Computers and Chemical Engineering 34 (2010) 825–845

Contents lists available at ScienceDirect

Computers and Chemical Engineering

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

Automated targeting technique for concentration- and property-based total

resource conservation network
Denny Kok Sum NG a , Dominic Chwan Yee Foo a,∗ , Raymond R. Tan b , Mahmoud El-Halwagi c
a
Department of Chemical and Environmental Engineering, University of Nottingham Malaysia, Broga Road, 43500 Semenyih, Selangor, Malaysia
b
Chemical Engineering Department, De La Salle University, 2401 Taft Avenue, 1004 Manila, Philippines
c
Chemical Engineering Department, Texas A&M University, College Station, TX 77843, USA

a r t i c l e i n f o a b s t r a c t

Article history: Resource conservation networks (RCNs) are among the most effective systems for reducing the consump-
Received 29 July 2009 tion of fresh materials and the discharge of waste streams. A typical RCN involves multiple elements of
Received in revised form resource pre-treatment, material reuse/recycle, regeneration/interception, and waste treatment for final
29 December 2009
discharge. Due to the close interactions among these individual elements, simultaneous synthesis of a
Accepted 21 January 2010
total RCN is necessary. This paper presents an optimisation-based procedure known as automated target-
Available online 4 February 2010
ing technique to locate the minimum resource usage or total cost of a concentration- or property-based
total RCNs. This optimisation-based approach provides the same benefits as conventional pinch analy-
Keywords:
Process integration
sis techniques in yielding various network targets prior to detailed design. Additionally, this approach
Resource conservation offers more advantages than the conventional pinch-based techniques through its flexibility in setting an
Automated targeting objective function and the ability to handle different impurities/properties for reuse/recycle and waste
Waste minimisation treatment networks. Furthermore, the concentration-based RCN is treated as the special case of property
Optimisation integration, and solved by the same model. Literature examples are solved to illustrate the proposed
Property integration approach.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction
The process industries have traditionally focused on conven-
tional end-of-pipe waste treatment in order to comply with
environmental legislation. This approach has been gradually
replaced by the use of pollution prevention strategies. This is mainly
due to the increase of public awareness of environmental sustain-
ability, the rising cost of raw material, and the increasingly stringent
environmental legislation. One of the cost effective solutions
is resource conservation, wherein materials are reused/recycled
within processes without adversely affecting the process perfor-
mance. Process integration has been commonly accepted as an
effective tool in developing and evaluating various resource con-
servation alternatives.
El-Halwagi (1997, 2006) defined process integration as a holistic
approach to process design, retrofitting and operation which empha-
sises the unity of the process. Within the framework of process
integration, pinch analysis technique has emerged as a promis-
ing tool in identifying various network targets prior to detailed
∗ Corresponding author. Tel.: +60 3 8924 8130; fax: +60 3 8924 8017.
E-mail addresses: Denny.Ng@nottingham.edu.my (D.K.S. NG),
Dominic.Foo@nottingham.edu.my (D.C.Y. Foo), raymond.tan@dlsu.edu.ph
(R.R. Tan), el-halwagi@tamu.edu (M. El-Halwagi).
design. Over the past decades, extensive pinch analysis works have
been reported for the synthesis of water (e.g., Almutlaq, Kazantzi,
& El-Halwagi, 2005; Bandyopadhyay, Ghanekar, & Pillai, 2006; El-
Halwagi, Gabriel, & Harell, 2003; Foo, Manan, & Tan, 2006; Manan,
Tan, & Foo, 2004; Ng, Foo, & Tan, 2009a; Wang & Smith, 1994a)
and utility gas networks (e.g., Agrawal & Shenoy, 2006; Alves &
Towler, 2002; Bandyopadhyay, 2006; El-Halwagi et al., 2003; Foo
& Manan, 2006). The net result of resource conservation activities
is the simultaneous reduction of both fresh resource consumption
and waste discharge.
After the opportunities for maximum material recovery are
exhausted through direct reuse/recycle, fresh resource flowrate
may be further reduced with interception or regeneration pro-
cesses (e.g., Agrawal & Shenoy, 2006; Bai, Feng, & Deng, 2007;
Bandyopadhyay & Cormos, 2008; Feng, Bai, & Zheng, 2007; Foo,
Manan, & Tan, 2006; Kuo & Smith, 1998a; Ng, Foo, & Tan, 2009b;
Ng, Foo, Tan, & Tan, 2007, 2008; Wang & Smith, 1994a). In addi-
tion, before waste is discharged to the environment, it needs to
be treated to meet the requirements given in the emission legis-
lation. Various works have also been dedicated to this area (e.g.,
Bandyopadhyay et al., 2006; Bandyopadhyay & Cormos, 2008; Kuo
& Smith, 1997; Ng, Foo, & Tan, 2007a, 2007b; Wang & Smith, 1994b).
As pointed out by Kuo and Smith (1998b), there are close inter-
actions between water reuse/recycle, regeneration system, and
effluent treatment system. Hence, an overall framework that inte-

