Documente Academic
Documente Profesional
Documente Cultură
The most used process for biological nitrogen removal from municipal and industrial wastewaters is the
activated sludge process. Because of the importance of this process, as well as the large number of existing
facilities, a lot of research effort has been focused on optimizing the operation strategies or improving the
individual plant design. However, the systematic optimization of the process structure (process synthesis)
and operation conditions based on rigorous process models has not been presented in the literature. The
objective of this work is to address the simultaneous optimization of the process configuration and equipment
dimensionssi.e., process synthesis and designsand the operation conditions of activated sludge wastewater
treatment plants for nitrogen removal based on a superstructure model. The model embeds up to five reactors
and a secondary settler, and allows flow distribution of the main process streams, i.e., nitrate and sludge
recycle streams and fresh feed, along the reaction zone. The objective function is to minimize the net present
value formed by investment and operating costs, while verifying compliance with the effluent permitted limits.
The investment cost computes the reaction tanks, aeration systems, secondary settler, influent pumping station,
and sludge pump costs. The operation cost computes the cost for pumping, aeration, dosage of an external
carbon source, excess sludge treatment for disposal, and fines according to pollution units discharged. Influent
wastewater flowrate and composition are assumed to be known. The activated sludge model no. 3 and the
Takács model are selected to describe the biochemical processes and the secondary settler, respectively. This
results in a highly nonlinear system with nonsmooth functions. Because of the problem complexity, in this
first approach, a nonlinear programming (NLP) problem (specifically a nonlinear programming with
discontinuous derivatives (DNLP) problem) is proposed and solved to obtain some insights for future models.
It was implemented and solved using general algebraic modeling system (GAMS). Results for case studies
are presented and discussed.
cal process rates and settler model include highly nonlinear accommodate the biological reactions. Feeding strategies includ-
functions. Moreover, the settler model includes nonsmooth ing stream distribution to any point of the reaction zone and
functions such as min/max that lead to special models called water recycles from one zone to another are a common feature
nonlinear programming with discontinuous derivatives problems of combined stabilization-denitrification systems. The stream
(DNLPs), which, in general, may cause numerical problems. leaving the reaction zone is generally fed into a sedimentation
To avoid this, a possibility is the introduction of binary variables basin to separate the stream into the cleaned effluent and the
to model the nonsmooth functions. As mentioned, to contem- sludge, a fraction of which is recycled back to the reaction zone.
plate all structural possibilities, a superstructure must be A fraction of sludge, called waste sludge, is purged from the
introduced. The natural approach to handle superstructure recycle line to compensate for the increase in biomass concen-
models is the MINLP methodology. Moreover, rigorous invest- tration due to biomass growth during the biodegradation
ment cost functions usually consider fixed costs, which are processes.
related to the existence or not of a process unit. Those functions According to Van Haandel et al.,1,12 the most used activated
are discontinuous, having different parameter values for different sludge wastewater treatment plant (ASWWTP) configurations
ranges of the equipment characteristic dimension. Finally, in a providing the different environmental conditions for biological
design problem, the secondary settler dimensions (depth and nitrogen removal and organic matter oxidation are presented in
transversal area) and feed allocation point might be optimization Figure 1. The reduction of carbonaceous matter and the
variables. To consider all these aspects, additional binary nitrification process (ammonium is converted to nitrate by
variables must be introduced into the model. autotrophic bacteria) are favored by aerobic conditions, while
The objective of this work is to address the simultaneous the denitrification process (nitrate is converted to nitrogen gas
optimization of the process configuration and equipment by heterotrophic bacteria) is favored by anoxic conditions if
dimensionssi.e., process synthesis and designsand the opera- readily biodegradable organic matter is available. Anoxic zones
tion conditions of activated sludge wastewater treatment plants can be placed either at the beginning (predenitrification con-
for nitrogen removal based on a superstructure model. In this figuration) or at the end of the reaction zone (postdenitrification).
first approach, a DNLP problem is proposed to overcome most In a predenitrifying system, an internal recirculation flow is
of the above-mentioned difficulties and to obtain some insights usually introduced to transport the nitrate-rich liquid back to
for the construction of more rigorous models in the future. The
the anoxic zone. The anoxic zone may require an external carbon
model embeds up to five reaction compartments and a secondary
dosage to facilitate denitrification if the influent readily
settler and allows for flow distribution of the main process
biodegradable matter is consumed by aerobic microorganisms
streams, i.e., nitrate and sludge recycle and fresh feed streams
in the nitrification zone. This is even worse when the fresh
and external carbon source dosage along the reaction zone. The
wastewater stream has a low carbon/nitrogen ratio.
