Applied Thermal Engineering 235 (2023) 121412

Contents lists available at ScienceDirect

Applied Thermal Engineering

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

Research Paper

A co-simulation method of refrigerator system and temperature field

Shijie Xu a, Min Wang a, Lili Han b, Shoujie Zhang b, Jianlin Yu a, Suxin Qian a, *
a
Department of Refrigeration & Cryogenic Engineering, School of Energy and Power Engineering, Xi’an Jiaotong University, Xi’an 710049, China
b
Hisense Refrigerator Co., Ltd., Qingdao 266000, China

A R T I C L E I N F O A B S T R A C T

Keywords: There are mainly two approaches to simulating refrigeration equipment such as refrigerators. The refrigeration
Refrigerator system model can reasonably simulate the transient cyclic performance of the refrigeration system, while the
Simulation computational fluid dynamics (CFD) model focuses more on temperature and velocity fields inside cabinets. To
Computational fluid dynamics
bridge the gap between the two modeling approaches, this study proposes a novel co-simulation method, namely
Refrigeration system model
the reduced-order temperature field model (ROM). Two special parameters, time constant and temperature
Reduced-order model
Temperature field deviation, are extracted from the full-order CFD results to characterize the temperature transients of user-
specified locations . Then the ROM extends the lumped cabinet temperature into temperature distributions
through the aforementioned two parameters in system simulation. Thus, temperature field transients can be
approximately presented in the dynamic system simulation. System pressures, power, and temperature distri­
butions in different compartments were measured to validate the ROM. The majority of time-average temper­
ature errors in the refrigeration compartment are within 1.0 ◦ C in the pull-down stage and 1.1 ◦ C in the on–off
stage. The majority of time-average temperature errors in the freezer compartment are less than 2.5 ◦ C in the
pull-down stage of the lower layer and 3.0 ◦ C in the upper layer, while all temperature errors are within 1.5 ◦ C in
the on–off stage.

models focused on refrigeration system performance in steady-state or

transient conditions.
1. Introduction Researchers have conducted many studies on refrigerator system
performance through simulation, which streamlines simulation tech­
The duty of a refrigerator is to keep the target temperature niques and frees researchers from cumbersome experiments. Due to
economically for any stored product. IIR estimated that there are 2 insufficient computer performance, early refrigerator system simula­
billion operational domestic refrigerators worldwide, which consumed tions were limited to steady-state. Jung conducted a steady-state simu­
about 4 % of global electricity [1]. As climate change has become one of lation to investigate the alternative to R12 refrigerant in 1991 and
the biggest non-traditional security challenges threatening human discovered the R142b refrigerant could be mixed with R32 or R22 to
development [2], any reasonable ways to reduce carbon emissions are replace R12 [6,7]. In the same year, Vidmar and Gaspersic developed
essential, where developing environmental-friendly refrigerator prod­ the first dynamic simulation model to study the start-up and shut-down
ucts plays a key role. Although the function of a refrigerator is relatively transient behavior of a refrigerator [8]. With the rapid development of
simple, the refrigerator itself is complex because it contains multiple computer technology, system simulation played a role in refrigerator
physical fields at different scales and different locations. The traditional studies. Bansal [9], Ding [10], and Hermes [11,12] individually devel­
wisdom of developing a new product relies heavily on engineering oped domestic refrigerator models with their respective teams. How­
experience, i.e. lessons learned from numerous failed experiments. ever, it is still a challenge to accurately predict the compartment
However, with the development of computer technology, two ap­ temperature transients using this approach. One way to improve accu­
proaches emerged in the past decades to partially settle this difficulty, a. racy is by incorporating a more complex calculation for the cabinet heat
k.a., the computational fluid dynamics (CFD) model and the refrigera­ load, from the lumped parameter model of walls [13] to distributed
tion system simulation. CFD models have been widely utilized to parameter model of walls [10,12]. Nonetheless, the lumped description
simulate physical fields such as temperature [3], humidity [4], and ve­ of temperature cannot obtain important metrics such as temperature
locity [5] in refrigerator compartments, while the refrigeration system

* Corresponding author.
E-mail address: [email protected] (S. Qian).

https://doi.org/10.1016/j.applthermaleng.2023.121412
Received 8 May 2023; Received in revised form 23 July 2023; Accepted 19 August 2023
Available online 22 August 2023
1359-4311/© 2023 Elsevier Ltd. All rights reserved.
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Nomenclature w velocity of refrigerant [m⋅s− 1]

x static quality [-]
Symbols x′ flow-weighted quality [-]
A area [m2] z length in z direction [m]
ai time constant of i th user-defined location [s]
b coefficient of fitted function [-] Greek letters
Ck heat capacity per unit area of k th CV [J⋅m− 2⋅K− 1] α angle [◦ ]/heat transfer coefficient [W⋅m− 2⋅K− 1]
c specific heat capacity [J⋅kg− 1⋅K− 1] β angle [◦ ]
c, q, r, s parameters proposed by Zivi [-] γ void fraction
dout outlet Diameter of the accumulator [m] ε effectiveness of heat exchanger [-]
FC freezer compartment εf fin effectiveness [-]
FCL lower layer of FC η efficiency [-]
FCU upper layer of FC μ dynamic viscosity [Pa⋅s]
G(s) transfer function π variable in Buckingham π theorem [-]
H liquid level [m] ρ density [kg⋅m− 3]
hk density-weighted enthalpy of the k th CV [J⋅kg− 1] ρk average density of kth CV [kg⋅m− 3]
Δhc enthalpy correction [J⋅kg− 1] Subscripts
h′k flow-weighted enthalpy of k th CV [J⋅kg− 1] 1,2 hot fluid, cold fluid (ε − NTU method)
k thermal conductivity [W⋅m− 1⋅K− 1]/coefficient of fitted Al aluminum
function [-] air air
Lct heat transfer length of the capillary tube [m] amb ambient
Lsuc heat transfer length of the suction line [m] AVG/avg average
Lcav mean thickness of the cavity [m] ct capillary tube
l length [m] ch compressor chamber
ṁ mass flow rate [kg⋅s− 1] cav cavity
NTU number of transfer unit [-] con contact
n rotate speed [rpm] dis discharged pipe
p pressure [Pa] eq the equivalent value
Q heat transfer rate [W] f saturated liquid/fin
qm mass flow rate [kg⋅s− 1] g saturated vapor
R thermal resistance [K⋅m− 2⋅W− 1] h isenthalpic
RC refrigeration compartment i inner
r radius [m] in inlet
S the slip-ratio [-] liq liquid
Scav cross-sectional area of the cavity [m2] m mass
T temperature [K] min the minimum value
ΔTi temperature deviation of ith user-defined location [K] o outer
Ti temperature of i th point [◦ C] out outlet
Ti,0 steady-state temperature of i th point [◦ C] p isobaric
δT time-average temperature error [K] ρ isopycnal
δTi absolute temperature error in ith time step [K] r refrigerant
tcon contact thickness [m] suc suction line
t time [s] s isentropic
Δt time step of simulation model [s] TC thermocouple
u specific internal energy [J⋅kg− 1] t tube
Vth stroke volume [m3] v volumetric
Vk volume of the kth CV [m3]
vsuc specific volume at compressor suction port [m3⋅kg− 1]

homogeneity in each compartment. To tackle the aforementioned challenges of the standalone refriger­
To investigate the temperature distribution, researchers usually used ator system simulation or CFD simulation, it is intuitive to couple the
CFD models. For example, Ding et al. studied the influence of clearance two models. Although the coupling algorithm has not been studied for
between shelves, doors, and the back wall on temperature distribution in refrigeration applications, technical approaches from other fields can be
a direct cool refrigerator [14]. More recently, Wie et al. studied the inspiring, including Simulink as the main simulation body, CFD model
pathway to improve temperature uniformity in a top-freezer refrigerator as the main simulation body, and collaborative simulation. Wang et al.
system with a tube-fin evaporator and a multi-duct system [15]. Addi­ couple the Simulink-based system model and CFD model of a marine gas
tionally, food transpiration affected by temperature and humidity dis­ turbine. At the end of each simulation time step in Simulink, S-Function
tribution [4], compartment position configuration [16], unsteady saves the present results and then upgrades boundary conditions in the
temperature distribution in different property assumptions [5], etc., Journal file, which will be read by Fluent at the beginning of the next
were also studied using CFD. However, it is difficult to accurately assign time step [17]. Alternatively, the system model in Simulink can be
the dynamic boundary conditions for CFD models, e.g. inlet temperature compiled into a C/C++ function and called in Fluent via the User-
from the evaporator for an open-loop model or evaporating temperature defined Function (UDF) [18]. For the collaborative approach, Han
for a closed-loop model. et al. coupled Simulink and the Fluent models in an alternant way

