Chemical Engineering Journal 286 (2016) 447458

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

Optimal WWTP process selection for treatment of domestic

wastewater A realistic full-scale retrofitting study
Hande Bozkurt a, Mark C.M. van Loosdrecht b, Krist V. Gernaey a, Grkan Sin a,
a
b

CAPEC-PROCESS Research Center, Department of Chemical and Biochemical Engineering, Technical University of Denmark, Building 229, DK-2800 Kgs. Lyngby, Denmark
Department of Biotechnology, Delft University of Technology, Julianalaan 67, Delft, The Netherlands

h i g h l i g h t s

A superstructure optimization framework is presented for WWTP process selection.

The framework is applied to a retrofitting study of 2 full-scale plants.

Emerging wastewater treatment technologies are compared at their optimality.

High-rate oxic reactor and anammox are selected as the optimal concept.

Selection of optimal side stream treatment technology depends on local preferences.

a r t i c l e

i n f o

Article history:
Received 11 June 2015
Received in revised form 7 October 2015
Accepted 27 October 2015
Available online 9 November 2015
Keywords:
Anammox
MILP
Retrofitting
Side-stream nitrogen removal
Superstructure optimization
Wastewater treatment

a b s t r a c t
Retrofitting existing wastewater treatment plants has become a major challenge in the wastewater engineering area, in particular due to the increasing number of competing technologies and concepts (treatment versus resource problem). In order to support the expert-based approach and provide a rational
design for WWTPs, we proposed the use of a mathematical programming based framework to manage
the complexity and solve the multi-criteria retrofitting problem. The treatment alternatives were represented in a superstructure coupled with a database containing information on performance and economics of treatment units. The optimization problem was solved for different objective functions and
constraints corresponding to different retrofitting scenarios. The framework was evaluated on two fullscale domestic wastewater treatment plants (265,000 PE and 750,000 PE respectively). To this end, the
design space included 9 side-stream treatment technologies and 9 main line technologies resulting in
more than 200 retrofitting alternatives. The retrofitting problem resulted in the selection of a high-rate
oxic reactor coupled with anammox technology in the main line as the optimal solution, due to the
low utility consumption and high biogas production associated with this network. On the other hand,
most of the side stream treatment alternatives have similar technicaleconomic benefits in terms of
the whole plant retrofitting cost function. Therefore the selection of the optimal alternative is more based
on local conditions and preferences in that case.
2015 Elsevier B.V. All rights reserved.

1. Introduction
The wastewater treatment industry faces many challenges such
as increasing energy costs, the presence of trace organics which has
become more critically investigated, the depletion of resources, the
increasing demand for water conservation as well as more stringent regulations [20]. As a result, domestic wastewater is now
being considered more as a resource than as a waste with the
recovery possibilities for clean water, energy and various materials
Corresponding author. Tel.: +45 45 25 28 06.
E-mail address: gsi@kt.dtu.dk (G. Sin).
http://dx.doi.org/10.1016/j.cej.2015.10.088
1385-8947/ 2015 Elsevier B.V. All rights reserved.

satisfied by novel approaches and emerging technological

developments [17].
Economics, i.e. mainly operational and capital cost for the
treatment facilities, has always been one of the key drivers when
making decisions on which treatment methods and technologies
to be applied. Wastewater treatment plants are significant energy
consumers, mainly for aeration, mixing, pumping etc., which is a
challenge in view of rising energy costs and growing concerns
and restrictions on emission of greenhouse gases; thus, wastewater specialists focus increasingly on effective energy management
and alternative energy strategies. Current initiatives include
increasing the biogas production (e.g. [32,30]), managing oxygen

448

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

demand by controlling oxygen concentration (e.g. [33]) or by

decreasing the oxygen demand of the microbial activities (e.g. Partial nitritation described in Hellinga et al. [10] and anaerobic
ammonium oxidation published in Mulder et al. [19]) as well as
controlling the equipment for efficient power use [20]. Moreover,
the increasing construction cost and decreasing space availability
promoted the development and use of more compact systems
which require less footprint (for instance, biological aerated filters
[18] and granular activated sludge [7,9]).
A major necessity in wastewater treatment is the removal of
nutrients especially nitrogen and phosphorus to acceptable
limits prior to effluent discharge. For instance, more stringent
nitrogen limitations in the regulations for the WWTP effluents gave
rise to development of innovative nitrogen removal technologies
mostly used for nitrogen rich streams resulting from sludge treatment [14]. On the other hand, there are also recent developments
for recovering phosphorus in the form of struvite [15] and research
is still ongoing for assimilating excess nutrients by making use of
microalgae as well [5]. As regards plant design and retrofitting,
both the multi-criteria nature of the decision making (economics,
technical feasibility, legislation, etc.) and the large number of alternatives of wastewater treatment technologies bring us to the question How do we take strategic decisions on the wastewater
treatment technologies and process networks to use?
In this study, a systematic framework based on mathematical
programming is used to handle the complex process synthesis
problem by means of a superstructure optimization approach for
generating novel and optimal WWTP process flowsheets for
domestic wastewater in early stage design and retrofitting studies.
The tool consists of the superstructure covering all relevant treatment alternatives and a database storing design parameters (i.e.
volumes, utility consumptions etc.) and performances (removal
efficiencies, sludge productions etc.) for each alternative technology. The solution of the optimization problem provides an optimal
process selection and the optimal flows through the selected network as well as the operational and capital cost component breakdown. In order to highlight the application of the tool, we
formulate a retrofitting problem focusing on the challenge that
many, if not the majority, of wastewater treatment plants in EU
and North America are facing due to tightened legal effluent discharge limits which put a considerable demand on increasing the
nitrogen removal performance. To this end, a Danish WWTP was
used as a case study and the tool was applied to represent the
design space for retrofitting alternatives (novel technologies, side
stream treatment processes as well as main stream) and identify
the optimal solution for various objective functions. The solutions
were further analyzed under various sources of uncertainties
including influent quantity and quality as well as expected utility
costs and effluent limit variations within the project investment
horizon.
Therefore, in this study, we built on the theoretical fundamentals of the superstructure optimization method developed earlier
[4], and comprehensively expanded and validated the methodology by including the following novel features: (1) data collection,
generic model development and validation for the selected treatment technologies (the superstructure includes more than 200
potential alternatives); (2) sensitivity analysis of the optimal solutions; and (3) verification of the applicability of the methodology
on a full-scale and realistic WWTP design/retrofitting problem.

