Computers and Chemical Engineering 84 (2016) 546–557

Contents lists available at ScienceDirect

Computers and Chemical Engineering

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

A decomposition methodology for dynamic modeling of cold box in

offshore natural gas liquefaction process
Chul-Jin Lee a , Kiwook Song b , Yongseok Lee b , Chonghun Han b,∗
a
Engineering Development Research Center, Seoul National University, Gwanak-ro, Gwanak-gu, Seoul 151-744, Republic of Korea
b
School of Chemical and Biological Engineering, Seoul National University, Gwanak-ro, Gwanak-gu, Seoul 151-744, Republic of Korea

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

Article history: Natural gas liquefaction process using mixed and/or cascade refrigerant is popular in onshore LNG (liq-
Received 29 December 2014 uefied natural gas) plant. Similar attempt has been adopted for FLNG (floating LNG) but still needed for
Received in revised form the improvement of the process to enhance its efficiency as well as reliability. The dynamic modeling of
27 September 2015
cold box which is a core equipment in LNG/FLNG plant enabling to liquefy natural gas is crucial in order
Accepted 28 September 2015
to develop or improve a liquefaction process concerning operability and controllability. A decomposition
Available online 9 October 2015
methodology for dynamic modeling of cold box in the case of lack of internal design data at early design
stage is presented. The proposed methodology is validated through the industrial application of offshore
Keywords:
Cold box
natural gas liquefaction process and expected to be extensively applied to the various process designs
LNG which require dynamic simulation of cold box unit.
FLNG © 2015 Elsevier Ltd. All rights reserved.
Liquefaction
Dynamic simulation
Decomposition approach

1. Introduction liquefaction processes. PFHEs usually have a higher heat transfer

Natural gas is the fastest growing energy source in the world the
consumption of which is expected to increase at an average rate of
1.4–1.6% per year from 2008 to 2035, owing to its lower environ-
mental impact compared to other fossil fuels (Gruenspecht, 2010).
With this growing demand of natural gas, a number of researches
have been investigated on the perspectives of how to reduce the
production cost and improve the efficiency of the liquefaction pro-
cesses which is the core part in the LNG value chain (Kikkawa et al.,
1997; Alabdulkarem et al., 2011; Roberts and Brostow, 2014; Minta
et al., 2008).
The natural gas liquefaction processes using mixed refrigerants
(MR) utilize the plate-fin type heat exchangers (PFHEs), sometimes
called the cold box, to associate all the streams in the heat exchange
into a single piece of equipment. This greatly improves the effi-
ciency and lowers capital cost, the number of equipment, space
requirement, etc.
The greatest usefulness of PFHEs exhibited in handling more
than two streams, usually up to 12 streams. Plate-fin, coil-wound,
and multi-pass shell-and-tube types are all multi-stream heat
exchangers extensively used in cryogenics, gas separation, and
∗ Corresponding author. Tel.: +82 2 880 1887; fax: +82 2 880 1887.
E-mail address: chhan@snu.ac.kr (C. Han).
area to volume ratio (300–1000 m2 /m3 ) compared to coil-wound
heat exchangers (50–150 m2 /m3 ). The high degrees of compactness
and flexibility in stream arrangements are their main advantages
(Pacio and Dorao, 2011).
Most of the research area dealing with LNG liquefaction pro-
cess and cold box consists of the evaluation of the performance
of the process focusing on the thermodynamic efficiency using
exergy minimization technologies (Remeljej and Hoadley, 2006;
Kanoğlu, 2002), and the optimization of the process design and
operation through rigorous simulations and algorithms, and plant
performance improvement by process integration and equipment
enhancements (Lim et al., 2013; Skaugen et al., 2014).
Despite the existence of prior work, very few studies have
focused on the dynamic modeling and simulation of the natural
gas liquefaction process. Particularly for offshore plants, the pro-
cess design has to be performed in a dynamic manner owing to
their highly dependent nature on the operational environments
like fluctuating disturbances, which are different from those for
onshore plants (Song et al., 2012; Peric et al., 2009).
Dynamic modeling and simulation of heat exchangers has been
mainly focused in the viewpoint of dynamic behaviors as well as
design aspects and. Sharifi et al. (1995) have studied to develop
a dynamic simulation of plate fin heat exchangers, but coupling
of the flow and temperature which shows the system response to
flow rate change still remains. Pingaud et al. (1989) and Averous