2
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

through a dynamic read-write data file. A similar technique has been

applied to missile separation [19].
The aforementioned three coupling methods are regarded as tight
coupling, which consumes excessive computation resources and results
in very low simulation speed, not to mention the challenge of storing
huge transient data in memory and hard drive. In this study, a looser
coupling method named the reduced-order model (ROM) has been
proposed to approximately simulate the temperature transients in
compartments. Using the ROM, temperature dynamic features at finite
(e.g., a few dozen or a few hundred) user-defined locations in com­
partments can be traced in the systematic simulation. Temperatures at
locations that are not specified by the user are not simulated by the
ROM. Compared to the regular CFD model that contains 10 million or
more degrees of freedom (variables), the degrees of freedom in a ROM
equals the number of user-specified locations in the compartment,
which is many magnitudes less than the conventional full-order CFD Fig. 2. Illustration of the simulated refrigerator.
model. Therefore, the ROM can be explicitly computed at a much faster
speed than the conventional CFD model. The architecture of the pro­ connected in series, while the evaporator is a tube-fin heat exchanger.
posed dynamic refrigerator model including the ROM is shown in Fig. 1. The major parameters of the refrigerator components are listed in Ap­
The major novelty of this study is this reduced-order model for co- pendix A.1 (Supplementary Materials). The simulated system and its log
simulation of the dynamic temperature distributions with the refriger­ (p)-h diagram are shown in Fig. 3. The condenser receives the high-
ation system. Furthermore, in the refrigerator system model, a new two- pressure, high-temperature vapor refrigerant (2) from the compressor.
part diabatic capillary tube model was implemented to physically The refrigerant in the condenser rejects heat to the ambient air and
describe the regenerative heat and the pressure drop. In the cabinet condenses into subcooled liquid (3). The subcooling degree then
model, state-space description was applied to model heat transfer possibly increases (3 → 4) as the refrigerant goes through the capillary
through the compartment walls. The above modeling novelties were tube-suction line heat exchanger (CT-SLHX) along with a moderate
experimentally validated by system performance and temperature dis­ pressure drop. As the refrigerant partially flashes into vapor, the pres­
tribution measurement during the pull-down and on–off stages in a sure in the capillary tube sharply decreases as the refrigerant continues
typical frost-free domestic refrigerator. to flow. The depressurized two-phase refrigerant flows into the evapo­
rator (5) and absorbs the heat from the cabinet (5 → 6). Finally, the
2. Description of refrigerator and experiment evaporated low-pressure and low-temperature refrigerant absorbs heat
via CT-SLHX (6 → 1) from the capillary tube (3 → 4) before being sucked
2.1. System profile into the compressor.

As depicted in Fig. 2, the simulated device is a fan-and-damper sin­

2.2. Experiment condition and measurement equipment
gle-evaporator frost-free refrigerator with a 458 L total storage volume.
The refrigerator cabinet contains a refrigeration compartment (RC), a
The experiment has been conducted to validate the proposed ROM
variable temperature compartment (VC), and a freezer compartment
for the refrigerator model, following the GB/T 8059-2016 Standard in an
(FC). The model number of the variable-speed piston compressor is
environmental chamber. All the measurement equipment information is
VTB1113YA, with a refrigerant charge of 62 g of R600a. The condenser
compiled in Table 1. The testing dry-bulb and wet-bulb temperatures are
consists of a micro-channel condenser and a hot-wall condenser that are
set to 32 ◦ C and 26.5 ◦ C, respectively, corresponding to a relative

Fig. 1. Overview of the architecture of the proposed dynamic refrigerator co-simulation model, including the refrigeration system sub-models, the cabinet model,
and the reduced order model for the temperature field.

3
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 3. Refrigeration system: (a) Schematic, (b) Cycle in log(p)-h diagram.

Table 1 2.3. Thermocouple calibration

Measurement equipment and propagated uncertainties in the experiment.
Item Range Uncertainty To minimize the uncertainty from thermocouples as much as
possible, all the thermocouples must be calibrated before the experi­
T type thermocouple − 200 to 350 ◦ C ±0.5 ◦ C
Pressure transducer 0–1.6 MPa ±1.6 kPa ment. Thermocouples are mounted with the copper pillar before the
Watt meter 0–600 V & 0–40 A ±0.2 % of calculated value calibration to simulate the real test condition. The calibration range of
Refrigerated and heating − 30 to 200 ◦ C ±0.01 ◦ C temperature is from − 30 ◦ C to 40 ◦ C with a 2 ◦ C step because the un­
circulator certainty varies with the tested temperature. The calibration results are
Hot-wire anemometer 0–30 m⋅s− 1, − 20 ±(0.1 m⋅s− 1 + 5 % of the
to 60 ◦ C displayed value)
fitted into linear functions in the end. The detailed calibrating process is
as follows:

humidity of 65 %. Before the test starts, the refrigerator is heated into (1) Put the thermocouples with copper pillars into the refrigerated
thermal equilibrium with the ambient air. The cabinet temperature of and heating circulator (RHC), which was full of liquid ethylene
the refrigerator is controlled by the following actuating components: an glycol. Set the temperature of RHC Ttrue as 40 ◦ C;
evaporating fan, two air dampers, and a compressor. The two air (2) Read the temperature of thermocouples on the computer. When
dampers respectively control the temperature of RC and VC. And FC the display results reach the set temperature and keep unchanged
temperature is directly controlled by the compressor as well as the for 15 min, record the temperature Ttest.
evaporating fan. The VC air damper is not independent but is synchro­ (3) Turn the set temperature down by 2 ◦ C, and repeat the process
nized with the RC air damper. The setpoint temperatures of RC, VC, and (2).
FC are 5 ◦ C, 0 ◦ C, and − 18 ◦ C, respectively. We tested the following two (4) All the recorded data is transformed into the form of ΔT = Ttest −
modes and each test was repeated two or three times to avoid Ttrue = f(Ttrue), and f(Ttrue) is a linear function, the coefficient k
inconsistency: and b of which is shown in Appendix A.3 in the Supplementary
Pull-down stage When the refrigerator turns on, the condensing fan Materials.
starts instantly, and the compressor starts with a 3 min delay. The
evaporating fan starts 10 min after the compressor turns on. The tem­ 3. Methodology and refrigerator sub-models
perature of each compartment starts to cool down before reaching its set
point, and thus the two air dampers are both in the open state. As the 3.1. Fundamental principle of the ROM
temperature pulls down, RC temperature first reaches the lower limit of
the set temperature (4 ◦ C), and the air dampers of RC and VC simulta­ Considering a compartment with N cells, the lumped temperature of
neously close. RC and VC rewarm while FC keeps cooling until it reaches a compartment, denoted as Tavg , is the volumetric average temperature
the lower limit of the setpoint (− 19 ◦ C). of all the air region cells. There exists a specific relation between the
On-off stage When FC first reaches the lower limit of the set tem­ temperature of any specific location Ti and the compartment lumped
perature (− 19 ◦ C), the compressor and evaporator fan shut down. When temperature Tavg . For example, when the supply air temperature varies,
RC and VC rewarm to the upper limit of setpoint (6 ◦ C), the air dampers
Ti varies slower than the lumped temperature Tavg , which is the case for
open again. However, at this moment FC is still in the rewarming process
areas that are closer to the returning vents or areas that are far away
and the compressor is still off. Thus, RC and VC rewarm beyond the
from the mainstream. Alternatively, Ti may respond faster than the
upper limit temperature. When FC also rewarms to the upper limit
lumped temperature Tavg , which is the case for areas that are closer to
(− 17 ◦ C), the compressor starts again and all the compartments receive
the supply vents or the mainstream. In either case, a transfer function
cooling.
G(s) can be used to describe the time-response difference between Tavg
In the simulation, The initial conditions represent the state of the
and Ti . For the first case, if Ti varies slower, the transfer function G(s) is
experimental refrigerator that has been shut down and placed in the
applied to the lumped temperature Tavg , i.e. Ti (s) = G(s)Tavg (s). For the
experimental environment for sufficient time to reach thermal equilib­
second case, if Ti varies faster, the transfer function G(s) is applied to Ti ,
rium with the ambient. All the initial conditions of the system model,
i.e. Tavg (s) = G(s)Ti (s), and the inverse transfer function can be used to
time constants and temperature deviations utilized in the ROM, and
correlations are in Appendix A.2 in the Supplementary Materials. compute Ti , i.e. Ti (s) = G− 1 (s)Tavg (s). Therefore, with the compartment

