Professional Documents
Culture Documents
203–219
of the scrolls, requiring higher torque than would ef ciency of the compressor. The actual power
have been required if the pressure had built up input for the compressor is calculated by the fol-
gradually to the condensing pressure. This oper- lowing equation:
ation mode is shown in Figure 3, and again, the Çt
W
lost work is represented by the shaded area. Ç=
W Ç loss
+W (6)
Both over-compression and under- h
compression may lead to energy losses in the Where W Ç is the compressor power input, W Ç t is
compressor. The theoretical power consumption the theoretical power, h is the electro-
of the compressor will be a little higher than the mechanical ef ciency, and W Ç loss is the constant
power required to perform a pure isentropic part of the electro-mechanical power losses.
compression. Due to the absence of suction and discharge
If external pressure ratio matches the ‘built- valves and theoretically negligible re-expansion
in’ pressure ratio, the isentropic power consump- volumes, the refrigerant volume ow rate at the
tion is: beginning of the compression, VÇ r, is equal to the
Wt =
g
g21 F
g 21
PeVÇ r p g 2 1 G (4)
product of the volume of the pockets that seal
the suction gas at the beginning the compression
and the rotational speed. Thus, the refrigerant
If the external pressure ratio does not match the mass ow can be determined as:
‘built-in’ pressure ratio, the power consumption
VÇ r
will be higher than that of the isentropic process: Çr=
m (7)
F G
nsuc
g
Wt = P VÇ · Where nsuc is the speci c volume at the suc-
g21 e r
F G
tion state.
g 2 1 p 1 g 21 In practice, leakage will reduce the refrigerant
+ p g 21 (5)
g ni g i mass ow rate in the scroll compressor calcu-
A simple linear representation, which follows lated by Equation (7). In the scroll compressor
the same modeling approach as the reciprocating model presented by Chen et al.,17 the leakage
compressor discussed previously, has been used due to the gap between the bottom/top plate and
to account for the electrical and mechanical the scrolls and the gap between the anks of the
two scrolls has been discussed quantitatively.
Based on the analysis of Chen et al.,18 the leak-
age was found to be a function of the ratio of
discharge pressure and suction pressure. To sim-
P2 plify the calculation, the discharge pressure is
assumed to be equal to condensing pressure and
Pi
the suction pressure is equal to the evaporating
pressure. Hence, the leakage rate is de ned as
follows:
Pi < P2
Pc
Ç leak = C ·
m (8)
Pe
P1
Where Pc is condensing pressure, Pe is evaporat-
ing pressure, and C is the coef cient to de ne
V the relationship between pressure ratio and leak-
Figure 3 Thermodynamic cycle of a scroll compressor age rate.
with under-compression loss The refrigerant mass ow rate calculated in
WCalc(kW)
VÇ r, ni, C, Wloss and h. These parameters, along 11
-10%
with the UA values for both heat exchangers and 10
the superheat, are estimated from catalogue heat 9
pump data. (As described later, use of antifreeze
further increases the parameters.) 8
7
2.1 Scroll compressor model validation 7 8 9 10 11 12 13 14
pressor model is only used within a heat pump Figure 5 Calculated power versus catalogue power
model, it is dif cult to directly validate its oper- (hypothetical reciprocating compressor)
ation. One indirect validation is to apply two
variations of a water-to-water heat pump capacity and power consumption when the heat
model—one with reciprocating compressor, the pump is modeled with a hypothetical reciprocat-
other with scroll compressor, to a water-to-water ing compressor. Figures 6 and 7 show the heat-
heat pump model that actually has a scroll com- ing capacity and power consumption when the
pressor. The scroll compressor model may be heat pump is modeled with a scroll compressor.
judged successful if it improves the overall per- The RMS errors are summarized in Table 1. In
formance of the heat pump model. A Florida addition, Table 1 shows the RMS errors of the
Heat Pump model WP 120, which utilizes a heat pump model prior to adding the leakage
Copeland scroll compressor, has been modeled parameter ‘C’ to the model. It noticeably
with both variations of the water-to-water heat improves the model performance.
