United States Air and Radiation EPA420-R-01-054
Environmental Protection November 2001
Agency M6.ACE.001
&EPA Air Conditioning Activity
Effects in MOBILE6
$5b Printed on Recycled
Paper
-------
EPA420-R-01-054
November 2001
Air Conditioning Activity in
WI6.ACE.001
John W. Koupal
Assessment and Standards Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
NOTICE
This technical report does not necessarily represent final EPA decisions or positions.
It is intended to present technical analysis of issues using data that are currently available.
The purpose in the release of such reports is to facilitate the exchange of
technical information and to inform the public of technical developments which
may form the basis for a final EPA decision, position, or regulatory action.
-------
1 ABSTRACT
MOBILE6 includes revised estimates of exhaust emissions resulting from air conditioning
operation. This will require air conditioning behavior and the resulting emission levels to be
predicted over a wide range of ambient conditions. Emission test data used to develop these
factors were gathered under conditions meant to represent a single extreme set of conditions.
This report addresses EPA's proposed methodology for applying the "extreme" data over the
broader range of ambient conditions in which air conditioner operation occurs in-use. Using air
conditioning activity data collected in Phoenix, a methodology has been developed which relates
temperature and humidity levels to air conditioner load using a combined measure known as the
heat index. This methodology also incorporates some solar load impact and will allow
adjustments for cloud cover if desired by the user. Estimates have also been developed for the
fraction of vehicles equipped with air conditioning systems, and of those, the fraction of
malfunctioning systems.
2 INTRODUCTION
The emission data to be used in the development of the MOBILE6 air conditioning factors were
gathered using a test procedure intended to represent extreme ambient conditions. From these
data emission factors will be developed which represent emission levels at full air conditioning
load (referred to as "full-usage" emission factors). These emission factors cannot appropriately
be applied to all ambient conditions, since less severe conditions will result in only partial A/C
loading and hence lower emissions. The development of the full-usage emission factors will be
the topic of a separate report.1 This report presents EPA's proposed methodology for applying
the full-usage emission factors across the broad range of ambient conditions for which estimates
of air conditioning emissions are required. A second aspect of air conditioning activity is market
penetration; namely, the fraction of vehicles equipped with air conditioning systems, and of
those, the fraction of malfunctioning systems which have not undergone repair. Proposed
estimates for these factors are also presented and discussed.
Subsequent to publication of the draft version of this report in January 1998, the document was
put out for stakeholder review. Formal peer review comments were also solicited from two
independent sources. No comments were received through the stakeholder review process,
hence peer review comments represent the only external feedback received on this report. A
summary of peer review comments and responses is contained in Appendix B.
3 OVERVIEW OF PROPOSED APPROACH
As detailed in M6.AC.002, the "with air conditioning" emission levels used as the basis for
MOBILE6 were generated using a test procedure meant to induce the level of A/C system load
on the vehicle which would occur in the real world under extreme ambient conditions. However,
1 "Air Conditioning Correction Factors in MOBILE6", M6.AC.002, July 2001
M6.ACE.001Page2
-------
MOBILE6 will need to model air conditioning emissions at less severe conditions where the
majority of vehicle operation will occur. A method for modeling air conditioning effects under
intermediate ambient conditions is therefore required.
The method proposed for use in MOBILE6 is to link emissions directly with the operation of the
vehicle's air conditioning compressor, which is propelled by the engine using a belt in a similar
manner to the vehicle's alternator. The compressor is the focus rather than driver behavior
because it is the direct cause of additional load on the engine and is therefore the best indicator of
how A/C system operation impacts emissions. Compressor load varies and the compressor
cycles on or off (i.e. is engaged or disengaged) depending on user demand and the vehicle's
response to ambient conditions. As a result, it is generally not inducing full load on the engine
100% of the time the A/C is turned on under intermediate ambient conditions. With this
approach driver behavior is accounted for implicitly; however, because the compressor will only
engage when the A/C system is on, compressor engagement over the course of a vehicle trip is
strongly driven by A/C demand from the user.