4
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

lumped temperature Tavg , the temperature at any user-specified location ⎧ ∫

Tavg − Ti
can be computed, if the detailed expression of the transfer function G(s) ⎪
⎪
⎪
⎨ ( ) dt + ΔTi ai > aavg
ai − aavg
and its inverse function G− 1 (s) are known. Ti = (8)
⎪
Following the above analysis, the transfer function describes the ⎪
⎪ dT ( )
⎩ avg aavg − ai + Tavg + ΔTi ai ≤ aavg
physical delay between the temperature at one location and another. dt
The simplest type of transfer function has the following form: According to GB/T 8059-2016 Standard, thermocouples embedded
G(s) = 1/(as + 1) (1) in copper pillars should be used to measure air temperature. With the
thermal mass and estimated time constant aTC of these copper pillars, the
where the variable a, namely the relative time constant, denotes the simulated temperature of the copper pillar that corresponds to the local
lagging of Ti relative to Tavg (or the other way around). In heat transfer, it air temperature can be computed by Eq. (9).
is called time constant and has the unit of [s].
The time constant a represents the lagging caused by the compre­ 1
Ti,TC (s) = Ti (s) (9)
hensive competition between velocity field, flow rate, intensities of 1 + aTC s
cooling power versus heat leak, etc. If Ti (s) = G(s)Tavg (s), the larger time where the subscript TC represents the copper pillar thermocouple.
constant means that the user-specified location is away from the main­ When comparing against experimental data, the simulated copper pillar
stream, e.g. close to the interior wall or in an air-tight drawer. temperature, Ti,TC , should be used, instead of the local air temperature
The next step is to derive the expression of the time constant a. Since Ti . Based on the used copper pillars in the test, the estimated aTC equals
Ti is the volume-weighted temperature of the user-specified control 460 s in this study.
volume with the local volume of Vi . The temperature variation at the The proposed ROM in this study should not be confused with the
sample site can be viewed as the result of heat transfer between the conventional ROM technique that has been implemented in commercial
control volume and the lumped temperature, as demonstrated by the software (i.e. proper orthogonal decomposition, POD) [20,21]. It is
following equation. worth noting that both the proposed ROM and the POD method extract a
dTi ( ) subset of information from the full-order CFD model. The POD method
ρcVi = UA Tavg − Ti (2) requires a group of orthogonal basis to describe features of a physical
dt
field and can be extended to dynamic simulations through rational
where the time constant a can be expressed as Eq. (3), which is the interpolation or extrapolation algorithms, which is completely data-
definition of the time constant in heat transfer. driven. The novel ROM requires only two characteristic variables, the
ρcVi relative time constant ai and the temperature deviation ΔTi , at each
a= (3) user-defined location, which has sufficient physical significance. In this
UA
way, the temperature prediction accuracy of the refrigerator’s model
In reality, it is difficult to obtain the arbitrary numbers UA and Vi and
can be elevated from the lumped temperature to the primary feature of
thus the time constant cannot be analytically computed. Nonetheless,
the temperature field. The ROM receives transients from other compo­
the time constant can be obtained from the full-order CFD model. Using
nent models as the boundary condition, which accounts for changes in
the full-order CFD model, the absolute time constant ai and aavg were
the on/off of fan and compressor, as well as variation in ambient tem­
equivalent to the time required to reduce from the initial excess tem­
perature, etc.
perature to e-1 (an example can be found in the Appendix A.4). The
relative time constant a Eq. (3) can be determined as follows.
3.2. Diabatic capillary tube model and validation
a = ai − aavg (4)

In the first case, when Ti varies slower than Tavg , i.e. ai > aavg , it can Diabatic capillary tubes are prevailing in domestic refrigerators. A
be computed as diabatic capillary tube is composed of a capillary tube and a suction line.
∫ One of these two tubes is affixed to another with aluminum tape, which
[ /( )] is named capillary tube-suction line heat exchanger (CT-SLHX). Based
Ti = G(s)Tavg = Tavg − Ti ai − aavg dt (5)
on the flow and heat transfer characteristics along the capillary tube, an
In the second case, when Ti varies faster than Tavg , i.e. ai ≤ aavg , it can explicit diabatic simulation model may be built. In general, the phase
be computed as change occurs near the exit region of the capillary tube, hence single-
phase flows are typically observed inside the entrance non-adiabatic
dTavg ( )
region [22]. Regarding the pressure drop characteristics, more than
Ti = G− 1 (s)Tavg = ai − aavg + Tavg (6)
dt 50 % of the overall pressure drop occurs in the last 20 % of the tube
The time constant introduces the transient characteristics of the length, especially near the exit, where the pressure declines abruptly
temperature at user-specified locations. To account for the steady-state [23]. Considering these characteristics, a diabatic capillary model was
temperature inhomogeneity, the variable, denoted as temperature de­ proposed here by combining the recuperator section and the adiabatic
viation ΔTi , is introduced to the equation. ΔTi measures the temperature pressure drop section.
inhomogeneity due to inhomogeneous heat transfer from the envelope In the recuperator section, the heat transfer is physically modeled to
structure to different locations of the compartments, as shown in Eq. (7). update the properties of the refrigerant entering the second adiabatic
Thus, the complete form of ROM is shown in Eq. (8). pressure drop section. The heat transfer between the recuperator section
and the suction line is computed by the ε − NTU method [24]:
ΔTi = Ti (t = ∞) − Tavg (t = ∞) (7)
Q = ε(qm c)min (T’1 − T’2 ) (10)

The value of effectiveness, ε, can be calculated in different flow

regimes.

⎧
⎨ 1 − exp(-NTU) two-phase
ε = [1 − exp( − 2NTU) ]/2 parallel flow, single-phase, ​ assume ​ equal ​ water ​ equivalent (11)
⎩
NTU/(1 + NTU) counter flow, single-phase, ​ assume equal water equivalent