http://dx.doi.org/10.1016/j.compchemeng.2015.09.020
0098-1354/© 2015 Elsevier Ltd. All rights reserved.
 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557
Nomenclature
FLNG floating LNG
JT Joule–Thomson    equipment function in accordance to expectations. The noble
k pressure flow coefficient kg/h/ kg/m3 kPa
LNG liquefied natural gas
MR mixed refrigerant
mref mass, flow of rate of streams in original steady state
design (ton/h)
msim mass, flow of rate of streams in dynamic simulation
(ton/h)
PFD process flow diagram
PFHE plate-fin type heat exchanger
PI proportional-integral
PRSV Peng–Robinson–Stryjek-Vera
Tref temperature of streams in original steady state
design ( ◦ C)
Tsim temperature of streams in dynamic simulation ( ◦ C)
UA heat transfer coefficient (kg/ ◦ C h)
Xref temperature, flow rate of values of streams in orig-
inal steady state design
Xsim temperature, flow rate of values of streams in
dynamic simulation
WEDT warm end delta temperature
et al. (1999) emphasized the importance of dynamic studies in com-
plement with steady-state ones for plate fin exchangers. Luo et al.
(2003) developed own dynamic simulator for PHTE. Singh and Hovd
(2007) studied the effect of the simplification of the heat exchanger
model on systematic control structure design in the PRICO process
in order to develop a dynamic model, illustrating the effect of the
operating strategy (maximum production and given production)
on the number of unconstrained degrees of freedom. Michelsen
et al. (2010) developed a dynamic, control relevant, and mechanis-
tic model for operability analysis of the TEALARC process. Although
the thermodynamics of this model are simplified, it has sufficient
complexity for both steady-state and dynamic operability analyses.
Applications to other processes for cold box with controllability
have been studied (Shin and Lee, 2009; Mandler, 2000; Boehme
et al., 2003).
Even though all of these relevant studies deal with the operabi-
lity and dynamic performance of the liquefaction processes through
control modeling based on the structure design procedure, any sys-
tematic research on the general methodology of the internal design
of the LNG multi-stream heat exchanger is yet to be developed
in accordance to the dynamic behavior of a fluid like LNG. Fur-
thermore, the refrigerants participating in the multi-stream heat
exchanger have not been obviously interpreted or evaluated.
The internal design of the PFHEs, mightily important for the
dynamic modeling of the liquefaction process, has an impact on the
thermodynamic and hydrodynamic behavior of the fluids passing
through the plate-fins where complicated heat transfer and phase
changes occur. It aims to match the results of the dynamic model-
ing calculated sequentially through several discrete grids of each
PFHE to those of the steady-state modeling using an error mini-
mization feedback algorithm. The main advantage of the proposed
decomposition methodology is that it can handle the coupling
between the flow rate and temperature effectively. When the
parameter regarding flow changes to obtain the target flow rate,
the temperature also changes due to the flow variation and vice
versa. The proposed methodology enables to handle the coupling
highly correlated between the flow rate and temperature using the
decomposition algorithm.
547
In this research, the decomposition methodology for dynamic
modeling of cold box is suggested. When the basic design in the
static mode is completed, an engineering study of the dynamic
behavior must be performed to verify that the process and
methodology is validated with the application of offshore natural
gas liquefaction process.
2. Theory
2.1. Dynamic modeling of a PFHE
Proper equipment sizing, specifying the resistance of unit oper-
ations, and adding accurate boundary conditions for the entire
flowsheet, are critical for the dynamic simulation of the plant.
Dynamic behaviors of streams are affected by the volume of the
equipment and hence the geometry information is essential. Rat-
ing of a multi-stream PFHE (i) the length and the width of the
exchanger, (ii) the layer configuration, (iii) fin types and properties,
(iv) zone configuration, and (v) the heat transfer configuration.
In many cases, dynamic simulation is used to check on whether
the plant is easy, safe, and cost-efficient to operate, and on whether
the equipment functions as expected and in accordance to the
design phase of the flowsheet. Therefore, it is common that the
exact internal design of the equipment is not prepared a priori and is
further finalized according to the simulation results. In these cases,
one can use simple exchanger rating tools, i.e. ASPEN MUSE, for
preliminary assumption of the geometry.
2.2. Layer configuration
The arrangement of streams in the exchanger affects the total
heat load distribution. A single pattern of layers repeated over the
height of the exchanger block is defined as a set. For optimal layer
configuration, single banking of cold and hot layers, with a counter-
current flow pattern, and with repeated sets, is assumed. For each
layer, fin properties, such as the fin height, thickness, density and
others, should be specified.
2.3. Zone configuration
A zone is a partitioned length of the exchanger. Each zone fea-
tures a stacking pattern with one feed and one product connected
to each representative layer in the pattern. It is recommended to
specify 10 or more heat zones to remove the wiggle temperature
profiles. Zone metal properties such as thermal conductivity, and
specific heat capacity should also be specified.
In a pressure–flow dynamic model solver, the flow rate in the
heat exchanger is calculated using resistance equations. The resis-
tance equation modeled based on the equation of turbulent flow
is