An ideal model of this sort would link ambient conditions and emissions by modeling changes in
compressor load (torque) as a function of changes in ambient conditions. Unfortunately, activity
data which would allow such a link does not appear to exist. Available activity data does not
include a direct measure of compressor load, but only the total time the compressor was engaged
over a single vehicle trip. Therefore, the methodology developed for MOBILE6 is based on the
relationship between emission response and the percentage of time the compressor is engaged.
Since the fraction of time the compressor is engaged over a trip (compressor-on fraction) has a
direct impact on the additional load experienced by engine during a trip, it is assumed that the
impact of compressor engagement on overall engine load is linearly proportional (1:1). The
second assumption is that changes in emission response correlate 1:1 with changes in engine
load, and hence with compressor-on fraction. With this methodology, the compressor-on
fraction is equal to the factor by which the full-usage emission factor is scaled in MOBILE6 to
derive the emission factor appropriate for the ambient condition. In other words, a compressor
which is engaged 100% of the time would result in the full-usage emission factor. If the
compressor is engaged only 50% of the time, 50% of the full-usage emission factor would be
applied. This scaling factor is termed the "demand factor".
The key to this approach is the assumption that the relative emission impact due to A/C
correlates 1:1 with compressor-on fraction for all pollutants. It should be noted that air
conditioning system experts from the automotive industry have identified several limitations with
this assumption. Specifically, compressor load fluctuates significantly when the compressor is
engaged, depending on a number of factors including ambient conditions, vehicle speed, vehicle
cabin temperature, and A/C system setting (e.g. fan setting, recirculation vs. outside air); A/C
system response to changes in all of these factors is highly vehicle-specific.2 Using only
compressor-on fraction is a rough estimate of actual compressor load since it assumes that
2 EPA/AAMA/AIAM meeting on MOBILE6 air conditioning issues, November 6, 1997
M6.ACE.001 Page 3
-------
fluctuations in relative compressor load average out over periods when the compressor is
engaged. Further, the impact of compressor load on emissions is likely not linear. However, the
complexity and data demands of a compressor-load based model are prohibitive within the
timeframe and scope of MOBILE6. Future research activity will need to address this lack of
information.
4 DEVELOPMENT OF DEMAND FACTORS
4.1 Phoenix Activity Survey
The activity data used in the development of the proposed demand factors are based on an
instrumented vehicle survey conducted as part of the Supplemental Federal Test Procedure
(SFTP) rulemaking process on 20 vehicles over almost 1000 trips in Phoenix, Arizona from
August-October 19943. Data gathered for each trip included time and date, total trip time, total
time the air conditioner was on, and total time the compressor was engaged. The datalogger also
recorded summarized trip information including trip distance, total idle time, and time spent in
five mile-per-hour trip bins. Hourly weather information taken from Phoenix's Sky Harbor
International Airport available through the National Climatic Data Center (NCDC) was used to
estimate dry bulb temperature and relative humidity at the start of each trip.
4.2 Treatment of Data
The initial dataset used for this analysis contained 987 trips. It did not include the first and last
trips for each vehicle if the trip was less than 0.25 miles; these cases were removed as part of an
earlier analysis because they represented trips taken by the contractor during the datalogger
installation and deinstallation process. For the MOBILE6 analysis, this trip file was further
modified to improve the representativeness in the following manner:
a. Trips with a duration of 30 seconds or less were removed to eliminate stalls and other
potential queuing-related cases.