5
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

The value of NTU can be expressed as The angles α and β are expressed as
( )
kA r − rct,o
NTU = (12) α = arccos suc,o (18)
(qm c)min rsuc,o + rct,o
where kA is the product of heat transfer coefficient and heat transfer ( )
rsuc,o − rct,o
area along the flowing direction. And (qm c)min is the product of mass flow β = arcsin (19)
rsuc,o + rct,o
rate and specific heat, where the minimum value of hot and cold fluid is
chosen. Lct and Lsuc are written as
Moreover, computing kA is essential for determining the effective­ ( )
1 β
ness and heat transfer performance. Because modern personal com­ Lct = πrct,o + (20)
2 π
puters have significant surplus computational power, the recuperator
section model is built as detailed as possible to guarantee that heat (α)
transfer is calculated physically. This research analyzed the heat transfer Lsuc = πrsuc,o (21)
π
between the capillary tube and the suction line using the thermal
d) Conduction thermal resistance of the aluminum tape
resistance network diagram. In the studied refrigerator, the capillary
( )
tube and suction line are laterally attached and secured with aluminum rsuc,o + rct,o sinα
tape, making the CT-SLHX highly symmetrical, as shown in Fig. 4. Thus, R4 = (22)
0.00006kAl lct
we can easily construct the thermal resistance network diagram in Fig. 5.
As described in Fig. 5, heat is transferred from hot liquid refrigerant where the thickness of the aluminum tape is 0.06 mm.
in the capillary tube to the copper tube wall. Ignoring the thermal e) Contact thermal resistance between the capillary tube and the
contact resistance between the aluminum tape and the tube walls, the suction line
temperature of the aluminum tape on both ends is identical to the tcon
R5 = (23)
temperature of the wall. Heat is then transferred from the capillary tube kAl lct rct,o
wall to the suction line wall via the aluminum tape, the air cavity, and
where tcon is the contact thickness.
the contact surface. Finally, heat is transferred from the suction line wall
f) Conduction thermal resistance of the suction line
to the cold vapor refrigerant flowing in the suction line.
( )
The following describes the calculation process of the thermal 1 rsuc,o
resistance of each part in the CT-SLHX (Item a ~ h). R6 = ln (24)
πksuc lsuc rsuc,i
a) Convection thermal resistance of liquid phase refrigerant in the
capillary tube g) Convection thermal resistance of vapor phase refrigerant in suc­
tion line
1 1
R1 = = ( ) (13) 1
αct,in Act,in αct,in πrct,i lct R7 = ( ) (25)
αsuc,in πrsuc,i lsuc
b) Conduction thermal resistance of capillary tube wall
( ) Due to the symmetry of the cross-section of the recuperator, half of
R2 =
1
ln
rct,o
(14) the total thermal resistance is expressed as follows.
πkct lct rct,i
R3 R4 R5
Req = R1 + R2 + + R6 + R7 (26)
c) Conduction thermal resistance of the air cavity R4 R5 + R3 R5 + R3 R4
2Lcav 2Lcav Considering all aforementioned equations, NTU can be computed as
R3 = = (15)
kair (Lsuc + Lct )lsuc kair (Lsuc + Lct )lct /
2 Req
Assuming that the cavity was a trapezoid, Lcav can be calculated NTU = (27)
(qm c)min
through
After obtaining the heat transfer from the ε − NTU method, the exit
Lcav =
2Scav
(16) enthalpy of refrigerant leaving the recuperator section can be used as the
rsuc,o + rct, o input boundary condition for the adiabatic pressure drop section. In this
where rsuc,o and rct,o are external radiuses of the suction line and section, Yang’s empirical correlation for adiabatic capillary tubes is
capillary tube, respectively, and Scav is the cross-sectional area of the implemented. This empirical correlation was proposed based on the
cavity, as described in Eq. (17). approximate analytic model [25], as shown in Eq. (28).
( )√̅̅̅̅̅̅̅̅̅̅̅ 0.7338 0.2220 0.4671 0.1226 1.5956 0.7061
( )2 ( ) ṁ = 4.2579 × 103 πr2 Pin ρin π2 π3 π4 π5 π6 π7 /3600
sinα rct,o + rsuc,o α 1 β
Scav = − πrsuc,o 2 − πrct,o 2 + (17) (28)
2 2π 4 2π
This empirical correlation was developed for seven refrigerants
including R600a and thus can be used for this study. The validation
dataset consists of 898 data points. After comparing the simulation and
experimental results, it can be determined that the majority of mass flow
rate deviations fell between − 20 % to +15 % [25]. Besides, this corre­
lation is numerically robust over a wide range of operating conditions,
which is a good candidate for cyclic start-up and shutdown simulations.
As the key component providing mass flow rate signal in the refrig­
eration system model, the precision of the diabatic capillary tube model
has a great influence on the system transients. Hence, it is necessary to
validate the model performance of the diabatic capillary tube. Here, the
predicted mass flow rate is compared against experimental data from the
literature [26], which was measured under 30 different working
Fig. 4. Heat transfer geometries of capillary tube suction line heat exchanger.

6
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 5. Thermal resistance network diagram of capillary tube suction line heat exchanger.

where lgasket is the total length of the gaskets.

3.3.2. Envelope model

Simulation of refrigeration systems requires a quick calculation of
the heat leak through the envelope. Here, a one-dimensional model for
the envelope wall is applied. In Fig. 7, it is assumed that the heat transfer
only occurs in the thickness direction of the envelope wall.
Each envelope wall is divided into N +1 equal control volumes as
shown in Fig. 7. The 1st and (N +1) th control volumes are only half the
size of the others. The (N +1) th control volume near the inner air is
directly affected by the air convective heat transfer. The energy con­
servation law of the envelope can be written as
⎧
⎪
⎪ C1 dT1 T2 − T1
⎪
⎪ = + αout (Tamb − T1 ) k=1
⎪
⎪ 2 dτ R1
⎪
⎪
⎨ Ck− 1 + Ck dTk Tk+1 − Tk Tk− 1 − Tk
= + 1<k <N+1 (31)
⎪
⎪ 2 dτ Rk Rk− 1
⎪
⎪
⎪
⎪
⎪ CN dTN = TN − TN+1 + αin (Tair − TN+1 ) k = N + 1
⎪
⎩
Fig. 6. Experimental validation of the diabatic capillary tube. Experimental 2 dτ RN
data were retrieved from Melo et al. [26]. where Tamb is the ambient air temperature and Tair is the lumped air
temperature of the compartment. Rk is the diffusive thermal resistance
conditions with R600a. The comparison results between the experi­ of the kth layer. Ck is the heat capacity per unit area of the kth layer.
mental dataset with the predicted mass flow rate are illustrated in Fig. 6. This one-dimensional envelope model is conventionally solved
In all 30 cases, mass flow rate simulation deviations are in the range of segment-by-segment, which will significantly slow down the simulation.
− 20 % to +20 %, and 80 % of the errors are within ±15 %. Therefore, Actually, expressing the control equations in state-space forms and just
the proposed bisection diabatic capillary tube model possesses a fairly leaving it to the computer does work well. Linear ordinary differential
acceptable performance. Compared with the traditional empirical cor­ equations (ODEs) systems can be written as
relations of a diabatic capillary tube [27], the proposed diabatic capil­ ⎧
lary tube model physically modeled the heat transfer part and thus is ⎨ ẋ = Ax + Bu
appropriate for a wider range of operating conditions. y = Cx + Du (32)
⎩
xt=t0 = x0

where A is a coefficient matrix of (N +1) × (N +1), B is a coefficient

3.3. Cabinet model
matrix of N × M, C is equal to a unit matrix I, D is equal to 0, u is a
coefficient matrix of 1 × 1, in this case. And details of these coefficients
3.3.1. The lumped air temperature model
can be found in Appendix A.5.
The air average temperature in each compartment is modeled by the
lumped-parameter method. The air from the evaporator is transported to
4. CFD model and validation
the compartment by the centrifugal fan, and there are multiple heat
exchanges between the cabinet air and other sources, including the
4.1. Model description
envelope, the shelf, and the air duct. Then the air flows into the air-
returning vent to the evaporator. Assuming the temperature at the air-
In Section 3.1, it is mentioned that the variables UA and Vi are
returning vent is equal to the compartment lumped temperature, and
impractical to derive the time constant for the ROM. However, there is
the air mass flow rate at the air-returning vent is equal to which at the
another way to estimate the time constants. The time constants and
air-supply vent, the energy conservation law of the air in each
temperature deviations can be directly obtained from CFD results.
compartment is written as
Before the ROM and the refrigerator model are validated, it is necessary
∑N dTair to ensure that the CFD model is reliable. Airflow rate and temperature
ṁin hin − ṁout hout + Qwall,i + Qsup + Qgasket + Qleak = ρVcp (29)
i dτ distribution are chosen for the reliability demonstration of the CFD
where Q is the heat flow, and the subscript wall, sup, gasket, and leak model. Relevant results are shown in Section 4.2 and Section 4.3.
respectively denote heat transfer with support structures (e.g. shelves), Additionally, the thermocouple locations, the cabinet parameter and
gaskets, walls, and other compartments. The Qgasket can be computed by corresponding boundary conditions can be found in Fig. 8, and Appen­
dix A.6 and A.7 in detail.
Liu’s empirical correlation [28].
The relative time constants and the temperature deviations are the
Qgasket = 0.045lgasket (Tamb − Tair ) (30) primary goals of the numerical calculations. To minimize the error

7
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Continuity equation:
∂ρ ∂(ρu) ∂(ρv) ∂(ρw)
+ + + =0 (33)
∂t ∂x ∂y ∂z
Momentum equations:
Component x
∂u 1 ∂p
+ div(uU) = div(ηgradu) − + Fx (34)
∂t ρ ∂x
Component y
∂v 1 ∂p
+ div(vU) = div(η grad v) − + Fy (35)
Fig. 7. Illustration of the one-dimensional envelope model. ∂t ρ ∂y
Component z
sources that do not produced by the ROM algorithm itself as far as
possible, it is necessary to ensure the precision of CFD results. Therefore, ∂w 1 ∂p
+ div(wU) = div(ηgradw) − + Fz (36)
a proper turbulent model is chosen then the calculation results are ∂t ρ ∂z
validated by experiments. The standard k-ε turbulent model is applied in Energy equation:
this paper, and following assumptions are involved in the CFD model.
∂(ρh)
+ div(ρUh) = − ρdivU + div(λgradT) + ST (37)
1) Incompressible flow ∂t
2) Boussinesq assumption The k-ε equations of the turbulent model are as follows:
3) Ignore the radiant heat transfer in compartments [( ) ]
∂(ρk) η
+ ρ(divU)k = div η + T ⋅divk + pk − ρε + Sk (38)
The conservation laws of the CFD model were as follows: ∂t σk

Fig. 8. 3D illustration of compartments for CFD models.

