Article

pubs.acs.org/IECR

Process Retrofitting via Intensification: A Heuristic Methodology and

Its Application to Isopropyl Alcohol Process
Mark Wendou Niu and G.P. Rangaiah*
Department of Chemical & Biomolecular Engineering, National University of Singapore, Singapore, 117585

ABSTRACT: Although process retrofitting is essential for

existing chemical plants to remain sustainable, studies on
systematic retrofitting methodologies for chemical processes are
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

limited and also do not focus on the possibility of replacing a

few unit operations with an intensified unit. Hence, in this study,
a heuristic-based systematic methodology is developed for process
retrofitting via intensification/integration. Compared with the
Downloaded via UNIV DE ANTIOQUIA on July 27, 2019 at 20:38:26 (UTC).

conventional method, it focuses on exploring integrated retrofit

solutions. Fast screening with heuristics, process knowledge, and
process simulation is used to avoid infeasible solutions in the
early stage. The proposed methodology is applied to retrofitting
the conventional isopropyl alcohol (IPA) process. An improved
solution without capital investment is obtained by operation
optimization; it results in reducing the manufacturing cost per
unit product by 5.5%. Then, two retrofit solutions, both involving reactive distillation, are developed. One of them reduces the
manufacturing cost per unit product by 23% and results in a reduction in the number of equipment in the process.

1. INTRODUCTION schemes including feed location, reflux ratio, feed conditioning

The chemical industry, comprising a number of distinctive and side condensing/reboiling. It was found that total exergy
sectors such as petroleum refining, petrochemicals, polymers, loss can be reduced from 7492 kW to 3628 kW for the three
agrochemicals, biotechnology products and more, is one of the columns in the biodiesel process. Long et al.4 explored DWC to
most energy intensive industries. Most of the chemical plants increase the throughput of an existing acetic acid purifica-
were built at a time when energy costs and environmental tion process. Several column arrangements were analyzed,
concerns were not on the agenda. Further, process and and results showed that DWC requires less energy cost than
computational technologies continue to be developed, thanks conventional columns. Long and Lee5 converted an extractive
to ongoing research in various parts of the world. Hence, the distillation (ED) to a thermally coupled ED sequence, where
design of existing plants may not be the most efficient and solvent recovery column was used as the side rectifier column.
optimal for the present situation. All these things call for In their first case study on separation of acetone and methanol
modifying and/or replacing the relevant process equipment in with water as the extractive solvent, retrofit design achieved 13%
the plant with more efficient units having lower operating cost energy savings and 10% additional capacity. For the second case
and/or emissions. Thus, process retrofitting is essential for study on separation of heptane and toluene with aniline as the
existing chemical plants to remain sustainable. Many studies extractive solvent, results show a 32% energy savings. Park et al.6
on chemical process retrofitting have been reported in the proposed a retrofit design for a boil-off gas (BOG) handling
literature. These can be classified into two categories: individual process, and the retrofit design was optimized using the SQP
case studies and systematic retrofitting methodologies. Recent program in MATLAB. Optimal results show 23% energy savings
studies in the former category are briefly reviewed below. and a payback period of only 2.2 months.
Heat exchanger network retrofitting is a specialized area not Development of new technologies motivates and accelerates
considered in this study. The review of techniques for heat retrofitting of conventional chemical processes. For example,
exchanger networks and their applications is available in with a better performance, improved life, and lower cost,
Sreepathi and Rangaiah.1 membrane separation is increasingly studied for process
Premkumar and Rangaiah2 investigated dividing wall columns retrofitting. Motelica et al.7 investigated the potential of adding
(DWC) for retrofitting conventional 2-column systems. Their a membrane unit for butadiene separation from the C4-fraction
results on a number of applications suggest 25% to 48% reduc- in an olefin plant. Their simulation results show that, with a
tion in total annual cost. Ngyuen and Demirel3 used column
grand composite curves (CGCCs) and exergy loss profiles to Received: July 23, 2015
generate retrofit solutions for columns in a biodiesel plant. Revised: February 14, 2016
Exergy loss profiles help to examine energy degradation on all Accepted: March 2, 2016
internal trays, and CGCCs help to identify possible retrofit Published: March 3, 2016

© 2016 American Chemical Society 3614 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