b. Since the location of each trip was not known, it was necessary to assume that the linked
weather information was appropriate in characterizing the conditions experienced by the
vehicle on every trip. To reduce the chance that a trip was taken outside of the greater
Phoenix area, all trips greater than 60 miles were deleted from the trip file. In addition, all
trips for a given vehicle which followed a trip of more than 60 miles were also deleted, to
reduce the chance that a vehicle made one long trip outside of the Phoenix area and
remained outside the area for the remainder of the monitored trips. 60 miles was chosen
as a cutpoint based on the estimated radius of the greater Phoenix area, the distance to
higher altitude locations outside Phoenix along highway routes, and the distribution of
3 "Study of In-Use Air Conditioner Operation in Phoenix, Arizona", Automotive Testing Laboratories, Inc. report
to EPA (EPA Docket No. A-92-64 Item IV-A-1)
M6.ACE.001Page4
-------
trips below and above 60 miles in the dataset. For two of the vehicles for which trip
distance data was not gathered, a trip duration cutpoint of 45 minutes was used.
c. Preliminary analysis of the trip data indicated that trips that were comprised solely of idle
had radically different behavior than all other trips. As shown in Figure 1, the average
compressor-on fraction for all-idle trips is much lower than trips consisting of even very
high percentages of idle. MOBILE6 will predominantly need to predict A/C emissions
over trips consisting of non-idle driving, and we were concerned that inclusion of all-idle
trips would skew the non-idle results. All-idle trips were therefore dropped from the trip
file for the purposes of developing the demand factor relationships. It is important to
clarify that this does not mean the trip dataset does not contain idle events; as indicated in
Figure 1, many of the remaining trips in fact have significant percentages of idle
operation.
These modifications reduced the number of trips in the dataset to 672. All subsequent analyses
were performed on this dataset.
4.3 Temperature, Humidity and Heat Index
Temperature and humidity are the most important drivers of A/C system demand. While
temperature is a widely recognized influence, the load placed on the air conditioning system by
humidity can account for over half of the total load under the ambient conditions of the SFTP air
conditioning test procedure2. It is considered important, therefore, to develop a demand factor
methodology which incorporates both temperature and humidity. This was supported by several
comments received following the March 1997 MOBILE6 workshop that advocated the inclusion
of humidity in the MOBILE6 air conditioning component.
To assess the effect of ambient variables (temperature, humidity, heat index) on air conditioning
demand, the Phoenix dataset was analyzed by "binning" trips according to the variable being
analyzed. For example, for temperature, all trips at a given temperature were combined, and the
compressor-on fraction was calculated at each temperature as total time with the compressor
engaged at that temperature divided by total trip time at that temperature. This aggregation step
was taken to reduce the variability from individual vehicle and driver behavior, since the
operation of air conditioning over the "composite" fleet and population is of more concern for
MOBILE6. Total trip time (including trips when the air conditioner was not turned on) was
used as the denominator in order to characterize air conditioning usage directly from ambient
conditions. This is a more direct approach than trying to model both a) air conditioning usage as
a function of ambient conditions, and then b) compressor operation when the air conditioning is
on, which is less supportable from the Phoenix data.
Analyzed in this way, the Phoenix dataset indicates a strong correlation between temperature and
compressor-on fraction, as shown in Figure 2 (compressor-on time is expressed as the fraction of
time the compressor is engaged over the total trip time at each temperature point). The
M6.ACE.001 Page 5
-------
relationship between humidity and compressor-on fraction is weak, however (Figure 3), and
ANOVA results indicate that when humidity is modeled with temperature it is not a significant
variable. Because humidity does physically affect overall A/C system load, this is judged to be
an artifact of the limited humidity range in Phoenix. Figure 4 shows relative humidity at the start
of each trip as a function of temperature (the reference lines show SFTP temperature and
humidity conditions); the average relative humidity for temperatures greater than 80° F was only
28%. By contrast, historical data in Houston indicate that during the summer months (when the
average daily maximum temperature is over 90° F) average relative humidity at noontime is
around 60%4. The development of demand factors which could be applied to humidity levels
like those observed in Houston would require significant extrapolation of the Phoenix data.