objective function is to minimize the NPV considering invest-
ment and operating costs. In all case studies, a chain of reaction Two basic process design classes can be identified, depending
compartments in series followed by a decanter are the available on the type of energy source that is utilized by the heterotrophic
pieces of process equipment, whose dimensions (continuous organisms to accomplish denitrification: internal or self-
design variables) are to be optimized. The selection of a com- generated. In processes that use an internal energy source, the
partment type, i.e., an aerated or anoxic unit, is to be chosen organisms make use of the influent biodegradable material to
with a continuous variable: the aeration flowrate. When contin- extract the energy they need. This requires that the anoxic
uous variables (reaction compartment volume, decanter area, reaction compartment has to be placed first in the network. Since
and flowrate of aeration, fresh wastewater, recycles, and external nitrate is not present in the influent, it has to be brought into
carbon source dosage to each reaction compartment) take a zero the anoxic reactor by recycling part of the effluent of an aerobic
value at a solution point, the corresponding unit and/or stream reactor. A process known as Ludzack Ettinger (Figure 1a)
is removed from the superstructure. In future works, new utilizes two reactors with partial communication, with the sludge
mathematical models based on MINLP or general disjunctive being recycled to the aerobic part. Recirculation of nitrified
programming (GDP) methodologies will be presented. liquor takes place in a rather indeterminate fashion by the mixing
action in the two reactors. As a consequence, the performance
2. Process Description of this process with respect to the reduction in total nitrogen
content is very poor. Another process, the modified Ludzack
In general, activated sludge systems employ aerated, tubular, Ettinger (MLE) (Figure 1b), uses separate anoxic and aerobic
well-mixed reactors or a series of well-mixed reactors to reactors and recycles both sludge and nitrified liquor to the
Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007 7499
Table 1. ASM3 Model Compounds (3) a process superstructure model with a maximum of five
symbol component unit reaction compartments and one secondary settler, and (4) a
SI soluble inert organics (g of COD) m-3
defined cost model computing operation and investment costs.
SS readily biodeg. substrates (g of COD) m-3 In this first approach, a NLP (DNLP) problem is proposed and
XI inert particulate organics (g of COD) m-3 solved for different case studies.
XS slowly biodeg. substrates (g of COD) m-3
XH heterotrophic biomass (g of COD) m-3
XA autotrophic biomass (g of COD) m-3 4. Process Model
XSTO internal storage product (g of COD) m-3
XSS suspended solids (g of SS) m-3 The considered process superstructure is shown in Figure 2a.
SO dissolved oxygen (g of O2) m-3 In order to facilitate the model description, a more detailed
SNOX nitrate and nitrite N (g of N) m-3
SN2 dinitrogen (g of N) m-3
superstructure is given in Figure 2b.
SNH ammonia and ammonium N (g of N) m-3 Basically, the plant superstructure model consists of a
SALK alkalinity (g of COD) m-3 maximum of five reaction compartments, a secondary settler,
pumps and stream mixers, and splitters. The design variables
anoxic reactor in order to achieve a better performance. By (the volume of each compartment (Vi) and the secondary settler
default, the processes using an internal energy source cannot cross-area (Asett)) are to be optimized together with the oper-
produce nitrate-free effluents. This is because both recycle and ation variables (flowrate of aeration and process streams). The
effluent flows come from the aerated reactor and contain a T
fresh feed (Qfresh ), the recycle streams, and the external carbon
concentration of nitrate. Figure 1 depicts other process con- source dosage (uECSD) can be distributed into one or more of
figurations attempting to obtain low nitrate concentrations in the five reaction compartments. The superstructure has two
the effluent by using different recycle strategies and by possible internal (nitrate) recycle streams and one external
alternating aerobic and anoxic zones. T
(sludge) recycle stream. The external recycle stream (Qr,ext )
Here, a model for optimal synthesis and design that considers pumps a fraction of sludge from the secondary settler under-
a superstructure that embeds a vast number of activated sludge flow back to the reaction zone. The two possible internal (nitrate)
process configurations, given specified design criteria and cost recycle flowrates are QTr,int,1 and QTr,int,2. The first recycles a
data, is presented. fraction of the mixed liquor from the last to the rest of the
reaction compartments, and QTr,int,2 recycles from the fourth to
3. Problem Definition the preceding compartments, as is shown in Figure 2b.
The problem addressed in this paper is the simultaneous The reaction compartment volumes can range from zero to a
optimization of the system structure (process configuration), the given arbitrary maximum value. A zero reaction compartment
design (equipment dimensions, i.e., reaction compartments volume indicates that it is eliminated from the superstructure.
volume and secondary settler transversal area), and the operating Finally, reaction compartments can operate under anoxic or
conditions (flowrates of aeration, recycles, and fresh feed aerobic conditions, depending on the optimal value computed
wastewater to each reaction compartment and external carbon for the oxygen transfer coefficient kLa. If the kLa value for a
source dosage) of activated sludge WWTPs for nitrogen given compartment is zero, an anoxic reactor is selected.
