SF 1 − e−kF 0 F (t) dt Both filtration units were monitored in terms of
membrane permeability according to a standardized
The coefficients kC and kF have to be fitted to the measuring procedure. During those tests the mem-
short-term and long-term characteristics of total mem- brane unit was operated in a 40 s filtration interval
brane resistance, respectively. with a constant “norm-flux” of 27 l/m2 h and the
The model parameter kF in particular is assumed trans-membrane pressure was measured. The flux
to be strongly dependent on the concentration of po- to pressure ratio gives an indication of the current
tentially pore-blocking and membrane-adhesive com- permeability of the membrane filtration unit for the
ponents in the feed suspension. The amount of EPS given set of operation parameters (e.g. temperature
might have a crucial contribution here, which could and total suspended solids content).
not be quantified in this particular study so far due to A long-term decrease of system permeability could
the lack of appropriate data. be observed on a time scale of several months. The
Another parameter SF is a membrane constant permeability decrease was partly reversible by intense
and, quite similar to adsorption processes, repre- chemical cleaning but re-appeared in a similar way
senting a factor for the specific surface area of the during the subsequent months (Fig. 5).
membrane material, which ultimately can be covered Based on the analysis of the operating data, per-
by fouling products. Thus, SF is also an indicator meability cycles were defined as the period of time
for the maximum increase of fouling resistance to between two major chemical cleaning procedures.
be expected. Dependency upon the properties of The characteristic decrease of permeability in an
the membrane material like pore-size distribution, asymptotic curve, approaching a nearly stable mini-
hydrophobicity, and porosity can also be assumed mum value, was attributed to fouling phenomena on
here. the membrane surface and within the pores. Periodic
In its final form, an implicit expression for the per- increases of the permeability could be explained by
meate flux can be written as: corresponding increases in water temperature, altering
the feed and permeate viscosity.
pTM System permeability data, the corresponding tem-
F (t) =  
t  peratures and total suspended solid contents in the
RM +kC cb eF (t)/kP +SF 1−e−kF 0 F (t) dt ηP
filtration units as well as the total throughput of
T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65 61

Fig. 4. Flow sheet of the wastewater treatment plant Rödingen.

Fig. 5. Permeability and hydraulic loads.

62 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65

permeate within a filtration cycle along with extensive Table 1

information on the general system set-up were used Model parameters used in the simulation study
to calibrate the model presented in Chapter 2. Model parameter Value

RM (1/m) 5.00 × 1011

kC (m2 /kg) 4.82 × 1010
4. Model implementation simulation results SF (1/m) 1.78 × 1012
kF (1/m) 8.30 × 10−2
The model comprising the biological and filtration
part were implemented using a Matlab/Simulink® pro- The set of parameters given in Table 1 was deter-
gramming package and a flow-sheet simulation inter- mined by a least-square-error method to achieve a op-
face was developed (Fig. 6). timal curve fitting.
Simulation studies were performed to reproduce The validity of these parameters was tested in
particular permeability measurement data for a char- a second simulation study in which the data from
acteristic filtration interval (27 November 2000 to another filtration unit and the same period of time
25 November 2001). One set of model parameters was taken and a simulation was performed with the
was used and the particular operation conditions (e.g. parameters determined in first study. In course of
temperature and total suspended solids content) were the simulation, an external chemical cleaning of the
taken into account. The simulation results were com- membrane modules in this unit was accounted for by
pared with the operating data (see Fig. 7). The model a reset of the fouling resistance RF . As illustrated in
is able to reproduce the general trend of permeability Fig. 8 the permeability evolution is well described
decrease and the fluctuations in operating condi- with the simulation, but difficulties exist to reproduce
tions. The mean error is around 11%. Taking average the huge deviations in the early stage of the filtration
measurement errors and unaccounted fluctuations in interval. The overall mean deviation for the simula-
operating conditions into account, a simulation per- tion of the second filtration unit is 15.7% and hugely
formance within acceptable error margins in complex determined by some peak deviations, which might
technical systems, processing heterogeneous feed be attributed to major operational problems, which
mixtures, can be claimed. cannot be considered in the model.

Fig. 6. Simulation flow-sheet of the Rödingen plant.

 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65 63

Fig. 7. Measurements and simulation of permeability in filtration unit I.

The general system behaviour in terms of biomass simulated by the integration of membrane and ac-
concentration and effluent parameters, e.g. chemical tivated sludge models for a steady-state case. The
oxygen demand (COD), ammonia–nitrogen (NH4 –N), simulation was based on average feed characteris-
nitrate–nitrogen and nitrite–nitrogen (NOx –N), was tics and using standard ASM3 parameters with COD

Fig. 8. Measurements and simulation of permeability in filtration unit II.