pump model. Figures 4 and 5 show the heating As can be seen from Figures 4–7, the perform-
QLCat(Btu/Hr)
85295 105295 125295 145295 165295 185295
55 185295
50
165295
QLCalcl(Btu/Hr)
45
+10%
QLCalcl(kW)
145295
40
-10% 125295
35
105295
30
25 85295
25 30 35 40 45 50 55
QLCat(kW)
Figure 4 Calculated heating capacity versus catalogue heating capacity (hypothetical reciprocating compressor)
50
165295
+10%
45
QLCalc(Btu/Hr)
QLCalc(kW)
145295
-10%
40
125295
35
105295
30
25 85295
25 30 35 40 45 50 55
QLCat(kW)
Figure 6 Calculated heating capacity versus catalogue heating capacity (scroll compressor)
14
ance of the heat pump model is signi cantly
improved, both for predicting heating capacity
13
+10%
and power, when the scroll compressor equipped
12 heat pump is modelled with the scroll com-
pressor version of the heat pump model. There-
WCalc(kW)
11
-10% fore, use of the scroll compressor model is rec-
10 ommended for a heat pump with scroll
compressors.
9
this degrades the heat pump performance. It and physical properties. Thus, it may vary con-
would be useful to be able to model the impact siderably as a result of changes in uid proper-
of antifreeze on the performance of the heat ties when antifreeze is added to water. The
pump. Unfortunately, very little information is geometry of the heat exchanger may not be
available from manufacturers with regards to the known, but is likely to be a co-axial heat
change in system performance when antifreeze exchanger. Some manufacturers also use com-
mixtures are used. The Trane Company has pro- pact heat exchangers. For the purpose of calcu-
vided the most detailed procedures to adjust the lating a degradation factor, it is reasonable to
catalogue data for pure water to the performance assume a co-axial heat exchanger geometry, with
with antifreeze mixtures.19 Correction factors for the coolant owing through the inner tube. In
heating capacity, cooling capacity and power this case, the Sieder-Tate correlation20 may be
consumption are presented versus the concen- taken as a reasonable approximation:
S D
tration of the antifreeze mixture in tables and 0.14
charts. Given the very limited amount of data Nu = 0.027 Re0.8Pr0.33 m/mw (9)
available, only a preliminary effort has been
made to develop a procedure for adjusting the However, since we are only interested in the
model. This procedure is described later along relative performance, we are only assuming that
with a check of how it performs compared to the the general form of the equation and the
Trane correction factors. exponents are correct.
Heat transfer taking place in the condenser or Substituting the de nitions of the dimen-
evaporator will be affected by the degraded heat sionless group into Equation (9) yields:
transfer coef cient, density and speci c heat of
the uid. In the parameter estimation model, the h = 0.027u0.8d2 0.2m2 0.47r0.8C0.33
p k
0.67
(10)
previously assumed constant heat transfer co- The relationship of the convection heat transfer
ef cient will be replaced by a variable heat coef cients between uids with different physi-
transfer coef cient that depends upon the uid cal properties can be found as follows, providing
side volumetric ow rate. This allows the separ- that the volumetric ow rate or velocity is the
ate estimation of the refrigerant side resistance same.
and coolant side resistance. Assuming the forms
of the correlation for the uid to heat exchanger
wall convection allows an estimate of a degra-
dation factor by substituting in correct values of
h1
h2
=
m1
m2 SD SD S D SD
2 0.47
r1
r2
0.8 Cp1
Cp2
0.33
k1
k2
0.67
(11)
density, speci c heat, viscosity and conductivity.
The degradation factor multiplies the uid-side A degradation factor of the convection heat
heat transfer coef cient estimated for pure water, transfer coef cients between antifreeze and pure
so that the heat pump performance with anti- water is de ned as:
freeze can be predicted. (hA)antifreeze
DF = (12)
(hA)water
3.1 Derivation of antifreeze degradation
factor Since the heat exchanger dimensions do not
The overall heat transfer resistance consists of change when antifreeze is used, the degradation
a coolant side convection heat transfer resist- factor can be reduced to:
ance, refrigerant side convection heat transfer hantifreeze
resistance, a tube wall resistance and, perhaps, DF =
hwater
S D S D
a fouling resistance. According to classical heat
transfer theory, the coolant side convection heat mantifreeze 2 0.47
rantifreeze 0.8
=
transfer resistance is a function of uid velocity mwater rwater
S Cpa n tifr e e z e
Cpw a te r D S
0.33
kantifreeze
kwater D 0.67
(13)
the fact that water source heat pump catalogues
give performance data for a range of water ow
rates. Using an analogy to the Wilson Plot22
The degradation factor for propylene glycol sol- allows the two resistances to be estimated separ-
ution is plotted in Figure 8 against the concen- ately when pure water is used. Then, the pre-
tration by per cent volume with different tem- viously described degradation factor may be
peratures. The propylene glycol properties were applied when antifreeze is used.
