Professional Documents
Culture Documents
International Journal of Heat and Mass Transfer
International Journal of Heat and Mass Transfer
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents a first-principle lumped-parameter dynamic model and experimental validation of
Available online 6 August 2013 vapor compression cycles for electronics cooling. The model couples the dynamics of the heat exchangers
with static empirical models for compressor and expansion valve. In contrast to past work on systems
Keywords: level modeling of refrigeration cycles, this paper focuses on imposed heat flux boundary condition, and
Vapor compression cycle the associated critical heat flux and critical vapor quality, in the evaporator. Using our vapor compression
Electronics cooling cycle testbed, we verify that the model prediction of the evaporator exit temperature and critical heat
Critical heat flux
flux matches well with experimental measurements. The model is also used to search for operating con-
Dynamic modeling
Critical heat flux enhancement
ditions to enhance the critical heat flux. Experimental results show that at an undesired operating con-
dition, even a small 5% change of heat flux could cause a wall temperature spike of over 100 °C, in
contrast to 15 °C at a more advantageous operating conditions. For large heat flux transients, the onset
of critical heat flux condition may be delayed, but its avoidance may require active refrigerant flow
control.
Ó 2013 Elsevier Ltd. All rights reserved.
0017-9310/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.06.075
912 J. Catano et al. / International Journal of Heat and Mass Transfer 66 (2013) 911–921
Fig. 1. Vapor compression cycle (a) Schematic of system components for traditional VCC (b) thermodynamic cycle of a traditional VCC in a Pressure vs. Enthalpy diagram
where the exit of the evaporator and inlet to the compressor is saturated vapor (c) Schematic of system components for VCC for electronics cooling (d) thermodynamic cycle
of VCC for electronics cooling in a Pressure vs. Enthalpy diagram where the exit of the evaporator is a liquid–vapor mixture.
type of system modification is required to guarantee safe operation lumped-parameter model is suitable for analysis and design itera-
of the compressor (Fig. 1(d)). tion as it is much less computationally demanding, but also at
This paper develops a comprehensive VCC dynamic model for much lower accuracy especially during start-up transients [18].
systems with imposed heat flux at the evaporators (as in the In traditional VCC systems, the evaporator is usually divided
cooling of high-power electronics). The heat exchangers, i.e., evap- into two regions: a two-phase region from the entrance of the
orators and condensers, are approximated by their spatially aver- evaporator until the flow reaches a saturated vapor condition,
aged behavior (one-zone model, since the distributed behavior is and a second region for superheated vapor. Similarly the condenser
approximated by a single set of lumped variables). The dynamic is divided into three regions: a superheated vapor region, two-
model includes prediction of the CHF condition by modeling the phase region and a subcooled region. The length of these regions
vapor quality at the onset of CHF (xcrit). Experimental validation varies with time and are modeled as additional state variables. This
is performed away and around the CHF condition. The refrigerant formulation is referred as the moving-boundary method [19,15].
used in the experimental system is R134a and the thermodynamic One challenge of this approach is that during large transients some
properties are obtained from published data by the National Insti- of these zones disappear or are created which result in varying
tute of Standards and Technology (NIST) [11]. The model is also number of dynamic states. In [20], a multi-model approach is
used to improve the operating condition by increasing CHF or proposed with switching between them as needed. In [21], a
reducing power consumption. The improved operating condition pseudo-quality is used that can be larger than one in the evapora-
is shown to withstand a step heat load change versus a nominal tor or lower than zero in the condenser, avoiding the need for mod-
operating condition which exhibits rapid temperature rise. el switching. This formulation is simpler to implement and is
expected to perform well for evaporators operating close to the
saturation condition, hence it is selected in our research.
2. VCC modeling We impose the following assumptions to simplify the physical
model:
Dynamic models for VCCs include four main components: evap-
orator, compressor, condenser and expansion valve [12]. The evap- Two-phase flows in evaporator and condenser are considered
orator and condenser are modeled using mass, momentum and homogeneous and in thermal equilibrium.
energy conservation equations. Expansion valve and compressor Axial heat conduction in evaporator and condenser is negligible.
are modeled using static equations involving empirical parameters Pressure drop in evaporator and condenser is neglected.