membrane selectivity of 7.5 for butadiene/mono-olefins, 23% Studies on systematic retrofitting methodologies are limited
reduction in energy required can be achieved. Ploegmakers et al.8 compared to individual case studies. Moreover, most of the
studied the addition of a membrane unit to an ethylene/ethane current retrofitting methodologies are limited to operation
fractionation column, in different configurations. Their results optimization of a single unit operation, restructure/repiping
show that with a minimum ethylene/ethane membrane selec- within a few unit operations, or adding a few unit operation(s)
tivity of 30, cost savings can reach as much as 16% for the in the upstream/downstream of the energy intensive unit. They
membrane unit in series. do not consider the possibility of replacing a series of unit
Although achieving attractive cost savings and/or capacity operations with an intensified unit operation. Hence, they may
increases, studies based on individual applications is case- not achieve attractive improvements for complicated chemical
specific, and their procedure is not general and difficult to processes such as complex reactions followed by azeotropic
extend to other applications. Therefore, systematic retrofitting distillation. Therefore, a new retrofitting methodology, which
methodologies are required. Fischer et al.9 presented the first can achieve substantial improvements, needs to be developed.
systematic procedure for developing and screening retrofit If a new concept and/or technology is brought into process
solutions. Their procedure includes an analysis step to identify retrofitting, significant improvements are possible. In this
“bottlenecking” equipment, and a retrofitting step that considers context, process intensification (PI) is attractive. Defined as
modifications in both structure and equipment. The case study “any chemical engineering development that leads to a
of retrofitting the hydro-dealkylation (HDA) process indicated substantially smaller, cleaner, safer, and more energy efficient
that retrofitting by this methodology could achieve significant technology”,18 it can bring significant benefits in terms of
savings. Another early study was by Dantus and High,10 process capital/operating costs, safety, product quality, etc.19
who developed a superstructure-based methodology for waste PI covers both process equipment types and methods, which
minimization and cost savings. It starts with sensitivity and respectively refer to novel intensified equipment (e.g., rotating
hierarchical analysis to identify waste minimization options packed bed/HiGee and DWC), and to integration of unit
and modified configurations, respectively. Then, the alternatives operations, that is, process integration.20 According to these
identified are used to construct a superstructure for formulating authors, process integration includes multifunctional reactors
and solving a mixed integer nonlinear programming problem. (e.g., reactive distillation, reactive extraction) and hybrid separa-
Dantus and High10 applied their proposed methodology to the tions (e.g., membrane absorption, membrane distillation).
manufacture of methyl chloride. When PI is introduced into process retrofitting, it usually implies
Kralj and Glavic11 developed a mathematical model based process integration rather than novel intensified equipment.
method to sequentially optimize superstructure, material, and Hence, PI and process integration are used synonymously in the
energy flow rates; subsequently, Kralj et al.12 extended it to a rest of this paper.
stepwise simultaneous optimization scheme. A methanol plant If introduced into retrofitting methodology, PI-based
was retrofitted by both the earlier and the latter methods, solution(s) will be able to cross the boundary of single unit
and the results show that the former leads to extra profit of operation, and that is exactly the focus of this study. This is
2.8 million USD/year compared 5.17 million USD/year by the probably the first study that specifically includes PI into process
latter method.12 retrofitting methodology. Furthermore, a complicated chemical
Researchers from Gani’s group developed an indicator- process, namely, the isopropyl alcohol (IPA) process, will be
based methodology for assessing the cost-saving potential of improved by applying the methodology. So far, retrofit of the
continuous chemical processes.13 The methodology consists of IPA process has not been reported in the open literature.
decomposition analysis, retrofit solution generation, and rough The rest of this paper is organized as follows. Section 2 presents
economic evaluation, and it was tested by retrofitting the HDA the new retrofitting methodology. Section 3.1 analyses the
process. The methodology was later extended to include environ- conventional IPA process by this methodology. Section 3.2
mental indicators so as to achieve sustainable retrofit targets.14 discusses modifications within each unit operation (i.e., local
Compared with its earlier version in Uerdingen et al.,13 optimal solution). Sections 3.3 and 3.4 develop the two
one major improvement was the use of an industrial simulator PI-based retrofit solutions. Section 3.5 compares all the retrofit
(PRO-II) in both process analysis and optimization. Uerdingen solutions to choose the best one. Conclusions of this study are
et al.15 improved the methodology further by distinguishing presented in section 4.
retrofit options that require new investment with those options
without any investment. They implemented the improved
2. METHODOLOGY
methodology for a fine chemical process. Carvalho et al.16
developed an Excel-based software (Sustain-Pro), which In this section, a heuristic-based methodology for retrofitting
eliminated manual calculation of selected indicators. Moreover, via process intensification/integration is discussed. Unlike the
an indicator-based sensitivity analysis algorithm was added into previous methodologies in the literature, the new methodology
the original methodology to accelerate generation of retrofit focuses on generating integrated retrofit solutions. As part of the
solutions. methodology, optimization of operating conditions (referred to
Some other researchers studied retrofitting methodology as ‘improved solution without capital investment’ in this study)
for a specific type of processes. For example, Simon et al.17 is also covered and compared to the integrated solution(s).
developed a systematic framework for retrofitting batch This is because the former does not require any new investment,
processes; it consists of top-down analysis to cover three levels: and so it may be attractive in some applications. The proposed
the plant, the process, and the unit operation. Investigating methodology consists of four steps: (1) base case analysis,
interaction between different levels helps to get a deeper retrofit target, and contributing units; (2) generation of an
understanding of the bottlenecks in the original process. Simon improved solution without capital investment; (3) generation of
et al.17 successfully applied their methodology to a fine chemical integrated solution(s); and (4) comparison of retrofit solutions
batch plant. and decision making.
3615 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Step 1: Base Case Analysis, Retrofit Target, and Note that the metrics eventually contribute to the chosen
Contributing Units. In the first step, the original process is objective function. For example, if the cost is the objective
simulated using a commercial simulator (e.g., ASPEN Plus), function, each of these metrics will have to be multiplied by
process data from the simulation are collected, and the relevant unit cost (such as of utility and component lost).
performance of each unit operation is evaluated by selected Although contribution of a unit operation to the objective function
metrics. Table 1 gives an example of a retrofit target and the (such as cost) can be used, metrics allow analysis to be focused
on a fundamental quantity characterizing the performance of the
Table 1. Metrics for the Retrofit Target with an Example for unit operation, which is useful in developing integrated retrofit
TAC solutions. The retrofit of heat exchanger networks (HENs) is
generally solved as a separate problem; therefore, heat exchangers
target minimize total annual cost ($/yr) between processes streams are not evaluated in Table 2.
metric A utility cost ($/yr) After evaluating the performance of each unit operation, it is
metric B component loss ($/yr) essential to find the main contributors for each metric. For this,
metric C conversion of reactants taking metric A as an example, the maximum contributor is
metric D selectivity of products identified as follows:
metrics used for evaluating the process performance. As can be dmax, A = max{d1A , d 2A , ..., dm A } (1)
seen, the retrofit target is to minimize the manufacturing
cost per unit product ($/kg), which is the total annual cost A less important contributor could be neglected if it satisf ies:
(TAC, $/yr) divided by annual production rate. The TAC
consists of amortized capital cost (ACC) and annual diA < 0.2dmax,A (i = 1, 2, ..., m , and i ≠ max) (2)
production cost (APC).21 ACC is not involved in evaluating
the original process. Therefore, metrics are generally developed The factor of 0.2 is used here instead of 0.1 in Douglas
for APC. Three metrics A and B (such as utility cost and et al.22 This is mainly to identify retrofit opportunities with
component loss), directly related to annual production cost, are greater benefit and hence more potential for implementation.
used to evaluate the process performance. Metrics C and D are Note that the proposed methodology can be applied with any
related to reactor performance, and they indirectly influence reasonable factor except that more effort, perhaps without
TAC; these metrics are important because reactions usually much benefit, will be required if the factor is low. Screened
play an essential role in chemical processes. by eq 2, those metric values less than 1/5 of their respective
Next is to use the metrics from Table 1 to evaluate the process maximum are not considered; in contrast, metric values
performance. Unlike the classical operating cost diagram,9,22 the between 0.2dmax and dmax are not negligible, and their associate
entire process is decomposed into reaction and nonreaction parts units are referred to as contributing units, which are considered
(e.g., separation units, compressors, etc.). This classification of for retrofitting. Similarly, abmax,B and commax,A could be
unit operations is vital for exploring integrated retrofit solutions calculated. Some of the metrics may have the same units and
in the later steps. In Table 2, nonreactive unit operations (e.g., significance (e.g., $/yr for metrics A and B). Such metrics could
distillation columns, absorber, and heater) are evaluated while be evaluated together to further remove the less contributing
reactors 1 to s are evaluated in Table 3. Note that each type of units. Equation 3 provides an example for such circumstances;
suppose dmax,A satisfies
Table 2. Evaluation of Nonreaction System
dmax,A < 0.2 × max{abmax,B , com max,A} (3)
unit operation metric A metric B
D1 d1A Then, Di (i = 1 to m) metrics are no longer considered as the
⋮ ⋮ contributing unit operations. If the retrofit target is to maximize
Dm dmA the metric, then max in eqs 1 and 3 should be changed to min.
AB1 ab1B Step 2: Generation of an Improved Solution without
⋮ ⋮ Capital Investment. As noted by Grossmann et al.,23 the
ABn abnB simplest modification is to optimize the operating conditions
⋮ ⋮ ⋮ without adding any new equipment, as it requires no extra
COM1 com1A capital investment. Further, the typical implementation period
⋮ ⋮ (excluding design, analysis, safety study, approval etc., which
COMk comkA can be done in advance without affecting the existing process
operation) for adding a new unit operation is a few weeks.
Table 3. Evaluation of Reaction System Similarly, changes in an operating variable can be implemented
in a few hours or a day, mainly to bring the process to the new
unit operation metric C metric D steady state. Therefore, as long as modifying the operating
R1 r1C r1D conditions can improve the retrofit targets, other more com-
⋮ ⋮ ⋮ plicated retrofit designs (such as integrated retrofit solutions
Rs rsC rsD mentioned later) should be compared with the former, and the
integrated retrofit solution is attractive only when it provides
unit operation has its own metric; for example, Di is evaluated by cost savings more than the retrofit solution without capital
metric A (e.g., utility cost), while ABj is evaluated by metric B investment. Therefore, step two focuses on optimizing the
(e.g., component/solvent loss), and COMk is again evaluated by operating conditions, generating an improved solution without
metric A (utility cost). Reactor R is evaluated by conversion capital investment, which is later used for evaluating the more
of reactants (metric C) and selectivity of products (metric D). complicated retrofit designs.
3616 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Figure 1. Process scheme with contributing unit operations circled.

As mentioned in the above section, the retrofit target is to savings can be significant. Another important aspect is that
minimize manufacturing cost per unit product ($/kg); when reactors usually play an essential role in chemical processes, and
capital investment is not considered, it becomes minimization integrated units such as reactive distillation, reactive absorption,
of production cost per unit product ($/kg), given by and the reactor−membrane hybrid system have been well
studied.20,24 If no reactor is involved, integration happens
Min E = (C U + CO + C M + COPO + C P + C D + CGEN)/P
among contributing units since the purpose of integration in
(4) retrofit design is to simultaneously resolve drawbacks of the
Here, CU is the cost of utilities (steam, electricity, and cooling involved units. The second rule is straightforward, and a similar
agents, $/yr); COP represents cost of operations ($/yr); CM statement can be found in Lutze et al.25 The above two rules
refers to cost of maintenance ($/yr); COPO, CP, and CD are, provide basic guidelines; in the following, when developing root
respectively, cost of operating overhead ($/yr), cost of integration, heuristics will be introduced to generate integrated
property, taxes, and insurance ($/yr), and cost of depreciation solutions.
($/yr); CGEN represents general expenses ($/yr); and P is the Step 3 contains several substeps; to better understand these
annual production rate of the desired product (kg/yr).21 substeps, consider the process flow sheet shown in Figure 1.
Decision variables for optimization are selected after analyzing The first substep is to summarize the contributing units and
the degrees of freedom for the relevant unit operations as well their associated metric values, and highlight these units in the
as their effect on the objective function. Constraints are defined process flow sheet (Table 4 and Figure 1). The analysis of the
for meeting both product specifications (e.g., product purity) reactors is presented in Table 3.
and process requirements (e.g., outlet flow rate should not
exceed 10% of its original value). Note that it may not be Table 4. Contributing Unit Operations (Other than
possible to improve some unit operations through simple Reactors) and Their Metrics for the Process in Figure 1
modification of operating conditions for one reason or other.
If the optimization is successful, a solution without capital unit operation metric value
investment is achieved, which could later be used as a bench- D2 d2A= v1
mark for evaluating the integrated retrofit solutions. In case D3 d3A = v2
there is no improved solution without capital investment, the AB1 ab1B = v3
integrated retrofit solution will be directly compared with the D5 d5A = v4
original process.
Step 3: Generation of Integrated Solution(s). Some The second substep is to develop Level One integration
mathsize="9.5pt"complicated chemical processes tend to have (root integration), which refers to possible integration between
reversible reactions, and require energy intensive separation two unit operations. The following heuristics help to decide
units (e.g., azeotropic distillation). To address this challenge, the feasibility of root integration: (1) Reactor is first considered
it is possible to bring the idea of process intensification into to be integrated with its downstream separation unit. This
retrofit solutions, which breaks the boundary of each single unit happens when the main reaction(s) is equilibrium controlled
operation and may result in significant savings. Theoretically, and at least one of the reactants is provided in excess. (2) Root
any unit operation may have a chance to be integrated with integration should not cross the reactors. (3) Integrated unit
another unit operation. However, to avoid generating too many operations should preferably be those being used in the
infeasible solutions, two basic rules are proposed: industry.
I. Integration happens among reactors and other contribu- The first heuristic provides a guideline for exploring the initial
ting units; if no reactor is involved, integrations should root integration involving a reactor. A typical equilibrium-based
happen only among contributing units reaction is
II. Do not integrate units inhibiting each other’s performance K
α A + β B ↔ γC + δD (5)
According to the first rule, a reactor has higher priority than
other unit operations. One reason is that, although a reactor Suppose the reactor is integrated with its downstream separa-
may or may not directly consume utilities, chemicals generated tion unit, then the new reactive−separation unit must have the
or reacted during chemical reactions could heavily influence ability to continuously remove products C and/or D, thus
the subsequent separation process. If the reactor is integrated pushing the equilibrium toward the products and allowing more
with certain separation unit(s) in a proper way, then the energy reactants to be converted into products. As the conversion of
3617 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