0098-1354/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2010.01.018
826 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Nomenclature
FRP flowrate of purified stream
Indices and sets FRR flowrate of reject stream
i index for source FSRi flowrate of SRi
I sets of source FSKj flowrate of SKj
j index for sink FTRDv flowrate of treated source v
J sets of sink FTRv flowrate of individual treatment source v
k index for property level FTRv,k flowrate of individual treatment source v at level k
K sets of property level FTR,RR treatment flowrate from reject stream
u index for regeneration FUPW flowrate of UPW
U sets of regeneration FWi flowrate of waste from SRi
v index for treatment system FWRP flowrate of wastewater generated from purified
V sets of treatment system stream
FWRR flowrate of wastewater generated from reject
Parameter/acronyms stream
ACAS annualised cost of air stripping UF
FIn inlet flowrate to UF system
ACBio annualised cost of biotreatment FPUF permeate flowrate for UF system
AOT annual operating time FRUF reject flowrate for UF system
CD maximum allowable discharge concentration FPRO permeate flowrate for RO system
COSTClay unit cost of clay
FRRO reject flowrate for RO system
COSTFW unit cost of fresh water
L mass flowrate of air
COSTR unit cost of regeneration
¯ mean property
COSTUPW unit cost of ultra pure water
pdischarge property discharge limit
COSTTR unit cost of wastewater treatment
pRu property of regenerated source u
DI deionising unit
pRP property of purified stream
m fractional interest rate per year
pRR property of reject stream
n number of property levels
pSRi property of source i
OCAS operating cost of air stripping
pSKj admissible property of sink j
OCBio operating cost of biotreatment
pmin lower bound of admissible property of sink j
pLBE lower bound of property discharge limit SKj
pUBE upper bound of property discharge limit pmax
SKj
upper bound of admissible property of sink j
RCCD resource conservation cascade diagram RSRi resistivity of source SRi
RCN resource conservation network R̄ mean resistivity
RO reverse osmosis x mass concentration of methanol in air
RP purified stream of filtration unit zSRi fractional contribution of SRi in the total mixture
RR reject stream of filtration unit flowrate
RRN reuse/recycle network ık net material flowrate from level k
RRT fixed removal ratio ın net material flowrate at final level n
SRi source i εk residue of the property load from property operator
SKj sink j level k
TAC total annualised cost (pD ) linearly-additive operator on the final discharge
TOC total operating cost stream
UF ultrafiltration (pRu ) linearly-additive operator on the property of regen-
UPW ultra pure water erated source u
WTN waste treatment network (pRP ) linearly-additive operator on the property of puri-
y number of years fied stream
property operator (pRR ) linearly-additive operator on the property of reject
k property operator level k stream
D outlet property operator for final discharge (pSRi ) linearly-additive operator on the property of SRi
(p̄) linearly-additive operator on the mean property
Variables (pTR ) linearly-additive operator on the waste treatment
C concentration effluent
CR outlet concentration of regeneration/interception R outlet property operator for regeneration
unit TR outlet property operator for waste treatment
CRu concentration of the individual regeneration SRi density of SRi
streams feeding to regeneration unit u ¯ mean density
CTR outlet concentration of treatment system k net waste flowrate at level k
CTv concentration of the individual waste streams feed- n net waste flowrate at final level n
ing to treatment unit v  minimum allowable concentration difference of
FClay flowrate of clay mass exchanger
FD waste discharge flowrate k residual waste load at level k
FFW flowrate of the fresh water
FRu flowrate from regeneration u
FRu,k flowrate from regeneration u at level k
FREGu total flowrate of regeneration u
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
grates all three elements within an overall framework known as
the total water network was proposed (Kuo & Smith, 1998b). How-
ever, this seminal contribution to the total network concept was
limited to the fixed mass load problem. More recently, Ng et al.
(2007a, 2007b) presented a sequential approach to locate the differ-
ent network targets for a total water network in the fixed flowrate
problem. However, both methods are essentially developed for
water networks. Hence, there is a clear need to develop a more
generic tool for all types of RCNs other than water, which is the sub-
ject of this paper. Further recent developments in the fixed load and
fixed flowrate problems are described in detail in a recent review
by Foo (2009).
Note that however, most previous works have been restricted
to “chemo-centric” or concentration-based systems where the
quality of streams are described in terms of the concentration
of one or more contaminants. However, there are many appli-
cations in which stream quality is characterised by physical or
chemical properties, such as density, vapor pressure or electri-
cal resistivity, rather than contaminant concentration. This led to
the concept of componentless design (Shelley & El-Halwagi, 2000)
and property integration. The latter is defined as “a functionality-
based, holistic approach to the allocation and manipulation of streams
and processing units, which is based on the tracking, adjustment,
assignment, and matching of functionalities throughout the process”
(El-Halwagi, Glasgow, Eden, & Qin, 2004). Several graphical, alge-
braic, and optimisation-based techniques have also been developed
for reuse/recycle (El-Halwagi et al., 2004; Foo, Kazantzi, El-Halwagi,
& Manan, 2006; Kazantzi & El-Halwagi, 2005; Ponce-Ortega,
Hortua, El-Halwagi, & Jiménez-Gutiérrez, 2009; Qin, Gabriel, Harell,
& El-Halwagi, 2004) and interception problems (El-Halwagi, 2006;
Grooms, Kazantzi, & El-Halwagi, 2005; Ng, Foo, Rabie, & El-Halwagi,
2008; Ng, Foo, Tan, Pau, & Tan, 2009).
Most recently, Ng et al. (2009a, 2009b) adopted the concept
of automated targeting technique (El-Halwagi & Manousiothakis,
1990) to locate the minimum flowrate/cost targets for a RCN with
material reuse/recycle and interception placement. The technique
was then extended to property-based RCN (Ng, Foo, Tan, Pau, et
al., 2009). Besides, Ng, Foo, Tan, Pau, et al. (2009) introduced a
new variant known as the bilateral property integration prob-
lem in which the property operator values of the process sinks
and sources exist in between that of the external fresh resources.
This concept is important whenever a nominal desired or tar-
geted sink property operator value exists that is neither the highest
(superior) nor the lowest (inferior) among all operator values of
the process sources and sinks, as in the conventional cases (Foo,
Kazantzi, et al., 2006; Kazantzi & El-Halwagi, 2005). An exam-
ple of such a property is pH, wherein values near neutral (pH 7)
correspond to high stream quality, while values deviating either
above or below the central value are of poor quality. Besides, the
work of Ng, Foo, Tan, Pau, et al. (2009) also solved the case where
interception for fresh resource (e.g., water pre-treatment unit) is
incorporated into the material reuse/recycle network, which has
been neglected by many researchers (Bagajewicz & Faria, 2009).
However, note that these early works on automated targeting tech-
niques are limited to RCN with reuse/recycle and interception
placement alone, without the simultaneous consideration of waste
treatment.
In this work, the above-mentioned automated targeting tech-
nique (Ng et al., 2009a, 2009b; Ng, Foo, Tan, Pau, et al., 2009)
is extended to the concentration- and property-based total RCN
wherein material reuse/recycle and waste interception (for further
recovery and/or for final discharge within environmental limits) are
considered simultaneously. The targeting technique also addresses
the bilateral problem where the property operator values of the
process sinks and sources exist in between the property operator
values of external fresh resources.
827
Based on the concept of insight-based targeting, the automated
targeting technique is formulated as a linear programming (LP)
model, for which any solution found is globally optimal. The main
limitation of this work, as with majority of pinch-based approaches,
is that it is restricted to single impurity or property problems.
However, as will be shown in the following sections, different
impurities or properties can be handled in the reuse/recycle and
interception/waste treatment sections. This is an added advantage
of this newly developed technique over the conventional pinch-
based approaches (e.g., see the work of Bandyopadhyay & Cormos,
2008; Ng et al., 2007a, 2007b).
2. Problem statement
The problem of a total RCN is stated as follows. Given a set of
process sources, SRi (i ∈ I) that may be reused/recycled to a set of
process sinks, SKj (j ∈ J). Each source has a flowrate, FSRi and is char-
acterised by a single constant property, pSRi . Each sink requires a
flowrate of FSKj and is restricted to comply with the predetermined
allowable property constraints as follows:
pmin
SKj
≤ pSKj ≤ pmax
SKj
(1)
where pmin
SKj
and pmax
SKj
are the specified lower and upper bounds of
the admissible properties for sink j. External fresh resource(s) are
readily available to supplement the flowrate required by the sinks,
and its flowrate (FFW ) is to be determined as part of the optimisa-
tion result. Interception unit(s) may be used to purify external fresh
resource and process sources. The former is commonly termed as
pre-treatment, which is used for external fresh resource that does
not fulfil the quality requirement of the process sinks. Besides, pro-
cess sources may also be regenerated/intercepted in unit u (u ∈ U)
for further reuse/recycle to the process sink, or treated in treatment
system v (v ∈ V ) to comply with emission limit prior to environ-
mental discharge. In addition, environment legislation restricts the
quality of waste to be discharged as follows:
pLBE ≤ pdischarge ≤ pUBE (2)
where pLBE
and pUBE
are the lower and upper bounds of property
discharge limit.
Based on the previous works of property integration (Foo,
Kazantzi, et al., 2006; Kazantzi & El-Halwagi, 2005; Ng, Foo, Tan,
Pau, et al., 2009; Shelley & El-Halwagi, 2000), a linearised property
mixing rule is needed to define all possible mixing patterns among
the individual properties in synthesising the RCN. The mixing rule
takes the form of (Shelley & El-Halwagi, 2000):