(discharge coefficient, and volumetric and isentropic efficiencies). Dynamics of compressor and expansion valve are much faster
The evaporator and condenser are distributed parameter systems than that of the heat exchangers and are modeled as static
modeled by a set of partial differential equations [13,14]. There components.
are a number of lumped-parameter approximation approaches The complete system is perfectly insulated (no heat losses or
driven by systems level design and control needs [15–17]. The heat gains from the ambient).
J. Catano et al. / International Journal of Heat and Mass Transfer 66 (2013) 911–921 913
Table 1
Comparison of evaporator modeling for control-oriented models of vapor compression cycles.
kg hms ha
Hedry ¼ 0:023Re0:8 0:4
g Pr g ; ð12Þ hm ¼ ha þ ; ð20Þ
De gs
where Pr is the Prandtl number and Re is the Reynolds number de- where gv and gs are given by the following correlations:
fined as
b3
cp lg Po
Pr ¼ g ; ð13Þ gv ¼ b1 þ b2 ; ð21Þ
kg Pi
_ e
mD 1 Po 1 1
Reg ¼ : ð14Þ gs ¼ c1 þ c2 þ c3 þ c4 þ c5 ðqi qsat Þ: ð22Þ
lg x Pi Pi _
m
The selection between the correlations is determined by xcrit: The constant parameters in (21) and (22) are identified using exper-
( imental data. The details may be found in [25].
Hewet if xe < xcrit ;
He ¼ ð15Þ
Hedry if xe P xcrit : 2.4. Condenser
The correlation to predict xcrit is presented in Section 4.2. A more The condenser is typically modeled as two or three zones, but
complete model for He would include multiple segments, e.g., sep- for simplicity we use only one zone for the model. As the evapora-
arately for dryout inception and dryout completion as in [23,24]. tor, the condenser is modeled using mass balance, refrigerant en-
We choose the simplified form here for the ease of stability analysis ergy balance, and wall energy balance. After integrating over the
and control design. length of the condenser, we obtain the lumped-parameter
Putting the state equations together, the final lumped-parame- approximation:
ter model for the evaporator is given by:
2 3 2 _ o Þqf qg =ðAcs Lqe Þ
_ im 3 dqc 1
x_ e ðm Lc ¼ ðm_ im
_ o Þ; ð23Þ
6 7 6 7 dt Ac
Ze6 _ 7 6 _ o he Þ=ðAc Lqe Þ 7
_ i hv m
4 Pe 5 ¼ 4 ðHe Ase ðT we T e Þ þ m 5; ð16Þ
_T we ðqe He Ase ðT we T e ÞÞ=ðcp V qÞ dðqc hc Pc Þ 1 1
w Lc ¼ ðm_ i hi m
_ o ho Þ þ Hc Asc ðT wc T c Þ; ð24Þ
dt Ac Ac
where the matrix Ze is given by Table 2.
dT wc
2.2. Accumulator cpwc qwc V wc ¼ Hc Asc ðT wc T c Þ Hcool Ascool ðT wc T cool Þ: ð25Þ
dt
The accumulator is modeled as a static heat exchanger using The two-phase flow properties are calculated as in (7) and (8) with
refrigerant energy balance and assuming that the pressure drop (xe,Pe) replaced by (xc,Pc). The heat transfer coefficients for the
is negligible: refrigerant and cooling side are model as:
qa _ o þ 246:3;
Hc ¼ 122108m ð26Þ
ha ¼ he þ ; ð17Þ
m_
kcool
Pa ¼ Pe : ð18Þ Hcool ¼ 0:023Re0:8 0:4
cool Pr cool ; ð27Þ
Dcool
We have ignored the dynamics of the accumulator in this paper,
where the constant parameters in (26) have been obtained by fit-
which would account for some of the mismatch in the pressure pre-
ting the experimental data [25]. The overall lumped-parameter
dictions. This dynamics will be addressed in our future work.