F =k  · p (1)

where k = pressure flow coefficient, kg/h kg/m3 kPaF is the flow
rate through each zone in the heat exchanger and p is pressure
difference in the same section, which has a role as driving force
of flow. With pre-specified k values, the resistance equation calcu-
lates the flow rates from the pressure differences of the surrounding
nodes.
For a heat exchanger consisting of n layers with m zones in a
set, n × m k values should be specified. However, the temperature
and pressure drops for each zone are not known, thus exact k val-
ues cannot be specified. For simplification, we propose to fix the k
values to a single value for each layer in every zone. This means
that a stream in a layer has equal k values for all m zones, i.e.
548 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557

kn1 = kn2 = · · · = knm = kn for all n, where knm is the k value for a layer
n and a zone m. Even though the exact k value of each zone in the
same layer may differ due to the phase change occurring in the
channel inside the heat exchanger, the final k value obtained after
parameter estimation should meet the target flow rate.
The initial value of kn for layer n is estimated and then manipu-
lated by a parameter estimation technique. The choice of a proper
initial point is crucial. Therefore, we propose herein to estimate
the initial k value with the density at the mean temperature of the
inlet and outlet temperatures, and the pressure drop as the total
pressure drop divided by the number of zones, m.

2.4. Heat transfer configuration

The heat transfer coefficient, specified UA value, is used to calcu-

late the convective heat transfer between the stream and the metal
that surrounds it. The overall duty of each stream is influenced by
the presence of metal fins, fin efficiencies, direct heat flow between
metal layers and other factors.
For a heat exchanger consisting of n layers with m zones in a set,
n × m UA values should be specified. In this study, we simplify the
problem by fixing equal UA values for each layer in every zone. This
means that a stream in a layer has equal UA values for all m zones,
i.e., that UAn1 = UAn2 = · · · = UAnm = UAn for all n, where UAnm is the
UA value for each layer n and zone m. Like the k value estimation Fig. 1. Decomposition algorithm to estimate k and UA values in cold box.
step described in Section 2.3, this simplification can also be applied
even though different liquid fraction in each zone caused by phase
external optimization solvers, it is suggested to use a PI controller
change would result in the variation of UA values.
in HYSYS to change the k values. Set the set point to the PFD flow
It is also difficult to calculate exactly the UAn value for layer n.
rate value and assign the k values of each layer as the output
This value must be found using trial-and-error parameter estima-
object. It is convenient to use spreadsheets to give equal k values
tion technique. Initial values can be estimated from the correlation
for each zone in each layer. Secondly, adjust the UA values until
value commonly available in the literature (Neeraas et al., 2004) or
the temperature values equal those of the PFD. Then, the flow
from steady-state UA values.
rates would change since the density is a function of temperature.
We now go back to the first step to adjust k values again. The
2.5. Update k values and UA values
iteration stops when the flow rate and temperature values of every
stream are equivalent to the reference PFD values. In this manner,
Based on the initial estimates of k and UA values, the dynamic
minimization step in Eq. (2) is done separately for each variable of
model of the heat exchanger will yield a different performance
temperature and flow rate through the proposed decomposition
compared to the original design. The values should be updated
methodology. The overall algorithm is depicted in Fig. 1.
until the temperature, flow rate, conditions of streams around the
exchanger are equal to the values of the process flow diagram. This
3. Modeling and simulation
is a typical parameter estimation problem. It can be formulated as
follows:
 3.1. Process flow diagram of cold box
Min z = (Xref − Xsim )2 (2)
Fig. 2 shows the process flow diagram of natural gas liquefaction
where Xref = temperature, flow of rate values of streams is original process using mixed refrigerant (Lee, 2012). The heat and material
steady state design andXsim = temperature, flow of rate values of balance are based on in-house data. Natural gas at −30 ◦ C, pre-
streams is original steady state design. cooled by propane, is further cooled and liquefied by the MR in
It is an arduous work to handle n variables consisting of k values, the cold box (Core 1, Core 2 and Core 3). The final temperature of
and n variables consisting of UA values, simultaneously because the natural gas is −147.8 ◦ C. The cold box consists of three cores,
their effects on stream conditions are highly correlated. Pressure, and each is consisted of a separate PFHEs. MR that is also precooled
temperature, and flow rate of streams are represented by equa- by propane is flashed in the MR separator, further cooled and liq-
tions that must be obeyed. k and UA values are variables that can uefied in the cold box. Light MR (MR2–MR3–MR4–MR5), chilled to
be adjusted. In order to satisfy the degree of freedom, we set the −147.8 ◦ C, undergoes isenthalpic expansion in the JT valve (JT2)
pressure specification to all boundary streams of each core to be and is then used as the cold stream to cool hot natural gas and MR.
constant. Thus the objective is to search for the k and UA val- Heavy MR (MR8–MR9–MR10) is chilled to −128 ◦ C and expanded
ues, while the outlet stream temperature and flow rate values are to be utilized as the cold stream together with MR7.
equal to those set in the PFD. Dynamic simulation of each core is ASPEN HYSYS (DYNAMICS) v 7.2 is used for steady-state and
performed separately, as the main concept of parameter tuning is dynamic modeling using PRSV as the property package. Stream
adjusting the simulation values to the PFD values of inlet and outlet conditions of the flowsheet are shown in Table 1.
stream for each core.
In this study, the decomposition approach was applied to the 3.2. Dynamic modeling of Core 1
dynamic modeling of the cold box by two steps, which adjusted
k value and UA values step by step. At first, adjust the k values Core 1 exchanges heat between one cold stream and three hot
until the flow rate values equal those of the PFD. Other than using streams. Without internal design data of the PFHE, ASPEN MUSE is
 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557 549

Fig. 2. Process flow diagram of the liquefaction process.

used to estimate the exchanger geometry and the number of layers.