(p̄) = zSRi (pSRi ) (3)
SRi
where (pSRi ) and (p̄) are linearly-additive operators on the
source property pSRi and the mixture property p̄, respectively; while
zSRi is the fractional contribution of SRi in the total mixture flowrate.
Note that the mixing rule in Eq. (3) is generic in nature, and hence
covers for cases where integration is based on impurity concen-
tration. In other words, various mass integration problems such as
water and hydrogen network syntheses may be treated as special
cases of property integration.
3. Automated targeting technique
The automated targeting technique was originally developed for
mass exchange network synthesis (El-Halwagi & Manousiothakis,
1990). It was then extended for concentration- (Ng et al., 2009a,
2009b) and property-based RCNs (Ng, Foo, Tan, Pau, et al., 2009)
based on the concept of cascade analysis technique (Foo, Kazantzi,
828 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
Fig. 1. Resource conservation cascade diagram, RCCD (Ng et al., 2009a; Ng, Foo, Tan, Pau, et al., 2009).
et al., 2006; Foo, Manan, & Tan, 2006; Manan et al., 2004). How-
ever, these previous works did not consider waste treatment for
discharge. This additional aspect is now included in this work. In the
following sub-sections, the automated targeting technique for each
category of RCN problems is illustrated. For the first two categories,
automated targeting model is illustrated for the reuse/recycle net-
work (RRN). It was then be extended to cater for waste treatment
for a total RCN.
3.1. Model for conventional RCN problem
Following the previous approach (Ng et al., 2009a, 2009b; Ng,
Foo, Tan, Pau, et al., 2009), the first step of the automated targeting
technique is to construct a resource conservation cascade diagram
(RCCD), as shown in Fig. 1. As shown, the sinks and sources are first
arranged in an ascending order based on their property operator
values ( k ), from the lowest (k = 1) to the highest level (k = n). In
cases where the operator levels for fresh resource(s) and zero level
do not coincide with any of the process sinks and sources present,
additional levels are added. An arbitrary value is also added at the
final level (highest among all operators) of the RCCD to allow the
calculation of residual property load. Next, the flowrate and prop-
erty load cascades are performed across all operator levels, given
as the constraints in Eqs. (4) and (5), respectively.
ık = ık−1 + (˙i FSRi − ˙j FSKj )k k∀K
εk = εk−1 + ık ( k+1 − k) k ∀K
As shown in Eq. (4), the sum of the net material flowrate cascaded
from the earlier operator level k − 1 (ık−1 ) with the flowrate balance
(˙ i FSRi − ˙ j FSKj ) at operator level k form the net material flowrate
of each k-th level (ık ). Meanwhile, the property load at each oper-
ator interval is given by the product of the net material flowrate
from level k (ık ) and the difference between two adjacent operator
levels ( k+1 − k ). The residual property load of each operator level
k (εk ) is to be cascaded down to the next operator level, as dictated
in Eq. (5). The residual property load, ε, must always take a positive
value, in order for a feasible RCN to take place (Ng et al., 2009a,
2009b; Ng, Foo, Tan, Pau, et al., 2009), given as in Eq. (6):
εk ≥ 0 k∀K
4. Model for the bilateral problems
As mentioned earlier, a bilateral problem occurs when the prop-
erty operator values of the process sinks and sources exist in
between of two external fresh resources. For this case, two reverse
sets of cascades are carried out simultaneously for the RRN (Ng,
Foo, Tan, Pau, et al., 2009). Note that each set of cascades consists
of both material flowrate and property load cascades. In the first set
of cascades, the operators are arranged in ascending order where
fresh resource is located at the lowest operator level. This set of
cascades is termed as the superior cascade (referring to the fresh
resource that is found at the superior operator level). In contrast,
operators are arranged in descending order in the second set of cas-
cades (inferior cascade), where a second fresh resource is located at
the highest operator level (Ng, Foo, Tan, Pau, et al., 2009). Thus, Eq.
(5) is modified as Eq. (7), so that the difference between the oper-
ator levels k and k + 1 is set as reverse as compared to Eq. (5). This
ensures a feasible property load cascade to be generated.
εk = εk−1 + ık ( k − k+1 ) k ∀K (7)
Since the superior and inferior cascades are optimised simul-
taneously, and the property load is cascaded in two directions, all
process sinks receive property load corresponding to the poorest
quality level they are able to tolerate. The accumulation of flowrate
for waste discharge at the final operator level is disallowed. Thus,
(4) in each level where a source exists, a new sink is added to allow the
waste (FWi ) to be discharged from the same level to avoid its accu-
(5) mulation when the net material flowrate (ık ) is cascaded. Thus, Eq.
(4) is modified for both superior and inferior cascades of property
operator level as follows:
ık = ık−1 + (˙i FSRi − ˙j FSKj − ˙i FWi )k k ∀K (8)
5. Model for the total RCN
To incorporate waste treatment into automated targeting tech-
nique, modification is required on the original automated targeting
model (Ng et al., 2009a, 2009b; Ng, Foo, Tan, Pau, et al., 2009). In
principle, two sub-networks will have to be analysed. This includes
the RRN as well as the waste treatment network (WTN). The
RRN model for the conventional and bilateral problems has been
(6) described earlier, while model for the WTN is described as follows.
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
Fig. 2. Generic cascade diagram for WTN.
As shown in Eq. (8), waste (˙ i FWi ) is discharged from a process
source of RRN (when it is not being recovered). These waste sources
are then sent to the WTN to improve its property before discharge.
In order to determine the minimum treatment flowrate, flowrate
and property load cascades for WTN are needed.
Fig. 2 shows a generic cascade diagram for the WTN. The identi-
fied waste flowrates from the RRN hence become sources in the
WTN cascades. Besides, the individual treatment flowrate (FTRv )
exists as a sink at each operator level where a source exists, while
their outlet flowrate becomes a treated source (FTRDv ) at the outlet
operator level (or calculated based on the removal ratio of the treat-
ment units). For a given waste treatment system v without flowrate
loss/gain,
summation
 of its individual treatment flowrate across all
level k ( k v FTRv,k ) should equal to its treated source (FTRDv ), as
given in Eq. (9).
FTRDv = ˙k ˙v FTRv,k v∀V (9)
Finally, the discharge flowrate (FD ), given as the sum of the indi-
vidual waste flowrates, i.e. ˙ i FWi , is added as a new sink at the
operator level of the final discharge limit. The flowrate balance at
each level takes the revised form of Eq. (8) as follows:
k = k−1 + (˙i FWi + ˙v FTRDv − ˙v FTRv − FD )k k∀K (10)
where ␨k and ␨k−1 are net waste flowrate at level k and k − 1.
Besides, the load cascades and their non-negativity of the resid-
ual waste load (k ) are given as in Eqs. (11) and (12), respectively,
similar to that in Eqs. (5) and (6) for the RRN.
k = k−1 + k ( k+1 − k) k∀K (11)
k ≥ 0 k∀K (12)
To ensure that no waste flowrate is being discharged from the
final operator level n, as that level is only used for the calculation
of residual waste load, a new constraint (as shown in Eq. (13)) is
needed.
n = 0 (13)
Since different sets of cascades are formulated for RRN and WTN,
different properties can be used for each of these sub-networks. In
other words, it is possible that the property used as a measure of
829
stream quality for purposes of resource recovery (e.g., total dis-
solved solids—TDS) does not have to be the same as the property
needed to meet legal effluent standards (e.g., turbidity). Thus, in
such a scenario, processes within the RRN will have specifications
for TDS, but no restrictions for turbidity, while the reverse will true
for the WTN. This feature is an advantage as compared to the con-
ventional pinch analysis techniques where only single property is
allowed (e.g., see the work of Bandyopadhyay & Cormos, 2008; Ng
et al., 2007a, 2007b).
It is worth noting that in the special case where property con-
straint of RRN and WTN being the same, a single set of flowrate and
property load cascades is sufficient to determine the flowrate/cost
targets. Fig. 3 shows a generic RCCD for such a case. As shown,
three property operator levels are added in the RCCD, i.e. outlet
operators for regeneration ( R ), waste treatment ( TR ) and final
discharge ( D ). At each property operator level where a source
exists, flowrate terms for regeneration (˙ u FRu ) and treatment
(˙ v FTRv ) are added as sinks, along with other sink flowrates (˙ j FSKj )
that exist. Besides, at the level where the regenerated (˙ u FREGu )
or treated (˙ v FTRDv ) flowrate exist, their flowrate terms are to be
added too. Thus, Eq. (8) is modified as follows:
ık = ık−1 + (˙i FSRi − ˙j FSKj − ˙u FRu − ˙v FTRv
+˙u FREGu + ˙v FTYRDv )k k ∀K (14)
Similar to the waste treatment system, for a regeneration u with-
out flowrate loss/gain, summation of its individual regeneration
flowrate across all level k (˙ k ˙ u FRu,k ) should equal to its regener-
ated source (FREGu ), as given by Eq. (15).
FREGu = ˙k ˙u FRu,k u∀U (15)
Note that the total flowrates of the individual treated (˙ u FRu )
and waste (˙ v FTRv ) streams at every level k should not be more
than what is available flowrate in SRi, hence Eq. (16) is needed:
(˙u FRu + ˙v FTRv )k ≤ FSRi,k k∀K (16)
In addition, the final discharge wastewater flowrate (FD ) is
placed as a sink at its operator level of D (Fig. 3). As mentioned
previously, the final operator level in RRN is only used for the calcu-
830 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
Fig. 3. Generic RCCD for total RCN with single property constraint.
lation of residual property load. Therefore, to ensure that no waste
flowrate is discharged from the final operator level from RRN, Eq.
(13) is revised as follows:
ın = 0
Meanwhile, other process constraints such as the property load
cascade (given as Eq. (5)) and non-negativity (Eq. (6)) shall remain
the same.
The objective of this work is to locate the minimum network
targets, prior to the detailed design of a RCN. Depending on the
problem, objective function of some cases may be set to determine
the minimum resource flowrate(s) of the network; while others are
set to minimise total costs. Five literature examples are adopted to
illustrate the proposed approach.
6. Example 1
Fig. 4 shows an industrial wafer fabrication process taken from
Ng, Foo, Tan, Pau, et al. (2009). This example is used to illustrate the
application of the automated targeting technique for a property-
based total RCN. In this example, the pre-treatment system is used
to generate ultra pure water (UPW) for the process requirements.
It is assumed that 70% of the inlet flowrate to ultrafiltration (UF) is
recovered as permeate, while the remainder (30%) is rejected. The
same assumption applies for the reverse osmosis (RO) membrane
unit. The effluent from RO unit is then sent to deionising (DI) unit
to produce the UPW. It is assumed that no water losses occur in the
DI. The following equations are hence included in the optimisation
model.
UF
Fln = FPUF + FRUF
FPUF = FPRO + FRRO
UF , F UF and F UF represent the inlet, permeate and reject
where FIn P R
flowrates for the UF system; while FPRO and FRRO are the permeate
and reject flowrates for the RO system. Since no water loss occurs
in the DI unit, the permeate from the RO system is then treated in
the DI unit to generate the UPW. In other words, the UPW flowrate
(FUPW ) is equivalent to FPRO . In order to reduce the externally pur-
chased municipal fresh water (FW) consumption, the recovery of
(17)
UF and RO reject streams from the pre-treatment system is also
considered, which means that their flowrate terms are also added
as sources in the model.
In addition, Eqs. (20) and (21) are also included to represent the
rejection in the UF and RO systems
FRUF = 0.3FIn
UF
(20)
FRRO = 0.3FPUF (21)
In this case study, the most significant water quality factor of this
example is reported as resistivity (R), which constitutes an index of
the total ionic content of aqueous streams. The general mixing rule
for resistivity is given as (El-Halwagi, 2006):
1  zSRi
= (22)
R̄ SRi RSRi
Note that the lowest resistivity also corresponds to the poor-
est quality level. Table 1 summarises the pertinent data for the
sinks and sources. On the other hand, heavy metal concentration
is chosen as the main constraint for final discharge, which is given
as 2 mg/L. A fixed outlet concentration of 0.5 mg/L is assumed for
the treatment unit. In order to synthesise a cost effective total
RCN, it is necessary to minimise the total operating costs (TOC) of
fresh water, UPW and waste treatment. In this work, the unit costs
for fresh water (COSTFW ), UPW (COSTUPW ) and waste treatment
(COSTTR ) are assumed as 1$/t, 2$/t and 0.5$/t, respectively.
(18)
TOC = (COSTFW FFW + COSTUPW FUPW + COSTTR ˙v FTRDv )AOT (23)
(19)
where FFW and FUPW are the flowrates of fresh water and ultra pure
water respectively; AOT is the annual operating time of the process.
The optimisation model is solved to minimise TOC (Eq. (23)),
subject to the constraints in Eqs. (5), (6) and (8) for RRN, Eqs.
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 831