8
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

[( ) ]
∂(ρε) ηT ε ε2
+ ρ(divU)ε = div η+ ⋅divε + Cε1 pk − Cε2 ρ + Sε (39)
∂t σε k k
The turbulent kinetic viscosity ηt is computed as follows.

k2
ηt = c μ ρ (40)
ε
The five coefficients of the k-ε model, Cε1 , Cε2 , cμ , σ k and σε , are set to
be 1.44, 1.92, 0.09, 1.00, and 1.30, respectively, as suggested by Xiong
[29].
An open-loop CFD model is adopted here, where the inlet of the
model is set as the discharge port of the centrifugal fan and the outlet of
the model is set as the inlet of the evaporator. Boundary conditions can
be found in Appendix A.6.
Since the experiments are carried out following the GB/T 8059-2016
Standard, thermocouples with copper pillars should be used for
measuring temperature. Therefore, copper pillars must be included in
the CFD model to account for their impact on streamlines and the ve­
locity field (a comparison of CFD results with and without considering
the copper pillar thermocouples can be found in Appendix A.8. The 3D
model of the refrigeration compartment and the freezer compartment
that marks the locations of the test points are illustrated in Fig. 8.

4.2. Validation for airflow rate distribution

The modeling and simplification of the air duct determine the airflow
rate distributions in compartments, further affecting the cooling ca­
pacity distributions. In the compartment, cooling capacity distribution
in air-supply vents has a great influence on the temperature pull-down
rate in different tested locations in the pull-down stage, and the tem­
perature profile in the on–off stage.
The validation process can be summarized as:

1) Number the air-duct vents according to the compartment type and

vent location (Fig. 9);
2) Fix the Bluetooth hot-wire anemometer oppositely facing the airflow
vents. Close the refrigerator door and wait for the reading value to
reach steady. Keep measuring for five minutes then record the mean
value of air velocity;
3) Calculate the airflow rate based on the mean air velocity and the vent
area;
4) Iterate over all the numbered vents in the same way;
5) Compared the calculated airflow rate with the CFD results. The
comparison results are shown in Fig. 10 and Table A10 in the Sup­
plementary Materials.

Fig. 9. Locations of numbered air-supply vents.

Fig. 10 illustrates the comparison of simulated and tested airflow
rates, as well as their absolute and relative errors. Except for RC-L2, RC-
L3, and VC-L1, the relative errors of the simulated airflow rate are all simulated value.
within ±20 % (tabulated data in Table A10). The airflow rate mea­
surement contains some errors, for example, the errors from the hot-wire 4.3. Validation for temperature distribution
anemometer not well oppositely facing the air-supply vents and the
error due to turbulent variation of readings. Besides, the magnitude of In this section, the comparison of temperature distribution is con­
airflow rates is small, which may cause an excess relative error even with ducted to further validate the CFD model. The temperature in refriger­
a small airflow rate deviation. The airflow rate absolute errors at all the ator compartments periodically fluctuates, and thus the average
vents except for FC-R1 are within ±1 m3⋅h− 1. In conclusion, the temperature at each location over several on–off periods is selected for
simplification and modeling of the air duct are acceptable. comparison.
Moreover, for the refrigerator system model, it is not necessary to Fig. 11(a) illustrates the comparison results of RC. The simulated
specify the airflow rate of each vent. Instead, only the total airflow rate temperature profile in RC, except for Point 8, is in decent accordance
in each compartment is needed, because the inhomogeneous air distri­ with the experimental temperature profile. The absolute error between
bution is already accounted for in the ROM. At the level of system experimental and simulated temperature is within ±1.0 ◦ C. Point 8 is
simulation, the total measured airflow rate of RC is 14.05 m3⋅h− 1, which located near the vent in the 3rd layer of RC, where the local inner wall
is 0.78 m3⋅h− 1 (− 5.3 %) less than the simulated value. The total temperature could contribute to the extra cooling of this location. Since
measured airflow rate of VC is 1.88 m3⋅h− 1, which is 0.38 m3⋅h− 1 the CFD model does not account for the conduction of walls, the simu­
(− 16.8 %) less than the simulated value. And the total measured airflow lated temperature at RC8 is higher than the measured temperature. A
rate of FC is 45.14 m3⋅h− 1, which is 1.04 m3⋅h− 1 (− 2.3 %) less than the more sophisticated CFD model can theoretically improve such

9
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 10. Airflow rate distribution of experiment and simulation results.

Fig. 11. Temperature distribution comparison of the experimental and simulated results. (a) Temperature distribution in RC. (b) Temperature distribution in FC.

deficiency. Fig. 11(b) illustrates the temperature distribution compari­ significant pressure overshoot in the condenser. Meanwhile, the evap­
son of FC. FC has many differences from RC. The structure of FC is orator depletes its refrigerant, and the pressure of which plunges. 2) The
simpler while the space of FC is more compact, and the cooling supply of evaporating pressure sensor is arranged on the suction tube. In the
FC is more straightforward, which together makes the temperature starting stage, there is a sharp pressure gradient in the evaporating coils,
distribution in FC more uniform. The simulated temperatures in FC are which is caused by the imbalance of the refrigerant flow. The lowest
consistent with measured temperatures, where the absolute temperature pressure drop is the suction tube.
error is within ±1.0 ◦ C. The comparison results of RC and FC indicate As the system runs, the refrigerant in the condenser has been grad­
that the CFD model is valid for reduce-order modeling. ually condensing with the sub-cooling degree growing in the outlet of
the condenser. And the pressure difference between the condenser and
5. Simulation results and validation the evaporator increases. The high pressure in the inlet of the capillary
tube also induces the mass flow rate out of the condenser to increase
5.1. Cooling transients until it equalizes the inlet flow. As the system runs, RC will first reach its
setpoint temperature. At this moment, the RC damper closes, which
5.1.1. System transients induces a decrease in thermal load and is reflected in Fig. 12 (a) as a
The speed of the variable speed compressor in the experiment during sudden pressure drop, forming a ‘pit’ on the pressure curve. This char­
the initial turn-on process is difficult to obtain. Thus, a fixed speed with acteristic is not obviously observed in the experimental curve, partially
relaxation is applied to the compressor model for starting inertia: because the transients of air in the heat exchanger model and air duct are
not considered in the simulation. Additionally, the pressure in the
1
G(s) = (41) condenser changes more severely than that in the evaporator, as shown
τs + 1
in Fig. 12 (a). This is a result of the difference in inner volume between
where the time constant τ in Eq. (41) is set to be 500 s. The 500 s was the condenser and evaporator coils, which is 1.11 × 10-4 m3 and 2.14 ×
chosen based on matching the measured pressure transient with the 10-4 m3, respectively. A smaller inner volume leads to a more sensitive
simulated pressure transient to reproduce the initialization process for response to pressure.
the variable speed compressor. Fig. 12 (b) represents the absolute errors between simulated pressure
Fig. 12 (a) represents the pressure transients at the suction tube and and experimental pressure in the cooling stage. It is denoted that the
the discharge tube. At the first start of the refrigerator, the suction absolute errors of most points are within ±50 kPa, excluding the data
pressure will rapidly decrease to a considerably low value, and then points in the first hour of the pull-down stage. The excessive errors at the
slowly lifts, which is due to two reasons: 1) The evaporating fan is still very beginning of the curve are mainly caused by neglecting the pressure
off in the first 10 min after the compressor turns on. The inactive drop in the evaporator. For the condenser, a pressure error of 50 kPa will
evaporating fan leads to insufficient heat transfer and results in two- not affect much. For example, when the refrigerant (R-600a) is at 1000
phase refrigerant at the exit of the evaporator. Therefore, a great kPa, the corresponding saturated temperature is 66.2 ◦ C, as well as
amount of refrigerant, much more than that flows out from the capillary 63.9 ◦ C at 950 kPa and 68.4 ◦ C at 1050 kPa. The absolute error of the
tube, is compressed into the condenser in a very short time and causes a

10
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 12. System transients in the pull-down stage. (a) Comparison of simulated pressure and experimental pressure. (b) Absolute pressure error between simulated
pressure and experimental pressure. (c) Comparison of simulated evaporating temperature and experimental evaporating temperature. (d) Evaporating temperature
errors. (e) Comparison of simulated compressor power and experimental total power. Total power includes compressor power and power of auxiliary components
such as fans, anti-sweat heater, etc. (f) Absolute error of compressor power.