removal, aimed at minimizing the net present value (NPV), The preference for a given plant configuration over the others
given the following: (1) defined influent wastewater specifica- depends on the influent wastewater flowrate and composition
tions (composition and flowrate), (2) effluent permitted limits, and the used cost functions, as well as on several economical
7500 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
[ ]
F1 hydrolysis XS/XH
F 1 ) kH X
KX + XS/XH H
[ ][ ]
F2 aerobic storage of SS SS SO
F2 ) kSTO X
KS + S S KO 2 + S O H
[ ][ ][ ]
F3 aerobic storage of SS KO 2 SNO SS
F3 ) kSTO‚ηNOX X
KO2 + SO KNOX + SNO KS + SS H
[ ][ ][ ][ ]
F4 aerobic growth SO SNH SALK XSTO/XH
F 4 ) µH X
KO2 + SO KNH + SNH KALK + SALK KSTO + XSTO/XH H
[ ][ ][ ][ ][ ]
F5 anoxic growth (denitrification) SO SNO SNH SALK XSTO/XH
F 5 ) µH X
KO2 + SO KNOX + SNO KNH + SNH KALK + SALK KSTO + XSTO/XH H
[ ]
F6 aerobic endogenous respiration SO
F6 ) bH,O2 X
KO 2 + S O H
[ ][ ]
F7 anoxic endogenous respiration KO2 SNO
F7 ) bH,NOX X
KO2 + SO KNOX + SNO H
[ ]
F8 aerobic respiration of XSTO SO
F8 ) bSTO,O2 X
KO2 + SO STO
[ ][ ]
F9 anoxic respiration of XSTO KO 2 SNO
F9 ) bSTO,NOX X
KO2 + SO KNOX + SNO STO
[ ][ ][ ]
F10 anoxic growth of XA (nitrification) SO SNH SALK
F10 ) µA X
KA,O2 + SO KA,NH + SNH KA,ALK + SALK A
[ ]
F11 aerobic endogenous respiration SO
F11 ) bA,O2 X
KA,O2 + SO A
[ ][ ]
F12 anoxic endogenous respiration KA,O2 SNO
F12 ) bA,NOX X
KA,O2 + SO KA,NOX + SNO A
and technological aspects and trade-offs. It is clear that The ASM3 model considers 13 compounds (Cx), which are
conventional processes described in Section 2 are embedded in divided into soluble compounds and particulate compounds, for
the superstructure and, hence, are candidates for the optimal which concentrations are indicated by S and X, respectively
flowsheet resulting from the assumed hypotheses. (Table 1), i.e., in eq 1, C can be S or X. The ASM3 involves 12
4.1. Reactor Model. For the aeration tanks, steady-state transformation processes, which are listed in Table 2 together
continuous stirred tank reactors (CSTRs) are considered, and it with their process rate equations.
is assumed that no biological reactions take place in the For dissolved oxygen, eq 1 is modified to account for gas-
secondary settler. The activated sludge model no. 3 (ASM3)12 liquid mass transfer,
is chosen as the biological process model. This model considers
both the elimination of the carbonaceous matter and the removal Qi
of the nitrogen compounds. The ASM3 is presently the most (S - SO,i) + kLai(SO,sat - SO,i) + rSO,i ) 0, ∀i
Vi O,i,in
widely accepted model for description of biological nitrogen
removal in activated sludge systems. For each model component where SO,sat is the oxygen saturation constant at 15 °C (SO,sat )
x and reactor i, 8 (g of O2)‚m-3).
Finally, the volume of reaction compartment i is defined as
Qi a positive variable. The following constraint is introduced,
(C - Cx,i) + rx,i ) 0, ∀i,x * O2 (1)
Vi x,i,in
Vi e Vmax, ∀i (3)
where Qi is the volumetric flowrate that enters and leaves
reaction compartment i, Vi is the volume of reaction compart- where Vmax is a sufficiently large upper bound for reactor
ment i, Cx,i and Cx,i,in are the concentrations of component x volumes. To avoid numerical problems (e.g., division by zero),
inside and at the inlet of the reactor i, respectively. The reaction very small lower bounds for reaction compartment volumes are
term rx,i, for each compound x and reactor i is computed as set (Vi,min ) 0.01 m3); however, when an optimal reaction
follows, volume Vi reaches the lower bound, it is considered as a zero
volume reaction compartment and is consequently “deleted”
rx,i ) ∑K υk,c .Fk,i,
x
∀i,x (2) from the superstructure. Finally, the following constraints are
considered for the mass transfer coefficient kLai in each
compartment i, which is defined as a positive variable,
where Fk,i is the kth process rate in reactor i and υk,c are the
stoichiometric coefficients. kLai e kLai,max, ∀i (4)
Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007 7501
Xx,sett,in Xx,m
) , ∀x,m (5)
Xsett,in Xm
νs,m(X) ) max{0,min[νo′,νo(e-rh(Xm-Xmin) -
e-rp(Xm-Xmin))]}, ∀m (8)
Xmin ) fnsXsett,in (9)
{ }
modeled as a nonreactive settling tank subdivided into 10 layers
of equal thickness, using the double-exponential settling velocity νs,mXm if Xm-1 e Xt
, m ) 7, ..., 10 (11)
model of Takács et al.14 According to a comparative study of min (νs,mXm,νs,m-1Xm-1) otherwise
several sedimentation models,15 this settler model provides the
most reliable results. Here, both a fixed settler depth of 4 m The threshold concentration Xt is adopted in such a form to
and a feed point allocation at the sixth layer from the bottom limit the solids downward flux to that which can be handled by
are adopted.11 However, its cross-area (Asett) results from the layer below. For example, above the feed layer, the flux
optimization. leaving layer m is restricted, if the concentration in layer m -
In order to shorten the notation, the total suspended solids 1 is greater to or equal than some threshold value (Xt), in which
concentration XSS is renamed to X in the settler model equations. case the flux leaving layer m is set equal to the min
Figure 3 schematizes the settler model, which consists of five (νs,mXm,νs,m-1Xm-1). According to Takács model,14 Xt is equal
different groups of layers depending on their relative position to 3000 g m-3.