Core 1 has dimensions of 2600 mm × 600 mm (length × width) with
800 layers. The pattern of ADADADBDCDCDCDCD(A = NG1–NG2,
B = MR2–MR3, C = MR8–MR9, D = MR13–MR14) is configured as a set,
hence with 16 layers in a set, and 50 sets in total. Cold and hot lay-
ers flow in counter directions. The total length of the exchanger is
partitioned to 10 zones (Fig. 3).
If modeled correctly, the dynamic model of the cold box, before
any simulation is completed, should yield equal stream conditions
as the PFD. Here, we assume that equal k and UA values are used
for layers with the same feed stream. In addition, equal k and UA
values are used for all zones in a layer. Therefore, four k values and
four UA values must be specified for the dynamic modeling of Core
1. The variables are defined as ki and UAi for (i = A, B, C, D).

3.2.1. Step 1: using initial estimates of k and UA values

First, we assume equal pressure drops for all zones in Core 1.
Since the total pressure drop is 0.3 bar, the pressure drop per zone is
0.03 bar. Then, we calculate the density of the stream at the average
temperature of the inlet and outlet conditions. Finally, the k value
is calculated with the flow rate of the stream in accordance to the
PFD value, using Eq. (1). UA values of 8 × 104 kJ/◦ C h are used as the
initial estimated points for all streams. The initial estimates of k and Fig. 3. Exchanger geometry and layer configuration of Core 1.
UA values are shown in Table 2, Step 1.

Table 1
Heat and material balance.

Stream NG1 NG2 NG3 NG4 NG5

Vapor fraction 0.980 0.336 0.0000 0.0000 0.088

Pressure (bar) 61.0 60.7 60.4 60.1 1.2
Temperature (◦ C) −30.0 −64.0 −128.0 −147.8 −158.3
Mass flow (ton/h) 118.0 118.0 118.0 118.0 118.0

Stream MR1 MR2 MR3 MR4 MR5

Vapor fraction 0.215 1.000 0.460 0.000 0.000

Pressure (bar) 64.0 64.0 63.7 63.4 63.1
Temperature (◦ C) −30.0 −30.0 −64.0 −128.0 −147.8
Mass flow (ton/h) 227.5 40.4 40.4 40.4 40.4

Stream MR6 MR7 MR8 MR9 MR10

Vapor fraction 0.118 0.651 0.000 0.000 0.000

Pressure (bar) 4.5 4.2 64.0 63.7 63.4
Temperature (◦ C) −155.8 −132.7 −30.0 −64.0 −128.0
Mass flow (ton/h) 40.4 40.4 187.1 187.1 187.1

Stream MR11 MR12 MR13 MR14

Vapor fraction 0.097 0.209 0.728 1.000

Pressure (bar) 4.2 4.2 3.9 3.6
Temperature (◦ C) −134.3 −133.0 −74.3 −33.1
Mass flow (ton/h) 187.1 227.5 227.5 227.5
550 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557

Table 2
k and UA values at each step for Core 1.

Stream Variable Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8

A (NG1-NG2) kA 58.53 66.98 64.41 64.46 64.36

UAA 80.00 708.00 710.92 723.65 725.13
B (MR2-MR3) kB 18.83 20.34 20.15 20.18 20.20
UAB 80.00 199.00 188.36 181.36 176.22
C (MR8-MR9) kC 52.96 52.92 53.12 53.31 53.37
UAC 80.00 87.42 83.57 77.29 76.50
D (MR13-MR14) kD 508.01 503.36 516.68 517.07 517.54
UAD 80.00 96.00 109.36 113.68 114.82

Units: k [1000 kg/h/ bar kg/m3 ], UA (MJ/◦ C h).

3.2.2. Adjusting k and UA values

Based on the initial guesses for the k and UA values, the stream
conditions are far away from the design point. As proposed in Sec-
tion 2, k values are updated first to meet the flow rate conditions
at the outlet streams without varying the UA values. This is Step 2.
Then, the UA values are updated to meet the temperature condi-
tions at the outlet streams without varying the k values (Step 3).
Because the density varies with temperature, the flow rate would
change at this point even if the k values are constant as those of Step
2 (remember that the flow rate is a function of the k value, density
and pressure drop). Thus, we go back to adjusting k values. Above
process is repeated until the relative differences between the PFD
values and the simulation values are under the specified tolerance
value (ε) of 0.01 (Steps 4–7). At Step 7, the resultant relative dif-
ference between the temperature and the flow rate values of the
PFD and the dynamic model is less than 1%. Therefore, at Step 8,
we adjust both the k and the UA values simultaneously. For each
step, it converged to the target flow rate and temperature within a
minute.