Given the weakness of the relationship between humidity and compressor-on fraction within the
boundries of the Phoenix humidity levels, extrapolation of these data is not desirable.
In an attempt to more accurately assess the relative impacts of temperature and humidity on air
conditioning load, a metric known as the heat index will be used in developing the demand
factors. Heat index is used by the National Weather Service to quantify discomfort caused by
the combined effects of temperature and relative humidity. The basis of the index is the human
body's ability to maintain thermal equilibrium through perspiration, taking into account
numerous factors including clothing thickness, atmospheric pressure and ambient conditions.
Equations have been developed which allow heat index to be calculated using only temperature
and relative humidity; these equations are proposed for use in MOBILE6 to compute heat index
based on temperature and humidity values input by the user (Appendix 1)5'6. Heat index as a
function of temperature and humidity is shown in Figure 5.
The approach for addressing intermediate conditions is to develop demand factors by modeling
compressor-on fraction as a function of heat index based on user input of temperature and
humidity. An attractive feature of this approach is that the air conditioning activity component
of MOBILE6 would be based directly on driver discomfort, the most likely factor impelling a
driver's A/C behavior and thus a strong determinant in the vehicle's emission response. As
shown in Figure 6, the Phoenix data exhibits a strong correlation between compressor-on fraction
and heat index, after "binning" trips by heat index as discussed. Using heat index instead of
temperature does not necessarily improve predictions of compressor-on fraction for the Phoenix
data, because the Phoenix results are driven almost completely by temperature. Heat index does
not detract from the relationship between temperature and compressor usage, either; for low
humidity conditions, the heat index is generally identical to the temperature, so using heat index
Gale Research Inc., The Weather Almanac, Sixth Edition (1992)
5 Meisner and Graves, "Apparent Temperature", Weatherwise, August 1985
6 The base humidity correction factor currently in MOBILES will be carried over to MOBILE6. The computation
of this factor and the air conditioning demand factor will be based on the same humidity data. A default specific
humidity value of 75 grains/pound (as in MOBILES) is proposed. Users will be able to input alternate humidity
levels in either specific humidity or relative humidity (see Section 6), with appropriate conversions made within
MOBILE6.
M6.ACE.001 Page 6
-------
or temperature would provide the same result for low humidity conditions. The intent of using
the heat index is to introduce a more equitable balance in the effect of temperature and humidity
on air conditioning load not provided by the Phoenix data. The underlying assumption of this
methodology is that the temperature-driven effects seen at high temperatures in Phoenix would
be replicated under lower temperature but higher humidity conditions seen elsewhere in the
country. Given the stated importance of humidity on air conditioning load, this assumption is
believed to be more reasonable than ignoring or understating the impact of humidity altogether.
4.4 Solar Load
The proposal for air conditioning effects in MOBILE6 presented at the October 1997 workshop
did not include any accounting for solar load. Comments received subsequent to the workshop,
however, expressed a strong desire for the inclusion of solar load, and automotive industry
experts have indicated that it is a contributing factor. Subsequent analysis of the Phoenix data
indicates that a solar load impact can be discerned, and consequently a method which accounts
for solar load and cloud cover (addressed in Section 4.6) is being used in MOBILE6.
Since solar load or cloud cover data were not available in the NCDC dataset linked to the
Phoenix survey, an empirical relationship between these factors and compressor activity could
not be developed directly. As an alternative, the impacts of solar load were isolated by binning
all of the trips based on time of day at the start of the trip. Four "period" bins were created: night
(sunset-sunrise), morning (sunrise - 10 am), peak sun (10 am - 4 pm), and afternoon (4 pm -
sunset). Sunrise and sunset times for each day in the survey as reported by the U.S. Naval
Observatory7 were used to determine appropriate trip bins. The bin definitions were determined
by analyzing solar radiation data gathered as part of the National Oceanic and Atmospheric
Administration's (NOAA) Surface Radiation Budget Project (SURFRAD)8. A regression across
all trips of compressor-on fraction by heat index was performed within each bin, with the results
shown in Figure 7. Because this analysis considered all trips instead of combining them by heat
index level, the regression was weighted by trip length to give more creedance to longer trips.