to the feed point. It also shows the streams due to the bulk The resulting steady-state compound balances around each
movement of the liquid and to gravity settling involved in the layer are the following:
mass balance around each layer. This balance depends on
whether the component is particulate or soluble. The movement Mass balances for the sludge (particulate components)
of soluble compounds across the settler is only due to the bulk For the feed layer
movement of the liquid, whereas the movement of particulate
compounds is due to the bulk liquid movement and to gravity (Qsett,inXsett,in/Asett) + Jclar,m+1 -
settling. (νup + νdn)Xm - Jsed,m
0) , m)6 (12)
The particulate component flux depends on the solid con- hm
centration but not on the solid composition. For any particulate
compound concentration Xx, the following holds where Qsett,in and Xsett,in are the volumetric flowrate and the
7502 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
Asett e Asett,max (20) where Qi,in,g is the gth entering stream flowrate and Qi is the
stream flowrate leaving mixer i.
where Asett,max is a maximum design limit, which is here set at The following are the component mass balances,
1500 m2.
4.3. Splitter Mass Balances. For the feed stream, ∑
G
Qi,in,gCx,i,in,g + ux,i,in ) QiCx,i,in, ∀i,x (30)
[ ( )
equipment characteristic dimension; for instance, for the reaction
kLaiVi 2
system investment cost, the volume Vi of each compartment i
is considered, and for the aeration system, the oxygen capacities
Ea ) 24 ∑I (2267 × 10-7)
24
+
( )]
OxCai(OxCai ) BkLaiVi) are used as characteristic dimensions. kLaiVi
For the secondary settler, the cross-area Asett is considered, (5.612 × 10-3) (50)
computing two different investment costs, namely, the decanter 24
tank construction cost and the corresponding electromechanical
system cost. For the influent pumping station investment cost, The effluent quality index EQ, which is related to the fines
which computes costs related to concrete, screws, and screening, to be paid due to contaminant discharge, is computed by
the characteristic dimension is the influent wastewater flowrate weighting the compounds loads having an influence on the water
T T quality that are usually included in the legislation. It is defined
(Qfresh ). Finally, the influent wastewater flowrate (Qfresh ) is
used as the characteristic dimension for the sludge pump cost. as
The investment cost of the plant is computed as follows,7
Effluent quality index ((kg of contaminating unit) d-1)
ICT ) ICt + ICa + ICsett + ICips + ICsr (37) 1
EQ ) (1000 )(β SSXSS,ef + βCODCODef + βBODBODef +
where βTKNTKNef + βNOSNO,ef)Qef (51)
5
∑
where XSS,ef, SNO,ef, CODef, BODef, and TKNef are the concen-
ICt ) ( btViδ ) t
(38)
tration of suspended solids, the concentration of nitrate and
i)1
nitrite nitrogen, CODef, BODef, and the total Kjendal nitrogen
5 in the clarified effluent; Qef is the flowrate of the clarified liquid;
ICa ) ( ∑ baOxCaiδ ) a
(39) and βy are the weighting factors to convert the contaminant y
i)1 into contaminating units. In addition,
7504 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
CODef ) SS,ef + SI,ef + XS,ef + XI,ef + XH,ef + Table 4. Effluent Threshold Values
XA,ef + XSTO,ef (52) contaminant, Cy threshold value, Cy,lim
SNH,ef ((g of N) m-3) 4
BODef ) 0.25(SS,ef + XS,ef + 0.8(XH,ef + XA,ef + XSTO,ef)) (53) NTOT,efa ((g of N) m-3) 18
BODef ((g of COD) m-3) 10
TKNef ) (0.01SI,ef + 0.03SS,ef + SNH,ef + 0.0426XS,ef + CODef ((g of COD) m-3) 100
XSS,ef ((g of SS) m-3) 30
0.02XI,ef + 0.7(XH,ef + XA,ef)) (54) a NTOT,ef is the total nitrogen: NTOT,ef ) TKNef + SNO,ef ((g of N) m-3).