model for the condenser in matrix form is
2 3 2 3
2.3. Compressor x_ c ðm_ im _ o Þqf qg =ðAcsc Lqc Þ
6 7 6 7
Z c 4 P_ c 5 ¼ 4 ðHc Asc ðT wc T c Þ þ m _ i hv m_ o hc Þ=ðAc Lqc Þ 5;
Because the dynamics of the compressor are expected to be
much faster than that of the heat exchangers in the system T_ wc ðHc Asc ðT c T wc Þ Hcool As ðT wc T col ÞÞ=ðcp V qÞ
cool w
2 3
Table 3 x_ e
Values of Matrix Zc for the condenser model. 6 P_ 7
6 e 7
6 7
Zc(1,1) ðA Lq Þ
qc ðqf qg Þ qc q c Z e 0 6 T_ we 7
f g
dq 6 7
Zc(1,2) q
qf qc ð1 xÞ dPcg þ ðqg xqc Þ dPfc ðAqc Lqqc Þ 0 Z c 6 x_ c 7
6
7
f g
6 _ 7
Zc(1,3) 0
4 Pc 5
Zc(2,1) ðhg hf Þqf qg hc qc ðqf qg Þ ðAqc Lqqc Þ
f g
T_ wc
Zc(2,2) dhf
qf qg ð1 xÞ dP dh dq dq
þ x dPgc þ q1 þ hc ðqg xqc Þ dPcf þ ðqf qc ð1 xÞÞ dPcg 2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3
c c ðkv Av qc ðPc Pe Þ qa V m xgv Þqf qg =ðAcs Lqe Þ
ðAc Lqc Þ 6 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7
qf qg 6 ðHe As ðT we T e Þ þ kv Av qc ðPc Pe Þhv qa V m xgv he Þ=ðAc Lqe Þ 7
Zc(2,3) 0 6 7
6 ðqe He As ðT we T e ÞÞ=ðcp V qÞw 7
Zc(3,1) 0 6 7
¼6 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7:
Zc(3,2) 0 6 ð q V xg k v A v q ðP P Þ Þ 7
Zc(3,3) 1 6 a m v c c e
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7
6 7
4 Hc Asc ðT c T wc Þ þ qa V m xgv hm kv Av qc ðPc Pe Þhc 5
Hc Asc ðT c T wc Þ Hcool Ascool ðT wc T cool Þ
ð32Þ
2.6. Complete dynamic model
3. Experimental setup
The complete lumped-parameter model of the VCC is given by
the nonlinear differential equation below with 6 states (xe,Pe,Twe,-
Fig. 2(a) presents the experimental system designed and con-
xc,Pc,Twc) and 4 inputs (qe,qa,x,Av):
structed to validate the mathematical models and demonstrate
the control algorithms. It consists of three cartridge heaters im-
Fig. 2. Experimental setup (a) Picture of multiple-evaporator vapor compression refrigeration cycle experimental setup for electronics cooling (b) Schematic of experimental
setup.
916 J. Catano et al. / International Journal of Heat and Mass Transfer 66 (2013) 911–921
To validate the model away from the CHF condition, step pulses
with varying amplitudes are imposed in x and Av, starting from a
steady state condition. Fig. 3 shows the dataset used for parameter
identification of system constants such as diameters and lengths of
the heat exchangers (geometry details are not provided by the con-
denser manufacturer). Fig. 3(a) shows multiple simultaneous ran-
dom pulses for x and Av with a constant heat load and the
measured and simulated mass flow rate. Fig. 3(b) shows the mod-
eled variables at the exit of the condenser and at the exit of the
evaporator. Fig. 4 shows additional datasets not used for identifica-
tion. The model is capable of capturing the overall trends for most
of the system dynamics. And is sufficient for its main purpose of
feedback control design and operating condition selection. The
modeling errors are attributed to the simplifications made to the
condenser (one-zone) and accumulator (static component).
Fig. 6. Validation of critical quality calculation from Eq. (33) (a) Critical quality
identification (b) Critical quality prediction measurements not used for the
identification.
2 3 !
P P P
xcrit ¼ ð8=DÞ0:15 0:55 0:642 þ 0:356 0:0637
98 98 98
Fig. 4. Experimental data used for model prediction, (a) Pulses in x with constant 0:5
G
qe and Av and measured and calculated mass flow rate (b) Measured and simulated : ð34Þ
values at the exit of condenser and evaporator. 1000
Note that most of the coefficients in (34) have different signs. This is
not surprising due to the different fluid properties and geometry
used to fit the empirical correlation.