calculated with the functions developed by The overall heat transfer coef cient is a func-
Rees.21 tion of coolant side heat transfer coef cient,
refrigerant side heat transfer coef cient, tube
3.2 Volumetric ow-dependent heat wall resistance and fouling resistance.
exchanger model
1 1 1
In order to apply the degradation factor = +
developed in the previous section, it is necessary (UA)total (hA)refrigerant (hA)coolant
to split the heat transfer resistance between the + Rwall + Rfouling (14)
coolant side convection and the other parts
(refrigerant side convection, tube wall resistance If the sum of the resistances of refrigerant side,
and fouling). The heat pump model described tube wall and fouling is assumed constant and
previously1 only used a single UA value for each de ned as:
heat exchanger. This did not lend itself to differ- 1
entiating between the two resistance compo- R9 = + Rwall + Rfouling (15)
(hA)refrigerant
nents.
The modi ed procedure takes advantage of Then Equation (14) can be reduced to:
Figure 8 Degradation factor for propylene glycol/water mixture by per cent volume
SD
(UA)totalFwater = (21)
VÇ 2 0.8
3.3 Antifreeze model performance
C1 + C2
A Regrettably, there is little data available from
manufacturers of water-source heat pumps
Let C3 = C1A0.8; then the overall heat transfer regarding performance with antifreeze. What is
coef cient takes the form: available only covers water-to-air heat pumps.
1 As a preliminary check on the effect of anti-
(UA)totalFwater = Ç (22) freeze on the water-to-water heat pump model,
C3V 2 0.8
+ C2
we have used it to model a water-to-air heat
This part of the procedure is analogous to a tech- pump for which the manufacturer has provided
nique introduced by Wilson in 1915.22 The tech- correction factors for use with antifreeze (a
nique, which is often called ‘Wilson Plot’ by Trane GSUJ 018, which utilizes a scroll
later authors, is used to infer heat transfer co- compressor), operating in heating mode only. In
ef cients on both sides of a heat exchanger when heating mode, the water-to-water heat pump per-
the total resistance is measured for series of tests forms much like the water-to-air heat pump,
3.0
1.4
2.5
1.2
Rtotal (F-Hr/Btu)
2.0
Rtotal (C/kW)
1.0
0.21 l/s (3.3 GPM)
1.5 0.8 0.32 l/s (5.0 GPM)
0.38 l/s (6.0 GPM)
0.6
1.0
R’ 0.4
0.5
0.2
0.0 0.0
0 10 20 30 40 50
Concentr ation in Pe rcent Volum e
Figure 9 Evaporator overall heat transfer resistance (Trane GSUJ 018 water-to-air heat pump)
rection factor that multiplies the heating capacity freeze. Further experimental work is needed to
with pure water to give the heating capacity with establish the accuracy of the model.
an antifreeze mixture. This is given only as a Since propylene glycol properties vary sig-
function of the concentration of propylene gly- ni cantly with temperature, the sensitivity to
col. Presumably, the reduction in heating heat pump entering uid temperature was also
capacity also varies with the ow rate and examined. For a single, intermediate ow rate,
entering uid temperature. We have calculated the correction factors were calculated for three
the reduction in heating capacity for a range of different entering uid temperatures. The results
propylene glycol concentrations, uid ow rates, are shown in Figure 11. While some variation
and entering uid temperatures. The calculated in correction factor may be observed, the differ-
correction factor is taken as the ratio of the heat- ences over the expected range of temperature are
ing capacity calculated with antifreeze to the not as signi cant as the differences over the
heating capacity calculated (with the model) for allowable range of uid ow rates.
pure water.