Table 10. Main Variables Optimal Values for Cases (a) I.A, (b) I.B,
and (c) I.C
contaminant effluent values costs
(a) Solution I.A
SNH,ef, (g of N) m-3 0.69 OCT,annual 386 148.53
NTOT,ef, (g of N) m-3 2.18 ICT 2 662 505.98
BODef, (g of O2) m-3 1.41 NPV 7 473 916.66
CODef, (g of O2) m-3 48.18
XSS,ef, (g of SS) m-3 14.36
(b) Solution I.B
SNH,ef, (g of N) m-3 1.94 OCT,annual 345 130.02
NTOT,ef, (g of N) m-3 3.82 ICT 2 717 580.57
BODef, (g of O2) m-3 0.90 NPV 7 017 900.62
CODef, (g of O2) m-3 49.58
XSS,ef, (g of SS) m-3 15.15
(c) Solution I.C
SNH,ef, (g of N) m-3 0.38 OCT,annual 501 878.11
NTOT,ef, (g of N) m-3 1.82 ICT 2 864 144.88
BODef, (g of O2) m-3 1.92 NPV 9 117 546.13
CODef, (g of O2) m-3 47.77
XSS,ef, (g of SS) m-3 14.31
As shown in Figure 4a, the optimization of case I.A resulted decreases 26 and 33%, respectively, while fines paid OCTEQ
in three reaction compartments with volumes of 1083, 9096, increase 28%. The increment in fines, as a consequence of
and 6099 m3, respectively. That is, the optimal configuration effluent quality deterioration compared to case I.A (higher SNH,ef
includes three of the five available reaction compartments. The and NTOT,ef), is probably due to an unfavorable C/N ratio from
last two compartments resulted to have zero volume (in fact, a treatment efficiency point of view. The other cost variations
the lower bound 0.01 m3 was achieved) and are not represented are less relevant.
7506 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
Figure 6. Two alternative configurations for case I.A: solutions I.Aalt(1) and I.Aalt(2).
Table 11. Three Equivalent Solutions for the “Real or Physical Solution”, Solution I.A
Seq(1) Seq(2) Seq(3)
Ri Vi, m3 kLai, d-1 Qfresh,i, m3 d-1 Vi, m3 kLai, d-1 Qfresh,i, m3 d-1 Vi, m3 kLai, d-1 Qfresh,i, m3 d-1
1 1 083 218 10 360 0 0 5 690 0 0 10 360
2 9 096 36 8 086 0 0 4 009 1 083 218 0
3 6 099 27 0 1 083 218 661 9 096 36 8 086
4 0 0 0 9 096 36 8 086 0 0 0
5 0 0 0 6 099 27 0 6 099 27 0
Table 12. Main Variables Optimal Values for Solutions I.Aalt(1) and I.Aalt(2)
I.Aalt(1) I.Aalt(2)
contaminant effluent values costs contaminant effluent values costs
SNH,ef, (g of N) m-3 0.87 OCT,annual 394016.37 SNH,ef, (g of N) m-3 1.11 OCT,annual 407977.51
NTOT,ef, (g of N) m-3 2.53 ICT 2570061.6 NTOT,ef, (g of N) m-3 2.69 ICT 2455421.69
BODef, (g of O2) m-3 1.45 NPV 7479505.62 BODef, (g of O2) m-3 1.33 NPV 7538821.52
CODef, (g of O2) m-3 48.44 CODef, (g of O2) m-3 48.11
XSS,ef, (g of SS) m-3 14.56 XSS,ef, (g of SS) m-3 14.24
compartment of 1 085 m3 followed by two less aerated ones volumes are fixed at 1 333 m3, and the influent wastewater
summing around 16 000 m3 were also found for this case. flowrate is kept at 18 446 m3/d as in case I. The secondary settler
Solution I.Aalt(2) is a plant with two aerated reactors and feed cross-area is also fixed at 1 500 m2. Thus, three cases are
distribution. The NPV increases only 0.86%, OCT increases 6%, optimized, named cases II.A, II.B, and II.C. Despite the fact
and ICT decreases 8%, with respect to solution I.A. The OCT that the reaction compartment volumes and the secondary settler
increases mainly due to an increment on the aeration energy cross-area are fixed, the rest of the design and operation
demand cost (OCTAE increases 11%), and the decrease of the variables result from optimization.
ICT is mainly attributed to the reaction tanks (ICt decreases As in previous cases, a multiple starting point strategy for
10%). optimizing each case was used, leading to different solutions.