above saturation temperature is within ±3 ◦ C. fan, condenser fan, anti-sweat heater, and electronic control system.
Fig. 12 (c) and Fig. 12 (d) represent the comparison of the experi­
mental and simulated evaporating temperature and their absolute error, 5.1.2. Temperature field transients
respectively. The experimental evaporating temperature is tested by a To clearly describe the actuation state of components in the refrig­
thermo-couple arranged in the geometric center of the evaporator, erator, active and inactive components are designated to be “1″ and “0”,
sticking to the coil wall. The simulated evaporating temperature is the respectively. For example, ‘RC = 1’ means the air damper of RC is open,
saturated temperature corresponding to the homogeneous evaporating and ‘FC = 1’ means the compressor is running. At the beginning of the
pressure. Fig. 12 (c) shows that the tendency of simulated and experi­ pull-down process, RC = 1, FC = 1, since all compartments need to be
mental temperatures are consistent, and most of the temperature errors cooled. The evaporating fan will not start until 10 min later. Therefore,
are within ±5 ◦ C (Fig. 12 (d)). Two reasons cause the error: 1) evapo­ the cabinet air temperature will not reduce at once, which explains why
rating temperature is more sensitive to pressure since evaporating there is a plateau at the beginning of the temperature curves. At 180
pressure is very low; 2) the temperature in the air-return vent is min, RC reached the lower limit of the setpoint temperature and starts to
approximated by the lumped temperature in the simulation, which rewarm, with the air damper closing (RC = 0, FC = 1). When RC
affect the accuracy of evaporating temperature and pressure. rewarms to the upper limit of the setpoint temperature at 200 min, the
Fig. 12 (e) represents the comparison of the experimental and damper reopens (RC = 1). Meanwhile, RC keeps rewarming, because the
simulated compressor power. The experimental power curve periodi­ pull-down stage of FC finishes at 198 min. At 200 min, FC is still
cally fluctuates because the electrical anti-sweat heater operates peri­ rewarming and the compressor and the evaporating fan are off (FC = 0).
odically. In Fig. 12 (f), the absolute error of the simulated power with At about 220 min, the compressor and the evaporating fan restart (RC =
respect to the experimental power is from − 40 W ~ -10 W. The trend of 1, FC = 1).
the simulated power curve agrees well with that of the experimental Fig. 13 (a-i) illustrate the temperature transients of all nine tested
one. And the error is mainly from ignoring the power of the evaporator points and Fig. 13(j) shows the time-averaged temperature errors of

11
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 13. Temperature transients of RC in the pull-down stage. (a–i) Temperature transients of the nine tested points. (j) Time-average temperature error of all the
tested points.

these nine points. The definition of the time-average temperature error Fig. 14 (a–e) describe the temperature transients in the lower layer of
of the tested points can be described as: FC in the pull-down stage, and Fig. 14 (f–j) describe the temperature
transients in the upper layer of FC in the pull-down stage. In the
1 ∑t2
δT = δTi Δt (42) beginning, RC and FC both need cool capacity (RC = 1, FC = 1). At 180
t2 − t1 t1
min, RC reaches the lower limit of the setpoint temperature and rewarms
where t1 is the start time; t2 is the stop time; δTi is the absolute (RC = 0, and FC = 1). This action accelerates the cooling rate of FC
temperature error at i th time iteration, and Δt is the time step. because all the cool capacity is distributed to FC, which is reflected by
Six simulated temperature curves follow closely with the experi­ the increasing slope of the FC temperature curve. At 198 min, FC reaches
mental results. They are Point 2, Point 3, Point 6, Point 7, Point 8, and the lower limit of the setpoint temperature and starts rewarming. All the
Point 9; the time-average temperature error of these six points are all simulated points in FC have an acceptable fit to the experimental curves.
within 1 ◦ C. The time-average temperature error of Point 1, Point 4, and Fig. 14 (k) and Fig. 14 (l) show the time-average temperature error of
Point 5 are within 2.5 ◦ C, which is acceptable. The errors are mainly FC in the lower and upper layers. The temperature errors of all the
generated in the intermediate of the pull-down process. Taking Point 1 testing points, except Point 5 in the upper layer, are within 3.0 ◦ C. Four
as an example, it can be seen that the temperature error increases when points in the lower layer have errors within 2.0 ◦ C. The error of FCU5 is
the air starts to be cooled and converges to zero towards the end of the mainly observed in the intermediate of the pull-down stage, and the
pull-down stage. This observation implies that despite there being some overall temperature errors in the upper layer of FC are larger than those
uncertainty in the process of obtaining the time constant, the predicted in the lower layer. Because the drawers in the upper layer feature more
temperature deviation is acceptable. And only Point 1, Point 4, and Point airflow vents and less space, the turbulent airflow is more complex,
5 have non-trivial temperature errors than other points during the pull- leading to more simulation errors. Nonetheless, in the overall trend view
down stage. In summary, 6 out of 9 points reach excellent accuracy, and of the curves, the temperature transients of tested points are tracked
3 out of 9 points reach acceptable results. The ROM performs well in the well.
temperature transients simulation of RC in the cooling stage. Detailed
analysis of error sources will be addressed in the discussion section.

12
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 14. Temperature transients of FC in the pull-down stage. (a–e) Temperature transients of the five tested points in the lower layer of FC. (f–j) Temperature
transients of the five tested points in the upper layer of FC. (k) Time-average temperature error of points in the lower layer. (l) Time-average temperature error of
points in the upper layer.

5.2. On-off transients However, in the simulation, these thermal inertias are not modeled, and
thus the drastic change in load is immediately reflected by the change in
5.2.1. System transients pressure. Fig. 15 (c) shows the comparison results of evaporating tem­
Fig. 15 (a) and Fig. 15 (b) compare the simulated and experimental perature. The simulated trend of evaporating temperature is in excellent
results of the discharge pressure and suction pressure in the on–off stage, accordance with experimental results, where the evaporating absolute
respectively. It can be seen that the absolute error of discharge pressure error is within ±5.0 ◦ C in Fig. 15 (d). Fig. 15 (e) and Fig. 15 (f) represent
ranges from − 50 kPa to 25 kPa. There are mainly two error sources. The the comparison of the simulated compressor power and experimental
first one is neglecting the pressure drop. During the first half of each total power. The absolute error of simulated power to experimental
cycle, the simulated homogeneous evaporating pressure is always higher power is in the range of –20 W ~ 0 W, the reason for which is the same as
than the measured pressure at the suction line. The second one appears the pull-down stage. Additionally, the discharge pressure, as well as the
when RC is rewarming and FC is still in the cooling stage. In this stage, compressor power in the on–off stage is lower than their values at the
all cooled air is supplied to FC, generating a significant imbalance be­ end of the pull-down stage. This is because the rotation speed of the
tween load and cooling power. In the real system, this imbalance fades compressor in the pull-down stage is set to a higher value to shorten the
away gradually due to the thermal inertia of the air and air duct. consuming time in the first start.

13
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 15. System transients in the on–off stage. (a–b) Comparison of discharge pressure to suction pressure, and their absolute errors. (c–d) Comparison of simulated
evaporating temperature to experimental values and their absolute errors. (e–f) Comparison of simulated compressor power to experimental values and their
relative errors.

5.2.2. Temperature field transients phenomenon is explained in the discussion section.

Fig. 16 (a-i) show the trend of the experimental and simulated Fig. 17 (a–j) show the trend of the experimental and simulated
temperature of tested points in RC in the on–off stage. In the tested temperature of tested points in FC in the on–off stage. In these sub-
points of RC, there are about eight points in excellent accordance with figures, Fig. 17 (a–e) show the temperature transients in the lower
the experimental results. The mean errors of these points, except Point 8, layer, and Fig. 17 (f–j) show the temperature transient in the upper
are about 1.0 ◦ C (Fig. 16 (j)). Point 8 has an error of 2.5 ◦ C, which may be layer. In these subfigures, at 0 min, the refrigerator restarts, and
foreseen in Fig. 11. From Fig. 11, it is shown that the temperature de­ simultaneously RC and FC begin to cool down. The measured temper­
viation of Point 8 in RC is the largest at − 1.7 ◦ C. Other points, for ature curves of FC nearly keep stable or increase slowly. Because in the
example, Point 1, Point 4, and Point 5 have lower time-average tem­ on–off stage, the compressor works at a lower rotation speed and pro­
perature errors in the on–off stage than those in the pull-down stage. vides less cooling capacity to the refrigerator. The cooling capacity
This is mainly because the temperature varies more than 27 ◦ C in the distributed to FC hence is insufficient. However, the majority of simu­
pull-down stage but only fluctuates around 4 ◦ C in the on–off stage. The lated temperature curves of FC increase slowly, mainly due to high
wide temperature variation is more likely to induce an inaccurate evaporating temperature in this high-capacity condition (Fig. 15 (c)).
temperature tracking of simulation, which appears during the interme­ After RC reaches the lower limit of setpoint temperature, the FC cooling
diate of the pull-down process. In general, The ROM obtains acceptable rate of measured curves increases, which is well captured by the simu­
results in simulating the distributed temperature transients in RC at lated curves. This is because after the RC damper closes, all of the
finite user-defined locations. The time-average error of Point 8 is more cooling capacity from the evaporator is transported into FC. During the
noteworthy than that in the on–off stage. The reason for this on–off stage, the time-average temperature errors are within 1.5 ◦ C.