the reactants is increasing, the amount of reactant(s) (in excess) it is inappropriate. One such example is the root integration
can be closer to reaction stoichiometry. Note that use of excess between D2 and D5 in Figure 1; as can be seen, D2 and D5 are
reactant (or low-conversion reactant) in the current process is on different sides of reactor R2, if they are integrated,
essential for integration. Capital cost of introducing a new component changes in R2 are likely to be ignored and result
reactive−separation unit is not negligible, and it should be offset in a faulty solution. The purpose of the last heuristic is to
by savings from the operating cost, which usually comes from eliminate integrated solutions which are not currently in use
reducing the separation cost of the excess reactant. The larger is in the chemical process industries. Finally, root integration
the amount of excess reactant required in the process (or the R1−D2, for example, follows the basic rules (I and II), and also
lower is the conversion of this reactant), more will be the assumed to satisfy heuristics 1 and 3; and so it is recorded for
potential improvement through integration. The first heuristic further consideration.
suggests under what circumstances the integrated reactive- Introducing an integrated unit operation will bring changes
separation unit should be considered. However, the operating to the base case; in most of the applications, some of the
window of a reactive−separation unit may be limited since both original unit operations will be replaced. In some applications,
separation and reaction must occur within the same ranges of increased flow rate may require additional unit operation(s)
temperature and pressure, which must also meet the mechanical (e.g., when flow rate exceeds 10% of its original value). So, the
design requirements (Figure 2).20 third substep is to figure out these changes to the process
scheme. Considering the process example in Figure 1, after
R1 → D2 integration, there are two possible new schemes shown
in Figure 3, in which definition of each unit/block is the same as

Figure 3. Process retrofit scheme for (a) root R1 → D2 (top) and (b)
root R1 → D2 (bottom).

that in Figure 1. In Figure 3a, R1 and D2 are replaced by a new

unit operation RD. Suppose the new integrated unit operation
satisfies certain operating conditions (e.g., operating temper-
Figure 2. Feasible operating range of a reactive−separation unit.
ature and feed ratio), and D2 and D3 are no longer required,
allowing RD to be directly connected to AB1 (absorber). All the
The first heuristic, however, does not indicate which remaining downstream unit operations after AB1 remain the same
integrated unit to be used. The following guidelines from as they are in the original process, which are represented by “ORI”
the literature will be useful in choosing the integrated unit. (for original). On the other hand, when RD satisfies some other
Reactive distillation (RD) is considered when (a) volatility of operating conditions, there is a different retrofit process scheme
the product and that of the remaining component are different18 shown in Figure 3b. As can be seen, R1 and D2 are replaced by the
to ensure that the product is not contaminated by the remaining same unit operation RD; the difference is that D3 (and not D2) is
component; (b) the reaction and catalyst must be active at the still required, and absorber AB1 and desorption tower D4 are no
set temperature and pressure; and (c) the downstream separa- longer required, allowing D3 to be connected to the other reactor
tion after the considered reactor contains distillation column(s) R2; the remaining unit operations are the same as in the original
and at least one distillation column is the contributing unit. process. Note that the two ORIs in Figure 3 are different.
Reactive absorption/extraction: RA (or RE) is appropriate Both options in Figure 3 obey rules and heuristics developed.
when (a) the absorbent does not absorb components other Since they are derived from the same root integration, they are
than the target component; (c) the absorbent is not involved in represented as
the reaction and is not toxic to the catalyst (if any); and (d) the
{(R1 → D2 ): H1‐RD‐AB1‐ORI ∪ H1‐RD‐D3‐R 2‐ORI}
downstream separation after the considered reactor contains
an absorber and at least one of the desorption columns is the where R1 → D2 indicates unit operations involved for root
contributing unit. Note that the contributing unit is usually integration; H1-RD-AB1−ORI and H1-RD-D3-R2-ORI are the
the desorption column. RD and RA/RE are the commonly used two retrofit process schemes. Before they could be added into
reactive−separation units. Hence, the criteria for implementing the retrofit solution pool, fast screening should be implemented
them are given above. Other systems include reactive−adsorp- to reject the inferior retrofit designs in their early stage. The
tion and reaction−crystallization, and their selection criteria are criterion for fast screening is that value of retrofit target for
available in the literature.18,20 the integrated solution should be no worse than that of the
The second heuristic helps to reduce infeasible root integra- improved solution without capital investment (if exists) or at
tions, which can be explained as follows. Products/reactants are least 10% better than that of the original process. The following
generated/consumed during chemical reactions. If root integra- equation represents this criterion in a mathematical form.
tion crosses the reactor, it may ignore these reactions and the
associate changes of components in the process, and therefore Vr ≤ Min{V0 , 1.15V 0OP} (6)

3618 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Figure 4. Possible root integrations.

Here, Vr, V0, and VOP 0 are the retrofit target values of the Up to now, all root integrations have been explored. Before
integrated solution, original process, and improved solution moving ahead, root integrations need to be compared to
without capital investment, respectively. Using the above eliminate the duplicate designs. Two designs are considered
criterion to evaluate H1-RD-AB1-ORI and H1-RD-D3-R2-ORI, duplicate when the process scheme is exactly the same. Using
assume that both of them are acceptable (i.e., they improve the this criterion for comparing all the candidates shows that
retrofit target). {(R1-D3): H1-RD-AB1-ORI} and {(R1-D2): H1-RD-AB1-ORI}
Up to now, one root (R1 → D2) is considered. The next step are duplicates, and one of them should be removed. Finally, the
is to consider all other root integrations, which are shown in possible root integrations are listed in the retrofit solution pool,
Figure 4 where the definition of each unit/block is the same as as shown in Table 5.
that in Figure 1). As can be seen, it starts with exploring the
remaining root integrations, in which D2 is involved. Note that Table 5. Possible Retrofit Schemes from Root Integrations
D2 → D5 is not considered as it violates the second heuristic.
D2 is located in the upstream of R2 (Figure 1), which means the root modified process
reactant needs to be purified before entering the reactor; R1 → D2 H1-RD-AB1-ORI
however, reactive distillation usually requires the reactor to be H1-RD-D3-R2-ORI
integrated with the downstream distillation column, and not the D2 → D3 H1-R1-D1-DWC-AB1-ORI
other way. Therefore, it is reasonable to assume that there is R2 → D5 ORI′-D4-RD
no root integration between R2 and D2. In contrast, possible
integration between D2 and D3 exists, namely, DWC. Repeating The next substep is extending the root integration and
the analysis as for R1 → D2 integration, D2 → D3 integration exploring further integration, which occurs by combing root
yields (Level 1) integrations. Note that this combination is among
different roots such as (R1 → D2) + (D2 → D3); (R1 → D2) +
{(D2 → D3): H1‐R1‐D1‐DWC‐AB1‐ORI}
(R2 → D5); and (D2 → D3) + (R2 → D5), and it can be referred
Further, the root integration D2 → AB1 is rejected based as Level 2 integration. However, retrofit schemes from the same
on eq 6. Until now, all the root integrations involving D2 root are not considered (e.g., H1-RD-AB1-ORI and H1-RD-D3-
are explored. There are totally two roots, and three possible R2-ORI) because they usually contain different subschemes or
schemes. have different operating conditions and it may not be possible
After exploring all possible root integrations for D2, it is to integrate them further.
necessary to move to other contributing unit operations. It can There are two types of combinations for Level 2 integration.
be seen from Figure 4 that, for D3, only D3 → R1 and D3 → AB1 In the first type, one unit operation may be involved in multiple
are considered (assuming that they all satisfy the respective rules (root) integrations (e.g., (R1 → D2) + (D2 → D3)), where a
and heuristics) as D2 → D3 has already been evaluated; D3 → R2 new integrated unit operation could be introduced to replace
and D3 → D5 are not considered for the reasons mentioned all these unit operations in the retrofit scheme. For example,
earlier while exploring root integration of D2 → R2 and a reactive dividing wall column (RDWC) may be used as
D2 → D5. Assuming the root integration of D3 → AB1 is not the integrated unit operation instead of RD or DWC alone.
attractive based on eq 6, there is only D3 → R1 left. The analysis Note that this RDWC should also satisfy the rules and heuristics
as for R1 → D3 integration gives the following retrofit scheme: given above. The new retrofit scheme is represented as