It is interesting to note that, from the 41 initial points The WWTP configurations that result from the proposed model
considered in case study I.A, 34 trials rendered feasible (7 initial showing the minimal NPV values for cases II.A, II.B, and II.C
points that were tried resulted infeasible). From the 34 successful are represented in parts a, b, and c of Figure 7, respectively,
trials, 29 different “mathematical solutions” were computed. named for simplicity solutions II.A, II.B, and II.C. Parts a, b,
After solution analysis, 6 different real or physical solutions and c of Table 13 show the contaminant effluent values, main
(WWTP configurations) were found. Twelve from the 29 variables optimal values, and costs. A detailed list showing costs
different mathematical solutions represent the real or physical for the optimal solution for each case can be found in Table
solution I.A, which shows the lowest NPV. 14. As expected, as the reaction compartment volumes and the
The solution of the problem using different initial points secondary settler cross-area are fixed, the feasible solution
allows one to ensure a good locally optimal solution and to regions are reduced with respect to cases I.A, I.B, and I.C, and
find alternative configurations and their characteristics. It is consequently, the NPV values are comparatively higher than
convenient for the designer to gain insight on locally optimal in previous cases.
properties and to know when different structures or operating Figure 7a shows that, for case II.A, a feed distribution with
conditions have similar costs. It also allows the further selection decreasing flowrate along the first three reactors (around 44%
among these alternatives according to several aspects that are of the flowrate is fed to the first compartment, 36% is fed to
difficult to introduce simultaneously into this DNLP model, such the second, and 20% is fed to the third one) is optimal. The
as flexibility (considering trends and predictions of future first, second, and fourth compartments resulted aerobic, while
requirements), reliability (minimum risk of failure due to a the third and fifth ones are anoxic. That is, this solution presents
complex treatment systems), and controllability, among others. aerated zones followed by anoxic zones. No external recycle
6.2. Case Study II. Here, the DNLP model is solved for distribution is present, and no external carbon source is dosed
operation variables optimization assuming given reaction com- to the process. As can be appreciated in Table 13a, none of the
partment volumes and secondary settler cross-area for the three effluent component concentrations achieve the effluent threshold
wastewater specifications listed in Table 9. Compartment value. As mentioned, the NPV increases 15% compared to that
7508 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
6.2.1. Costs Distribution for Solutions II. The average costs solution II are the sludge treatment cost, which increases 23 (
distributions for solutions II.A, II.B, and II.C are illustrated in 4%, and the aeration energy demand cost, which increases or
Figure 8. Around 72 ( 2% of NPV corresponds to OCT and 28 decreases depending on the wastewater characteristics. The two
( 2% corresponds to ICT. In order of relevance, each cost last cost variations are particularly significant in the case of
contributes to NPV as follows (average from solutions II.A, wastewater C. Finally, the sludge recirculation pump investment
II.B, and II.C): fines 30 ( 9%, sludge treatment 20 ( 7%, cost increases 86 ( 9% but has a low impact on the NPV
aeration energy demand 19 ( 4%, tanks investment 18 ( 1%, variation.
settler investment 5 ( 0.5%, influent pumping station investment The model resulted in being flexible and robust and able to
3 ( 0.5%, pumping energy demand 3 ( 0.5%, aeration system be used for process synthesis, for process optimization (fixing
investment 2 ( 0.5%, and, finally, sludge recirculation pump the structural variables), and also for simulation of a given plant
investment 0.5 ( 0.1%. and operating conditions. For example, the optimal configura-
Summarizing, for the assumptions made and the parameter tion, design, and operating conditions obtained for influent A
values used in the model, it was shown that: in case I [see Figure 4a, main optimal values: Vi ) (1083, 9096,
• For case I, the lowest NPV value achieved corresponds to 6099, 0, 0); Asett ) 1500; KLai ) (218, 36, 27, 0, 0); Qfresh,i )
solution I.B, followed by solutions I.A and I.C. (10360, 8086, 0, 0, 0); QTr,int,1 ) 0; QTr,int,2 ) 0; Qr,ext,i ) (11841,
• For case II, the lowest NPV value corresponds to solution
T
0, 0, 0, 0); uECSD ) 0] were simulated by feeding influents B
II.A (despite wastewater A not being the less contaminated one), and C, without imposing effluent threshold values, since
followed by solutions II.B and II.C. simulation runs were performed instead of optimization. Effluent
• It is convenient to choose a large plant with lower fines quality conditions were not met (effluent contaminant concen-
and, consequently, lower OCT than a small plant with SNH,ef at trations are higher than the effluent permitted limits), and the
the effluent threshold value (increasing OCT through fines). NPV increased 71 and 83%, with respect to the optimal
• As in previous works,5,6 it was observed that treatment configurations for those case studies (solutions I.B and I.C,
performance benefits from operational features such as the use respectively).