2. Methodology
The superstructure optimization framework developed in our
earlier study [4] for the optimization based design of WWTPs is
used. The framework is illustrated in Fig. 1 and individual steps

Fig. 1. Framework for the superstructure based optimization methodology [4].

are briefly described for the sake of completeness. Wastewater

characterization, receiving environment limitations and the objective function are defined in the first step prior to the superstructure
generation in step two, which contains the base case (i.e. the existing treatment units, only relevant for the retrofitting studies) and
several other treatment unit alternatives. In the third step, the necessary design data for treatment technologies are collected. After
designing each treatment unit based on the collected data and
information, the parameters of the generic process interval model
are calculated. Each treatment unit in the superstructure is defined
using a generic model based on mass input/output and incorporates several phenomena namely mixing, reaction, phase and flow
separation. Therefore, the generic process model can function as
any treatment unit, and is capable of representing either a reactor
or a separation unit, as well as the combination of the two. The
superstructure is coupled with a database containing the parameters of the generic model, e.g. process performance (stoichiometry,
efficiency etc.), utility consumptions, volumes, sludge productions,
together with the cost related data (for instance, electricity cost,
price of a specific chemical, biogas price etc.). In the next two steps,
the optimal wastewater network problem is formulated as a mixed
integer (non)linear programming (MI(N)LP) problem and solved
for deterministic conditions and under uncertainty with respect
to defined/chosen uncertain parameters by defining several scenarios. The mathematical formulation and solution of the optimization problem together with the detailed data collection and
validation of the design procedure can be found elsewhere [4].

2.1. Case study for retrofitting problem definition

Two full-scale plants are used in this study namely: (i) Avedre
WWTP mainly used for investigating the retrofitting problem,
and (ii) Lynetten WWTP used for the validation of the concept.
The design and operation details are given here for the Avedre
plant, while the details for the Lynetten plant can be found in
Appendix 1.
Avedre WWTP is located west of Copenhagen (Denmark), and
receives wastewater from 10 suburban municipalities with a population of approximately 265,000 people. Annually, around 25
30 million m3 of wastewater is treated in the plant by physical,
biological and chemical means of treatment. The produced sludge
is incinerated in a fluidized bed incineration plant. The layout of
the existing plant is shown in Fig. 2.

449

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

Fig. 2. Avedre WWTP layout.

Table 1
Avedre WWTP influent and effluent composition (2012 yearly average values).

Flow rate
COD
BOD
NH4-N
NOx-N
Total-N
Total-P
SS

Unit

Influent

Effluent

m3/d
g COD/m3
g COD/m3
g N/m3
g N/m3
g N/m3
g P/m3
g/m3

72,037
476
180
33

43
6.5
209

67,113
23
2
1.2
3
4.8
0.6
4.6

The WWTP receives organics, nutrients and solids with the

influent wastewater. The yearly average pollutant loads in terms
of concentration units are shown in Table 1 as averaged data for
the year 2012. The table also summarizes the average effluent
water composition of the same year.
For practical purposes, the influent wastewater composition
was characterized in terms of ASM1 components [11]. The conversion procedure and influent characterization after conversion are
illustrated below in Table 2. The influent wastewater was assumed
not to contain any biomass (XBH, XBA), no particulates from biomass decay (XP), nitrate/nitrite nitrogen (SNO), soluble organic
nitrogen (SND) and oxygen (SO).

Accordingly, all the effluent COD was assumed to be in the form

of soluble inert materials SI (Eq. (1)). The ratio of BOD to
biodegradable COD was taken as 0.595 and used in the conversion
procedure [8]. The effluent BOD concentration (i.e. 2 mg/L which
corresponds to approximately 3.4 mg/L of biodegradable COD)
was assumed to be negligible and therefore was not included in
the COD calculation. COD was assumed to be composed of
biodegradable COD and soluble and particulate inert COD;
therefore, Eq. (3) was used to calculate particulate inert
COD XI. Biodegradable COD includes soluble and particulate fractions SS and XS respectively. It was assumed that it consists of
30% soluble and 70% particulate biodegradable COD. It is assumed
that the total nitrogen is composed of ammonium nitrogen SNH
and particulate biodegradable organic nitrogen XND (Eqs. (5)
and (6)). It should be noted that the influent fractionation procedure provided a base case on which comprehensive uncertainty
analysis is further performed to robustify the solutions as part of
the systematic framework in step 5.
The retrofitting problem was defined so that the feasibility of
extending the existing treatment line will be analyzed in two
ways: (1) addition of a new task responsible for nitrogen removal,
and several alternative technologies will be considered in that
respect in the sludge reject water line; and (2) evaluating alternative treatment technologies in the main wastewater treatment

Table 2
Avedre WWTP influent characterization in terms of ASM1 components (2012 yearly average values).
Component

Unit

Value

Description

Equation

SI
SS
XI
XS
XBH
XBA
XP
SO
SNO
SNH
SND
XND

g COD/m3
g COD/m3
g COD/m3
g COD/m3
g COD/m3
g COD/m3
g COD/m3
g COD/m3
g N/m3
g N/m3
g N/m3
g N/m3

23.48
107.85
93.22
251.65
0
0
0
0
0
32.85
0
9.8

SI CODout
SS CODbiodegradable  0:3
XI COD  SI  CODbiodegradable
XS CODbiodegradable  0:7

1
2
3
4

SNH NH4 N

XND Tot N  NH4 N

450

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

line. The objective function was defined such that it covers the
operational and capital cost (only for the new treatment units)
i.e. total annualized cost (TAC) which was to be minimized. The
operational cost covers utility cost (i.e. aeration, electricity consumption, chemical addition etc.), sludge production via landfill
price, biogas price, mixing and pumping.
2.2. Design space formulation superstructure definition
A superstructure can be defined as the representation of different treatment alternatives and their interconnections in a systematic and compact way. The superstructure developed for this
specific retrofitting problem is illustrated in Fig. 3; Table 3 summarizes and describes the treatment units selected in the design space
within the context of this problem.
The process intervals (i.e. individual treatment units) shown in
gray in Fig. 3 are the existing treatment units in the Avedre
WWTP. The wastewater source was defined in the source column
WW, the water line is composed of a primary clarifier PC and
a biological treatment unit AS. The sludge treatment line, on
the other hand, is composed of a sludge thickener, anaerobic digestion unit and a dewatering unit (i.e. Thick, AD and Dewat in the
superstructure). The sink intervals for water and sludge effluents
were represented as Water and Sludge, respectively.
In this specific retrofitting problem, the alternative technologies
were located under three different treatment tasks: Primary Treatment, Secondary Treatment and Reject Water Treatment. Together
with the primary clarifier of the base case, a high-rate oxic reactor
(A-stage) and an Upflow Anaerobic Sludge Blanket reactor (UASB)
were placed under Primary Treatment as alternative treatment
units. Several different biological nutrient removal (BNR) type of
systems with different SRTs (i.e. BNR10, BNR14 and BNR28), a
low-rate oxic system in a pre-denitrification mode (B-stage), an
Anaerobic Ammonium Oxidation reactor operating at low temperatures (i.e. main stream wastewater temperature) Anammox, an
activated sludge type of reactor with granular sludge (Granular
AS) and a membrane bioreactor (MBR) were selected among the
secondary treatment technologies and placed in the superstructure. Finally, within the context of the sludge reject water treatment task, several different treatment units were included in the