The experiments to estimate xcrit are performed at different
operating conditions by varying Av,x,qe, and qa. This procedure al-
lows for a reasonable range in m,_ and Pe (3–7 kg min1 and 158–
230 kPa). However, due to the coupling dynamics of the cycle,
other system variables, such as the inlet quality to the evaporator,
vary as well, making a complete parametric study impractical with
the current experimental setup. From the experimental data xcrit is
2
estimated at the onset of the temperature spike, that is when d dtT2we
is at a maximum. The correlation (34) is used to predict data not
used in the identification as shown in Fig. 6(b) and the predicted
values agree well with measurements.
To validate the response of the dynamic model (32) near CHF,
the heat load qe is increased until CHF is reached as shown in
Fig. 7(a). The response of the system at the exit of the evaporator
Fig. 5. Experimental boiling curve around the critical heat flux, the red line
indicating the conditions obtained during increasing qe and the black line shows the
is presented in Fig. 7(b). The model is capable of predicting the on-
path of decreasing qe. set of CHF and the magnitude of the temperature rise. However,
the model shows an instantaneous drop in Pe and xe that does
not corresponds to the actual physics of the system. The drop in
imental measurements as shown in Fig. 6(a). The identified equa- both variables is due to the drop in heat transfer coefficient once
tion for our refrigeration system is the CHF condition is reached. In the physical system this drop is
918 J. Catano et al. / International Journal of Heat and Mass Transfer 66 (2013) 911–921
Fig. 7. Identification of critical heat flux model (a) Steps in qe for dynamic model
validation of CHF prediction (b) Dynamic response at the exit of evaporator during Fig. 8. Prediction of critical heat flux model with data not used for model
CHF with xcrit = 0.894. identification (a) System inputs (b) Dynamic response at the exit of evaporator
during CHF prediction.
not present since CHF is a local phenomena that only affects the lo-
cal wall temperature, with small effect on Pe and xe. This effect
could be avoided by using a distributed model or by incorporating
an additional state in the evaporator which would model the local The effect of xv is as expected where a lower value represents
wall temperature, which couples to the remaining states of the more liquid in the two-phase region and lower xe for a fixed value
evaporator in only one direction. In this formulation changes in of qe. The effect of Pe is not as expected [9], since from (34) an in-
Pe and xe would affect the new state however changes in the new crease in Pe decreases xcrit. However, higher Pe usually results in a
state would not affect Pe or xe. Additional experiments have been higher Pc, hence a higher temperature difference between the
conducted for model validation as shown in Figs. 8(a) and 8(b). refrigerant and the heat sink. At this new operating condition the
The model is not capable of predicting both transitions in heat exit of the condenser can have a higher degree of subcooling,
transfer coefficient present in the experimental data. However, decreasing hc and xv at the evaporator inlet. Similarly, the effect
for controller design, the modeled dryout condition is sufficiently of m_ is not straightforward, where higher m _ decreases xcrit by
close to the dryout inception. increasing the liquid droplets entrapped in the vapor core, at the
same time it decrease xe for a fixed qe. This trends are presented
4.3. Critical heat flux enhancement in Figs. 9(a), 9(b) and 9(c).
The closed cycle structure of the system means changes in
From the experimental data near CHF as shown in Fig. 9, we operating conditions affect both the exit and inlet conditions at
make the following observations of CHF in a VCC with imposed the evaporator. For example, an input change can have a detri-
heat flux: mental effect on the exit conditions, such as a decrease in xcrit,
coupled with an positive effect on the inlet conditions which
1. Decreasing evaporator inlet flow quality, xv, increases CHF. can improve the overall CHF of the cycle. The CHF condition is
2. Increasing Pe increases CHF. also directly related to qa, since increasing qa means more active
3. Increasing m_ increases CHF. charge in the VCC, leading to rising Pe and Pc. Experimentally, an
J. Catano et al. / International Journal of Heat and Mass Transfer 66 (2013) 911–921 919
(a) 7.5
700
800
7 680
CHF (W/cm )
700 660
Wm (W)
2
6.5
640
600
620
6
500
4 600
6
5.5 40 580
8 50
10 60 560
Av (%) 70
5
0.23 0.24 0.25 0.26 0.27 0.28 0.29 0.3 0.31 12 80 ω (Hz)
xv
(a)
(b) 7.5 0.1
0.12
0.09
0.1
7 0.08
0.08
Xavail
CHF (W/cm2)
0.06 0.07
6.5
0.04 0.06
0.02
4 0.05
6 6 40
8 50
60 0.04
10 70
A (%) 12 80 ω (Hz)
v
5.5
(b)
5
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3
5 Fig. 10. Modeled system response for different operating conditions with
Evaporator Pressure (Pa) x 10
qe = 900 W and qa = 300 W (a) Compressor power (b) Available quality.