Fig. 4. Schematic diagram for Example 1 (Ng, Foo, Tan, Pau, et al., 2009).

(9)–(13) and (16) for WTN as well as Eqs. (18)–(21) for the pre-
treatment model; solution of which results in the RRN and WTN
cascades in Fig. 5. The TOC is targeted as $27,223,512 (assuming an
AOT of 8000 h). The optimum flowrates for FW (FFW = FUPW + FRUF +
FRRO ), UPW (FUPW ) and wastewater (FD ) are determined as 3095,
1516.55 and 2205 t/h, respectively. The flowrate targets match
those found in Ng, Foo, Tan, Pau, et al. (2009), which was done
by minimising flowrates. However, waste treatment was not con-
sidered in this earlier work. As shown in Fig. 5b, the improved
automated targeting technique in this work determines that the
minimum waste treatment flowrate (˙ v FTRDv ) as 739.68 t/h, orig-
inating from waste W2 (FTR2 = 559.68 t/h) and W9 (FTR9 = 180 t/h).
One of the feasible RCN designs that achieves the targets is shown
in Fig. 6; note that this configuration is only one of various alter-
native designs. As shown in this example, different properties are
used in the synthesis of the RRN and WTN. Therefore, the pro-
posed approach is superior to previous works (e.g., see the work
of Bandyopadhyay & Cormos, 2008; Ng et al., 2007a; Ng, Foo, Tan,
Pau, et al., 2009) where only single property can be considered.
A second scenario is next considered. Apart from the waste treat-
ment facility, a membrane unit with recovery factor of 0.95 and
permeate quality of 15 M m is used to regenerate water sources
for further recovery. Besides, the membrane unit will also reduce
the heavy metal content of the purified stream to 0.1 ppm, along
with a reject at 6.15 ppm. It is further taken that the unit cost of
regeneration (COSTR ) for the membrane unit is taken as 1.5 $/t.
With the inclusion of the membrane unit, the optimisation objec-
tive in Eq. (23) is modified to include the regeneration cost, i.e.:
TOC = (COSTFW FFW + COSTUPW FUPW + COSTR ˙u FRu
+COSTTR ˙v FTRDv )AOT (24)
As shown in Ng et al. (2009b) and Tan, Ng, Foo, and Aviso (2009),
a regeneration process with two outlet streams is termed as parti-
tioning regeneration system, where a feed stream is separated into
a higher quality product and a lower quality reject streams. Hence,
the material balances for a partitioning regenerator are similar to

Table 1
Limiting data for Example 1 (Ng, Foo, Tan, Pau, et al., 2009).

Process Flowrate (t/h) Resistivity, R (M m) Operator, (R) (M m−1 ) Heavy metalcontent (mg/L)

Sink
Wet (SK1) 500 7.000 0.143 –
Litography (SK2) 450 8.000 0.125 –
CMP (SK3) 700 10.000 0.100 –
Etc (SK4) 350 5.000 0.200 –
Cleaning (SK5) 200 0.008 125.000 –
Cooling tower makeup (SK6) 450 0.02 50.000 –
Scrubber (SK7) 300 0.01 100.000 –

Source
UF reject (SR1/UFR) 30% of UF inlet 0.010 100.000 1.50
RO reject (SR2/ROR) 30% of RO inlet 0.005 200.000 10.00
Wet I (SR3) 250 1.000 1.000 5.00
Wet II (SR4) 200 2.000 0.500 4.50
Litography (SR5) 350 3.000 0.333 5.00
CMP I (SR6) 300 0.100 10.000 10.00
CMP II (SR7) 200 2.000 0.500 4.50
Etc (SR8) 280 0.500 2.000 5.00
Cleaning (SR9) 180 0.002 500.000 15.00
Scrubber (SR10) 300 0.005 200.000 10.00
Ultra pure water (UPW) ? 18.000 0.056 0.00
Municipal fresh water (FW) ? 0.020 50.000 0.50
832 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 5. RRN and WTN cascades of Example 1.

those of a mixer, but working in reverse. Eq. (15) is revised and
added as Eq. (25) to represent for flowrate balance between the
inlet (˙ u FRu ) and the outlet streams (i.e. purified flowrate, FRP and
reject flowrate, FRR ) of one partitioning regeneration unit (assum-
ing no water generation/loss). Based on the given recovery factor,
95% of the inlet flowrate is recovered as purified stream for recov-
ery (as shown in Eq. (26)), while the balance of 5% is rejected as
wastewater (included as source in Eq. (28)). Besides, these flowrate
terms are also added in the flowrate balance of the flowrate cascade,
which leads to Eqs. (27) and (28) as the revised form of Eqs. (8) and
(10), respectively.
˙k ˙u FRuk = FRP + FRR (25)
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 833

Fig. 6. An alternative total RCN design for Example 1.