Since cloud cover information was not available, it could not be considered as a variable in the
analysis. Historical data from Phoenix indicates that at the time of year the survey was
conducted direct radiation from the sun (i.e. little or no cloud cover) is present close to 90% of
the time9, so for the purposes of this analysis all daytime trips were assumed to be taken during
periods of no cloud cover. The lines show a clear difference in compressor-on fraction between
nighttime and daytime at the same heat index level, indicating the importance of solar load and
meriting a separation of daytime and nighttime demand equations. In support of this conclusion,
ANOVA performed on the trip file with compressor fraction as the dependent variable and heat
7 U.S. Naval Observatory Sunrise/Sunset Web Site (http://riemann.usno.navy.mil/aa/data/docs/rs_oneyear.html)
8 NOAA SURFRAD Web Site (http://www.srrb.noaa.gov/surfrad/surfpage.htm)
9 Gale Research Inc., The Weather Almanac, Sixth Edition (1992)
M6.ACE.001 Page 7
-------
index and period as the independents indicated that period is significant to the 0.01 level.
A second question is whether a significant difference exists between the daytime periods. Figure
7 shows that as would be expected, the peak sun curve is higher than the morning and afternoon
curves above 75°, while the morning and afternoon curves are similar for the mid-range heat
index levels. To investigate this issue further, ANOVA analyses of compressor-on versus heat
index and period were performed for all daytime trips and again for trips taken only in the
morning and afternoon periods. The period of the day was again significant to the 0.01 level for
all daytime trips, but was not significant when only the morning and afternoon trips were
analyzed. From this it was concluded that the peak sun period is the cause of the difference
between the daytime curves and merits separate treatment.
4.5 Proposed Demand Equations
Three demand factor equations were developed for MOBILE6: nighttime, morning/afternoon and
peak sun. The "raw" equation for each period, as well as for all daytime trips and all trips, are
shown in Table 1; these equations were developed by fitting a quadratic equation through all trips
by period, weighting by trip length. The relatively low R2 values compared to the composite
result shown in Figure 5 are attributable to the regression being performed over the entire trip
sample. A quadratic curve form is favored over more complex forms because it provides a
balance between goodness of fit and more reasonable behavior at the high and low ends of the
heat index range. Still, because a smaller sample of trips occurred at the high and low ends (only
5% of trips occurred when the heat index was less than 75°) the behavior of the fitted curves at
these ends tend to defy engineering judgment. In particular the morning/afternoon curve is
higher than the peak curve below 75 °, and the night curve is higher than the daytime curves
above 100°. To rectify this, separate demand equations will be applied only in the middle of the
heat index range, while the higher and lower ends will be modeled with composite equations.
The "daytime combined" will be used for all daytime periods at the lower end, and all individual
curves will be modeled with the "all combined" equation at the high end. The heat index values
at which the composite equations and period-specific equations diverge (at the low end) or
converge (at the high end) are determined based on the respective points of intersection. This
progression is outlined in Table 2, with the revised equation forms for each period shown in
Figure 8.
Since MOBILE6 will calculate emission factors on an hourly basis, changes in solar load
throughout the course of a full day will be modeled by applying the appropriate demand
equations at each hour. The night equation will be applied from sunrise to sunset, the
morning/afternoon equation will apply from sunrise - 10 am and 4 pm - sunset, and the peak
equation will apply from 10 am - 4 pm. The peak sun cutpoints were determined based on
analysis of the NOAA data on different days throughout the summer months, which indicated
that direct solar radiation levels stay relatively high from 10 am to 4 pm but tend to drop off
rapidly before 10 am and after 4 pm. However, the user will be allowed to input alternate time
for which peak sun demand equations are applied if desired. Default sunrise/sunset times of 6
M6.ACE.001 Page 8
-------
am and 9 pm will be used in MOBILE6 to approximate a typical summer day with daylight
savings time. The user will also have the option of inputting alternative sunrise/sunset times in
order to alter the hours for which the morning/afternoon and nighttime demand equations are
applied.