3.3. Dynamic modeling of Core 2

Core 2 exchanges heat between one cold stream and three hot Fig. 4. Exchanger geometry and layer configuration of Core 2.
streams. Without internal design data of the PFHE, ASPEN MUSE
is used to estimate the exchanger geometry and number of layers. must be specified for dynamic modeling of Core 3. The variables are
Core 2 has dimensions of 2600 mm × 600 mm (length × width) with defined as ki and UAi for i = I, J, K.
800 layers. The pattern of EHEHEHFHGHGHGHGH (E = NG2–NG3,
F = MR3–MR4, G = MR9–MR10, H = MR12–MR13) is configured as a
4. Results and discussion
set, hence with 16 layers in a set, and 50 sets in total. Cold and hot
layers flow in counter directions. The total length of the exchanger
4.1. Dynamic models of each core
is partitioned into 10 zones (Fig. 4).
Here, we also assume that equal k and UA values are used for
4.1.1. Core 1
layers with the same feed stream. Equal k and UA values are used
The k and UA values at each step are shown in Table 2. At Step
for all zones in a layer too. So, four k values and 4 UA values must be
1, the mass flow rate values of streams A and B are lower than the
specified for dynamic modeling of Core 2. The variables are defined
PFD values, while those of stream C and D are greater than the PFD
as ki and UAi for i = E, F, G, H.

3.4. Dynamic modeling of Core 3

Core 3 exchanges heat between one cold stream and two hot
streams. Without internal design data of the PFHE, ASPEN MUSE is
used to estimate the exchanger geometry and the number of lay-
ers. Core 2 has dimensions of 1640 mm × 600 mm (length × width)
with 120 layers. The pattern of IKIKJK (I = NG3–NG4, J = MR4–MR5,
K = MR6–MR7) is configured as a set, hence with six layers in a set,
and 20 sets in total. Cold and hot layers flow in counter direc-
tions. The total length of the exchanger is partitioned into 10 zones
(Fig. 5).
Here, we also assume that equal k and UA values are used for
layers with the same feed stream. Equal k and UA values are used
for all zones in a layer too. So, three k values and three UA values Fig. 5. Exchanger geometry and layer configuration of Core 3.
 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557 551

Fig. 6. Flow rate and temperature conditions at each step for Core1.

values. Therefore, we can say that k values for streams A and B
should be higher, and those for C and D should be lower.
At Step 2, the temperature of the hot streams A and B are higher
than the PFD values. This means that the UA values should be higher
for these streams, hence more heat duty provided to lower the
outlet temperature. Meanwhile, the temperature of the cold outlet
stream D is lower than the PFD value. Thus, UAD should be higher,
providing more heat duty to the hot streams. Then, the outlet tem-
perature at stream MR14 is lowered (Fig. 6 and Table 3).
4.1.2. Core 2 and Core 3
In the case of Core 2, we can see that the initial k and UA val-
ues were poorly given. It is not easy to estimate the k value of the
cold stream in particular, because the cold MR is vaporizing in the
552 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557

Table 3
Flowrate and temperature conditions at each step for Core 1.

Stream Conditions Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8

NG2 (A) Flowrate 104.6 117.9 121.1 117.9 118.1 117.7 118.2 118.0
Temperature −60.4 −59.3 −63.2 −63.6 −63.9 −63.9 −64.0 −64.0
MR3 (B) Flowrate 37.6 40.3 40.6 40.5 40.3 40.3 40.5 40.4
Temperature −62.1 −61.0 −63.4 −63.9 −64.1 −64.1 −64.0 −64.0
MR9 (C) Flowrate 187.6 187.2 185.4 187.0 186.4 187.1 186.9 187.1
Temperature −65.6 −65.1 −63.9 −64.3 −64.1 −64.4 −64.0 −64.0
MR14 (D) Flowrate 230.8 227.7 221.1 227.5 226.3 227.6 227.4 227.5
Temperature −53.0 −52.4 −33.1 −35.1 −33.1 −33.4 −33.1 −33.1

Units: flowrate (ton/h), temperature (◦ C).

Table 4
k and UA values at each step for Core 2.

Stream Variable Step 1 Step 2 Step 3 Step 4 Step 5 Step 6

E (NG2-NG3) kE 39.07 40.88 39.01 38.76

UAE 80.00 Max 2184 2194
F (MR3-MR4) kF 12.63 13.46 12.99 12.88
UAF 80.00 Max 1104 1110
G (MR9-MR10) kF 47.98 48.49 48.50 48.36
UAF 80.00 Max 806 800
H (MR12-MR13) kH 256.07 316.85 326.36 325.20
UAH 80.00 80.09 100 115

◦
Units: k [1000 kg/h/ bar kg/m3 ], UA (MJ/ C h).