FRP = 0.95˙k ˙u FRu,k (26) 7. Example 2

ık = ık−1 + (˙i FSRi + FRP − ˙j FSKj − ˙u FRu − ˙i FWi )k
k = k−1 + (˙i FWi + FTRDv + FRR − Ev FTRv − FD )k k∀K
Note that FRP and FRR are added in the RRN (Eq. (27)) and WTN
(Eq. (28)), respectively. This is because the purified stream is of
higher quality for further recovery, while the reject stream is to be
treated and discharged as wastewater.
Solving Eq. (24) with the constraints in Eqs. (5), (6) and (27)
for RRN; Eqs. (11)–(13) and (28) for WTN; Eqs. (18)–(21) for
pre-treatment model; as well as Eqs. (9), (15), (16), (25) and
(26) for regeneration model yields the cascade results in Fig. 7.
The total operating cost target is determined as $20,543,976.
Note that this scenario represents a significant reduction of cost
($6,679,536 = $27,223,512 − $20,543,976) as compared to the ear-
lier scenario without the regeneration process (Fig. 6). As shown in
Fig. 7a, the minimum flowrates for the UPW and purified water from
the membrane unit are determined as 463.94 t/h and 1079.5 t/h,
respectively. In addition, 41.73 t/h of wastewater from the mem-
brane reject stream, FTR,RR (Fig. 7b) is treated prior to final discharge
(with total flowrate, FD of 56.81 t/h, see Fig. 7b). Note that the
total regenerated flowrate is hence determined using Eq. (25)
as 1136.31 t/h (=1079.5 + 56.81 t/h). An alternative network that
achieves the flowrate and cost targets is shown in Fig. 8.
k∀K (27) This example is adopted from Ng, Foo, Tan, Pau, et al. (2009) to
illustrate the application of automated targeting for bilateral prob-
(28) lem. Table 2 shows the limiting data for a water recovery problem
in palm oil production (Ng, Foo, Tan, Pau, et al., 2009). Clay bath
flotation technique (Fig. 9) is used to separate palm kernel from
the shell and uncracked nut. It consists of clay-water slurry whose
proportions are chosen to achieve the desired density. However,
the quality of the slurry degrades during use through inadvertent
separation of the clay particles from water. Besides, the cracked
mixture contains impurities that also affect the density of the solu-
tion in the clay bath. Therefore, make-up water and fresh clay are
fed into the system to compensate for losses and to maintain the
separation efficiency.
In order to achieve high separation efficiency of the cracked
mixture, the slurry in the clay bath should posses a density of
1120 kg/m3 . Thus, density is identified as the most critical property
for wastewater recovery. Shelley and El-Halwagi (2000) reported
the general mixing rule for density as follows:
1 z
SRi
= (29)
¯ SRi
SRi
where ¯ is the mean density of the mixture; SRi is the density of
SRi.

Table 2
Limiting data for Example 2 (Ng, Foo, Tan, Pau, et al., 2009).

Process Flowrate (kg/day) Density,  (kg/m3 ) Operator, () (×10−4 m3 /kg) TSS, C (ppm) Operator, (C) (ppm)

Sink
Clay solution (SK1) 10 1120 8.93 – –

Source
Over flow (SR1) 4 1018 9.82 800 800
Bottom flow (SR2) 6 1200 8.33 2000 2000
Reject stream (RR) ∞ 1600 6.25 7472.5 7472.5
Purified stream (RP) ∞ 1100 9.09 10 10
Fresh water (FW) ∞ 1000 10
Clay ∞ 2600 3.85
834 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 7. RRN and WTN cascades of Example 1 for complete network.

As shown in Fig. 9 and Table 2, two water sources, i.e. over flow unit that performs as a partitioning regeneration system is intro-
(SR1) and bottom flow (SR2) are considered for water recovery. duced to purify the clay bath bottom stream for further recovery.
Note that five percent of these sources are purged to prevent accu- Both purified (RP) and reject streams (RR) from the filtration unit
mulation of impurities in the clay bath. In this work, a filtration are allowed to be recycled to the clay bath, therefore, Eq. (27) is
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 835

Fig. 8. An alternative design for complete network of Example 1.

revised as below:
ık = ık−1 + (˙i FSRi + FRP + FRR − ˙j FSKj − ˙u FRu − ˙i FWi )k
this bilinear constraint converts the problem into a non-linear
model. In addition, flowrate and property load balances of the par-
k∀K
titioning regeneration are also included in the optimisation model,
(30) given as in Eqs. (25) and (32), respectively.

˙u [FRu (pRu )] = FRP (pRP ) + FRR (pRR ) (32)

Since the FRP and FRR exits as sources at concentration level k
and potentially discharge as wastewater, these sources are con-
sidered as part of the total wastewater (˙ i FWi ). Note that both
purified and reject stream should not recovery simultaneously, as
that would entail mixing streams that have just been separated. To
avoid remixing of these two streams, Eq. (31) is added:
(FRP − FWRP ) × (FRR − FWRR ) = 0
where FWRP and FWRR are flowrates of wastewater generated from
the purified and reject streams, respectively. However, including
where (pRu ), (pRP ) and (pRR ) are linearly-additive property
operators of the regenerated source, purified and reject streams,
respectively.
Meanwhile, concentration of total suspended solid (TSS) is cho-
sen as the main property for final discharge and is given as 250 ppm;
a fixed outlet concentration of 20 ppm is in assumed for the treat-
(31)
ment process. Besides, the treated effluent from WTN cannot be
reused/recycled to the RRN because the waste treatment process
does not improve the density of the effluent. To synthesise a cost

Fig. 9. Schematic diagram of Example 2 (Ng, Foo, Tan, Pau, et al., 2009).
836 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 10. Optimisation result for Example 2: (a) superior cascade; (b) inferior cascade of RRN; (c) WTN cascade.

Table 3
Limiting data for Example 3 (Wang & Smith, 1994a).

Sink, SKj Flowrate, FSKj (t/h) Concentration, CSKj (ppm) Sources, SRi Flowrate, FSRi (t/h) Concentration, CSRi (ppm)

1 20 0 1 20 100
2 100 50 2 100 100
3 40 50 3 40 800
4 10 400 4 10 800

˙ j FSKj 170 ˙ i FSRi 170

Table 4
Limiting data for Example 4 (Sorin and Bédard, 1999).