(c) 7.5
850
7
900
CHF (W/cm2)
W (W )
800
6.5 800
700
m
750
6 600
4
40 700
6
5.5 50
8 60
Av ( %) 70 650
10 80
ω (Hz)
5
0.006 0.007 0.008 0.009 0.01 0.011 0.012 0.013 0.014
Mass flow rate (kg/s) (a)
Fig. 9. Effect of various VCC operating conditions on CHF (a) Effect of xv (b) Effect of −3
−3 x 10
_
Pe (c) Effect of m. x 10
20
15
15
increase of 250 W in qa, with fixed values of Av and x, increased 10
xavail
CHF by 22.7%. 10
5
0
5
−5
5. VCC performance optimization 4
40
6 50
8 60
70 0
We use the lumped-parameter model to predict important sys- A ( %) 10 80
v ω (hz)
tem metrics for different operating conditions such as compressor
power consumption and available quality defined as (b)
Fig. 11. Modeled system response for different operating conditions with
xav ail ¼ xcrit xe : ð35Þ
qe = 1100 W and qa = 300 W (a) Compressor power (b) Available quality.
Optimal
100 the systems is away from CHF and avoid or delay CHF for operating
Nominal conditions with high heat loads.
50
However, this steady-state optimization is only useful from
0 small disturbances around the selected operating point. Large dis-
0 500 1000 1500 2000 2500 3000 3500
Time (s) turbances cause the system to deviate from the desired operating
condition and can lead to CHF as shown experimentally in
Fig. 12. Comparison between a nominal operating condition (ARI) and an operating Fig. 13. In this case, larger changes in qe are imposed to the maxi-
condition with maximum xavail. mum xavail operating condition at qe = 1000 W (Av = 4.5%, x = 45 Hz
and qa = 300 W). In Fig. 13, two step changes in qe are imposed
from steady-state condition. First qe is decrease to 700 W which
1500 decreases Twe accordingly. From the new steady-state condition
at qe = 700 W, a large step change is imposed to qe = 1330 W, rep-
q (W)
1000 resenting a 90% increase from the steady-state value. For this large
e
step the system reaches CHF almost immediately. For such large
500 changes in heat load, closed-loop control may be able to avoid or
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
delay the onset of the CHF condition. This will be described in a
companion paper.
150
Twe (°C)
100
6. Conclusions
50
[2] E. Zivi, Design of robust shipboard power automation systems, Annu. Rev. [16] B.P. Rasmussen, A.G. Alleyne, Control-oriented modeling of transcritical vapor
Control 29 (2) (2005) 261–272. compression systems, Trans. ASME 126 (2004) 54–64, 200.
[3] NRAC, Naval research advisory committee report: roadmap to an electric naval [17] P.T. Hewlett, T.M. Kiehne, Dynamic simulation of ship-system thermal load
force, Date last accessed 05/21/2011, <http://www.onr.navy.mil/nrac/ management, in: 2010 IEEE Conference on Automation Science and
reports.asp> (Online). Engineering (CASE), 2010, pp. 734 –741.
[4] R.R. Schmidt, B.D. Notohardjono, High-end server low temperature cooling, [18] E.A. Groll, S. Bendapudi, J.E. Braun, A comparison of moving-boundary and
IBM J. Res. Dev. 46 (6) (2002) 739–751. finite-volume formulations for transients in centrifugal chillers, Int. J. Refrig.