Table 5
Flowrate and temperature conditions at each step for Core 2.

Stream Conditions Step 1 Step 2 Step 3 Step 4 Step 5 Step 6

NG2 (E) Flowrate 111.3 118.1 121.4 118.1 118.2 118.0

Temperature −105.9 −111.8 −122.5 −125.0 −127.8 −128.0
MR3 (F) Flowrate 36.8 40.4 41.1 40.5 40.5 40.3
Temperature −109.3 −114.6 −122.7 −125.1 −127.9 −128.0
MR9 (G) Flowrate 182.9 186.3 186.3 187.1 187.3 187.2
Temperature −111.0 −116.7 −123.0 −125.2 −127.9 −128.0
MR14 (H) Flowrate 169.2 227.6 210.5 227.8 227.4 227.6
Temperature −75.9 −82.0 −74.0 −75.9 −74.4 −74.3

Units: flowrate (ton/h), temperature (◦ C).

Table 6
k and UA values at each step for Core 3.

Stream Variable Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8

I (NG3-NG4) kI 32.73 32.94 32.89 32.88 32.87

UAI 80.00 Max Max 1800.00 1800.00
J (MR4-MR5) kJ 10.47 10.53 10.54 10.54 10.54
UAJ 80.00 Max 950.00 315.00 270.00
K (MR6-MR7) kK 34.72 45.18 46.72 47.17 47.35
UAK 80.00 57.00 65.00 71.50 73.00

Units: k [1000 kg/h/ bar kg/m3 ], UA (MJ/◦ C h).

cold box and the density of the fluid changes greatly. With poor ini-
tial estimates, stream conditions at Step 1 were far away from the
design points. At Step 3, UA values for the hot streams (E, F and G)
were at maximum limits but still could not meet the temperature
condition. This is because of the increased flow rates of hot streams
and the lowered flow rate of the cold stream. As the temperature
values of the hot streams decrease, the density values are increased,
and hence the flow rates are also increased. Therefore, adjustment
of the k values is needed again. After five adjustments, the final val-
ues of k and UA were obtained (Figs. 7 and 8 and Tables 4–7). PI
control parameters to obtain k and UA parameters are summarized
in Table 8.

Table 7
Flowrate and temperature conditions at each step for Core 3.

Stream Conditions Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8

NG4 (I) Flowrate 116.6 118.0 118.1 118.0 118.0 118.0 118.0 118.0
Temperature −141.5 −144.9 −146.5 −147.0 −147.4 −147.5 −147.7 −147.8
MR5 (J) Flowrate 39.9 40.4 40.3 40.4 40.4 40.4 40.4 40.4
Temperature −143.9 −147.2 −147.1 −147.6 −147.8 −147.9 −147.8 −147.8
MR7 (K) Flowrate 26.5 40.4 38.1 40.4 39.7 40.4 40.2 40.4
Temperature −130.0 −135.9 −132.7 −133.7 −132.7 −133.0 −132.7 −132.7

Units: flowrate (ton/h), temperature (◦ C).

 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557 553

Fig. 7. Flow rate and temperature conditions at each step for Core 2.

4.2. Model validation through feed load variation simulation
For dynamic validation of the developed model, controllers are
added. FIC-LNG VLV is a flow rate controller with the LNG-VLV actu-
ator position defined as the output target object, and the mass flow
rate of NG5 as the process variable source. TIC-100 is a tempera-
ture controller with the JT2 actuator position as the output target
object, and the temperature of NG4 as the process variable source.
TIC-101 is also a temperature controller which varies the opening
of JT1 to keep the temperature of NG3 at the specified set point.
554 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557

Fig. 8. Flow rate and temperature conditions at each step for Core 3.

The process flow diagram with the added controllers is shown in
Fig. 9. Controller tuning parameters used in the model validation
summarized in Table 8.
At 5 min after the integration, the feed load of natural gas is
decreased from 100% to 80% by manual action. This is done by
changing the set point of FIC-LNG VLV from 118 ton/h to 94.4 ton/h.
Since the flow rate of natural gas is decreased, natural gas is
cooled further. Thus, the temperature of both NG3 and NG4 is low-
ered. Then, the temperature controllers closed the JT valves, hence
attaining a lower mixed refrigerant flow rate. The total flow rate of

Fig. 9. Process flow diagram of the liquefaction process with controllers.