64 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65

Table 2 ulation activities. On a long-term, a close connection

Biological parameters: measurement versus simulation between process control and simulation is envisaged
Parameter Measurement Simulation to derive methods for optimised process design and
(average) (steady-state) operation.
COD (mg/l) 23.0 23.4
NH4 –N (mg/l) 0.173 0.2
NOx –N (mg/l) 6.2 6.6 Acknowledgements

The presented ongoing research is support by

fractionation to calibrate the model [5]. Table 2 shows the Ministry for Environmental Affairs of North
the comparison of simulation results and operational Rhine-Westfalia (MUNLV). The corresponding au-
data. thor acknowledges the German Research Association
An interesting aspect is the simulation of COD (DFG) for funding.
turnover due to microbial growth processes in com-
parison to endogenous metabolism, two processes ac-
counted for in ASM3. The ratio of COD consumption References
due to growth and endogenous respiration of the het-
[1] M. Gander, B. Jefferson, S. Judd, Aerobic, MBRs for domestic
erotrophs in the bioreactors was calculated to be close
wastewater treatment: a review with cost considerations,
to one, indicating a strong endogenous metabolism in Sep. Purif. Technol. 18 (2000) 119–130.
the system as expected for membrane bioreactor sys- [2] S. Chang, A.G. Fane, Filtration of biomass with axial
tems with high biomass content. While the full-scale inter-fibre upward slug flow: performance and mechanisms,
operational plant operates without primary clarifica- J. Membr. Sci. 180 (2000) 57–68.
[3] R. Ghosh, Z.F. Cui, Mass transfer in gas-sparged ultra-
tion and with excess sludge removal, this ratio might
filtration: upward slug-flow in tubular membranes, J. Membr.
be even closer to one in plants with even lower organic Sci. 4196 (1999) 1–13.
sludge loads. [4] E. Aoustin, A.I. Schäfer, A.G. Fane, T.D. Waite, Ultrafiltration
of natural organic matter, Sep. Purif. Technol. 22–23 (2001)
63–78.
[5] W. Gujer, M. Henze, T. Mino M. van Loosdrecht, IAWQ Task
5. Conclusion Group, Activated sludge model no. 3, Water Sci. Technol. 39
(1) (1999) 183–193.
The presented study demonstrates that the outlined [6] H. Nagaoka, S. Kono, S. Yamanishi, A. Miya, Influence
of organic loading rate on membrane fouling in membrane
modelling concept for membrane bioreactors could
separation activated sludge process, Water Sci. Technol.
successfully be applied to describe the system be- 41 (10) (2000) 355–362.
haviour of a full-scale municipal wastewater treatment [7] H. Nagaoka, S. Yamanishi, A. Miya, Modeling of biofouling
plant, equipped with membrane filtration units. In par- by extracellular polymers in a membrane separation activated
ticular, the model describing filtration performance of sludge system, Water Sci. Technol. 38 (4–5) (1998) 497–504.
[8] H. Nagaoka, S. Ueda, A. Miya, Influence of bacterial
submerged capillary hollow fibre membranes could
extracellular polymers on the membrane separation activated
be applied to reproduce the characteristic decrease of sludge process, Water Sci. Technol. 34 (9) (1996) 165–172.
membrane permeability. [9] S. Chang, C.-H. Lee, Membrane filtration characteristics in
With respect to the biological status of the system, membrane-coupled activated sludge system—the effect of
only steady-state system behaviour was simulated so physiological states of activated sludge on membrane fouling,
Desalination 120 (3) (1998) 221–233.
far, but showed good results describing the tremendous [10] F. Jorand, F. Boué-Bigne, J.C. Block, V. Urbain,
importance of endogenous respiration in membrane Hydrophobic/hydrophilic properties of activated sludge
bioreactor systems. More attention will be paid to the exopolymeric substances, Water Sci. Technol. 37 (4–5) (1998)
transient behaviour of the activated sludge system and 307–315.
the evolution of permeability on a short time-scale. [11] H. Husain, P. Cote, Membrane bioreactors for municipal
wastewater treatment, Water Qual. Int. (2) (1999) 19–22.
Especially, the dependence of permeability on per- [12] S. Chang, A.G. Fane, Characteristics of microfiltration of
meate backflush-cycles and cleaning intervals will be suspensions with inter-fiber two-phase flow, J. Chem. Technol.
investigated as a subject of further modelling and sim- Biotechnol. 75 (2000) 533–540.
 T. Wintgens et al. / Journal of Membrane Science 216 (2003) 55–65 65

[13] S.J. Judd, P. Le-Clech, T. Taha, Z.F. Cui, Theoretical and [15] B. Günder, Das Membranbelebungsverfahren in der
experimental representation of a submerged MBR system, kommunalen Abwasserbehandlung, Dissertation, Stuttgart
Membr. Technol. 135 (2001) 4–9. University, Stuttgart, 1999.
[14] C. Wisniewski, A. Grasmick, A. Leon Cruz, Critical [16] N. Engelhardt, W. Firk, W. Warnken, Integration of membrane
particle size in membrane bioreactors—case of a denitri- filtration into the activated sludge process in municipal
fying bacterial suspension, J. Membr. Sci. 178 (2000) 141– wastewater treatment, Water Sci. Technol. 38 (4–5) (1998)
150. 429–436.