{(R1 → D3): H1‐RD‐AB1‐ORI} {(R1 → D2 → D3): H1‐RDWC‐R 2‐ORI}

In root (R1 → D2) in Table 5, only H1-RD-D3-R2-ORI is
For AB1, only AB1 → R1 needs to be considered; assuming that considered to be combined with {(D2 → D3): H1-R1-D1-DWC-
it violates heuristics for using reactive distillation, and hence AB1-ORI}. This is because D3 is not present in the retrofit
not included in the retrofit solution pool. For D5, following the scheme {(R1 → D2): H1-RD-AB1-ORI}, and so it is not possible
second heuristic, R2, is the only candidate and assume this to explore the integration between a RD and a DWC. Suppose
root integration satisfies the rules and heuristics. Repeating the H1-RD-D3-R2-ORI satisfies the rules and heuristics, it is then
analysis for exploring root integration design, the following evaluated by eq 6, after which, this level 2 retrofit scheme
retrofit scheme is obtained: can be added into the retrofit solution pool. The second type
{(R 2 → D5): ORI′‐D4 ‐RD} of combinations for Level 2 integration refers to combining
the root integrations with no shared unit operation such as
Similar to ORI, ORI′ means unit operations in the upstream of (R1 → D2) + (R2 → D5) and (D2 → D3) + (R2 → D5). This
D4 are the same as those in the original process. scenario is a simple combination and does not involve any
3619 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

new integrated unit operation. The new solution fulfilling the ing cost per unit product as the retrofit target. For retrofit
criterion in eq 6 will be compared to those in the solution pool; designs involving new unit operations, both APC and ACC
if it is not a duplicated one, the new solution will be added into should be considered for calculating the manufacturing cost per
the retrofit solution pool (Table 6). unit product (E):
Min E = (CPro + α × CTCI)/P (7)
Table 6. Retrofit Solutions Pool after Level 2 Integration
integration modified process CPro represents annual production cost ($/yr), which has the
same definition as described in step two; CTCI is the total
root R 1 → D2 H1-RD-AB1-ORI
capital investment ($) and α is the amortization factor (1/yr).
H1-RD-D3-R2-ORI
Suppose the value of total capital investment is M, plant life is
D2 → D3 H1-R1-D1-DWC-AB1-ORI
n years, and expected return is i. The amortized capital cost A is
R 2 → D5 ORI′-D4-RD
given by26
level 2 R 1 → D2 → D3 H1-RDWC-R2-ORI
combination R1 → D2 + R2 → D5 H1-RD1-AB1-D4-RD2 A i × (1 + i)n
D2 → D3 + R2 → D5 H1-R1-D1-DWC-AB1-D4-RD =
M (1 + i)n − 1 (8)

The next step is to explore Level 3 integration through For n = 15 years and i = 0.15, this equation gives A/M = 0.171,
combining Level 2 retrofit solutions (obtained from combining and so α = 0.171.
two root integrations) with another root integration (e.g., hybrid Evaluate all the integrated solutions in the retrofit solution
RDWC with membrane separation). One can repeat the pro- pool using eq 7, and choose the one with optimal E. Then,
cedure to generate Level 3, Level 4 to Level n integrated compare this E with EO of the original process or EOO of the
solutions until there is no further combination available. The retrofit solution without capital investment, if it exists. There
solution pool after adding level 2 root integration is shown in are several possibilities: if E is better than both EO and EOO,
Table 6; note that root integrations in Table 5 are also part of then the integrated solution is the final retrofit solution; if E is
this solution pool. In the combined retrofit solution H1-RD1- better than EO but worse than EOO, then the retrofit solution
AB1-D4-RD2 (Level 2), RD1 and RD2 are used to distinguish the without capital investment is the final retrofit solution. If E is
two different RD columns; otherwise, the design is the same as worse than EO and EOO does not exist, then there is no retrofit
that in the root integration. Figure 5 summarizes the steps in the solution for the original process.
development of integrated retrofit solutions.
Step 4: Optimization, Cost Estimation, and Solution 3. CASE STUDY
Comparison. The remaining integrated retrofit solutions are The application and potential of the new methodology described
optimized and evaluated to select the best one as the final in the previous section are illustrated in this section using a case
solution for retrofitting. Generally speaking, the selected retrofit study, namely, the process for isopropyl alcohol (IPA)
design is the one that brings the largest improvement in the production. IPA process is selected because (1) it involves
retrofit target, and so the final solution may vary when the both main and side reactions, which are governed by chemical
retrofit target changes. Assume minimization of the manufactur- equilibrium, (2) the separation process includes complicated

Figure 5. Generation of Integrated Solutions.

3620 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Figure 6. Original IPA process scheme; see Table 7 for data on feed and product streams.

Table 7. Process Data of Key Streams

stream category component mole fraction molar flow (kmol/h) temp (°C) pressure (kPa)
propylene feedstock propylene 0.95 400 −47.6 101.325
propane 0.05
water feedstock water 1 654.2 25 101.325
IPA product IPA 0.9999 349.9 83.0 105
water 0.0001

azeotropic distillation, (3) the original process is complex involv- The main side product, diisopropyl ether (DIPE) forms after
ing seven main units (namely, propylene column, reactor, lights the etherification reaction:
column, preconcentrator, ether column, extractive distillation
2(CH3)2 CHOH ⇌ H 2O + [(CH3)2 CH]2 O (10)
column, and regenerator), and (4) there have been no studies on
retrofitting this industrially important process. The most popular propylene direct hydration is the trickled bed
3.1. Base Case Analysis, Retrofit Target and Contribu- process. This mixed vapor−liquid-phase process uses strongly
ting Units. Isopropyl alcohol (IPA), also known as isopropanol acidic proton-exchange resin as the catalyst.28 It has good
or 2-propanol, is a secondary alcohol with the chemical formula process performance, and sufficient process data are available in
of C3H7OH and molecular mass of 60.09502 kg/kmol. It is the open literature. Hence, it is selected as the original/existing
widely used as a solvent in many industries and finds applica- process for the case study.
tions in food processing, coatings to reduce flammability, as Simulation of the IPA process is conducted using Aspen Plus
thinner and additive in paints, as well as disinfectants such as 8.4. The process flowsheet with selected simulation results is
alcohol wipes in household products and medical applications.27 shown in Figure 6 while process data of feedstock, product, and
With water, IPA forms an azeotrope of 87.4 wt % alcohol, entrainer are given in Table 7. As can be seen, fresh propylene
having a boiling point of 80.3 °C at atmospheric pressure.28 (at −47.6 °C and 101.325 kPa)29 and recycled propylene are
IPA is manufactured by two major commercial processes: mixed in a molar ratio of 4.65:1, pressurized, and preheated
indirect and direct hydration of propylene, of which the latter is before feeding into the trickle bed reactor. Meanwhile, another
common due to less corrosion in unit operations. Direct hydra- reactant water (fresh/recycled = 1:7.63) is pressurized to
tion of propylene is an exothermic, reversible reaction carried 7500 kPa and also fed into the reactor. The product stream
out with an acid catalyst, which could be cation-exchange resins from the bottom of the reactor is depressurized to 902 kPa and
such as molybdophosphoric acid, titanium and zinc oxides.28 sent to the lights column to strip the dissolved propylene and
The main reaction is propane. The gaseous mixture leaving the top of the lights
column is transferred to the propane column to remove propane
CH3CH = CH 2 + H 2O ⇌ (CH3)2 CHOH (9) (in order to avoid its accumulation in the process) from
3621 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