 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557 555

(a) Flow rate variation of NG (Action) and MR (Result)

250
MR 100% → 84.5%
230

210

Mass flow rate (ton/h)

190

170 NG, ton/h

MR total, ton/h
150
NG 100% → 80%
130

110

90
NG 80% → 100%
70

50
0 20 40 60 80 100 120 140 160 180
Time (min)

(b) JT1 Valve opening & Flow rate variation (Result)

60 200
JT1 opening, %
180
55 JT1 mass flow, ton/h
160

140
Valve opening (%)

Mass flow rate (ton/hr)

50
120

45 100
JT1 opening
80
50% → 41.7%
40
60

40
35
20

30 0
0 20 40 60 80 100 120 140 160 180
Time (min)

(c) JT2 Valve opening & Flow rate variation (Result)

60 50
JT2 opening, %
45
55 JT2 mass flow, ton/h
40

35
Mass flow rate (ton/hr)
Valve opening (%)

50
30

45 25
JT2 opening
50% → 45.9% 20
40
15

10
35
5

30 0
0 20 40 60 80 100 120 140 160 180
Time (min)

Fig. 10. Action and results of feed load variation simulation.

556 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557

Fig. 11. Constructed dynamic model of C3 MR process.

MR decreased by 15.5%. The opening of JT1 changed the flow rate 4.3. Industrial case study of C3 MR process
from 50% to 41.7%, while the opening of JT2 changed the flow rate
from 50% to 45.9%. The general control structure for C3 MR process includes two
At 74 min after the integration, the feed load of natural gas is kinds of control logic in cold box operation (Sicinski et al., 2007).
increased back from 80% to 100% by manual action. This is also done One is LNG temperature control and the other is WEDT (warm
by changing the set point of FIC-LNG VLV from 94.4 to 118 ton/h. The end delta temperature) control. For LNG temperature control, the
temperature controllers open JT1 and JT2 to lower the temperature LNG temperature leaving cold box (‘LNG’ stream) should be main-
of NG3 and NG4. Soon, all the stream conditions are back at the tained at −148.7 ◦ C in order to satisfy LNG product temperature
initial conditions. Thus, the developed dynamic model exhibits the to −158.5 ◦ C after Joule–Thomson valve. The amount of light MR
expected performance as shown in Fig. 10. (LMR) is changed to keep this LNG temperature to −148.7 ◦ C. On
the other hand, WEDT means the temperature difference between
heavy MR (HMR) and MR leaving cold box (‘MR Mix-3’ stream).
Because the temperature of heavy MR is fixed to −29.9 ◦ C, con-
Table 8 stant WEDT means constant temperature of ‘MR Mix-3’ stream.
Controller tuning parameters. As the WEDT is assumed to about 1.5 ◦ C, the temperature of ‘MR
Type Item kC Ti Td Mix-3’ stream has to be −31.4 ◦ C. The amount of HMR was var-
ied to track this ‘MR Mix-3’ stream temperature of −31.4 ◦ C. The
Parameter estimation k Value 0.1 0.2 0
UA value 1 20 0 suggested process flow diagram with control structure is shown in
Model validation TIC-JT1 0.1 30 0 Fig. 11.
TIC-JT2 0.1 30 0 The set point tracking was conducted to validate the dynamic
FIC-LNG VLV 0.01 0.8 0 model. The temperature of ‘MR Mix-3’ stream was controlled to
 C.-J. Lee et al. / Computers and Chemical Engineering 84 (2016) 546–557
[kg/h]
590000
588000
586000
584000
582000
580000
578000
576000
574000
572000
LNG Flow: 1.5% ↑
570000
0 50 100 150
Fig. 12. Set point (SP) tracking in the case of increasing LNG production.
−31.4 ◦ C by manipulating light MR and heavy MR ratio using LMR
valve (‘VLV-LMR from MCHE’ valve) and temperature of LNG from
cold box was controlled to −148.7 ◦ C using HMR valve (‘VLV-
HMR’ valve). The LNG product was increased by 1.5 wt% from
572,772 kg/h to 581,364 kg/h and the time to chase desired set point
was 230 min as shown in Fig. 12.
5. Conclusions
A systematic approach to develop dynamic model of PFHEs in
the case of lack of internal design data was discussed. After the
physical configuration of the PFHEs including the equipment siz-
ing, zone and layer arrangement had been estimated, the proposed
decomposition methodology was applied. The values of k and UA of
each layer in each zone must be specified as a dynamic specification.
To simplify the problem, it was assumed that same specifications
are assigned to layers with the same feed. Additionally, equal values
of k and UA were specified at each zone in each layer. A decompo-
sition algorithm was suggested for adjusting the k values firstly,
followed by the UA values, at different steps. A PI control algorithm
was used to tune the values. The entire model development process
was applied to the cold box modeling of the natural gas liquefac-