4.6 Cloud Cover
As mentioned in Section 4.4, the daytime demand equations were developed from the Phoenix
data under the assumption that all trips were taken during periods of no cloud cover (an
assumption that likely serves to slightly understate solar load impact). Because of this
MOBILE6 will assume as a default that a sunny day is being modeled. Comments received
following the October 1997 MOBILE6 workshop advocated some accounting of cloud cover,
particularly for modeling seasonal emissions. MOBILE6 will incorporate an optional input for
percent cloud cover on a daily basis. The method for handling cloud cover input will be to scale
back the default daytime demand equations. Analysis of NOAA solar radiation data indicates
that direct solar radiation is reduced to zero when the sun is obstructed by clouds (Figure 9).
Based on these data, the nighttime demand equation is proposed to represent 100% cloud cover.
For intermediate cloud cover inputs, the model will interpolate between the appropriate daytime
demand equation and the nighttime demand equation. Thus, 50% cloud cover at noontime would
result in a demand factor halfway between the demand calculated with the peak and nighttime
equations at the appropriate heat index.
4.7 Other Factors Considered
While ambient conditions are the primary factors in determining A/C system demand, trip-related
characteristics are also likely to influence air conditioning behavior. Four such factors
investigated for this analysis were soak time prior to the vehicle trip, trip duration, average
vehicle speed, and percent of idle during a trip. A single Analysis of Variance was performed
with compressor-on fraction as the dependent variable and these four variables and heat index as
covariates, period as a factor, and vehicle as a random effect (as suggested in peer review
comments), with the results shown in Table 3. A technical basis exists for considering each
factor, and it is likely that the dataset implicitly contains the effects of each. However, none of
these variables showed enough significance to merit individual treatment; to a large extent this is
likely because the Phoenix dataset does not provide adequate resolution or sample size to discern
individual effects. A discussion of each factor follows.
4.7.1 Soak Duration
The length of soak time prior to a daytime trip could influence A/C system demand because of
the impact on cabin temperature. Vehicles parked in the sun for extended periods of time
experience elevated cabin temperatures compared to short soaks. However, information on
several factors which would greatly influence the impact of soak time were not available,
including whether a vehicle was parked in a shady location (such as a parking garage) or whether
M6.ACE.001 Page 9
-------
the windows were left open during the soak. Without this sort of information a meaningful
assessment of soak time is difficult; not surprisingly, the ANOVA results do not show
significance for this factor. A more thorough investigation of soak time impacts would require a
measure of cabin temperature and more detailed trip/soak information. It is likely however, that
the solar load impacts discussed in Section 4.4 are driven in part by this effect, so the impact of
differing soak times are subsumed in the solar load corrections if a representative soak
distribution is assumed.
4.7.2 Trip Duration
Trip duration could also be expected to impact air conditioning behavior. Cabin temperature
over the course of a longer trip will be reduced by the A/C system, thereby reducing the need for
cooling and hence the amount of time the compressor is engaged relative to the start of the trip.
Figure 10 shows a series of linear regressions for compressor-on versus trip duration over four
heat index ranges. Two trends emerge from these regressions: the expected downward trend as
trip duration increases, and a leveling of this slope as the heat index increases. The latter trend
suggests that trip duration has a more significant impact for the intermediate heat indices where
cooling needs can be met in the early stages of a trip, but for higher heat indices the cooling
demand remains high throughout the trip. ANOVA results do not indicate significance,
however. Since the Phoenix dataset contains a wide distribution of trip durations, these effects
are assumed to be accounted for implicitly in the demand equations.