propylene as the latter is recycled to the reactor. Note Table 8. Evaluation of Contributing Units: IPA Reactor
that exchanger E1 for preheating fresh propylene produces
chemical conversion selectivity
chilled water, which can be used in the condenser of the
propane column. This is shown in Figure 6 by the dashed line. propylene 0.814
Alternately, fresh propylene can be directly used as the coolant water 0.065
in the condenser of the propane column. IPA 0.962
Heavier components from the bottoms of the propane DIPE 0.038
column are depressurized to 120 kPa, mixed with recycled IPA
and then sent to the preconcentrator, where more than 90% of
water is removed in the bottoms stream; this water stream is (cooling water) and $4/GJ (chilled water), and $0.06/kWh.21,26
recycled to the reactor (Figure 6). The distillate stream of the Note that the unit price of steam is the average of that in Seider
et al.21 and Turton et al.26
preconcentrator contains mainly water, IPA, and DIPE. In this
From eq 3 and Table 9, it is clear that the cost of cooling
mixture, the DIPE mole fraction is only 0.043, and its boiling
and electricity is much lower than that of steam; therefore, only
point is lower than that of water and IPA; therefore, it is
unit operations with significant steam consumption need to
removed in the ether column before separation of IPA and
be considered. From eq 1 and eq 2, the contributing units
water. DIPE leaves the process from the top of the ether
identified are lights column, preconcentrator, ED column, and
column, while the IPA and water mixture is collected from the
regenerator. As can be seen from the reactor performance
bottom and fed to the extractive distillation (ED) column. As
data (Table 8), conversion per pass of water is merely 0.065
mentioned above, IPA and water form an azeotrope; to break
(compared to 0.814 for propylene), and hence over 90 mol %
this azeotrope, dimethyl sulfoxide (DMSO) is introduced as the
of the reactor outlet stream is unreacted water, leading to high
entrainer.30 After ED, IPA with a molar purity of 0.9999 is
separation cost in the subsequent process (e.g., high steam
collected from the column top as the main product, whereas
cost in the preconcentrator). Owing to this link between low
the bottoms stream is sent to the regenerator to recover the
conversion in the reactor and high utility cost in the separation
entrainer. DMSO is concentrated to 0.99999 (mole fraction) in
units, the reactor is also selected as a contributing unit.
the bottoms of the regenerator. After mixing with makeup
3.2. Generation of an Improved Solution without
solvent (in very small quantity, not shown in Figure 6), the
Capital Investment. Following the methodology, the
entrainer is sent back to the ED column. In Figure 6, E2 is a
objective function is to minimize manufacturing cost per unit
single heat exchanger with the cold stream coming from P2 and
product ($/kg, eq 4) and the decision variables are related to
the hot stream is the bottoms of the regenerator.
the contributing units identified, namely, reactor, lights column,
To verify the simulation of the original process, several key
preconcentrator, ED column, and regenerator. After analyzing
results are compared with those in the literature. The water
their degrees of freedom and the sensitivity of the objective
to propylene ratio at the inlet of the reactor in the current function, reflux ratio of the preconcentrator (Rpre), the reflux
simulation is 12 compared to the range of 12−15 stated by Xu ratio of the regenerator (Rreg), and DMSO circulation rate
et al.31 Conversion of propylene is 0.814 compared to ⩾0.75.31 (FDMSO) are selected as the decision variables. Some operating
Selectivity of IPA is 0.962 versus ⩾0.93.31 For the ED column, conditions of the contributing units such as reflux ratios of
reboiler duty is 14.66 kW/kmol of column inlet stream (with lights and ED columns are not selected because they are used
molar composition: IPA = 0.55 and water = 0.45), while to fulfill the process requirements (e.g., the former is used
Luyben and Chien30 gave a value of 14.99 kW (for inlet to limit the concentration of propylene in the bottom outlet,
composition: IPA = 0.5 and water = 0.5). Molar composition of and the latter is used to achieve the IPA purity in the top outlet
the ED bottoms stream is water = 0.328 and DMSO = 0.672, stream).
which is same as that in Luyben and Chien.30 Note that solvent Constraints in the optimization are based on process
loss (DMSO loss) in the water purge is very small (2.6 × 10−7 knowledge and retrofitting considerations. For example, bounds
kmol/h), and hence makeup solvent is not included in Table 7. for decision variables are specified as
This loss is similar to 1.1 × 10−8 kmol/h in Arifin and Chien,32
where 100 kmol/h of IPA and water mixture (mole fraction R pre ≥ 0 (11)
of 0.5 IPA and 0.5 water) is separated through ED and
regenerator using DMSO as the entrainer. Thus, the present RReg ≥ 0 (12)
simulation results are comparable to those in the literature.
FDMSO ≤ 605 kmol/h (13)
Following the methodology in the previous section, the next
step is to determine the retrofit target and then identify the Equations 11 and 12 ensure a positive reflux ratio. Equation 13
contributing units. The retrofit target chosen for the case study gives the upper bound of DMSO circulation rate, which should
is to minimize annual manufacturing cost per unit of product not exceed 10% of its original value.36 Besides FDMSO, the flow
($/kg). Based on this retrofit target and features of the process, rate of both top and bottom outlet streams from preconcen-
“conversion of reactants”, “selectivity of products”, “steam trator, ED column, and regenerator should also follow similar
cost”, “cooling cost,” and “electricity cost” are selected as the constraints as eq 13. Equality constraints which include mass
metrics for evaluating the original process, and their values and energy balances for each unit operation in the process are
are listed in Tables 8 and 9. In Table 9, the condenser duty for taken care by Aspen Plus simulation, which means these are
the propane column is 2105.8 kW, part of which is provided by satisfied once the process simulation converges.
the chilled water from E1, resulting in the duty of 1362.7 kW The built-in optimizer in Aspen Plus gave the following
to be supplied by an external cold utility. Utilities used are optimal values: Rpre changes from 1.15 (initial guess) to 0.73,
steam (LP: 3.5 bar and HP: 31.0 bar), coolants (cooling Rreg decreases from 0.45 (initial guess) to 0.39, and FDMSO
water and chilled water) and electricity; their unit prices are increases from 515 kmol/h (initial guess) to 545 kmol/h.
$16.9/1000 kg (LP) and $22.3/1000 kg (HP), $0.02/1000 kg A cost estimation of both original and the optimized processes
3622 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Table 9. Evaluation of Contributing Units: Unit Operations

cost of steam cost of cooling
unit operation reboiler duty (kW) steam type steam cost ($/h) condenser duty (kW) cooling agent cooling cost ($/h)
propane column 1624.6 LP 46.1 1332.1 chilled water 19.6
lights column 8175.8 LP 190.1 8280.4 cooling water 13.5
pre-concentrator 11156.9 LP 359.3 15303.6 cooling water 26.1
ED column 9212.9 LP 256.0 7573.3 cooling water 13.0
regenerator 5906.4 HP 257.7 4635.1 cooling water 7.9
ether column 1744.6 LP 54.0 1444.2 cooling water 2.5
cost of electricity
unit operation power consumption (kW) cost of electricity ($/h)
P1 10.5 0.6
P2 120.7 7.3
P3 0.24 0.01
P4 318.5 19.1
P5 0.04 0.002