Sink, SKj Flowrate, FSKj (t/h) Concentration, CSKj (ppm) Sources, SRi Flowrate, FSRi (t/h) Concentration, CSRi (ppm)

1 120 0 1 120 100

2 80 50 2 80 140
3 80 50 3 – –
4 140 140 4 140 180
5 80 170 5 80 230
6 195 240 6 195 250

˙ j FSKj 695 ˙ i FSRi 615

 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
Fig. 11. An alternative total RCN design for Example 2.
effective total property-based RCN, it is necessary to minimise the
TOC of the fresh, regenerated sources and waste treatment, given
by the optimisation objective:
TOC = (COSTFW FFW + COSTClay FClay + COSTR ˙k ˙u FRu
+ COSTTR ˙v FTRDv )AOT (33)
where FFW and FClay are the flowrates for fresh water and clay,
respectively; while COSTFW , COSTClay , COSTR and COSTTR reflect
the unit cost for each of these sources, regeneration and treatment,
which may be determined from historical data. In this work, the
costs of fresh water, clay, regeneration and waste treatment are
given as 1, 2, 0.5 and 0.5 $/t, respectively.
The optimisation objective in Eq. (33) is solved to minimise the
TOC, subject to the constraints for superior (Eqs. (5), (6) and (30))
and inferior cascades (Eqs. (6), (7) and (30)) of RRN; the cascade for
WTN (Eqs. (10)–(13)); as well as partitioning regeneration model
(Eqs. (9), (15), (16), (25), (26), (31) and (32)) which yields the result
in Fig. 10. The minimum flowrate targets of fresh water (FFW ) and
clay (FClay ) are found to each be 1.01 kg/day; with 5.7 kg/day of SR2
being regenerated to produce 4.18 kg/day of FRP and 1.52 kg/day of
FRR , which matches the previously reported solution based on mini-
mum flowrate (Ng, Foo, Tan, Pau, et al., 2009). Note that the purified
stream is recycled to the sink and the reject stream is sent for treat-
ment. Note also that 1.68 kg/day of waste (˙ v FTRDv ) is treated and
a total of 2.02 kg/day of waste (FD ) is discharged from the WTN.
The TOC target is determined as $1240 (with an AOT of 365 days).
Fig. 11 shows one potential total RCN design for Example 2; note
also that there may be alternative configurations that achieve the
various network targets.
8. Example 3
In order to illustrate the special case of single contaminant con-
straint, a literature example on water minimisation of the fixed load
problem (Wang & Smith, 1994a) is used. Table 3 shows the limit-
ing data for Example 3. As shown, there are four water sinks and
four water sources. In this example, contaminant concentration (C)
837
is the main constraint for resource conservation and wastewater
discharge. Based on the previous works (Ng et al., 2007a, 2007b),
the regeneration unit and the final treatment system are both fixed
outlet concentration type, with outlet concentrations of 5 ppm (CR )
and 10 ppm (CTR ), respectively. The maximum allowable discharge
concentration (CD ) is taken as 20 ppm. For comparison with the ear-
lier work of Ng et al. (2007a, 2007b), the objective function is set to
minimise fresh water flowrate (FFW ), given as in Eq. (34).
Minimise FFW (34)
Solving the objective function in Eq. (34) subject to the con-
straints in Eqs. (5), (6), (9) and (14)–(17), a total RCN cascade is
resulted, and is shown in Fig. 12. Note that the minimum fresh
water and wastewater flowrates of 20 t/h are targeted respectively,
which is consistent with the result reported by Ng et al. (2007b).
In contrast, the regeneration (˙ u FREGu = FR1/2 + FR3/4 = 71.58 t/h) and
waste treatment (FTR1/2 = 20 t/h) flowrates are different from those
reported in Ng et al. (2007b). This is because in the automated tar-
geting model, the effluent from the waste treatment unit (˙ v FTRDv )
is allowed for reuse/recycle in the water network. However, this
is not the case in Ng et al. (2007b), where the effluent from waste
treatment is only considered for final discharge. A water network
that achieves the flowrate targets is shown in Fig. 13. As shown,
there are slight differences between this network with that pre-
sented by Ng et al. (2007b) in Fig. 14. Water from SR3 and SR4 are
fully regenerated for further reuse/recycle in Fig. 13. Besides, 20 t/h
of SR2 is sent for final treated; and 2.22 t/h of this treated flowrate
is reused in SK3.
9. Example 4
Example 4 involves another water minimisation problem of
the fixed flowrate type. It is solved to target the minimum water
flowrates for the total water network with fixed removal ratio type
treatment system via automated targeting. Table 4 shows the lim-
iting data of Example 4 (Sorin & Bédard, 1999). It is noted that there
are five water sources that are available for recovery to six water
sinks. In this example, the discharge limit is assumed as 50 ppm
838 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 12. Total RCN cascade of Example 3.

and single type treatment system of fixed removal ratio (RRT) of formulation.
0.9 is utilised for both regeneration and final treatment, similar to
that presented by Bandyopadhyay and Cormos (2008). Since both FD ≤ ˙v FTRDv (36)
systems share a single treatment system, Eq. (14) is simplified as In this work, it is assumed that the unit operating costs of fresh
follows: water and treatment are 1 and 0.5 $/t, respectively. The optimisa-
tion objective is then set as minimise TOC, given by
ık = ık−1 + (˙i FSRi − ˙j FSKj + ˙v FTRDv − ˙v FTRv )k (35)
TOC = (COSTFW FFW + COSTTR ˙v FTRDv )AOT (37)

Note that the ˙ u FRu and ˙ u FREGu terms in Eq. (14) have been
removed and the regeneration and treatment systems are repre-
sented by only treatment flowrate terms (˙ v FTRDv and ˙ v FTRv ) in
Eq. (35). Since the treated flowrate can be reused/recycled in the
total water network, the waste discharge may be lower than the
treatment flowrate; hence, additional Eq. (36) is included in the
Following the proposed approach, Eq. (37) is solved subjected
to the constraints in Eqs. (5), (6), (9), (16), (17), (35) and (36). The
minimum fresh water (FFW ) and wastewater flowrate (FD ) targets
are located as 120 t/h and 40 t/h, respectively. Fig. 15 shows the
RCCD for the total RCN of Example 4. As shown, the total treatment
flowrate target (˙ v FTRDv ) is determined as 128.39 t/h, which con-

Fig. 13. Network design for Example 3.

 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 839

Fig. 14. Total water network for Example 3 (results of Ng et al., 2007b).

sists of three individual treatment flowrates, i.e. 16.78 t/h from SR2
(FTR2 ), 26.61 t/h from SR4 (FTR4 ) and 85 t/h from SR6 (FTR6 ), respec-
tively (see Fig. 15). The minimum TOC for the water network is
determined as $1,473,556 (with an AOT of 8000 h). An alternative
water network that achieves the targets is shown in Fig. 16. How-
ever, it is noticed that three treatment units (TR2, TR4 and TR6) are
used to treat sources SR2, SR4 and SR6, respectively. In order to sim-
plify the network structure (and to reduce capital cost indirectly)
in Fig. 16, a new scenario with the use of single treatment system
should be considered. Hence, modification on the automated tar-

Fig. 15. RCCD for the total RCN of Example 4.

840 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 16. Network design of total RCN for Example 4 (with multiple treatment systems).

geting model is required. In this new scenario, three sources (SR2,
SR4 and SR6) are allowed to mix before being treated.
Because of the infinite number of possible mixing ratios between
SR2, SR4 and SR6, both the regeneration flow rates and the inlet
and outlet concentrations of the three sources are treated as vari-
ables in the revised model. To avoid an infeasible RCCD, the lower
and upper bounds of the concentrations of the treatment outlet
stream is included in the model (Eq. (38)). Because SR2, SR4 and SR6
are available for regeneration, the boundaries of treatment outlet
concentration (CTR ) are defined by these sources, given as in Fig. 15:
14 ppm ≤ CTR ≤ 25 ppm
Next, the flowrates balance of treatment system (Eq. (16)) is also
included in the model. Besides, Eq. (39) shows the load balance of
the treatment system.
Note that the product of the unknown regeneration flowrate and
concentration in Eq. (39) leads to a bilinear term and, hence, an NLP
problem.
Solving Eq. (37) subject to constraints Eqs. (5), (6), (9), (16),
(17), (35), (36), (38) and (39) (via Lingo 10.0 with a global solver),
gives the cascade of the total RCN in Fig. 17. As shown, the fresh
water (FFW ) and wastewater flowrate (FD ) targets are located as
120 t/h and 40 t/h, respectively, with minimum treatment flowrate
(˙ v FTRDv ) of 128.39 t/h (at CTR of 22.11 ppm) and TOC of $1,473,556.
It is noted that the flowrate targets and TOC are identical with the
previous scenario where multiple treatment systems are consid-
(38) ered. A much simpler network for this scenario is shown in Fig. 18.
10. Example 5

˙v FTRDv CTR = (1 − RRT)˙k ˙v (FTRv,k CTv )
where CTv is the concentration of the individual waste streams. Note
that one can easily trace this concentration from the original source
concentration (CSRi ).
(39) Fig. 19 shows a water recovery problem in a Kraft pulping pro-
cess extracted from El-Halwagi (1997). Following the appropriate
segregation principles (Foo, Manan, & El-Halwagi, 2006) the limit-
ing water data for the problem is summarised in Table 5. It can be
seen that there is a water surplus (i.e. total source flowrates exceed

Table 5
Limiting data for Example 5 (El-Halwagi, 1997; 6).