tion process. The developed dynamic models yielded the expected
results in a dynamic validation study.
Acknowledgements
This research was supported by the second phase of the Brain
Korea 21 Program in 2014, by Engineering Development Research
Center (EDRC) funded by the Ministry of Trade, Industry and Energy
(MOTIE), Institute of Chemical Processes in Seoul National Univer-
sity, MKE and grant from the LNG Plant R&D Center funded by
the Ministry of Land, Transportation and Maritime Affairs (MLTM)
of the Korean government, the IT R&D program of MOTIE/KEIT
(10049155, Development of equipment control algorithm based
on plasma monitoring for efficiency improvement of 10 nm etch
process), by the Energy Efficiency & Resources Core Technology Pro-
gram of the Korea Institute of Energy Technology Evaluation and
Planning(KETEP) granted financial resource from the Ministry of
Trade, Industry & Energy, Republic of Korea (No. 20132010201760)
and (No. 20132010500050).
557
[°C]
-30
-30 .5
-31
SP: -31.4°C
-31.5
-32 LNGFlow
MRout Temp.
-32 .5
-33
-33 .5
-34
20 0 250 300 350
[min]
References
Alabdulkarem A, Mortazavi A, Hwang Y, Radermacher R, Rogers P. Optimiza-
tion of propane pre-cooled mixed refrigerant LNG plant. Appl Thermal Eng
2011;31:1091–8.
Averous D, Hammadi K, Pingaud H, Joulia X, Guittard P. Dynamic simulation of brazed
plate–fin heat exchangers. Comput Chem Eng 1999;23:S447–50.
Boehme R, Parise JAR, Marques RP. Simulation of multistream plate–fin heat
exchangers of an air separation unit. Cryogenics 2003;43:325–34.
Gruenspecht H. International Energy Outlook 2011. Center for Strategic and Inter-
national Studies; 2010.
Kanoğlu M. Exergy analysis of multistage cascade refrigeration cycle used for natural
gas liquefaction. Int J Energy Res 2002;26:763–74.
Kikkawa Y, Nakamura M, Sugiyama S. Development of liquefaction process for nat-
ural gas. J Chem Eng Jpn 1997;30:625–30.
Lee C-J. Dynamic modeling and simulation for LNG value chain. Korea: Seoul National
University Seoul; 2012, Unpublished doctoral dissertation.
Lim W, Choi K, Moon I. Current status and perspectives of liquefied natural gas (LNG)
plant design. Ind Eng Chem Res 2013;52:3065–88.
Luo X, Guan X, Li M, Roetzel W. Dynamic behaviour of one-dimensional flow
multistream heat exchangers and their networks. Int J Heat Mass Transf
2003;46:705–15.
Mandler JA. Modelling for control analysis and design in complex industrial separa-
tion and liquefaction processes. J Process Control 2000;10:167–75.
Michelsen FA, Halvorsen IJ, Lund BF, Wahl PE. Modeling and simulation for control of
the TEALARC liquified natural gas process. Ind Eng Chem Res 2010;49:7389–97.
Minta, M., Stone, J.B., Feist, R.S., 2008. U.S. Patent Application 12/668,811.
Neeraas BO, Fredheim AO, Aunan B. Experimental data and model for heat transfer,
in liquid falling film flow on shell-side, for spiral-wound LNG heat exchanger.
Int J Heat Mass Transf 2004;47:3565–72.
Pacio JC, Dorao CA. A review on heat exchanger thermal hydraulic models for cryo-
genic applications. Cryogenics 2011;51:366–79.
Peric M, Zorn T, el Moctar O, Schellin TE, Kim YS. Simulation of sloshing in LNG-tanks.
J Offshore Mech Arctic Eng 2009;131:031101.
Pingaud H, Le Lann JM, Koehret B, Bardin MC. Steady-state and dynamic simulation
of plate fin heat exchangers. Comput Chem Eng 1989;13:577–85.
Remeljej CW, Hoadley AFA. An exergy analysis of small-scale liquefied natural gas
(LNG) liquefaction processes. Energy 2006;31:2005–19.
Roberts, M.J., Brostow, A.A., 2014. U.S. Patent No. 8,656,733. Washington, DC: U.S.
Patent and Trademark Office.
Sharifi F, Narandji MG, Mehravaran K. Dynamic simulation of plate heat exchangers.
Int Commun Heat Mass Transf 1995;22:213–25.
Shin Y, Lee YP. Design of a boil-off natural gas reliquefaction control system for LNG
carriers. Appl Energy 2009;86:37–44.
Sicinski, M.A., Johnston, B.K., Trautmann, S.R., Roberts, M.J., 2007. U.S. Patent Appli-
cation 11/782,990.
Singh A, Hovd M. Model requirement for control design of an LNG process. Comput
Aided Chem Eng 2007;24:533–8.
Skaugen G, Hammer M, Wahl PE, Wilhelmsen Ø. Constrained non-linear opti-
misation of a process for liquefaction of natural gas including a geometrical
and thermo-hydraulic model of a compact heat exchanger. Comput Chem Eng
2014;73:102–15.
Song K, Lee C-J, Jeon J, Han C. Dynamic simulation of natural gas liquefaction pro-
cess. In: 22nd European Symposium on Computer Aided Process Engineering
30; 2012. p. 882.