Table 10. Cost Estimation of both Original and Optimized stoichiometric ratio. According to Heuristic 1, exploration of
Processes root integration will start from integrating reactor with its
downstream separation units. In the proposed case study, only
item original optimized
distillation columns are involved in separation; therefore, the
cost of utilities ($/yr) 10,182,496 9,224,027 potential integrated unit is the reactive distillation column.
cost of operations ($/yr) 3,100,937 3,100,937 As an RD column is currently used in chemical process
cost of maintenance ($/yr) 1,426,194 1,426,194 industries,33 heuristic 3 is satisfied too. Following the
cost of operating overhead ($/yr) 844,810 844,810 methodology, guidelines for applicability and potential of RD
property taxes and insurance ($/yr) 310,042 310,042 should be considered. In Table 11, the preconcentrator, used
cost of depreciation ($/yr) 1,240,169 1,240,169 for removing the excess water, is the most energy intensive
general expense ($/yr) 1,204,961 1,204,961 unit (contributing unit). According to simulation results, the
Summary reaction temperature is 129 to 135 °C (comparable to the
total manufacturing cost ($/yr) 18,309,609 17,351,140 temperature of the IPA reactor28), while the pressure is 1751.4
annual IPA production (kg/yr) 168,207,880 168,207,880 to 1752.9 kPa (lower than that of the IPA reactor shown in
manufacturing cost per unit product ($/kg) 0.109 0.103 Figure 6). Thus, most of the guidelines for applicability and
potential of RD are satisfied; the requirement on relative
is conducted following the steps provided by Seider et al.,21 volatility of components will be later verified when the root
and the results are shown in Table 10. As can be seen, the integration is analyzed.
objective function value decreases from $0.109/kg to $0.103/kg To develop the initial reaction−separation root integration,
(by 5.5%), mainly benefitting from less utility consumption the most relevant separation unit should be selected. As
(dropped from $10,182,496/yr to $9,224,027/yr). This suggests mentioned above, the preconcentrator is the most relevant unit
the existence of an improved solution without capital invest- for integration; hence, root R → PRC is first considered and
ment, which will be used as the benchmark to evaluate the this integrated design is shown below:
integrated retrofit solutions.
3.3. Generation of Integrated Retrofit Solutions. From {(R → PRC): PC‐RD‐LC‐ED‐RE}
the previous section, it can be seen that, although there is one In the above process notation, PRC is the preconcentrator,
retrofit solution without capital investment, the improvement PC is the propane column, LC represents the lights column,
is limited with manufacturing cost per unit product decreasing and RE is the regenerator. This integration does not cross any
by only 5.5%. Therefore, it is necessary to explore integrated reactor, and therefore heuristic 2 is fulfilled. Note the ether
retrofit solutions, which may bring larger process improve- column is not included; this is because the side reaction is also
ments. From section 3.1, the identified contributing units are controlled by the reaction equilibrium, and simulation suggests
listed in Table 11 for exploring possible integrations. that DIPE generation is very little when IPA is continuously
removed from the system.
Table 11. Summary of Contributing Units in the Case Study Similarly, two more roots, which also satisfy heuristic 2, are
unit operation metric value developed and their integrated retrofit designs are
reactor water conversion 0.065 {(R → LC): RD‐PRC‐ED‐RE}
lights column steam cost 231.6 $/h
pre-concentrator steam cost 316.1 $/h {(R → ED&RE): PC‐RD}
ED column steam cost 261.0 $/h
In the first design, {(R → LC): RD-PRC-ED-RE} assumes that
regenerator steam cost 264.3 $/h
the conversion of propylene is 100%; as a result, the lights and
propane columns are no longer required. The inert component
The reactor is identified as one of the contributing units, (propane) in the propylene feed is removed from the top of RD
the main reaction (eq 10) is equilibrium controlled, and the column, while the water−IPA azeotropic mixture is removed from
water to propylene ratio at the reactor inlet is 12 times the the bottom. The boiling points of key components (Table 12)
3623 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Table 12. Boiling Point of Key Components (at 1 atm)28 The physical explanation of this is that, when less water is fed into
the RD, water content in the RD outlet stream containing IPA
component boiling point (K)
is at a manageable level, this RD outlet stream could be directly
propane 231.15 fed to the ED column without the need for preconcentrator and
propylene 225.43 lights column. Note that the unreacted propane leaves in the
IPA 355.65 overheads stream of RD, and will not affect the performance of
water 373.13 ED. Simulation of this modified process is successful indicating
the feasibility of the integrated retrofit solution. Now, a
comparison of {(R → LC&PRC): RD-ED-RE} with its root
are used to assess the feasibility of this retrofit solution. As can
{(R → LC): RD-PRC-ED-RE} shows that the former contains
be seen, at 1 atm the proposed product (IPA−water mixture)
fewer unit operations than the latter, which is achieved by
has a boiling point between 355.65 and 373.13 K, while the
reducing the flow rate of water fed to the RD (operating
main remaining component (propane) has a boiling point
parameter). Since less water is fed into the RD for the former, the
of 231.15 K. With such significant difference, even when the
overall utility consumption of {(R → LC&PRC): RD-ED-RE} is
pressure increases to 1770 kPa, the remaining component
expected to be lower than that of {(R → LC): RD-PRC-ED-RE}.
will not affect the proposed product. Simulation results also
Therefore, the latter is dropped. Similarly, comparing {(R →
confirm this.
PRC): PC-RD-LC-ED-RE} with {(R → LC&PRC): RD-ED-
The second integrated retrofit design, {(R → ED&RE):
RE}, the latter can be achieved when propylene conversion is
PC-RD} requires full consumption of water in the RD, which
100%, which has been validated through simulation. Therefore,
means propylene should be fed in excess. Since there is no
water left unreacted, the ED and regenerator columns are no {(R → PRC): PC-RD-LC-ED-RE} can also be eliminated. On
longer needed. Normal boiling points of both the proposed the other hand, the combination of {(R → PRC): PC-RD-LC-
product (IPA) and the remaining components (propylene and ED-RE} and {(R → ED&RE): PC-RD} could not generate any
propane) indicate this RD-based retrofit solution is feasible. new integrated solution, and so the final level two integrated
In fact, this type of design has already been studied.31,34 Note retrofit design is {(R → LC&PRC): RD-ED-RE}.
that, according to the proposed methodology, roots R → ED Next, level three integration, which combines level two
and R → RE should be considered separately. However, in this integration with a separate root retrofit, should be considered.
case study, the regenerator is dependent on the ED, which means According to the methodology, any level two retrofit solution
that regenerator is not needed if ED is not required. Therefore, will not be integrated with the root retrofit it was derived from.
the root is marked as R → ED&RE rather than R → ED. On As such, the only candidate is {(R → LC&PRC): RD-ED-RE}
the other hand, R → RE also leads to the same root, namely, integrated with {(R → ED&RE): PC-RD}. However, recall that
R → ED&RE and the same integrated retrofit design. the former requires excess water in the RD while the latter
All the root integrations involving the reactor have been needs excess propylene, and so it is inappropriate to combine
considered. The next step is to investigate root integrations them. Hence, there is no level three integration possible.
between the remaining contributing units. The lights column is Up to now, all the possible retrofit designs are explored. After
operating at a much higher pressure (902 kPa) than other screening by heuristics, process knowledge and simulation results,
contributing distillation columns (120−150 kPa), and is only two remain for final cost estimation and optimization.
therefore unlikely to be integrated with them. The remaining These two retrofit designs are {(R → ED&RE): PC-RD} (root
contributing units are preconcentrator, ED, and regenerator integration) and {(R → LC&PRC): RD-ED-RE} (level two
columns. Liang et al.35 have combined the preconcentrator and integration). Table 13 summarizes the number of theoretical
regenerator into a dividing wall column (DWC), which is then
followed by an ED column. However, DIPE is not considered Table 13. Theoretical Retrofit Designs and Screening
in their system; if DIPE is involved and not removed before item number
azeotropic distillation, it can form an azeotropic mixture theoretical number of root retrofit designs 10
with DMSO and might affect the purity of the IPA product number eliminated based on heuristics 1
collected from the top of ED column. Therefore, root PRC → number eliminated based on process knowledge 5
RE is not considered. Thus, all possible root integrations number eliminated based on process simulation 1
have been analyzed, and the feasible options are {(R → PRC): duplicated designs 0
PC-RD-LC-ED-RE}, {(R → LC): RD-PRC-ED-RE} and number of root designs remaining 3
{(R → ED&RE): PC-RD}. Thus, heuristics, guidelines for theoretical number of level two retrofit designs 3
using integrated units and process understanding, provide fast number eliminated based on heuristics 1
screening, removing infeasible roots in the early stage. The next number eliminated based on process knowledge 0
step is to explore a combination of two root integrations among number eliminated based on process simulation 0
the feasible ones. duplicated design 1
Level two integration is then considered. {(R → LC): number of level two retrofit designs remaining 1
RD-PRC-ED-RE} and {(R → ED&RE): PC-RD} are unlikely theoretical number of level three retrofit designs 1
to be combined, because the former requires excess water feed number eliminated based on heuristics 1
into the RD while the latter calls for excess propylene as the number of level three retrofit designs remaining 0
water will be fully consumed in the RD. {(R → PRC): PC-RD- number of total retrofit designs remaining 3+1+0=4
LC-ED-RE} and {(R → LC): RD-PRC-ED-RE} are possible to root retrofit designs eliminated due to level two integration 2
be combined, resulting in a level two integrated retrofit solution level two retrofit designs eliminated due to level three 0
as below: integration
number of retrofit designs remaining for optimization 4−2=2
{(R → LC&PRC): RD‐ED‐RE} and cost estimation

3624 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

solutions and candidates remaining after each step of screening. DIPE. The lights mixture (mostly propylene and propane) is
Note that the “root retrofit designs eliminated due to level two rectified into the top stream. After depressurizing to 905 kPa,
integration” refers to removal of {(R → PRC): PC-RD-LC-ED-RE} it is split into two streams (split ratio of 1.49:1), which are
and {(R → LC): RD-PRC-ED-RE} from the solution pool due respectively sent to propane column-1 and propane column-2
to the level two integrated retrofit solution {(R → LC&PRC): for recovering and recycling unreacted propylene. Two propane
RD-ED-RE}. columns are required in the retrofit scheme {(R → ED&RE):
3.4. Optimization and Cost Estimation of Promising PC-RD} for the following reasons. First, the original propane
Integrated Solutions. Two promising integrated retrofit column is proposed to be reused (to minimize capital
designs, namely, {(R → ED&RE): PC-RD} and {(R → investment for retrofitting). Second, propylene is in excess
LC&PRC): RD-ED-RE} are remaining. In the last step, in the retrofitted process, resulting in a higher flow rate of the
optimization and cost estimation are conducted for choosing propylene/propane mixture to the propane column (175.5 kmol/h
one of these designs. The complete process flow diagram of compare to 112.2 kmol/h in the original process). Third, the
{(R → ED&RE): PC-RD} is shown in Figure 7. Similar to the original propane column will not be able to handle this
original process, fresh propylene (400 kmol/h) and recycled increased flow rate. Therefore, a new propane column (propane
propylene are mixed and pressurized to 1755 kPa, and then column-2) is added while the original propane column is labeled
fed to stage 5 of the RD column; water (380.3 kmol/h) is as propane column-1 in the retrofitted process (Figure 7).
also pressurized to 1750 kPa, and fed to stage 3 of this column. The retrofitted process in Figure 7 is optimized to minimize
The water to propylene feed ratio in the RD column is 1:1.39. the manufacturing cost per unit product ($/kg); and the
Reactions occur on stages 3 to 5 (reaction zone), where the objective function should include the amortized capital invest-
temperature is between 126 and 133 °C to minimize generation ment ($/yr) as in eq 7. Decision variables for optimization are
of DIPE. IPA with a mole fraction of 0.9999 is collected from selected after analyzing the degrees of freedom for the relevant
the bottom of the RD column. Note that the IPA production unit operations as well as their effect on the objective function.
rate increases from 349.9 kmol/h to 379.5 kmol/h, mainly The chosen decision variables are feed flow rate of water (f water),
because RD-based retrofit design minimizes generation of reflux ratio of RD (RRD), distillate rate of propane column-1

Figure 7. RD-based integrated retrofit design of IPA process using propylene in excess; values shown correspond to the optimized case.

3625 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

Figure 8. RD-based integrated retrofit design of IPA process using water in excess; values shown correspond to the optimized case.