Sink, SKj Stream Flowrate, FSKj (t/h) Methanol concentration, Sources, SRi Operation Flowrate, FSRi (t/h) Methanol concentration,
CSKj (ppm) CSRi (ppm)

1 Pulp washing 467 20 1 W3 12.98 419

2 Chemical recovery 165 20 2 W5 9.7 16,248
3 Condenser 8.2 10 3 W7 10.78 9900
4 W8 116.5 20
5 W9 48 233
6 W10 52 311
7 W11 52.2 20
8 W13 300 30
9 W14 140 15

˙ j FSKj 640.2 ˙ i FSRi 742.16

 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 841

Fig. 17. Cascade of total RCN for Example 4 (with single treatment system).

total sink flowrates) for this set of processes. As shown, there are
nine process sources that may be considered for water recovery to
three process sinks. Methanol concentration is taken as the main
constraint in water recovery. Besides, the maximum discharge limit
of methanol to the river is given as 15 ppm.
In this example, air stripping may be used to regenerate the
water sources for further recovery (El-Halwagi, 1997). According
to Ng et al. (2009b), this mass exchanger is categorised as a single-
pass regeneration process with fixed outlet concentration. Eq. (40)
shows the equilibrium equation for air stripping of methanol from
aqueous stream. A minimum allowable concentration difference
() of 4.275 ppm is assumed for this example (El-Halwagi, 1997).
CR = 0.38(x + ) (40)
where CR and x are the mass concentrations of methanol in water
(i.e. outlet concentration for regeneration unit) and air, respec-
tively.

Fig. 18. Network design of total RCN for Example 4 (with single treatment system).
842 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845

Fig. 19. Example 4: a Kraft pulping process (basis: 1 h; T refers to ton).

 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845 843

Fig. 20. Cascade of total RCN for Example 5.

Since ambient air is used as stripping agent for this case, x in Eq. tively (El-Halwagi, 1997). With this constraint, Eq. (45) is included
(40) is taken as 0 ppm. Eq. (40) then determines that the minimum in the mathematical formulation.
outlet concentration of regeneration system (CR ) as 1.625 ppm.
The operating cost of the stripping unit (OCAS ) is given by Eq. ˙k ˙v (FTRv,k CTv )
≤ 1000 ppm (45)
(35) (El-Halwagi, 1997): ˙k ˙v FTRv,k