4.7.3 Average Speed
Average speed could have an impact on A/C system load because higher rates of air flow across
the vehicle's A/C condenser will reduce the work required to cool the ambient air (although this
could be offset to some degree by ambient air entering the cabin at a higher rate). Regression
analysis indicates a downward trend in compressor-on as average speed increases (Figure 11).
Again, this effect is more prevalent for the lower heat index levels, and drops off as the heat
index increases. The ANOVA results again do not indicate significance, however. These effects
are assumed to be accounted for implicitly in the demand equations since the equations are based
on a distribution of average speeds.
4.7.4 Idle Fraction
The fraction of idle during a trip could impact overall compressor operation because A/C
calibrations at idle appear to be unique, as discussed in Section 4.2. For many idle-only trips the
compressor is either engaged 100% of the time or 0% of the time, with most idles in the Phoenix
dataset exhibiting the latter. Not engaging the compressor at idle would presumably be used as a
strategy for driveability, because the relative load place on engine by the A/C system at idle is
high. As shown in Figure 1, the overall average compressor-on times for all trips with idle
fractions less than 100% appear similar, indicating that the effect seen on idle-only trips doesn't
carry over to idles during normal trips. The ANOVA results again do not indicate significance,
M6.ACE.001 Page 10
-------
but it is likely that idle fraction and average speed are higher correlated so an effect solely
attributable to idle is difficult to separate out. Again, to the extent there are impacts they are
assumed to be accounted for in the demand equations.
5 MARKET PENETRATION ESTIMATES
The second component of activity determining how many vehicles in the fleet are equipped with
air conditioning systems (market penetration), and of those, how many are functional. Three
steps go into the development of these estimates: determining base market penetration rates by
model year, estimating A/C system malfunction rate by vehicle age, and estimating how many
malfunctioning systems are not repaired. This section addresses each issue.
5.1 Base Rates
Base market penetration data by model year were gathered from Ward's Automotive Handbook
for light-duty vehicles and light-duty trucks through the 1995 Model Year. This information
was available from 1972 for cars and 1975 for trucks. Year-to-year rates are more variable in
the first few years of available data, so estimates for earlier model years will be estimated by
applying the 1972 and 1975 rates for cars and trucks, respectively. In the later years, the rate of
increase becomes more steady. Projections beyond 1995 were developed by taking the average
yearly rate of increase from the last five years of available data and applying them to each
subsequent year until a predetermined cap was reached. A cap of 98% was placed on vehicles
and 95% on trucks under the assumption that there will always be vehicles sold without air
conditioning systems, more likely on trucks than cars. The resultant base rates are shown in
Figure 12. The caps are in place by the 1999 model year, and will remain for subsequent years.
5.2 Malfunction Rates
Of all vehicles equipped with air conditioning, it is appropriate to assume that not all of the
systems are functional, requiring an estimate of the fraction of non-functioning systems by
vehicle age. Unfortunately, there appears to be little publicly available data upon which to base
these estimates. One available source is the annual Consumers Reports Automobile Purchase
Issue, which began reporting reader survey results on A/C system malfunctions starting in 1994.
The reported results from the 1997 survey were used to develop malfunction estimates by vehicle
age based on a yearly increase in absolute malfunction rate of 1.5 percent (Table 4). Starting at
age nine the malfunction rate will be held constant at 12.5 percent. This is based on the
assumption that the increased probability of malfunction as a vehicle ages will be offset by the
increased probability a vehicle will have already undergone repair as it grows older.