Wastewater
source
WW

Primary
Treatment
PC
A-stage

Secondary
Treatment

database and shown in the superstructure. These units are responsible for treating the nitrogen rich water stream resulting from
thickener and dewatering units. Finally, three by-pass units are
added under these three tasks having treatment alternatives; this
allows the optimizer to by-pass this task when it is not needed.
After defining the treatment tasks and alternative treatment
processes, the superstructure definition is finalized by identifying
the interconnections between the process intervals.
2.3. Data collection and design of treatment units
Treatment technologies in the superstructure, including the
base case treatment units, were designed by following the steps
of the previously developed systematic data collection and design
procedure [4]. While designing the technologies, the engineering
standards were used with the reported parameters that included
also safety factors [2,23,29,12]. For the side stream processes and
the innovative main stream processes, the reported efficiencies at
their optimality in the literature were taken into account. Therefore, all the technologies were evaluated at their reported optimality by its developers. Hence, the idea of this study is to cover the
concept screening step, where the decision makers are supported
to quickly generate and evaluate different technological alternatives. In the further steps, the short-listed concepts need to go
through a more rigorous evaluation (see the framework in Fig. 1).
In other words, the intention is to short-list a limited number of
promising candidates for further detailed analysis using more rigorous models, where detailed modeling, simulation and optimization can be done.
In this section, the design specifications of separate treatment
units are given in three different groups: base case treatment units,
main wastewater treatment line alternatives and sludge reject
water stream treatment technologies.
2.3.1. Base case treatment units
The base case units (i.e. the existing treatment technologies)
were designed based on the design data obtained from Avedre
WWTP according to the design models used in this study [4]. The
base case design is summarized in Table 4 in terms of the compositions of the influent, the effluent both sludge and water and

Reject water
Treatment

Sludge
Thickening

Sludge
Stabilization

Sludge
Dewatering

Sinks

Water

AS
BNR10

Thick

AD

Dewat

Sludge

Sharon
BNR14
UASB
BNR28

Shar/An-2st
Shar/An-SBR
Canon

By-pass1
B-stage
Anammox

Anitamox
Deammon
Demon

Granular AS

Canr
Panammox

MBR

By-Pass3

By-pass2
Fig. 3. Superstructure developed for the Avedre WWTP retrofitting study representing the design space considering novel primary, secondary and reject water treatment
technologies.

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

451

Table 3
Treatment technologies represented in the superstructure.
Unit

Description

Unit

Description

WW

Wastewater source

Sharon

PC
A-stage

Base case primary clarifier, removal of particulate materials

High-rate oxic reactor, low SRT system, removal of organics,
1-stage
Upflow anaerobic sludge blanket, removal of organics, biogas
production
Base case activated sludge type of reactor, removal of
organics and nutrients, 3-stage
Biological nutrient removal, SRT = 10, removal of organics
and nutrients, 3-stage
Biological nutrient removal, SRT = 14, removal of organics
and nutrients, 3-stage
Biological nutrient removal, SRT = 28, removal of organics
and nutrients, 3-stage
Low-rate oxic reactor, high SRT, removal of organics and
nutrients, 3-stage
Main stream anaerobic ammonium oxidation reactor,
removal of organics and nutrients, 1-stage
Granular activated sludge reactor, removal of organics and
nutrients, 1-stage
Membrane Bioreactor, removal of organics and nutrients, 3stage

Shar/An-2st
Shar/An-SBR

Thick

Single reactor system for high activity ammonium removal

over nitrite, removal of nitrogen
2 stage Sharon and Anammox reactors, removal of nitrogen
1 stage (SBR) Sharon and Anammox reactors, removal of
nitrogen
Completely autotrophic nitrogen removal over nitrite,
removal of nitrogen
Partial nitritation/Anammox in a moving bed biofilm reactor,
removal of nitrogen
Deammonification in a biofilm type of reactor, removal of
nitrogen
Deammonification in a SBR type of reactor, removal of
nitrogen
Complete autotrophic nitrogen removal in a SBR type of
reactor, removal of nitrogen
Two-step partial nitritation/anammox process, removal of
nitrogen
Base case sludge thickener

AD

Base case anaerobic digester

Dewat

Base case dewatering

Water
Sludge
By-pass

Water effluent
Sludge effluent
By-pass intervals

UASB
AS
BNR10
BNR14
BNR28
B-stage
Anammox
Granular AS
MBR

Table 4
Base case design summary.
Parameter

Flow rate
COD
BOD
NH4-N
NOx-N
Total-N
Total-P
SS
CH4

Unit

m3/d
g COD/m3
g COD/m3
g N/m3
g N/m3
g N/m3
g P/m3
g/m3
g COD/m3

Influent

72,037
476
180
33

43
6.5
209

Effluent

Sludge reject water

Water

Sludge

71,950
36.3

1.78
4.95
8.1
0.45
13

46.2
3.1 105
2 104
328

1.1 104
114
3.2 105
9.2 103

2,489
302
3.8
116
3.2
136
40
280

Canon
Anitamox
Deammon
Demon
Canr
Panammox

2.3.3. Sludge reject water stream treatment units

In the sludge reject water treatment task definition, the main
purpose is removal of nitrogen from this highly concentrated
stream so that the total nitrogen load into the main stream treatment unit is lowered while decreasing the operational cost of this
unit (mainly in terms of aeration requirements and extra carbon
dosage). In the context of this retrofitting study, two different
treatment mechanisms and nine different technology alternatives
are considered under the related task. The Sharon reactor operates
with the nitritation/denitritation mechanism, whereas all the other
treatment units treat nitrogen by a partial nitritation/anammox
mechanism either in a single reactor or separate reactors. The stoichiometry for the two mechanisms is given below in Eqs. (7)(10),
respectively.

Stoichiometry for nitritation/denitritation [10]

the sludge reject water streams. The details of the design procedure can be found in Appendix 2.

2.3.2. Main wastewater line treatment units

As stated earlier, the main wastewater line treatment alternatives are placed under the primary and secondary treatment tasks.
In this section, their design specifications and performance data
are summarized. In Table 5, the design data for conventional activated sludge (AS) type of treatment technologies are shown and
the design details are given in Appendix 2. These technologies were
designed following the steps of a systematic data collection procedure, which is detailed elsewhere [4].
The UASB technology is responsible for COD reduction at 20 C;
however, there is no nitrogen removal occurring in the system. The
Anammox technology operates at low temperatures partly with
partial nitritation/anammox and partly with denitrification mechanisms. Another technology placed under the Secondary Treatment
task is Granular AS which incorporates granulated microorganism
groups performing COD oxidation, nitrification and denitrification.
Finally, the MBR unit is designed as a submerged MBR system,
responsible for organics and nutrient removal. The design details
of the units are presented in Appendix 2.