OCAS ($/h) = 0.003L (41) The operating cost of the biotreatment facility (OCBio ) is given
where L is the air flowrate in the stripping unit, which is determined by the following relationship (El-Halwagi, 1997):
as (El-Halwagi, 1997):
   OCBio ($/h) = 0.00011 ˙k ˙v (FTRv,k CTv ) + 1.3 ˙v FTRv (46)
 (F C ) − F C
L = 0.5 FRu 
u Ru Ru i Ru R
(42)
(FRu CRu ) Besides, annualised cost of biotreatment (ACBio ) is taken from
u u
El-Halwagi (1997):
where FRu and CRu are the flowrate and inlet concentration respec-  
tively, of the individual regenerated streams. Note that one can Bio 240, 000 v FTRDv
AC ($/year) = AF (47)
easily trace CRu from the original source concentration (CSRi ). Note 0.35 3600
that the unknown terms in Eqs. (41) and (42) leads to bilinear term
and, hence, an NLP problem. The optimisation objective is set to minimise the total annu-
The annualised capital cost (ACAS ($/year)) of the air stripping alised cost (TAC), as shown in Eq. (48) (assuming an AOT of 8000 h).
unit is given by
 0.65 TAC = (COSTFW FFW + OCAS + OCBio )AOT + ACAS + ACBio (48)
AS
F
u Ru
AC ($/year) = 270, 000 AF (43)
3600 Solving Eq. (48) subject to the constraints in Eqs. (5), (6), (9),
(14)–(17), (36) and (41)–(47) using Lingo 10.0 (with global solver)
The annualising factor (AF) in Eq. (44) is defined as (Smith, yields the RCCD in Fig. 20. As shown, the minimum regeneration
2005): (FREGu ) and treatment (FTRDv ) flowrate targets are 95.05 t/h and
m(1 + m)y 101.96 t/h. Note that, the entirety of streams of SR1–SR3, SR6 and
AF = (44) 9.59 t/h of SR5 are regenerated via air stripping for further recov-
(1 + m)y − 1
ery. Meanwhile, 38.41 t/h of SR5 and 63.55 t/h of SR8 are treated
where m = fractional interest rate per year, y = number of years. In for discharge using biotreatment. Even though this biotreatment
this example, m and y are given as 10% and 5 years, respectively. effluent may be recycled to the process, it is determined as unde-
Meanwhile, in order to comply with environmental legisla- sirable by the mathematical model. Note also that no fresh water
tion, a biotreatment facility is employed for wastewater treatment. is required in the process since water is in surplus in the pro-
Note that upon treatment, the effluent from the biotreatment may cess. Hence, a total of 101.96 t/h of wastewater (FD ) is generated
be recycled to the process (its outlet flowrate is hence added as from the process. This is a type of threshold problem as reported
a source in Eq. (35)). It is also determined that the maximum by Foo (2008). The minimum TAC of this example is determined
inlet and average outlet (CTR ) concentration of methanol for the as $2,221,741. Fig. 21 shows a possible water network design for
biotreatment facility are given as 1000 ppm and 15 ppm, respec- Example 5.
 844
D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
Fig. 21. An alternative water network design for Example 5 (basis: 1 h; T refers to ton).
 D.K.S. NG et al. / Computers and Chemical Engineering 34 (2010) 825–845
11. Conclusion
In this paper, the automated targeting approach is extended for
concentration- and property-based total RCN, where the individ-
ual elements of reuse/recycle, regeneration/interception and waste
treatment are considered simultaneously. The automated target-
ing has the advantages of both insight-based and mathematical
optimisation approaches. However, an added advantage is that the
objective function may be modified to suit case-specific consider-
ations, rather than being limited to just minimising the external
resource flowrate. In addition, unlike previous pinch-based meth-
ods, the proposed approach is applicable for cases where different
impurities or properties can be used for reuse/recycle and inter-
ception sections. Literature examples were used to illustrate the
proposed approach.
Acknowledgments
The financial support from the Malaysian Ministry of Science,
Technology and Innovation (MOSTI) through Science Fund 03-02-
12-SF0018, and the De La Salle University Science Foundation are
deeply appreciated.
References
Agrawal, V., & Shenoy, U. V. (2006). Unified conceptual approach to targeting and
design of water and hydrogen networks. AIChE Journal, 52(3), 1071.
Almutlaq, A., Kazantzi, V., & El-Halwagi, M. M. (2005). An algebraic approach to
targeting waste discharge and impure-fresh usage via material recycle/reuse
networks. Clean Technologies and Environmental Policy, 7(4), 294–305.
Alves, J. J., & Towler, G. P. (2002). Analysis of refinery hydrogen distribution systems.
Industrial and Engineering Chemistry Research, 41, 5759.
Bagajewicz, M. J., & Faria, D. C. (2009). On the appropriate architecture of the
water/wastewater allocation problem in process plants. In Computer-aided
chemical engineering. Amsterdam: Elsevier., pp. 1–20.
Bai, J., Feng, X., & Deng, C. (2007). Graphical based optimisation of single-
contaminant regeneration reuse water systems. Chemical Engineering Research
and Design, 85 (A8), 1178.
Bandyopadhyay, S. (2006). Source composite curve for waste reduction. Chemical
Engineering Journal, 125, 99.
Bandyopadhyay, S., & Cormos, C.-C. (2008). Water management in process indus-
tries incorporating regeneration and recycle through a single treatment unit.
Industrial and Engineering Chemistry Research, 47, 1111.
Bandyopadhyay, S., Ghanekar, M. D., & Pillai, H. K. (2006). Process water manage-
ment. Industrial and Engineering Chemistry Research, 45, 5287.
El-Halwagi, M. M. (1997). Pollution prevention through process integration: Systematic
design tools. San Diego: Academic Press.
El-Halwagi, M. M. (2006). Process integration. Amsterdam: Elsevier.
El-Halwagi, M. M., Gabriel, F., & Harell, D. (2003). Rigorous graphical targeting for
resource conservation via material recycle/reuse networks. Industrial and Engi-
neering Chemistry Research, 42, 4319.
El-Halwagi, M. M., Glasgow, I. M., Eden, M. R., & Qin, X. (2004). Property integration:
Componentless design techniques and visualization tools. AIChE Journal, 50(8),
1854.
El-Halwagi, M. M., & Manousiothakis, V. (1990). Automatic synthesis of mass-
exchange networks with single component targets. Chemical Engineering Science,
9, 2813–2831.
Feng, X., Bai, J., & Zheng, X. (2007). On the use of graphical method to determine the
targets of single-contaminant regeneration recycling water systems. Chemical
Engineering Science, 62, 2127.
Foo, D. C. Y. (2008). Flowrate targeting for threshold problems and plant-wide inte-
gration for water network synthesis. Journal of Environment Management, 88,
253–274.
Foo, D. C. Y. (2009). State-of-the-art review of pinch analysis techniques for water
network synthesis. Industrial and Engineering Chemistry Research, 48(11), 5125.
845
Foo, D. C. Y., Kazantzi, V., El-Halwagi, M. M., & Manan, Z. A. (2006). Surplus diagram
and cascade analysis techniques for targeting property-based material reuse
network. Chemical Engineering Science, 61, 2626.
Foo, D. C. Y., & Manan, Z. A. (2006). Setting the minimum utility gas flowrate targets
using cascade analysis technique. Industrial and Engineering Chemistry Research,
45, 5986.
Foo, D. C. Y., Manan, Z. A., & El-Halwagi, M. M. (2006). Correct identification of
limiting water data for water network synthesis. Clean Technologies and Envi-
ronmental Policy, 8, 96–104.
Foo, D. C. Y., Manan, Z. A., & Tan, Y. L. (2006). Use cascade analysis to optimize water
networks. Chemical Engineering Progress, 102(7), 45.
Grooms, D., Kazantzi, V., & El-Halwagi, M. M. (2005). Scheduling and operation
of property-interception networks for resource conservation. Computers and
Chemical Engineering, 29, 2318–2325.
Kazantzi, V., & El-Halwagi, M. M. (2005). Targeting material reuse via property inte-
gration. Chemical Engineering Progress, 101(8), 28.
Kuo, W.-C. J., & Smith, R. (1997). Effluent treatment system design. Chemical Engi-
neering Science, 52(23), 4273.
Kuo, W. C. J., & Smith, R. (1998a). Design of water-using systems involving regener-
ation. Process Safety and Environment Protection, 76, 94.
Kuo, W. C. J., & Smith, R. (1998b). Designing for the interactions between water-
use and effluent treatment. Chemical Engineering Research and Design, 76,
287.
Manan, Z. A., Tan, Y. L., & Foo, D. C. Y. (2004). Targeting the minimum water flowrate
using water cascade analysis technique. AIChE Journal, 50(12), 3169.
Ng, D. K. S., Foo, D. C. Y., & Tan, R. R. (2007a). Targeting for total water network—Part
1: Waste stream identification. Industrial and Engineering Chemistry Research, 46,
9107.
Ng, D. K. S., Foo, D. C. Y., & Tan, R. R. (2007b). Targeting for total water network—Part
2: Waste treatment targeting and interactions with water system elements.
Industrial and Engineering Chemistry Research, 46, 9114.
Ng, D. K. S., Foo, D. C. Y., & Tan, R. R. (2009a). Automated targeting technique for
single-component resource conservation network—Part 1: Direct reuse/recycle.
Industrial and Engineering Chemistry Research, 48(16), 7637.
Ng, D. K. S., Foo, D. C. Y., & Tan, R. R. (2009b). Automated targeting technique
for single-component resource conservation networks—Part 2: Single pass and
partitioning waste interception systems. Industrial and Engineering Chemistry
Research, 48(16), 7647.
Ng, D. K. S., Foo, D. C. Y., Tan, R. R., Pau, C. H., & Tan, Y. L. (2009). Automated targeting
for conventional and bilateral property-based resource conservation network.
Chemical Engineering Journal, 149, 87.
Ng, D. K. S., Foo, D. C. Y., Tan, R. R., & Tan, Y. L. (2007). Ultimate flowrate targeting
with regeneration placement. Chemical Engineering Research and Design, 85(A9),
1253.
Ng, D. K. S., Foo, D. C. Y., Rabie, A., & El-Halwagi, M. M. (2008). Simultaneous synthe-
sis of property-based water reuse/recycle and interception networks for batch
processes. AIChE Journal, 54(10), 2624–2632.
Ng, D. K. S., Foo, D. C. Y., Tan, R. R., & Tan, Y. L. (2008). Extension of targeting procedure
for “ultimate flowrate targeting with regeneration placement” by Ng et al. Chem.
Eng. Res. Des., 85 (A9), 1253–1267. Chemical Engineering Research and Design,
86(10), 1182.
Ponce-Ortega, J. M., Hortua, A. C., El-Halwagi, M. M., & Jiménez-Gutiérrez, A. (2009). A
property-based optimization of direct-recycle networks and wastewater treat-
ment processes. AIChE Journal, 55(9), 2329–2344.
Qin, X., Gabriel, F., Harell, D., & El-Halwagi, M. M. (2004). Algebraic techniques
for property integration via componentless design. Industrial and Engineering
Chemistry Research, 43, 3792–3798.
Shelley, M. D., & El-Halwagi, M. M. (2000). Componentless design of recovery and
allocation systems: A functionality-based clustering approach. Computers and
Chemical Engineering, 24, 2081.
Smith, R. (2005). Chemical process design and integration. New York: John Wiley &
Sons.
Sorin, M., & Bédard, S. (1999). The global pinch point in water reuse networks. Process
Safety and Environmental Protection, 77, 305.
Tan, R. R., Ng, D. K. S., Foo, D. C. Y., & Aviso, K. B. (2009). A superstructure model for the
synthesis of single-contaminant water networks with partitioning regenerators.
Process Safety and Environmental Protection, 87, 197.
Wang, Y. P., & Smith, R. (1994a). Wastewater minimisation. Chemical Engineering
Science, 49(7), 981.
Wang, Y. P., & Smith, R. (1994b). Design of distributed effluent treatment systems.
Chemical Engineering Science, 49(18), 3127.