of different aeration and stream distributions patterns, mainly On the other hand, if reaction compartments and decanter
in cases where the reaction volumes are constrained. are given, the model can be used for optimization. For example,
• The average costs distributions for solutions I.A, I.B, and by setting volumes and decanter area to the optimal values
I.C show that around 65 ( 4% of NPV corresponds to OCT achieved for case I.A [Vi ) (1083, 9096, 6099, 0, 0); Asett )
and 35 ( 4% corresponds to ICT. On the basis of average values 1500] and optimizing the other variables, the following optimal
from solutions I.A, I.B, and I.C, the main contributors to NPV values feeding influent B were obtained: [KLai ) (238, 19, 20,
in order of relevance are as follows: fines (20 ( 6%), aeration 0, 0); Qfresh,i ) (12565, 5173, 708, 0, 0); QTr,int,1 ) 0; QTr,int,2 )
energy demand (24 ( 4%), tanks investment (23 ( 3%), sludge 0; Qr,ext,i ) (18042, 0, 0, 0, 0); uECSD
T
) 0]. Here, the effluent
treatment (19 ( 5%), settler investment (6 ( 1%), influent quality conditions were met and the NPV increased only 5%
pumping station investment (3 ( 0.5%), aeration systems with respect to the optimal configuration for this case study
investment (2 ( 0.2%), pumping energy demand (2 ( 0.5%), (solution I.B). Finally, feeding influent C, the main optimal
and, finally, sludge recirculation pump (0.5 ( 0.1%). values were as follows: [KLai ) (256, 53, 45, 0, 0); Qfresh,i )
• The average costs distributions for solutions II.A, II.B, and (9228, 9218, 0, 0); QTr,int,1 ) 0; QTr,int,2 ) 0; Qr,ext,i ) (10426, 0,
II.C show that around 72 ( 2% of NPV corresponds to OCT
T
0, 0, 0); uECSD ) 0]. Again, the effluent quality conditions
and 28 ( 2% corresponds to ICT. Each cost item contributes to were met and the NPV increased only 1% with respect to the
NPV as follows: fines 30 ( 9%, sludge treatment 20 ( 7%, optimal configuration for this case study (solution I.C).
aeration energy demand 19 ( 4%, tanks investment 18 ( 1%, It is clear that convenient aeration and recycle stream patterns
settler investment 5 ( 0.5%, influent pumping station investment and a suitable influent feed distribution can significantly improve
3 ( 0.5%, pumping energy demand 3 ( 0.5%, aeration system the system performance in the face of an influent composition
investment 2 ( 0.5%, and sludge recirculation pump investment variation. This also shows that it is important to optimize
0.5 ( 0.1%. operation modes of the process under variable influent condi-
• The NPV for solutions II with respect to solutions I increases tions. Design under uncertainty or flexible design will be
17 ( 9%. This fact is a consequence of a reduction of the addressed in a future work.
feasible solution regions, because the reaction compartment The models were implemented and solved using general
volumes and the secondary settler cross-area are fixed. Invest- algebraic modeling systems GAMS.20 The model results in 536
ment cost decreases 8 ( 2% while operation cost increases 31 single variables and 509 single equations. The code CONOPT21
( 16%. The main increment in OCT is in fines to be paid (74 was employed for solving the DNLP problem. The total CPU
( 17%), and the main decrease in ICT is on the reaction time needed to solve the models was among 8.7 and 16.3 s.
compartments investment cost (12 ( 3%). The remaining The average total CPU time needed to solve the models was
investment costs have low impact on NPV variation. The other 12.9 s. An Intel Pentium IV 2.40GHz CPU with 248 MB of
two operation costs that vary significantly from solution I to RAM was used.
7510 Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007
Acknowledgment
The financial support from Consejo Nacional de Investiga-
ciones Cientı́ficas y Técnicas (CONICET) and Agencia Nacional
de Promoción Cientı́fica y Tecnológica (ANPCyT) of Argentina
are gratefully acknowledged.
Notation
A ) area (m2)
Figure 8. Costs distribution for solutions II. b ) cost parameter
B ) constant (3000-1 (kg of O2) d (h m3)-1)
BOD ) biochemical oxygen demand ((g of O2) m-3)
7. Conclusions and Future Works C ) component concentration (C ) S or X)
COD ) chemical oxygen demand ((g of O2) m-3)
In the present work, a DNLP model was developed
E ) energy (kWh d-1)
for optimal synthesis and design as well as optimization
EQ ) effluent quality index ((kg of contaminating unit) d-1)
of the operation variables of wastewater treatment plants for
h ) height (m)
given influent wastewater specifications (composition and
I ) set of reactors/set of mixers
flowrate). Two different scenarios are selected as case studies.