NH4 1:5O2 ! NO2 H2 O 2H

NO2 0:5CH3 OH 0:5CO2 ! 0:5N2 HCO3 0:5H2 O

Stoichiometry for partial nitritation/anammox [25]

NH4 HCO3 0:75O2 ! 0:5NH4 0:5NO2 CO2 1:5H2 O

9

NH4 NO2 ! N2 2H2 O

10

Table 6 summarizes the design specifications for the treatment

alternatives of the sludge reject water treatment task. The
sizing of the units is done based on the given HRT for the
Sharon and the reported volumetric nitrogen loading / removal rate
values for the other units. In order to determine the performance of
the systems in terms of their nitrogen (total nitrogen or NH4
nitrogen) removal efficiencies, the removal efficiencies at optimal
operating conditions reported by the studies stated in Table 6 are
used.

452

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

Table 5
Design data for AS type of treatment technologies.
Technology

Mechanism

SRT (d)

HRT (h)

Reactor type

Performance

A-stage

COD oxidation

0.5

Aerated reactor + sedimentation

basin

60% biodegradable COD removal

B-stage

COD oxidation
Nitrification
Denitrification
Chemical P removal
COD oxidation
Nitrification
Denitrification
Chemical P removal
COD oxidation
Nitrification
Denitrification
Chemical P removal
COD oxidation
Nitrification
Denitrification
Biological and chemical P removal

13

4 (aerobic)
3 (anoxic)

Pre denitrification type of reactor

+ sedimentation basin

7.2 (aerobic)
2.4 (anoxic)

Pre denitrification type of reactor

+ sedimentation basin

9 (aerobic)
3 (anoxic)

Pre denitrification type of reactor

+ sedimentation basin

15 (aerobic)
5 (anoxic)
0.5 (anaerobic)

Pre denitrification type of reactor

+ sedimentation basin

78% COD
67% total
90% total
Removal
88% COD
84% total
90% total
Removal
87% COD
85% total
90% total
Removal
87% COD
86% total
90% total
Removal

BNR10

BNR14

BNR28

10

14

28

N
P

N
P

N
P

N
P

Table 6
Design data for sludge reject water treatment alternatives.
# of
reactors

Reactor type

Sharon

CSTR

Shar/An2st

Shar/AnSBR
Canon

CSTRgranular
sludge
SBR

Anitamox
Deammon

1
1

Demon
Canr
Panammox

1
1
2

Granular
sludge
MBBR
MBBR
SBR
SBR
SBR-granular
sludge

Volumetric N removal
rate (kg N/m3 d)

Volumetric N loading
rate (kg N/m3 d)

10

0.55

Comments

References

86% for 1st reaction

60% for second
reaction
50% in Sharon
80% in anammox

1 d HRT for
Sharon reactor

Hellinga et al. [10]

0.420.5

90% overall N
conversion to N2
95% for ammonium
81% for total N
85% for ammonium
70% for total N

0.61
0.75
1.8 PN
0.46 Anammox

90%
89%
89%
86%

1.5
0.8
0.30.4

Efficiency

2.4. Generic process interval model generation and MILP problem

formulation
The information about the generated superstructure and the
design of the treatment technologies, which are placed in the
superstructure, is converted into a generic mass inputoutput type
simple model (details are shown in Bozkurt et al. [4]). The optimization problem has a number of parameters defined, and the
values of these parameters are calculated based on the individual
treatment unit designs employed, stored in matrices in an MS
Excel based structure, and finally sent to the optimizer to be used
in the solution of the optimization problem. The table summarizing the parameter values of the generic mass inputoutput model
for each treatment alternative is shown in Appendix 3.
The superstructure optimization problem was formulated as a
Mixed Integer (non)Linear Programming (MI(N)LP) problem as
shown in Eqs. (11)(14). With respect to the nature of the problem,
the optimization problem can indeed result in a linear (MILP) or a
non-linear (MINLP) formulation. In this study, the optimization
problem was formulated as a linear problem, i.e. a MILP formulation. The models represent the mass inputoutput of the components within and in between each treatment technology together
with process constraints, structural constraints, effluent limit constraints, economic models and the objective function; the detailed
mathematical formulation can be found elsewhere [4].

Min OBJ

for
for
for
for

Van Dongen et al. [25]

Van der Star et al. [24]
Joss et al. [13]
Abma et al. [1]
Lemaire et al. [16]
Rosenwinkel and
Cornelius [22]
Wett et al. [31]
Vangsgaard [27]
Personal communication,
Jesus Colprim

ammonium
total N
ammonium
total N

X
CAPEX kk
OPEX kk
t
kk

11

subject to;

hai;kk ; li;ii;kk ; ci;kk;rr ; hreact;kk;rr ; W i;kk ; Split i;kk ; SW kk ; reckk

0 process model

12

gSk;kk 6 0 process constraints

13

X
ykk 6 1 where y 2 f0; 1gn

14

structural constraints

kk

3. Results and discussion

After formulating the optimization problem, it was solved both
under deterministic conditions and under uncertainty.
The results of different scenarios and uncertain domains considered in the Avedre retrofitting problem are summarized and discussed below.
3.1. Plant retrofitting deterministic case
The deterministic solution of the optimization problem results
in the optimal network, the fate of pollutants/components

453

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

throughout the selected treatment network and the value of the

objective function together with the cost breakdown into the components of the objective function (i.e. utility cost, product cost, capital cost etc.).
The Avedre WWTP retrofitting study has been conducted by
considering two different scenarios: (1) process selection for
sludge reject water treatment; and, (2) retrofitting for the entire
plant covering treatment alternatives for main wastewater treatment line and sludge reject water line as well.
3.1.1. Scenario 1 process selection for sludge reject water treatment
In the first scenario, in addition to the base case treatment layout, an additional task has been defined in the superstructure to
treat the nitrogen rich water stream resulting from sludge treatment. The optimization results (represented as the selected alternative providing the most optimal solution to the design
problem) together with the base case cost breakdown and a ranking of all the other alternatives with respect to their objective function values are given in Table 7. The biogas price, pumping cost and
mixing cost are not included in the cost breakdown since they do
not change with the selection of different reject water treatment
alternatives.
It is seen that the base case treatment units perform very well in
terms of nitrogen removal, since the effluent NOx-N and NH4-N
concentrations are very low; however, it should also be noted that
the base case design has been made with respect to the yearly
average influent characterization and performance of the units.
The effect of reject water treatment technologies is known to be
more significant when the treatment plant operates with effluent
total nitrogen concentrations greater than 10 g N/m3.
The Canon technology is a one stage partial nitritation/anammox unit working with granular sludge formation. Its required volume, which has been calculated with respect to the reported
volumetric nitrogen loading/removal rate, is very small and the
utility consumption is very low. Although the base case scenario
has no associated capital cost with it, the overall cost to the plant
becomes lower and the Canon process is therefore favored by the
optimizer. Moreover, a 5% improvement in the performance is
observed in terms of total nitrogen removal.
The Sharon technology, which is ranked as the last alternative
(i.e. the most costly alternative) by the optimizer, requires an extra
utility cost due to the requirement of methanol addition for denitritation and pH control; therefore, it is not feasible for full scale
applications, especially after the discovery of anammox microorganisms and their successful full-scale applications.
2-stage applications (i.e. Panammox and Shar/An-2) are known
to have advantages during operation (e.g. better control of nitritation and anammox) [28]; however, mostly because of the capital
cost requirements, they are not favored by the optimizer and have
also not been favored for full scale applications recently [28,14].