14
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 16. Temperature transients of RC in the on–off stage. (a–i) Temperature transients of the nine tested points. (j) Time-average temperature error of all the
tested points.

Additionally, the time-average temperature errors in FC are more uni­ reduced-order CFD model swiftly.
formly distributed than those in RC. Here, the authors discuss the possible simulation error sources. The
ROM features two key fitting parameters, time constant and temperature
6. Discussions deviation. The time constant describes the cooling or heating dynamic
behavior of specific points. By the definition of time constant, the
Overall, the ROM-based refrigerator model well describes the tran­ experiment-based time constant can be extracted from the measured
sient temperature field as well as the refrigeration system performance, temperature of the two points. Here the authors take the rewarming
thus demonstrating the feasibility of the proposed ROM technique for process of RC5 and RC8 as an example. The experimental time constants
future refrigerator development. Because the cabinet model and the of RC5 and RC8 are 16,974 s and 23,688 s, respectively.
refrigeration system model are simply coupled by the ROM, a validated Fig. 18 shows three kinds of temperature curves, the standard
cabinet model can be matched to different designs of the refrigeration exponential function curve (denoted as standard), the measured curve,
system, and instructive results can be obtained to optimize the system and the simulation curve, of RC5 and RC8. The measured curve is ob­
design parameters. The ROM-based refrigerator model predicts the tained from true experimental data and corrected by calibration results
temperature field transients that are impractical to be solely captured by of thermocouples. The simulation curve is the result of the refrigerator
the refrigeration system model. Moreover, compared to the conven­ model used in this article. And the standard curve is the curve of Eq.
tional transient CFD model, applying the ROM significantly shortens the (43).
simulation time, because calibrating the ROM parameters skips the
dTi Ti,0 − Ti
thermal mass of the solid structures, which dominates the simulation = (43)
dt τ
time. Additionally, a skilled structure engineer is able to establish the

15
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 17. Temperature transients of FC in the on–off stage: (a–e) Temperature transients of the five tested points in the lower layer. (f–j) Temperature transients of the
five tested points in the upper layer. (k) Time-average temperature error of points in the lower layer. (l) Time-average temperature error of points in the upper layer.

where τ are 16,974 s and 23,688 s, respectively, for RC5 and RC8. Ti,0 components of the refrigerator exchange heat with the surrounding air
is the temperature of RC5 and RC8 when the rewarming process ends. continuously, the temperature of user-defined points in the temperature
Fig. 18 shows the standard curve deviates from the experimental field is affected. This kind of error is amplified in the pull-down stage but
curve. Because the standard curve is calculated via fixed time constants, weakened in the on–off stage according to the temperature-varying
whereas the real-time constants vary over the whole rewarming process characteristics.
as a result of the thermal mass and variable heat leak of the envelope In the on–off stage, the errors due to the time constant are weakened,
components, the support structure, heat leakages, etc. There are also in- while the additional error due to the thermo-bulb becomes more sig­
conformities between the simulation curve and the measured curve, nificant. The air dampers are controlled based on the thermo-bulb. In the
possibly because the time constants calibrated from the CFD model simulation model, the thermo-bulb temperature is obtained by ROM and
ignore the influence of inhomogeneous temperature distribution of its thermal mass. In reality, in addition to the impact of local air tem­
supporting structures and walls in the cabinet. When the internal perature, the thermo-bulb temperature is also affected by heat

16
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

Fig. 18. Temperature transients comparison between experiment and experiment-based simulation, taking RC5 and RC8 as examples.

conduction in the air duct, which generates a new source of error that down stage. The pressure error significantly affects the evaporating
mainly affects the on–off stage. As a result, in the pull-down stage, the temperature and the supply air temperature because the saturation
measured temperature of RC5 is lower than the simulated temperature temperature of R600a is very sensitive to pressure at the low-
(Fig. 13 (e)). But in the on–off stage, the measured temperature of RC5 pressure condition. Moreover, the pressure drop in the evaporator
rises (Fig. 19 (a)), and the error decreases. In the pull-down stage, the also dictates the suction flow rate when the compressor restarts. In
measured temperature of RC8 is in good accordance with the simulated addition, the pressure drop influences the location of the transition
temperature (Fig. 13 (h)), while in the on–off stage, the measured from the two-phase region to superheat region in the evaporator.
temperature of RC8 reduces (Fig. 19 (b)), and the error increases. • Heat exchanger configuration In this study, a one-dimensional model is
In summary, the noticeable error characteristics of RC5 and RC8 in applied for heat exchangers. For the hot-wall condenser, this
different operation stages have their internal rationality. The errors are approach is well justified. However, the tube-fin evaporator features
mainly due to simplifications of the submodels involved in the refrig­ a cross-flow configuration. Thus, a higher dimensional heat
erator model. More efforts should be conducted in further work to exchanger model can improve the prediction accuracy for the
further improve the accuracy of these sub-models. evaporator in frost-free refrigerators.
• Frost thermal resistance Frost accumulating and its impact on evapo­
1) Refrigeration system model rating pressure and temperature should be accounted for in future
studies.
The refrigeration system transients have certain deviations from 2) Reduced-order temperature field model
experimental results, which contribute to part of temperature transient • Precision enhancement The proposed ROM in this study does not ac­
errors in compartments. Therefore, the refrigeration system model count for the temperature distribution of internal structures, which
should be improved. affects the air temperatures that are close to the wall. As discussed
above, it also determines the accuracy of the temperature readings of
• Refrigerant distribution The refrigerant distribution directly influences thermo-bulbs in different compartments, which in turn affects the
the operating characteristics, such as the transients of suction pres­ accuracy of the ROM. A form of modified ROM that addresses these
sure, discharge pressure, and evaporating temperature, of the issues will be tested in the future.
refrigerator system. In the current refrigeration system model, the • Application range To improve food storage quality, proper humidity
accumulator charge ratio is too high, and the liquid line charge control is very important. The humidity, as well as the coupling of
(including the filter dryer) is ignored, which affects the starting heat and moisture transfer, is meant to be studied. It will be a chal­
transients of the evaporator. Applying a more suitable void fraction lenging future task to develop a more sophisticated ROM that can
model and considering the liquid line charge can improve the ac­ simultaneously simulate the temperature field and humidity field.
curacy of refrigerant distribution and refrigeration system transients.
• Pressure drop In this study, the pressure drop in heat exchangers is
ignored. This simplification causes the pressure error in the pull-

Fig. 19. Measured temperature transients of RC5 (a) and RC8 (b).