(DISpro‑1), distillate rate of propane column-2 (DISpro‑2), and split optimizer in Aspen Plus gave the following optimal values of
ratio (SP). Note that the reflux ratio of each propane column is decision variables: RD reflux ratio = 29.9, distillate rate of propane
used to achieve the purity of the top stream: 0.95 propylene and column-1 = 62.1 kmol/h, distillate rate of propane column-2 =
0.05 propane, and hence it is not selected as a decision variable. 92.5 kmol/h, and split ratio =1.49:1, and the value of the objective
Bounds on decision variables and process constraints are function is 0.096 $/kg.
RRD > 0 (14)
The complete process flow diagram of {(R → LC&PRC):
RD-ED-RE} is shown in Figure 8. Fresh propylene (400 kmol/h)
DISPro ‐ 1 ≤ 1.1 × DIS 0Pro (15) is pressurized to 1815 kPa, and after producing chilled water in
E1, the stream temperature rises to 12.8 °C; then it is heated to
0 ≤ DISPro ‐ 2 ≤ 175.5 (16) 48.9 °C in heat exchanger E2, and fed to stage 4 of the RD
column. Water (819.4 kmol/h) is also pressurized to 1804 kPa,
0 < SP < 1 (17) and fed to stage 2 of this column. The water to propylene
feed ratio in the RD column is 2.05:1. Reactions occur on stages
fPro ‐ 1 ≤ 1.1 × f 0Pro (18) 2 to 4 (reaction zone). The lights mixture with a flow rate of
In the above equations, f Pro‑1 and are the inlet flow rate of
f 0Pro 30.0 kmol/h is vented from the top; it has about 30 mol %
propane column-1 and of the propane column in the original propylene (i.e., 9 kmol/h) and the rest is mostly propane. Note
process, respectively. that the propylene vented from the RD column is only 2% of
Equation 14 ensures a positive value for the reflux ratio of fresh propylene to the process, and therefore there is no need to
RD, eq 15 provides the upper bound of the distillate rate for recover and recycle it, thus avoiding the expensive propylene−
propane column-1, which should not exceed 10% of its original propane splitter. The bottom stream of the RD is depressurized
value (DIS0Pro).36 Equation 16 specifies the upper bound on to 135 kPa and then split into two streams (split ratio = 2.57:1)
distillate rate for propane column-2; as it is a new unit for respectively sending to ED column-1 and ED column-2 for
operation, this bound should be reasonably large. Here, it is removing water from the azeotropic mixture. Similar to the
based on the inlet flow rate before splitting. Equation 17 refers reasoning for two propane columns in {(R → ED&RE):
to the split ratio, which should be between 0 and 1. Equation 18 PC-RD}, two ED columns are required because water is in
is an extra constraint for the inlet follow rate of propane excess in the retrofitted process, resulting in a higher flow rate
column-1. Equations 15 and 18 are related to the reused unit of the azeotropic mixture. Therefore, a new ED column (ED
operation (distillation column) from the original process. When column-2) is added while the original ED column is labeled as ED
the original distillation column needs to be reused, it should not column-1 in the retrofitted process (Figure 8). IPA with a mole
exceed the design margin of the original unit (usually 10%).36 fraction of 0.9999 (and total production rate of 370.3 kmol/h) is
Finally, similar to the optimization for solution without capital collected from the top of both ED column-1 and ED column-2,
investment, equality constraints, which include mass and energy while rich solvent from the bottom of these columns is sent to
balances for each unit operation in the process, are taken their respective regenerators (regenerator-1 and regenerator-2),
care by Aspen Plus simulation and its convergence. The built-in where the lean solvent (DMSO), after cooling, is returned to
3626 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

the respective ED column. Note that regenerator-1 is the original explanation for higher capital investment is that, when
regenerator, while regenerator-2 is the newly added unit. Water propylene is fed in excess, unreacted propylene has to be
from the distillate streams of both regenerators is not recycled evaporated and recycled from the top, resulting in a large
because it contains a trace amount of DMSO that could affect the diameter and high reboiler duty for the RD column. Moreover,
catalyst in the RD column. another propylene/propane column, which is large and
Similarly, the retrofitted process {(R → LC&PRC): RD-ED- expensive, is required for {(R → ED&RE): PC-RD} using
RE} is optimized to minimize the manufacturing cost per unit propylene in excess. On the other hand, {(R → LC&PRC):
product ($/kg, eq 7). As before, decision variables for optimiza- RD-ED-RE} using water in excess requires a smaller RD column
tion are selected after analyzing the degrees of freedom for the with lower reboiler duty. Therefore, although two extra
relevant unit operations as well as their effect on the objective columns are required for azeotropic separation of IPA and the
function. The chosen decision variables are reflux ratio of RD water mixture, the manufacturing cost of {(R → LC&PRC):
(RRD′), split ratio (SP′), distillate rate of regenerator-1 (DISreg‑1), RD-ED-RE} is lower than that of {(R → ED&RE): PC-RD}.
and distillate rate of regenerator-2 (DISreg‑2). Constraints are Hence, the former integrated solution involving an RD column
similar to those for the process {(R → ED&RE): PC-RD}. The with excess water is recommended for retrofitting the original
built-in optimizer in Aspen Plus gave the following optimal IPA process.
values for the decision variables: RD reflux ratio = 30.1, split Table 15 summarizes the status of unit operations for
ratio = 2.57:1, distillate rate of regenerator-1 = 322.7 kmol/h, both {(R → ED&RE): PC-RD} and {(R → LC&PRC):
and distillate rate of regenerator-2 = 125.5 kmol/h, and the
value of the objective function is 0.082 $/kg. Table 15. Status of the Columns for the RD-Based IPA
Note that in {(R → ED&RE): PC-RD}, less fresh water is Processes
required as propylene is fed in excess (6845.4 kg/h compared
to 11775.6 kg/h in the original process). Taking the cost of retrofitted process: retrofitted process:
process water as 0.064 $/1000 kg,26 the total manufacturing {(R→ED&RE): column {(R→LC&PRC): column
PC-RD} status RD-ED-RE} status
cost decreases by 2524 $/yr due to less fresh water being
RD column new RD column new
required. Similarly, in {(R → LC&PRC): RD-ED-RE}, extra
propane column-1 existing ED column-1 and existing
fresh water is required (14749.2 kg/h compared to 11775.6 kg/h unit regenerator-1 units
in the original process), which increases the total manufacturing propane column-2 new ED column-2 and new
cost by 1594 $/yr. Both these values are included in the cost regenerator-2
estimation (Table 14).
RD-ED-RE} processes. As can be seen, the number of columns
Table 14. Cost Estimation of the RD-Based IPA Processes in {(R → ED&RE): PC-RD} is only three compared to seven
{(R→ED&RE): {(R→LC&PRC):
in the original process in Figure 8 (excluding pumps and
item PC-RD} RD-ED-RE} heat exchangers); on the other hand, the number of columns
capital investment for added unit 14,959,577 7,303,771 in {(R → LC&PRC): RD-ED-RE} is five. In both retrofitted
operations ($) processes, existing LC, PC, and ether column are not used.
cost of utilities ($/yr) 8,652,000 7,529,200 Hence, plot space should not be a limitation in retrofitting the
cost of operations ($/yr) 1,362,987 2,243,867 existing IPA process; however, this may require shutdown of
cost of maintenance ($/yr) 1,501,166 1,008,595 the existing process for several months, and so requires advanced
cost of operating overhead ($/yr) 472,336 602,182 planning and execution.
property taxes and insurance ($/yr) 326,340 219,260 3.6. Control of RD-Based IPA Processes. The RD
cost of depreciation ($/yr) 1,305,362 877,039 column is involved in both the base case and retrofit designs.
general expense ($/yr) 1,306,716 1,275,213 Compared to the conventional unit operations, it is more
cost of water reduced/increased ($/yr) −2,524 1,594 complicated and difficult to control. Even then, it has found a
Summary number of industrial applications. Further, there have been
total manufacturing cost ($/yr) 14,924,384 13,758,544 many studies on design and control of RD. See the book by
amortized capital investment ($/yr) 2,558,088 1,248,945 Luyben and Yu33 for details on these. In particular, Wang and
annual IPA production (kg/yr) 182,438,598 178,040,240 Wong34 investigated dynamics and control of a RD column for
manufacturing cost per unit product 0.096 0.084 IPA production, through rigorous simulations. They discussed
($/kg) multiplicities and nonlinearities involved, and then used variable
transformation to overcome the strong nonlinearity in the
3.5. Solution Comparison. Cost estimation of the two temperature−composition loops of the RD column for IPA
integrated retrofit solutions, after optimization, is summarized production. However, Wang and Wong34 considered only the
in Table 14, where {(R → ED&RE): PC-RD} uses propylene RD column with pure propylene feed, thus not requiring the
in excess and {(R → LC&PRC): RD-ED-RE} requires water in propylene−propane splitter. Hence, their study needs to be
excess. Both are better than the improved solution without extended to investigate dynamics and control of realistic and
capital investment (with a manufacturing cost of 0.103 $/kg, complete RD-based IPA processes.
which is 5.5% lower than that of the original process in In general, not all steady-state designs lead to controllable
Table 10). The integrated solution, {(R → ED&RE): PC-RD} processes and so control studies should be performed on one
achieves 11.9% savings in manufacturing cost of the original or more promising retrofit solution(s) before implementation.
process, whereas the integrated process {(R → LC&PRC): Usually, these are performed after steady-state design and
RD-ED-RE} achieves 22.9% savings. The capital investment analysis. Accordingly, the present work is focused on steady
for added unit operations and the cost of utilities makes the state design. Further studies are required to include control
former more costly than the latter (Table 14). One possible strategy into the proposed retrofitting methodology.
3627 DOI: 10.1021/acs.iecr.5b02707
Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