For the commercial and experimental applications of one stage

partial nitritation/anammox processes (i.e. Canon, Anitamox, Canr,
Shar/An-SBR, Demon and Deammon), the difference in terms of cost
and removal efficiency does not seem significant. When the objective function values of the most and the least favored alternatives
(i.e. Canon and Deammon) are compared, the difference seems to be
very low 1.2%. Therefore, in addition to the techno-economic performance metrics presented above, the selection of the optimal
process technology alternative by a decision maker can be based
on other technical and operational aspects of these technologies
such as flexibility, operability, risk for failure, site specific conditions, experience of the plant staff etc., [26,6].
3.1.2. Scenario 2 retrofitting of the entire plant
The second scenario considers a full-scale retrofitting study. In
other words, in this scenario, the alternative treatment technologies are placed in the superstructure not only under the sludge
reject water treatment task, but also under the primary and secondary treatment tasks as illustrated previously in Fig. 3.
Table 8 summarizes the cost breakdown information for the
selected alternative together with the simulation results for other
secondary treatment alternative units. Since all the treatment units
are designed in their optimality, their performances in terms of
removal efficiencies of key contaminants are satisfactory and very
similar to each other. Therefore, the main comparison is done here
based on the cost breakdown information. Among the reject water
treatment alternatives, the Canon technology was selected unless
the task is by-passed. Since the performance and cost related differences among one stage partial nitritation/anammox technologies are very minor as stated previously under the discussion of
scenario 1, the selection of the Canon technology reflects the selection of a one stage partial nitritation/anammox mechanism.
The optimizer favors the A-stage Anammox and Canon units for
primary, secondary and sludge reject water treatment tasks,
respectively. In other words, this network is selected as the best
network in the predefined design space with regard to its reported
cost and performance criteria at the optimum operational conditions of the technologies (i.e. reported removal efficiencies and
utility consumptions in the literature). This network is characterized by its low utility consumption, which mainly results from
the lower oxygen consumption of the anammox technology as
compared to the conventional nitrification/denitrification route.
Another important factor is that only a fraction of COD is oxidized
in the A-stage unit while the rest is directed to the sludge line to be
digested in the AD unit; this also explains the high biogas production. The capital cost associated with the selected network is relatively low as compared to most of the networks shown in the
summary table. The feasibility of the selected network has been
studied using a rigorous simulation model and demonstrated
indeed to be capable of high nitrogen removal without extra car-

Table 7
Summary of results for the deterministic solution of scenario 1 (all the cost parameters are given in unit cost and the concentrations of the nitrogen components are reported in
g N/m3).

Base case
Selected alternative
2
3
4
5
6
7
8
9

Process interval

Utility cost

Landfill cost

Capital cost

OBJ

Eff SNO

Eff SNH

By-pass3
Canon
Anitamox
Canr
Shar/An-SBR
Demon
Deammon
Panammox
Shar/An-2st
Sharon

891
883
884
882
883
883
884
882
883
926

1763
1762
1762
1762
1762
1762
1762
1762
1762
1762

8.08
12.89
16.17
19.85
19.87
24.25
33.09
36.72
35.74

1471
1470
1477
1478
1482
1483
1488
1495
1500
1542

4.79
4.54
4.58
4.53
4.54
4.56
4.58
4.54
4.56
4.61

1.01
0.95
0.96
0.95
0.95
0.95
0.96
0.95
0.95
0.96

454

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

Table 8
Summary of results for the deterministic solution of scenario 2 (all the cost parameters are given in unit cost).

Base case
Selected alternative
2
3
4
5
6
7
a

Network

Utility cost

Landfill cost

Biogas price

Pumping cost

Mixing cost

Capital costa

OBJ

PC-AS By-Pass3-Thick-AD-Dewat
A-stage Anammox Canon Thick-AD-Dewat
PCGranularAS Canon Thick-AD-Dewat
PC-BNR10 Canon Thick-AD- Dewat
A-stage B-stage Canon Thick-AD-Dewat
PC-BNR14 Canon Thick-AD-Dewat
PC-MBR Canon Thick-AD-Dewat
PC-BNR28 Canon Thick-AD-Dewat

891
761
1128
965
949
992
3083
827

1763
871
1145
1387
1165
1383
1060
1377

1914
2354
1830
2355
2584
2185
1832
1929

491
188
191
500
484
496
188
491

240
58

127
156
148
53
219

1630
745
2243
2736
2614
940
3849

1471
1154
1380
2868
2906
3450
3492
4834

The capital cost is calculated based on the volume of the units; no other equipment or supporting materials (i.e. membrane, carriers, filters etc.) are included.

Table 9
Summary of the uncertainty characterization.
Scenario

Parameter

Unit

Mean

Min.

Max.

Probability distribution

SS
SI
XI
SNH

g COD/m3
g COD/m3
g COD/m3
g N/m3

107.85
23.48
93.22
32.85

75.50
16.44
65.25
23.00

140.21
30.52
121.19
42.71

Uniform
Uniform
Uniform
Uniform

Electricity price
O2 transfer efficiency
Total N limit

Euro/kW h
kg O2/kW h
g N/m3

0.08
4.12
10

0.12
5.58
15

Uniform
Uniform
Uniform

bon addition and at low HRT hence corroborating the outcome of

the optimization study [21].
The Granular AS technology is coupled with the PC of the base
case and the Canon unit of the sludge reject water treatment task
and ranked as the second best network favored by the optimizer
with respect to the objective function value. The stoichiometric
oxygen requirement of the Granular AS technology is similar to
the conventional biological nutrient removal system (i.e. oxidation
of organic material and nitrification/denitrification route for nitrogen removal); therefore, the utility cost consisting of aeration
requirement is similar to the conventional BNR units. However,
the main advantage of this technology is that there is no sludge
return flow and internal recirculation flow; moreover, no mixing
is provided. Therefore, the pumping and mixing costs associated
with this network are significantly lower. Last but not least, it is
a rather compact technology requiring less space as can also be
observed from the low capital cost.
The low SRT BNR units (BNR10 and BNR14) selected together
with the PC and Canon units, as well as the combination of Astage and B-stage technologies incorporate chemical phosphorus
treatment which can be observed by the high utility cost reported
for those units; on the contrary, the high SRT BNR28 system has an
anaerobic compartment for biological phosphorus removal, which
increases its capital cost but decreases the utility consumption (i.e.
less chemical addition for chemical precipitation of phosphorus).
An increasing SRT results in less biogas production because less
organics are diverted to the sludge treatment line. When the Astage technology is selected in the network the biogas production
increases as compared to the other BNR units because of the diversion of more organics to the AD unit when this technology is used.
The pumping cost for those four networks is similar; however, the
mixing cost increases with the increased SRT, in other words with
the increasing volume.
The MBR technology is selected together with the PC and Canon
units. The MBR is known to have a high energy consumption, and
the utility cost in the cost breakdown evaluation reflects this: it
includes the energy consumption of the unit together with aeration requirements. On the contrary, the associated capital cost,
pumping and mixing cost are low. The MBR technology is designed
as a compact technology because it operates with high biomass