IC ) investment cost (euro)
In case study I, the reaction compartment volumes and settler id ) interest rate (discount rate)
cross-area are optimization variables. In case study II, J ) solids flux due to gravity settling
the reaction compartment volumes are fixed at 1 333 m3, and K ) set of process rates
the secondary settler cross-areas are fixed at 1 500 m2. As kLa ) oxygen transfer coefficient (d-1)
mentioned, the settler depth is fixed at 4 m in both cases. Three n ) life span of the WWTP (year)
different influent wastewater compositions characterized N ) nitrogen
by different carbon/nitrogen ratios were used to show the model NPV ) net present value (euro)
capabilities, mainly its robustness and flexibility. The dif- OC ) operation cost (euro)
ferent C/N ratios were obtained by varying the influent XS and OV ) operation variable
SS content. Optimal solutions resulting from increasing C/N OxCa ) oxygen capacity ((kg of O2)/h)
ratios (B < A < C) were compared and analyzed for both P ) set of equipment units
cases. Q ) volumetric flowrate (m3 d-1)
The models were implemented in GAMS. The code CONOPT r ) reaction rate (g m-3)
was employed for solving the DNLP problems. As a S ) soluble component concentration (g m-3)
consequence of the problem nature and the solver characteristics, TKN ) total Kjendal nitrogen
in order to verify the expected presence of local optimal u ) mass flowrate (g d-1)
solutions, a multiple starting point strategy was adopted. Indeed, V ) volume (m3)
different locally optimal solutions to the problem were found X ) particulate component concentration (g m-3)
for each case study. Because of the model flexibility and Z ) characteristic dimension
robustness, it is a useful tool to help the designer to make a
Subscripts
further selection among the solutions found considering other
design aspects such as flexibility, reliability, and controllability, a ) aeration system
among others. BOD ) biochemical oxygen demand
bottom ) settler bottom effluent
As mentioned, a DNLP model was developed as a first step
c ) component
of the modeling task. Although the existence of nonsmooth
clar ) clarification zone
functions, CONOPT solved the model quite well, providing
COD ) chemical oxygen demand
solutions for >80% of the different initial points tried. Also,
ECSD ) external carbon source dosage
the existence and type of reaction compartments and process
ef ) clarified effluent
streams were satisfactorily handled by the use of very small
EQ ) effluent quality index
lower bounds (practically zero), which introduce only a small
ext ) external
error in the objective function calculation if the associated fresh ) influent wastewater
equipment is eliminated to achieve a practical flowsheet. g ) generic flowrate
However, it is interesting to note that, in some cases, solutions i ) reactor/mixer prior to reactor
with small kLa values (but not the lower bound value) were in ) inlet
computed, which may be nonsensical from an engineering point int ) internal
of view. This limitation will be overcome with the implementa- ips ) influent pumping station
tion of MINLP models. k ) process
In future works, new mathematical models based on MINLP l ) operation variable
or GDP programming will be presented. This will also allow lim ) limit
the incorporation of the settler depth and the feed allocation m ) settler layer
point as optimization variables, as well as the use of a more max ) maximum
rigorous economic objective function. All the expertise gained min ) minimum concentration
and results obtained from the proposed model will be used as p ) unit
a starting point. Finally, design under uncertainty or flexible pump ) pump
design will also be addressed. r ) recycle
Ind. Eng. Chem. Res., Vol. 46, No. 23, 2007 7511
Treatment Plant. Proceedings of the 6th World Congress of Structural and for evaluating control strategies in wastewater treatment plants. Presented
Multidisciplinary Optimization, Rio de Janeiro, Brazil, May 30-June 3, at European Conference Control (ECC’99), Karlsruhe, Germany, Aug 31-
2005. Sept 3, 1999.
(11) Copp, J. The COST Simulation Benchmark: Description and (18) Vanrolleghem, P. A.; Jeppsson, U.; Carstensen, J.; Carlsson, B.;
Simulator Manual; Office for Official Publications of the European Olsson G. Integration of WWT plant design and operationsA systematic
Community: Luxembourg, 2002. approach using cost functions. Water Sci. Technol. 1996, 34 (3-4), 159.
(12) Van Haandel, A. C.; Ekama, G. A.; Marais, G. V. R. The Activated (19) Mussati, M; Gernaey, K; Gani, R.; Bay Jørgensen, S. Performance
Sludge Processs3. Single Sludge Denitrification. Water Res. 1981, 15 (10), analysis of a denitrifying wastewater treatment plant. Clean Technol.
1135. EnViron. Policies 2002, 4, 171.
(13) Gujer, W.; Henze, M.; Mino, T.; van Loosdrecht, M. Activated
Sludge Model No. 3. Water Sci. Technol. 1999, 39, 183. (20) Brooke, A.; Kendrick, D.; Meeraus, A. GAMSsA User’s Guide
(14) Takács, I.; Patry, G.; Nolasco, D. A Dynamic Model of the (Release 2.25); The Scientific Press: San Francisco, CA, 1992.
Clarification-Thickening Process. Water Res. 1991, 25, 1263. (21) Drud, A. S. CONOPT: A system for large scale non-linear
(15) Grijspeerdt, K.; Vanrollegham, P.; Verstraete, W. Selection of one- optimization, reference manual for CONOPT subroutine library; ARKI
dimentional sedimentation: Models for on-line use. Water Sci. Technol. Consulting and Development A/S: Bagsvaerd, Denmark, 1996.
1995, 31 (2), 193.
(16) Vanrolleghem, P. A.; Gillot, S. Robustness and economic measures ReceiVed for reView April 4, 2007
as control benchmark performance criteria. Water Sci. Technol. 2002, 45 ReVised manuscript receiVed August 7, 2007
(4-5), 117. Accepted August 8, 2007
(17) Alex, J.; Beteau, J. F.; Copp, J. B.; Hellinga, C.; Jeppsson, U.;
Marsili-Libelli, S., Pons, M. N.; Spanjers, H.; Vanhooren, H. Benchmark IE0704905