17
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

7. Conclusions Acknowledgment

A first principle transient model is established for domestic frost-free This research was supported by Hisense Refrigerator Co. Ltd. and the
refrigerators in this study. The key novelty of the proposed model is the Young Elite Scientists Sponsorship Program by CAST under grant No.
reduced-order description of the temperature field, which enables co- 2019QNRC001. We would also like to acknowledge Dr. Hongtao QIAO
simulation between the temperature field and the other parts of the and Binglong HAN for the valuable discussions on refrigeration system
refrigerator. The methodology of co-simulating the refrigerator system simulation techniques.
and the compartment temperature field, as well as full-scale validation,
are systematically demonstrated in this paper and summarized as Appendix A. Supplementary Materials
follows:
Supplementary data to this article can be found online at https://doi.
• The ROM was validated against a full-order CFD model. At the load- org/10.1016/j.applthermaleng.2023.121412.
free condition, the major errors between predicted temperatures
from the ROM and the conventional transient CFD results are within References
1.5 K. The validation reveals two facts: 1) The ROM can successfully
describe the temperature distribution transients in compartments. 2) [1] L. Dupont J., P. Domanski, P. Lebrun, F. Ziegler, The role of refrigeration in the
global economy, in: 38th Note on Refrigeration Technologies, IIF-IIR, 2019.
The ROM accuracy depends on the CFD model because the two ROM [2] D. Srikanth, Non-traditional security threats in the 21st century: a review, J. Dev.
parameters are derived from CFD results. Stud. 4 (2014) 60–68.
• A semi-physical bisection diabatic capillary tube model is proposed, [3] Y. Yang, D. Huang, R. Zhao, W. Guo, X. Wei, Analysis of temperature rapid rise
phenomenon during damper-off cycle in side-by-side frost-free refrigerator, Int. J.
which is the combination of a recuperator section and an adiabatic Refrig 133 (2022) 201–213.
pressure drop section. The adiabatic part is modeled by Yang’s cor­ [4] O. Laguerre, S. Benamara, D. Flick, Numerical simulation of simultaneous heat and
relation, and the recuperator section is modeled by the ε-NTU moisture transfer in a domestic refrigerator, Int. J. Refrig 33 (7) (2010)
1425–1433.
method with a significantly detailed thermal resistance network di­ [5] J.A. Alfaro-Ayala, A.R. Uribe-Ramírez, J.I. Minchaca-Mojica, J.D.J. Ramírez-
agram. The model is validated with an experimental data bank Minguela, B.U. Alvarado-Alcalá, O.A. López-Núñez, Numerical prediction of the
including 30 cases. The maximum predicted mass flow rate errors are unsteady temperature distribution in a cooling cabinet, Int. J. Refrig 73 (2017)
235–245.
within 20 %, and 80 % of the total errors are within 15 %. Addi­
[6] D.S. Jung, R. Radermacher, Performance simulation of a two-evaporator
tionally, a one-dimensional envelope model is developed and trans­ refrigerator—freezer charged with pure and mixed refrigerants, Int. J. Refrig 14 (5)
formed into the state-space form, which facilitates implementation. (1991) 254–263.
• The CFD model is proved to be reliable as the steady-state temper­ [7] D.S. Jung, R. Radermacher, Performance simulation of single-evaporator domestic
refrigerators charged with pure and mixed refrigerants, Int. J. Refrig 14 (4) (1991)
ature errors are within ±1 ◦ C compared to the experiment. In the 223–232.
pull-down stage, the condensing and evaporating pressure errors are [8] V. Vidmar, B. Gaspersic, Dynamic simulation of domestic refrigerators with
within ±50 kPa. The error in evaporating pressure caused an error in refrigerants R12 and R134a, in: Montreal, Quebec, Canada, 1991, pp. 1250–1254.
[9] P.K. Bansal, T.C. Chin, Design and modelling of hot-wall condensers in domestic
evaporating temperature of 5 ◦ C, which does not have a significant refrigerators, Appl. Therm. Eng. 22 (14) (2002) 1601–1617.
impact on the compartment temperature fields. The absolute error of [10] E. Lin, G. Ding, D. Zhao, Y. Liao, N. Yu, J. Yamashita, Dynamic model for multi-
refrigerator power is observed from − 40 W to − 10 W. Regarding the compartment indirect cooling household refrigerator using Z-transfer function
based cabinet model, Int. J. Therm. Sci. 50 (7) (2011) 1308–1325.
temperature field, the majority of time-average temperature errors [11] B.N. Borges, C. Melo, C.J.L. Hermes, Transient simulation of a two-door frost-free
are within 1.0 ◦ C in the refrigeration room, 2.5 ◦ C in the lower layer refrigerator subjected to periodic door opening and evaporator frosting, Appl.
of the freezing room, and 3.0 ◦ C in the upper layer of the freezing Energy 147 (2015) 386–395.
[12] C.J.L. Hermes, C. Melo, A first-principles simulation model for the start-up and
room. cycling transients of household refrigerators, Int. J. Refrig 31 (8) (2008)
• In the on–off stage, the condensing and evaporating pressure errors 1341–1357.
are within ±25 kPa. The evaporating pressure error led to a 5 ◦ C [13] R. Radermacher, E. Gercek, V.C. Aute, Transient simulation tool for refrigeration
systems, in: Vicenza, Italy, 2005.
error in evaporating temperature. The absolute error of refrigerator
[14] G.-L. Ding, H.-T. Qiao, Z.-L. Lu, Ways to improve thermal uniformity inside a
power was observed from − 20 W to 0 W. Regarding the temperature refrigerator, Appl. Therm. Eng. 24 (13) (2004) 1827–1840.
distribution, the time-average temperature errors of the refrigeration [15] J. Hyuk Wie, H. Woo Cho, Y. Gap Park, Y. Soo Kim, Y. Min Seo, M. Yeong Ha,
room are within 1.1 ◦ C, and the time-average temperature errors of Temperature uniformity analysis of a domestic refrigerator with different multi-
duct shapes, Appl. Therm. Eng. 188 (2021), 116604.
both layers of the freezing room are within 1.5 ◦ C. [16] E.M.A. Mokheimer, Y.S. Sanusi, Comparative analysis of different configuration
domestic refrigerators: a computational fluid dynamics approach, J. Energy Res.
The thorough experimental validation of the newly proposed Technol. 137 (2015), 062002.
[17] Z. Wang, J. Li, T. Li, S. Li, Research on Simulink/Fluent collaborative simulation
reduced-order temperature field model provides a new paradigm of co- zooming of marine gas turbine, Appl. Comput. Intell. S. 2017 (2017) 1–8.
simulating the temperature field CFD model and the refrigeration system [18] H. Xie, D. Zhang, L. Shen, Collaborative simulation based on MATLAB/Simulink
model, despite certain errors, that can be improved by advanced models and FLUENT, J. Syst. Simul. 19 (2007) (1856) 1824–1827 (in Chinese).
[19] T. Han, S.Q. Tang, C.Q. Huang, K.S. Dong, FLUENT/SIMULINK collaborative
that consider refrigerant pressure drop and thermal inertias of air duct, simulation for missile separation from cavity, Adv. Mat. Res. 591–593 (2012)
etc. The ROM is of significant application value for product 1902–1906.
development. [20] C. Allery, C. Béghein, A. Hamdouni, Applying proper orthogonal decomposition to
the computation of particle dispersion in a two-dimensional ventilated cavity,
Commun. Nonlinear Sci. 10 (8) (2005) 907–920.
Declaration of Competing Interest [21] A. Sempey, C. Inard, C. Ghiaus, C. Allery, Fast simulation of temperature
distribution in air conditioned rooms by using proper orthogonal decomposition,
Build. Environ. 44 (2) (2009) 280–289.
The authors declare that they have no known competing financial
[22] N. Ablanque, C. Oliet, J. Rigola, A. Oliva, Numerical simulation of non-adiabatic
interests or personal relationships that could have appeared to influence capillary tubes. special emphasis on the near-saturation zone, Int. J. Refrig 55
the work reported in this paper. (2015) 153–167.
[23] P.K. Bansal, G. Wang, Numerical analysis of choked refrigerant flow in adiabatic
capillary tubes, Appl. Therm. Eng. 24 (5-6) (2004) 851–863.
Data availability [24] S. Yang, W. Tao, Heat Transfer, Higher Education Press, Beijing, 2006.
[25] L. Yang, W. Wang, A generalized correlation for the characteristics of adiabatic
All data were listed in the paper capillary tubes, Int. J. Refrig 7 (2008).
[26] C. Melo, L.A. Torquato Vieira, R.H. Pereira, Non-adiabatic capillary tube flow with
isobutane, Appl. Therm. Eng. 22 (14) (2002) 1661–1672.

18
S. Xu et al. Applied Thermal Engineering 235 (2023) 121412

[27] D. Sarker, J.H. Jeong, Development of empirical correlations for non-adiabatic [29] L. Xiong, Y. Lin, S. Li, k-ε Turbulent model and its application to the FLUENT, Ind.
capillary tube based on mechanistic model, Int. J. Refrig 35 (4) (2012) 974–983. Heat. 36 (2007) 3 (in Chinese).
[28] G. Liu, G. Yan, J. Yu, Research on test method of heat transfer coefficient for
refrigerator gasket, Int. J. Refrig 110 (2020) 106–120.

19