0.098
4.85
12.5

concentrations, and also, it does not incorporate a settler, which

decreases the system volume (i.e. reactor + settler) significantly.
Since the membrane cost is not incorporated into the capital cost
calculation in this study, the resulting capital cost for the MBR unit
is very low as compared to other units. It is reported by the US EPA
that 1 kW h of electricity contributes to the emission of
6.89  104 metric tons CO2. Therefore, the calculated utility cost
reduction by the selected alternative as compared to the base case
treatment units accounts for a daily reduction of 902 kg CO2.
3.2. Optimal process selection under uncertainties
The uncertainty analysis was done by considering two different
scenarios for the full-scale superstructure shown in Fig. 3: taking
into account (1) the effect of influent composition; and, (2) effect
of cost and effluent limitation uncertainty. The selected uncertain
parameters together with the uncertain domain definition are
given in Table 9. Accordingly, a 30% variation around the deterministic value was assumed in the SS, SI and XI fractions of COD and the
resulting XS concentration was calculated assuming that the total
influent COD concentration is constant. The details of the Uncertainty Analysis procedure are presented elsewhere [4].
First, the deterministic problem was solved for the realization of
50 different future cases (generated using Latin Hypercube Sam-

Table 10
Summary of results for 50 different future cases.
Scenario

Network

60

20

3
4
5
6

8
8
2
2

100

Probability of
realization (%)

Selected intervals
A-stage Anammox
Canon
PC Granular AS ByPass3
PC AS Canon
PC AS By-Pass3
PC Granular AS Canon
A-stage Anammox ByPass3
A-stage Anammox
Canon

455

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

Fig. 4. Cumulative distribution of the objective function for scenario 1 (left) and scenario 2 (right).

Table 11
Summary of results for solution under uncertainty.

Network
Utility cost
Landfill cost
Capital cost
Biogas price
Mixing cost
Pumping cost
OBJ

Scenario 1: uncertainties
in influent characteristics

Scenario 2: uncertainties in cost

and effluent limits

Deterministic solution

A-stage-Anammox-Canon
764
877
1630
2376
58
188
1141

A-stage-Anammox-Canon
765
871
1630
2354
58
188
1157

A-stage-Anammox-Canon
761
871
1630
2354
58
188
1154

pling) and the results are presented in Table 10 together with the
cumulative distribution of the objective function value in Fig. 4.
It is indicated by the results of scenario 1 that the uncertainty on
the influent composition had a significant effect on the network
selection as well as on the value of the objective function. 60% of
the future realizations resulted in the selection of the same network as found for the deterministic solution; however, 5 alternative network selections seem to be possible, despite their low
chance of occurrence. When the distribution of the value of the
objective function is considered, it ranges from 513 to 1938 unit
cost. In other words, 40% of the future realizations have higher
TAC as compared to the deterministic problem solution.
When the results of scenario 2 are analyzed, which takes into
account cost related uncertainty as well as possible changes in
effluent nitrogen limitation, the network selection seems to be
more stable. For all of the future realizations, the deterministic network selection has not changed. Similarly, the variation in the
objective function value is not as significant as for scenario 1; it
changes from 978 to 1333 unit cost, which indicates that 56% of
the future realizations have higher TAC than the deterministic
problem solution.
In a second step, the optimization problem is formulated as a
stochastic programming problem and solved by using the
sample average approximation (SAA) technique [3] in order to calculate the expected value of the objective function within the
whole uncertain domain, i.e. 50 future realizations.
The results of this step are summarized in Table 11. Both
scenarios for the uncertainty domain resulted in the same
network selection as the deterministic solution. Although different
future realizations result in a significant variation, especially for
the objective function value, as previously shown in Fig. 4, when

the results of 50 future scenarios are averaged by solving with

the SAA technique, the effect of uncertainty seems to be insignificant to the selection of the optimal network. It is important to note
that the uncertainty characterization (i.e. selection of uncertain
parameters and definition of the uncertain domain) plays
an important role in the generation of the results and depends
on the users priorities. In scenario 1, the uncertainty had a positive
effect on the solution; in other words, the SAA solution resulted in
a lower objective function value as compared to the deterministic
solution. When the cost breakdown is analyzed, this mainly results
from the increase in the biogas production as a result of the variation in the COD fractionation.
When the results of scenario 2 are analyzed, it can be concluded
that the cost and effluent limitation related uncertainty, with their
defined domain, do not have a significant effect on the problem
solution in terms of both network selection and cost breakdown.
3.3. Impact of changing influent characterization and the treatment
plant scale on the network selection
In order to comment on the effect of changing the influent characterization and the plant capacity on the results of network selection for retrofitting purposes, another case was defined keeping the
design space constant (i.e. the same superstructure was used with
the same design data of the individual technologies). The Lynetten
WWTP is located east of Copenhagen, Denmark. It has a capacity to
serve approximately 750,000 PE by treating around 60 million m3
of wastewater annually (more details are given in Appendix 1).
Its existing treatment line is very similar to the one of the Avedre
WWTP as illustrated previously in Fig. 2. The main difference
between the wastewater quality and quantity are that the Lynetten

456

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

Table 12
Comparison of Lynetten WWTP and Avedre WWTP under deterministic conditions for the full-scale retrofitting case.
Network