Correction factors provided by the heat pump 4 Field validation of the water-to-water
manufacturer may be compared to correction heat pump model
factors calculated with the water-to-water heat
pump model in Figure 10. The model-calculated As mentioned earlier, additional experimental
correction factors are averaged for all catalogue- work is needed to validate the extensions to the
speci ed entering uid temperatures at each ow water-to-water heat pump model. It would be
rate. As can be seen, the correction factors for particularly useful to do so in a laboratory where
the three different ow rates show some diver- the ow rates and propylene glycol concen-
gence from the manufacturer’s single curve. trations could be readily varied. This has not yet
Assuming for a moment that the curves calcu- been done—this section of the paper reports on a
lated with the model are correct, the manufac- simpler ‘ eld validation’. Experimental data was
turer’s single curve probably provides a reason- collected from an operating water-to-water heat
able approximation for designers. However, it is pump, which contains a scroll compressor and
desirable for modelling purposes to be able to utilizes a 42% by weight propylene glycol on the
more accurately represent the effects of anti- condenser side and pure water on the evaporator
1.00
0.99
0.98
0.97
Correction Factor
0.99
0.98
Correction Factor
0.93
0.92
0 10 20 30 40 50
Concentration in Percent Volum e
Figure 11 Heating capacity correction factor with varying entering uid temperatures (Trane GSUJ water-to-air
heat pump)
side of the heat pump. (The application in this load side uid temperatures, entering source
case was a pavement heating system rather than uid temperatures, a 1.51 L/s (24 GPM) source
a building.) Additional information regarding the uid ow, and a 5.56°C (10°F) temperature rise
heat pump, instrumentation, other operating con- on the load side. With an entering source uid
ditions, etc. is given by Jin.23 temperature of 4.44°C (40°F) and a leaving load
The test was performed on a Florida Heat uid temperature of 48.9°C (120°F) the heating
Pump (FHP) model WP120 water-to-water heat capacity is 33.54 kW (114 442 Btu/h).
pump with nominal cooling capacity of 35.2 kW Catalogue data were used to t coef cients for
(10 tons). The unit is design to operate with the water-to-water heat pump model (with scroll
entering uid temperatures between 27°C compressor and two-parameter condenser and
(20°F) and 49°C (120°F). The manufacturer evaporator models). Table 3 gives the estimated
gives heating capacity, heat of absorption, and parameters for the model. Note that in this case,
power consumption data for a range of leaving both the evaporator and condenser are modelled
Table 3 Parameter estimation results for the FHP water-to-water heat pump model WP120
with the two-parameter heat exchanger model. sumably, this transient behavior is due to the
The resulting RMS errors for 20 points are: heat storage inside the heat pump and uid trans-
1.28%—heating capacity, 1.99%—heat extrac- port delays during start-up. During the last 5
tion rate, and 1.78%—power. hours, the temperature deviation is much
One data set is selected here for presentation. smaller, and results in a RMS error of about 6%
In this case, the heat pump was operated con- in the heating capacity and about 11% RMS
tinuously for over 9 h. The uid temperatures error in the source-side heat transfer rate. On
and ow rates vary over time. However, some- both sides, the model over-predicts the heat
thing approaching a quasi-steady-state operation transfer rate compared to the experimental
is achieved, particularly during the last 5 h. The measurement. Jin23 gives a detailed uncertainty
uid ow rates are shown in Figure 12. Exper- analysis of both the experimental measurements
imentally measured entering and exiting uid and the model predictions. The errors described
temperatures are shown in Figures 13 and 14. fall within the uncertainty analysis. There is also
The model has been used to determine leaving the possibility that the individual unit perform-
uid temperatures and power consumption, tre- ance has fallen below the catalogue speci -
ating the experimentally measured entering uid cations. (In the uncertainty analysis, the cata-
temperatures and ow rates as inputs. The leav- logue data were assumed to be accurate to within
ing uid temperatures are plotted in Figures 13 65%. Degradation of performance due to foul-
and 14, and power consumption is plotted in Fig- ing, refrigerant charge leakage, mechanical wear,
ure 15. etc. was not quanti ed.)