4. CONCLUSIONS (11) Kralj, A. K.; Glavic, P. Retrofit of complex and energy intensive
processes. Comput. Chem. Eng. 1997, 21 (Suppl), 517−522.
In this study, a heuristic-based, systematic method is developed (12) Kralj, A. K.; Glavic, P.; Kravanja, Z. Retrofit of complex and
for process retrofitting via process intensification/integration. energy intensive processes II: stepwise simultaneous superstructural
It consists of four main steps: (1) base case analysis, retrofit approach. Comput. Chem. Eng. 2000, 24, 517−522.
target, and contributing unit operations specification; (2) (13) Uerdingen, E.; Fischer, U.; Hungerbuhler, K.; Gani, R.
generation of improved solution without capital investment; Screening for profitable retrofit options of chemical processes: a new
(3) generation of integrated solutions; and (4) optimization method. AIChE J. 2003, 49 (9), 2400−2418.
and cost estimation of integrated solutions. This methodology (14) Jensen, N.; Coll, N.; Gani, R. An integrated computer-aided
is focused to achieve integrated retrofit solutions. It is illustrated system for generation and evaluation of sustainable process
for retrofitting a conventional IPA process. It leads to two alternatives. Clean Technol. Environ. Policy 2003, 5, 209−225.
integrated retrofit solutions, namely, {(R → ED&RE): PC-RD} (15) Uerdingen, E.; Fischer, U.; Gani, R.; Hungerbuhler, K. A new
and {(R → LC&PRC): RD-ED-RE} involving a RD column retrofit design methodology for identifying, developing, and evaluating
with propylene in excess and a RD column with water in excess, retrofit projects for cost-efficiency improvements in continuous
chemical processes. Ind. Eng. Chem. Res. 2005, 44, 1842−1853.
respectively. The former reduces the number of columns
(16) Carvalho, A.; Gani, R.; Matos, H. Design of sustainable chemical
from 7 in the original process to 3 and also the manufacturing processes: Systematic retrofit analysis generation and evaluation of
cost from 0.109 $/kg to 0.096 $/kg, whereas the latter has alternatives. Process Saf. Environ. Prot. 2008, 86, 328−346.
5 columns and a manufacturing cost of 0.084 $/kg. Hence, the (17) Simon, L. L.; Osterwalder, N.; Fischer, U.; Hungerbuhler, K.
integrated retrofit solution {(R → LC&PRC): RD-ED-RE} is Systematic retrofit method for chemical batch processes using
the recommended design. indicators, heuristics, and process models. Ind. Eng. Chem. Res. 2008,

■ AUTHOR INFORMATION
Corresponding Author
47, 66−80.
(18) Reay, D., Ramshaw, C. and Harvey, A. Process Intensification:
Engineering for Efficiency, Sustainability and Flexibility; Butterworth-
Heinemann: Oxford, 2013.
*E-mail: [email protected]. Tel.: (65) 6516 2187. (19) Creative Energy Energy transition. European Roadmap for
Notes Process Intensification; 2007: p 804−806; www.efce.info/efce_media
The authors declare no competing financial interest. (accessed May 5, 2015).

■
(20) Stankiewicz, A. and Moulijn, J. A. Re-engineering the Chemical
ACKNOWLEDGMENTS Processing Plant: Process Intensification; Marcel Dekker: New York,
2004.
M.W.N. thanks the National University of Singapore for (21) Seider, W. D., Seader, J. D., Lewin, D. R.; Widagdo, S. Product
supporting his doctoral research. Both authors acknowledge and Process Design Principles: Synthesis, Analysis and Evaluation, 3rd ed.;
Yonghui Lin for the initial study on reactive distillation options John Wiley & Sons: 2010.
for the IPA process. (22) Douglas, J. M. Cost diagrams and the quick screening of process

■
alternatives. Ind. Eng. Chem. Process Des. Dev. 1985, 24, 970−976.
REFERENCES (23) Grossmann I. E., Westerberg A. W., Biegler L. T. Retrofit design
of processes, CMU Research Showcase; 1987, http://repository.cmu.
(1) Sreepathi, B. K.; Rangaiah, G. P. Review of Heat Exchanger edu/cgi/viewcontent.cgi?article=1109&context=cheme (accessed May
Network Retrofitting Methodologies and Their Applications. Ind. Eng. 5, 2015).
Chem. Res. 2014, 53 (28), 11205−11220. (24) Luyben, W. L. and Yu, C. C.. Reactive Distillation Design and
(2) Premkumar, R.; Rangaiah, G. P. Retrofitting Conventional Control; John Wiley & Sons: 2008.
Column Systems to Dividing Wall Columns. Chem. Eng. Res. Des. (25) Lutze, P.; Roman-Martinez, A. R.; Woodley, J. M.; Gani, R. A
2009, 87, 47−60. systematic synthesis and design methodology to achieve process
(3) Nguyuen, N.; Demirel, Y. Retrofit of distillation columns in
intensification in (bio) chemical processes. Comput. Chem. Eng. 2012,
biodiesel production plants. Energy 2010, 35, 1625−1632.
36, 189−207.
(4) Van Duc Long, N.; Lee, S.; Lee, M. Y. Design and optimization of
(26) Turton, R., Bailie, R. C., Whiting, W. B., Shaeiwitz, J. A.,
a dividing wall column for debottlenecking of the acetic acid
Bhattcharyya, D.. Analysis, Synthesis and Design of Chemical Processes,
purification process. Chem. Eng. Process. 2010, 49, 825−835.
(5) Van Duc Long, N.; Lee, M. Y. Optimal retrofit design of 4th ed.; Prentice Hall PTR: NJ, 2012.
extractive distillation to energy efficient thermally coupled distillation (27) Dutia, P. Isopropyl alcohol: A techno-commercial profile. Chem.
scheme. AIChE J. 2013, 59 (4), 1175−1182. Weekly 2012, April 24, 211 http://www.chemicalweekly.com/Profiles/
(6) Park, C.; Song, K.; Lee, S.; Lim, Y.; Han, C. Retrofit design of a Isopropyl_Alcohol.pdf.
boil-off gas handling process in liquefied natural gas receiving (28) Kroschwitz, J. I. Isopropyl Alcohol. In Kirk-Othmer Encyclopedia
terminals. Energy 2012, 44, 69−78. of Chemical Technology, 5th ed.; Wiley-Interscience: New York, 2004.
(7) Motelica, A.; Bruinsma, O. S. L.; Kreiter, R.; den Exter, M.; (29) American Chemistry Council 2007. Propylene: Production
Vente, J. F. Membrane retrofit option for paraffin/olefin separation: A Stewardship Guidance Manual, http://www.lyondellbasell.com/
technoeconomic evaluation. Ind. Eng. Chem. Res. 2012, 51, 6977− techlit/techlit/3303.pdf (accessed May 2, 2015).
6986. (30) Luyben, W. L.; Chien, I. L. Design and Control of Distillation
(8) Ploegmakers, J.; Jelsma, A. R. T.; van der Ham, A. G. J.; Systems for Separating Azeotropes; John Wiley & Sons: 2011.
Nijmeijer, K. Economic evaluation of membrane potential for (31) Xu, Y.; Chuang, K. T.; Sanger, A. R. Design of A Process for
ethylene/ethane separation in a retrofitted hybrid membrane- Production of Isopropyl Alcohol by Hydration of Propylene in A
distillation plant using Unisim design. Ind. Eng. Chem. Res. 2013, 52, Catalytic Distillation Column. Trans IChemE 2002, 80 (A), 686−694.
6524−6539. (32) Arifin, S.; Chien, I. L. Design and Control of an Isopropyl
(9) Fisher, W. R.; Doherty, M. F.; Douglas, J. M. Screening of process Alcohol Dehydration Process via Extractive Distillation Using
retrofit alternatives. Ind. Eng. Chem. Res. 1987, 26, 2195−2204. Dminethyl Sulfoxide as an Entrainer. Ind. Eng. Chem. Res. 2008, 47,
(10) Dantus, M. M.; High, K. A. Economic evaluation for the retrofit 790−803.
of chemical processes through waste minimization and process (33) Luyben, W. L.; Yu, C. C.. Reactive Distillation Design and
integration. Ind. Eng. Chem. Res. 1996, 35, 4566−4578. Control; John Wiley & Sons: 2008.

3628 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629
Industrial & Engineering Chemistry Research Article

(34) Wang, S. J.; Wong, D. S. H. Control of reactive distillation

production of high-purity isopropanol. J. Process Control 2006, 16,
385−394.
(35) Liang, K.; Li, W.; Luo, H.; Xia, M.; Xu, C. Energy-Efficient
Extractive Distillation Process by Combining Preconcentration
Column and Entrainer Recovery Column. Ind. Eng. Chem. Res. 2014,
53, 7121−7131.
(36) Towler, G.; Sinnott, R. Chemical Engineering Design: Principles,
Practice and Economics of Plant and Process Design; Butterworth-
Heinemann: Oxford; Waltham, MA, 2013.

3629 DOI: 10.1021/acs.iecr.5b02707

Ind. Eng. Chem. Res. 2016, 55, 3614−3629