[5] P. E. Phelan, J. Swanson, F. Chiriac, V Chiriac, Designing a mesoscale vapor- 31 (2008) 1437–1452.
compression refrigerator for cooling high-power microelectronics, in: ITHERM [19] G.L. Wedekind, B.L. Bhatt, B.T. Beck, A system mean void fraction model for
2004: The Ninth Intersociety Conference on Thermal and Thermomechanical predicting various transient phenomena associated with two-phase
Phenomena in Electronic Systems, Las Vegas, NV, 2004. evaporating and condensing flows, Int. J. Multiphase Flow 4 (1978) 97–114.
[6] R. Shah, A.G. Alleyne, C.W. Bullard, Dynamic modeling and control of multi- [20] T.L. McKinley, A.G. Alleyne, An advanced nonlinear switched heat exchanger
evaporator air-conditioning systems, ASHRAE trans. 110 (2004) 109–119. model for vapor compression cycles using the moving-boundary method, Int. J.
[7] T.W. Webb, T.M. Kiehne, S.T. Haag, System-level thermal management of Refrig. 31 (2008) 1253–1264.
pulsed loads on an all-electric ship, IEEE Trans. Magn. 43 (1) (2007) 469–473. [21] B.D. Eldredge, B.P. Rasmussen, A.G. Alleyne, Moving-boundary heat exchanger
[8] B. Li, A.G. Alleyne, A dynamic model of a vapor compression cycle with shut- models with variable outlet phase, J. Dyn. Syst. Meas. Control 130 (2008)
down and start-up operations, Int. J. Refrig. 33 (2009) 538–552. 061003.1–061003.12.
[9] V.P. Carey, Liquid–Vapor Phase-Change Phenomena: An Introduction to the [22] R.L. Zhou, T.J. Zhang, J. Catano, J.T. Wen, G.J. Michna, Y. Peles, M.K. Jensen, The
Thermophysics of Vaporization and Condensation Processes in Heat Transfer steady state modeling and optimization of a refrigeration system for high heat
Equipment, Series in Chemical and Mechanical Engineering, Hemisphere Pub. flux removal, Appl. Therm. Eng. 30 (2010) 2347–2356.
Corp., New York, NY, 1992. [23] L. Wojtan, T. Ursenbacher, J.R. Thome, Investigation of flow boiling in
[10] J.G. Collier, J.R. Thome, Convective Boiling and Condensation, 3rd ed., Oxford horizontal tubes: Part i a new diabetic two-phase flow pattern map, Int. J.
Science Publication, New York, NY, 1996. Heat Mass Trans. 48 (14) (2005) 2955–2969.
[11] NIST standard reference data. Thermophysical properties of fluid systems, [24] L. Wojtan, T. Ursenbacher, J.R. Thome, Investigation of flow boiling in
Date last accessed 01/25/2011, <http://webbook.nist.gov/chemistry/fluid> horizontal tubes: Part II. Development of a new heat transfer model for
(Online) stratified-wavy dryout and mist flow regimes, Int. J. Heat Mass Transfer 48
[12] S. Bendapudi, J.E. Braun, A review of literature on dynamic models of vapor (14) (2005) 2970–2985.
compression equipment, Technical Report 4036-5, ASHRAE, 2005. [25] J. Catano, Dynamic modeling and advanced control of vapor compression
[13] E.W. Grald, J.W. MacArthur, A moving-boundary formulation for modeling cycles for electronics cooling, PhD thesis, Department of Mechanical,
time-dependent two-phase flows, Int. J. Heat Fluid Flow 13 (3) (1992) 266– Aerospace and Nuclear Engineering, Rensselaer Polytechnic Institute, 2011.
272. [26] Heating Air-Conditioning and Refrigeration Institute, Performance rating of
[14] T. Pfafferott, G. Schmitz, Modelling and transient simulation of co2- positive displacement refrigerant compressors and compressor units, ANSI/
refrigeration systems with modelica, Int. J. Refrig. 27 (2004) 42–52. AHRI Standard 540 (formerly ARI Standard 540), 2004.
[15] X.-D. He, S. Liu, H.H. Asada, Modeling of vapor compression cycles for
multivariable feedback control of hvac systems, Trans. ASME: J. Syst. Meas.
Control 119 (2) (1997) 183–191.