Figures 13 and 14 show quite good agreement Figure 15 shows a comparison of the meas-
between the measured leaving uid temperatures ured electrical power consumption and the
and the model predictions. The maximum devi- model-predicted electrical power consumption.
ation (around 4°C or 7°F) is observed at the rst Consistent with the over-prediction of heat trans-
two points after the heat pump is turned on. Pre- fer rates, the model over predicts the power con-
25
1.4
20
1.2
Source f low
Flow r ate (GPM)
1.0
Flow rate (l/s)
15
0.8
Load flow
0.6 10
0.4
5
Circulating pump on
0.2
0.0 0
12:00 14:24 16:48 19:12 21:36 0:00
Time
52
120
42
100
Load EFT & LFT (C)
System On 40
2
-8 20
-18 0
12:00 14:24 16:48 19:12 21:36 0:00
Tim e
Figure 13 Heat pump model validation in heating mode: load side entering/leaving uid temp
80
22
70
17
60
Source EFT & LFT (C)
20
-8
-13 10
-18 0
12:00 14:24 16:48 19:12 21:36 0:00
Tim e
Figure 14 Heat pump model validation in heating mode: source side entering/leaving uid temp
12
10
Heat pump power (kW)
6 Heat pump on
0
12:00 14:24 16:48 19:12 21:36 0:00
Time
sumption. During the last 5 h, the RMS error of heat pump start-up. For many types of energy
the model compared to the experimental power calculations this will result in a very small error
consumption is about 7%. This is slightly below in the prediction of energy consumption. How-
the estimated combined uncertainty of the ever, if the heat pump cycles on and off fre-
experimental measurement and the model. The quently, as in some residential applications, it
model assumes a perfect energy balance—in may be worthwhile to consider a simple dynamic
heating mode, the source side heat transfer rate model. In order to retain the physical basis of
plus the compressor electrical power consump- the model, and the ability to estimate all para-
tion will always equal the load side heat transfer meters from catalogue data, it is likely that a
rate. In practice, this may not be met due to dynamic model would be very simple—say, tre-
losses from the heat pump e.g., heat transfer ating the heat pump thermal mass as a lumped
from the compressor shell. Furthermore, errors capacitance.
in the experimental measurements necessarily
preclude a perfect measured heat balance. The 5 Conclusions and recommendations
relative energy imbalance, calculated as the dif-
ference between the load side heat transfer and A parameter estimation based model of water-
the sum of the source side heat transfer and com- to-water heat pumps has been revised to include
pressor heat, divided by the sum of the source scroll compressors and allow an accounting for
side heat transfer and compressor heat, in the last the effects of antifreeze. A ve-parameter scroll
5 h, varies between 3% and 7%. compressor model has been incorporated into the
To summarize, the heat pump model gives heat pump model. As the goal of the heat pump
reasonably good agreement to the experimental model is to produce a physically-realistic model
data. The most signi cant error occurs during the for which all parameters can be estimated from
catalogue data, the scroll compressor sub-model cant. However, further investigation into a
was validated by comparing the results of both dynamic model that retains the bene cial charac-
the scroll-compressor-equipped and reciprocat- teristics of this model—its physical basis and
ing-compressor-equipped heat pump models ability to estimate all parameters from catalogue
applied to catalogue data for a scroll- data, is recommended.
compressor-equipped heat pump. The scroll
compressor version of the heat pump model
gives noticeably improved results over the recip- Acknowledgements
rocating compressor version. It was noted that
the overall accuracy of the heat pump model was The revisions to the water-to-water heat pump
improved by including the fth parameter into model and experimental measurements were
the scroll compressor model: a differential funded by the U.S. Federal Highway Adminis-
pressure-dependent leakage. tration under Grant No. DTFH61–99-X-00067.
A preliminary investigation of a methodology Support by the Federal Highway Administration
for modelling the use of antifreeze solutions with does not constitute endorsement of the views
water-to-water heat pumps has been described. expressed in this paper.
The methodology retains the physically-realistic
approach of the heat pump model, and should,
in theory, allow the prediction of heat pump per- References
formance to be extended to a range of antifreeze
solutions, even when only pure-water-based 1 Jin H, Spitler JD. A parameter estimation based
catalogue data are available. To date, we can model of water-to-water heat pumps for use in
only say that the results appear to be approxi- energy calculation programs. ASHRAE Trans.
mately consistent with the correction factors 2002; 108: 3–17.
offered by one manufacturer. The model does 2 Fischer SK, Rice CK. The Oak Ridge heat
show sensitivity to ow rates and uid tempera- pump models: 1. A steady-state computer design
tures that are not re ected in the manufacturer’s model for air-to-air heat pumps. Report no.
recommendations. It would be desirable to per- ORNL/CON-80/R1. Oak Ridge, TN: Oak Ridge
form an experimental investigation of the anti- National Laboratory.
freeze effects on water-to-water heat pumps 3 Domanski PA, Didion DA. Mathematical model
of an air-to-air heat pump equipped with a
under laboratory conditions, where entering uid capillary tube. Int. J. Refrigeration 1984; 7:
temperature, ow rate and antifreeze concen- 249–55.
tration can be more easily changed and con- 4 Parise JAR. Simulation of vapor-compression
trolled. heat pumps. Simulation 1986; 46: 71–76.