Utility cost

Landfill cost

Biogas price

Pumping cost

Mixing cost

Capital cost

OBJ

Avedre WWTP

A-stage Anammox Canon Thick-AD Dewat

761

871

2354

188

58

1630

1154

Lynetten WWTP

A-stage Anammox Canon Thick-AD Dewat

2459

1928

7562

424

130

3668

1047

WWTP treats almost three times the amount of wastewater treated

in the Avedre WWTP; moreover, the incoming COD concentration
657 g COD/m3 is considerably higher in the Lynetten WWTP.
When the results of the deterministic solution for the full scale
retrofitting case for two different wastewater compositions are
compared as shown in Table 12 (i.e. Avedre WWTP vs. Lynetten
WWTP), the ranking of the network selection does not change significantly and the above-discussed points under Section 3.1 seem
to hold for both cases. However, one very important conclusion
is that since the influent organics content for the Lynetten WWTP
is higher compared to the Avedre WWTP, the biogas production
increases significantly. When comparing the selected network cost
breakdown for the two cases, the value of the objective function
decreases although the size of the treatment plant increases almost
three fold. The underlying reason is as follows: While Lynetten,
with the larger treatment capacity, has a higher utility cost, landfill
cost, capital cost as well as pumping and mixing costs, these higher
costs are offset by a much higher biogas production. The net effect
of this is that Lynetten has a lower economic cost function

compared to Avedre plant. It is noted that WWTPs concepts very

much depend on the influent wastewater characteristics; therefore
selection of the optimal concept for WWTP retrofitting could be
different for WWTPs with wastewater characteristics that are different from those analyzed here.
3.4. Sensitivity analysis
In order to comment on the effect of uncertain parameters and
their variation on the problem solution, a sensitivity analysis had
been conducted. The scatterplots shown in Fig. 5 represent the
variation of influent masses of three groups of contaminants (given
on the x-axis corresponding to the 50 LHS samples from the abovementioned uncertainty domain) and their effect on the objective
function value (given on the y-axis). Accordingly, in Fig. 5a, we
observe a pattern in which the increase in the biodegradable
COD resulted in a decrease in the objective function value. A closer
analysis of the objective function and its components indicates that
increasing biodegradable COD leads to an increase in the utility

Fig. 5. Scatterplots for sensitivity analysis.

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

cost but also a higher gain in the biogas production (and associated
increase in biogas price) that offsets the increase in the utility cost.
The net result is then a decrease in the objective function. On the
other hand, a possible increase in the influent particulate inert
COD (Fig. 5b), causes an increase in the objective function value.
This is mainly resulting from the increase in the sludge production
and related disposal cost. When the third plot (Fig. 5c) is considered, there is no systematic/strong pattern observed between the
range of variations in the influent ammonium nitrogen and the
overall objective function value. This suggests that the side stream
treatment constitutes a buffering effect against stochastic variation
in the influent ammonium nitrogen concentration, which in turn
treats the nitrogen to the desired levels without significantly
affecting the total economic objective function of the plant.
3.5. Critical summary
A design space covering novel treatment technologies was
defined and the superstructure was formulated accordingly. The
problem formulation was done so that the comprehensive design
space covers not only alternative technologies and network definitions for the already existing primary and secondary treatment
tasks, but also introduces a new task responsible for sludge reject
water treatment and several alternative technologies responsible
for it. Overall, the superstructure includes 18 new technologies
resulting in approximately 240 different possible network designs.
The optimization problem was formulated as a MILP problem and
solved for different scenario definitions under both deterministic
conditions and uncertainty. The main conclusions of the study
can be summarized as:

The addition of innovative side stream treatment alternatives is
intended to contribute to the discussion about the need for retrofitting the existing treatment line in order to overcome the
defined nitrogen removal problem. On the other hand, the main
line has been extended to cover not only innovative technologies such as granular activated sludge, cold-anammox and
MBR but also well-known, traditional BNR technologies in order
to let the user compare several concepts/technologies. The
application of a high rate oxic reactor (A-stage) coupled with
anammox technology in the main wastewater treatment line
and an anaerobic digester in the sludge treatment is ranked as
the best design concept by the optimizer with its low utility
consumption, high biogas production and relatively low area
requirement.

By solving the problem under uncertainty, the effect of the variations in influent wastewater composition, cost and effluent
limitation parameters on retrofitting solutions are investigated.

Among the sludge reject water treatment alternatives, the commercial one stage partial nitritationanammox technologies
considered in the design space is found to be better with respect
to economic criteria as compared to two stage partial nitritationanammox as well as nitritation/denitritation concepts as
a result of the plant-wide optimization. However it should be
noted that operational aspects such as stability, robustness,
maintenance etc., are crucial factors while selecting the appropriate technology for a defined facility, and they should be
taken into account in further steps of the decision making
process.

Finally, the defined uncertain parameters with their selected
domain did not show a significant impact on the problem solution; however, it was also observed that the future scenarios
could result in a high variation in terms of their TAC values,
which indicates the importance of considering uncertainty in
the early stage design/retrofitting studies.

457

It is important to recall that this study focused on the first step

in process design/retrofitting which involves the generation of
ideas and concepts, and emphasizes screening a larger space
of potential design alternatives. Hence other uncertainties
including changes in COD/N fractions in the influent or deviations in the reported removal efficiencies can be addressed in
the subsequent steps of process design. This can be done by performing detailed analysis using rigorous models (including
dynamics) as well as data from pilot or full-scale plants if available (as presented in the framework in Fig. 1).

The formulation of the superstructure (i.e. the selection of treatment technologies and definition of feasible connection
streams) is a step where expert knowledge and justification is
required. This step could be supported by expert knowledge
by using artificial intelligence methods indeed be done in a
more systematic way (for instance as in [6]), which is the currently a topic of ongoing research. However, this has not been
included in the current study. Our future work however
addresses this grand challenge by extending the optimization
based methodology validated here with artificial intelligence
based expert systems.
4. Conclusion
In this contribution, the early stage decision making for WWTP
design and retrofitting studies was addressed, since this has
become a formidable challenge considering the increasing number
of treatment technologies combined with other drivers such as the
need for nutrient recovery, energy efficiency, water and sludge
reuse. A previously developed methodology has been used to manage the complexity and support optimization based retrofitting on
a realistic full-scale wastewater treatment plant case study. The
framework is demonstrated to be a promising tool to help support
process design engineers in early stage screening and identifying
novel treatment concepts.
Acknowledgement
This study has been carried out with the main financial support
from the Technical University of Denmark (DTU) for the PhD project of Hande Bozkurt.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.cej.2015.10.088.
References
[1] W.R. Abma, W. Driessen, R. Haarhuis, M.C.M. van Loosdrecht, Upgrading of
sewage treatment plant by sustainable and cost-effective separate treatment
of industrial wastewater, Water Sci. Technol. 61 (7) (2010) 17151722.
[2] ATV-DVWK Standard A131E Dimensioning of single-stage activated sludge
plants (2000).
[3] J.K. Birge, F. Louveaux, Introduction to Stochastic Programming, Springer, New
York, USA, 1997.
[4] H. Bozkurt, A. Quaglia, K.V. Gernaey, G. Sin, A mathematical programming
framework for early stage design of wastewater treatment plants, Environ.
Model. Softw. 64 (2015) 164176.
[5] T. Cai, S.Y. Park, Y. Li, Nutrient recovery from wastewater streams by
microalgae: status and prospects, Renew. Sustain. Energy Rev. 19 (2013)
360369.
[6] J. Comas, J. Alemany, M. Poch, A. Torrens, M. Salgot, J. Bou, Development of a
knowledge-based decision support system for identifying adequate
wastewater treatment for small communities, Water Sci. Technol. 48 (11
12) (2003) 393400.
[7] M.K. de Kreuk, J.J. Heijnen, M.C.M. van Loosdrecht, Simultaneous COD,
nitrogen, and phosphate removal by aerobic granular sludge, Biotechnol.
Bioeng. 90 (6) (2005) 762769.

458

H. Bozkurt et al. / Chemical Engineering Journal 286 (2016) 447458

[8] C. Fall, E.-.. Millan-Lagunas, K.M. Ba, I. Gallego-Alarcon, D. Garcia-Pulido, C.

Diaz-Delgado, C. Solis-Morelos, COD fractionation and biological treatability of
mixed industrial wastewaters, J. Environ. Manage. 113 (2012) 7177.
[9] A. Giesen, L.M.M. de Bruin, R.P. Niermans, H.F. van der Roest, Advancements in
the application of aerobic granular biomass technology for sustainable
treatment of wastewater, Water Pract. Technol. 8 (1) (2013) 4754.
[10] C. Hellinga, A.A.J.C. Schellen, J.W. Mulder, M.C.M. vanLoosdrecht, J.J. Heijnen,
The SHARON process: an innovative method for nitrogen removal from
ammonium-rich wastewater, Water Sci. Technol. 37 (9) (1998) 135142.
[11] M. Henze, W. Gujer, T. Mino, M.C.M. van Loosdrecht, IWA Scientific and
Technical Report on Activated Sludge Models ASM1, ASM2, ASM2d and ASM3.
IWA Task group on Mathematical Modelling for Design and Operation of
Biological Wastewater Treatment, 2000.
[12] M. Henze, M.C.M. van Loosdrecht, G. Ekama, D. Brdjanovic, Biological
Wastewater Treatment Principles, Modelling and Design, IWA Publishing,
Glasgow, 2008.
[13] A. Joss, D. Salzgeber, J. Eugster, R. Knig, K. Rottermann, S. Burger, P. Fabijan, S.
Leumann, J. Mohn, H. Siegrist, Full-scale nitrogen removal from digester liquid
with partial nitritation and anammox in one SBR, Environ. Sci. Technol. 43
(2009) 53015306.
[14] S. Lackner, E.M. Gilbert, S.E. Vlaeminck, A. Joss, H. Horn, M.C.M. van Loosdrecht,
Full-scale partial nitritation/anammox experiences an application survey,
Water Res. 55 (2014) 292303.
[15] K.S. Le Corre, E. Valsami-Jones, P. Hobbs, S.A. Parsons, Phosphorus recovery
from wastewater by struvite crystallization: a review, Crit. Rev. Environ. Sci.
Technol. 39 (2009) 433477.
[16] R. Lemaire, I. Liviano, S. Ekstrm, C. Roselius, J. Chauzy, D. Thornberg, C.
Thirsing, S,. Deleris. 1-stage Deammonification MBBR process for reject water
sidestream treatment: investigation of start-up strategy and carriers design,
in: Conference Proceedings WEF Nutrient Recovery and Management 2011,
Miami.
[17] P.L. McCarty, J. Bae, J. Kim, Domestic wastewater treatment as a net energy
producer can this be achieved?, Environ Sci. Technol. 45 (17) (2011) 7100
7106.
[18] L. Mendoza-Espinosa, T. Stephenson, A review of biological aerated filters
(BAFs) for wastewater treatment, Environ. Eng. Sci. 16 (3) (1999) 201216.
[19] A. Mulder, A.A. van de Graaf, L.A. Robertson, J.G. Kuenen, Anaerobic
ammonium oxidation discovered in a denitrifying fluidized bed reactor,
FEMS Microbiol. Ecol. 16 (1995) 177184.

[20] R. Reardon, J. Davel, D. Baune, S. McDonald, R. Appleton, R. Gillette,

Wastewater treatment plants of the future: current trends shape future
plans, Florida Water Resour. J. 10 (2013) 814.
[21] P. Regmi, B. Holgate, D. Fredericks, M.W. Miller, B. Wett, S. Murthy, C.B. Bott,
Optimization of a mainstream nitritationdenitritation process and anammox
polishing, Water Sci. Technol. 72 (4) (2015) 632642.
[22] K.H. Rosenwinkel, A. Cornelius, Deammonification in the moving-bed process
for the treatment of wastewater with high ammonia content, Chem. Eng.
Technol. 28 (1) (2005) 4952.
[23] G. Tchobanoglous, F.L. Burton, H.D. Stensel, Wastewater Engineering:
Treatment and Reuse, McGraw-Hill Publishing, New York, 2003.
[24] W.R.L. Van der Star, W.R. Abma, D. Blommers, J.W. Mulder, T. Tokutomi, M.
Strous, C. Picioreanu, M.C.M. van Loosdrecht, Startup of reactors for anoxic
ammonium oxidation: experiences from the first full-scale anammox reactor
in Rotterdam, Water Res. 41 (2007) 41494163.
[25] U. Van Dongen, M.S.M. Jetten, M.C.M. van Loosdrecht, The SHARONanammox
process for treatment of ammonium rich wastewater, Water Sci. Technol. 44
(1) (2001) 153160.
[26] M.C.M. Van Loosdrecht, S. Salem, Biological treatment of sludge digester
liquids, Water Sci. Technol. 53 (12) (2006) 1120.
[27] A.K. Vangsgaard, Modeling, experimentation and control of autotrophic
nitrogen removal in granular sludge systems (PhD thesis), Technical
University of Denmark, Denmark, 2013. pp. 204.
[28] S.E. Vlaeminck, H. De Clippeleir, W. Verstraete, Microbial resource
management of one-stage partial nitritation/anammox, Microb. Biotechnol. 5
(3) (2012) 433448.
[29] WEF and ASCE/EWRI, Design of Municipal Wastewater Treatment Plants, WEF
Press, Virginia, 2010.
[30] P. Weiland, Biogas production: current state and perspectives, Appl. Microbiol.
Biotechnol. 85 (2010) 849860.
[31] B. Wett, M. Hell, G. Nyhuis, T. Puempel, I. Takacs, S. Murthy, Syntrophy of
aerobic and anaerobic ammonia oxidisers, Water Sci. Technol. 61 (8) (2010)
19151922.
[32] S. Yadvika, T.R. Sreekrishnan, S. Kohli, V. Rana, Enhancement of biogas
production from solid substrates using different techniques a review,
Bioresour. Technol. 95 (2004) 110.
[33] L. mand, G. Olsson, B. Carlsson, Aeration control a review, Water Sci.
Technol. 67 (11) (2013) 23742398.