A limited eld validation was performed for 5 Greyvenstein GP. A computer simulation model
a water-to-water heat pump that incorporated a for the design and optimization of heat pumps
scroll compressor, and utilized a propylene– and refrigeration systems. SA J. Science 1988;
glycol antifreeze mixture on the source-side. A 84: 494–502.
comparison between experimental data and 6 Cecchini C, Marchal D. A simulation model of
model prediction shows reasonably good agree- refrigerant and air-conditioning equipment based
on experimental data. ASHRAE Trans. 1991; 97:
ment between the predicted and the measured
388–93.
leaving uid temperatures and power consump- 7 Stefanuk NBM, Aplevich JD, Renksizbulut M.
tion. The maximum deviations are observed at Modeling and simulation of a superheat-
the beginning of the on-cycle. Presumably, this controlled water-to-water heat pump. ASHRAE
is due to heat storage inside the heat pump and Trans. 1992; 98: 172–84.
uid transit delay during the start-up period. For 8 Bourdouxhe J-PH, Grodent M, Lebrun JJ,
many energy calculations, this will be insigni - Saavedra C, Silva KL. 1994. A toolkit for
primary HVAC system energy calculation—part Experimental analysis and simpli ed modeling
2: reciprocating chiller models. ASHRAE Trans. of a hermetic scroll refrigeration compressor.
1994; 100: 774–86. Appl. Therm. Eng. 2002; 22: 107–20.
9 Gordon JM, Ng KC. Thermodynamic modeling 16 LeBrun J, Bourdouxhe J-P, Grodent M. HVAC
of reciprocating chillers. J. Appl. Phys. 1994; 1 toolkit: a toolkit for primary HVAC system
75: 2769–74. energy calculations. Atlanta: ASHRAE, 1999.
10 Morishita E, Sugihara M, Inaba T, Nakamura T. 17 Chen Y, Halm N, Groll EA, Braun JE. A
Scroll compressor analytical model: comprehensive model of scroll compressors,
Proceedings of the Purdue Compressor part I: compression process modeling:
Technology Conference/Proceedings of the 1984 Proceedings of the 2000 International
International Compressor Engineering Compressor Engineering Conference at Purdue.
Conference—At Purdue 1984; 487–95. Purdue University, West Lafayette, IN, 2000:
11 Qu Z, Li X, Li S. Analytical modeling of 715–24.
thermodynamic process for scroll compressor 18 Chen Y, Halm N, Groll EA, Braun JE. A
with back pressure optimization. Journal of comprehensive model of scroll compressors,
Xi’An Jiaotong University 1998; 32: (7) 51–55. part II: overall scroll compressor modeling.
12 Bush J, Elson J. Scroll compressor design Proceedings of the 2000 International
criteria for residential air conditioning and heat Compressor Engineering Conference at Purdue.
pump applications: Proceedings of the 1988 Purdue University, West Lafayette, IN, 2000:
International Compressor Engineering 725–34.
Conference (1), Of ce of Publications, Purdue 19 Newton H. Personal communication, 2001.
University, West Lafayette, IN, 1988: 83–97. 20 Kern DQ. Process heat transfer. New York:
13 Ikegawa M, Sato E, Tojo K, Arai A, Arai N. McGraw-Hill, 1950.
Scroll compressor with self-adjusting back- 21 Rees SJ. Personal communication, 2002.
pressure mechanism. ASHRAE Trans. 1984; 90: 22 Wilson EE. A basis for rational design of heat
314–26. transfer apparatus. ASME Trans. 1915; 37: 47–
14 Etemad S, Nieter J. Design optimization of the 70.
scroll compressor. Int. J. Refrigeration 1989; 23 Jin H. Parameter estimation based heat pump
12: 146–50. models. PhD thesis. Oklahoma: Oklahoma State
15 Winandy E, Saavedra C, Lebrun J. University, 2002.