The second component in developing malfunction estimates is rate of repair. In the absence of
concrete data, estimates were generated based on three qualitative assumptions: a) all vehicles up
to three years old (assumed to be the standard bumper-to-bumper warranty period) would receive
repair; b) after three years the majority of owners would still receive repair, but this percentage
M6.ACE.001 Page 11
-------
would decrease as the vehicle grew older, and c) vehicles built prior to the 1993 model year
(estimated as a cutpoint for which Freon was replaced with R-134a on most vehicles) would
experience a lower rate of repair due to the prohibitive cost of system recharging. From these
assumptions, it was estimated that 100% of R-134a systems would be repaired during the
warranty period, 90% in years four through eight, 80% in years nine through 13, 70% in years 14
through 18 and 60% in years 19 and up. The non-warranty period repair rates will be reduced by
a factor of 0.75 for Freon (pre-1993) systems, but only if the modeled calender year is 1995 or
later; if not, the R-134a estimates will be applied (in other words, lower repair rates for Freon-
equipped vehicles will not be invoked if recharging with Freon was viable during the modeled
calender year ). The resultant rate of unrepaired malfunctions combine the malfunction rates
from Table 4 with the rate of nonrepair in a given year. These estimates are shown in Figure 13.
For a given model year, the estimate of vehicles on the road with functional air conditioning
systems (referred to as adjusted penetration rates) will combine the base market penetration
estimates for that model year (from Figure 12) with the unrepaired malfunction rates in Figure 13
for the appropriate vehicle age.
6 HANDLING OF AMBIENT INPUT DATA
A significant change in MOBILE6 will be the ability to model on an hour-by-hour basis, whereas
MOBILES is geared towards providing daily estimates. Hour-by-hour temperatures can be
entered if the information is available to the user; otherwise, daily minimum and maximum
temperatures can be entered as was done with MOBILES. There are no temperature defaults in
MOBILE6, so at a minimum these daily temperatures must be entered. Humidity is entered as a
single daily number, and for this reason it will be entered as absolute humidity (grains per pound
of air) and converted to hourly relative humidity inside the model based on hourly temperature.
As a default, absolute is assumed constant at 75 grains/pound, no cloud cover would be assumed,
and the sunrise/sunset/peak sun times discussed in Section 4.5 would be applied.
7 ACKNOWLEDGMENTS
Several individuals contributed time, effort, ideas and consultation to this report. Rob French of
OTAQ developed and provided guidance on the Phoenix dataset and contributed to the heat
index concept. John Gilmore of OTAQ researched the base market penetration estimates and
malfunction survey information. John German of Honda provided general consultation and
input. Christine Dibble of the Office of Atmospheric Programs provided information on R-134a
phase-in. John Augustine of NOAA provided information on the SURFRAD data. Dennis
Kahlbaum of AIR, Inc. provided information on the derivation of the heat index measure.
Harold Haskew of Haskew Associates and Dr. Mohunder Bhutti of Delphi-Harrison provided
consultation on the theoretical aspects of air conditioner operation.
M6.ACE.001 Page 12
-------
Figure 2 - Compressor-On vs. Percent Idle
1.0
5 -9l
§ .8-
LL
6 71
i '6l
CD -5'
Q.
E .4-
0
0 3-
(D
O)
CO o i
i_ -^
Q)
^ .1-
0.0
100%
76-99% 51-75% 26-50% 0-25%
Percent Idle
Figure 3 - Compressor-On vs. Temperature
1.0
I .8
o
CD
LL
c .6
CD
Q.
O
.4
.2
0.0
60
70
80
90
100
Temperature (F) - Start of Trip
Non-idle trips (weighted by number of trips )
R squared = 0.87
110
-------
Figure 4 - Compressor-On vs. Specific Humidity
1.0
g .8-
1
0
b
1 -4'
Q.
0 0
O -2l
0.0,
2
D
D
D°o "
D D D ™ 00 0 QQ
D Q ° 0 D
D DQ D 0
" o n o ° n
DQD DDD D 0
D D
D D D
D «>
Q D Q D
a a a
a a
D
D D
a
a
0 40 60 80 100 120
Specific Humidity (Grains/Pound) - Start of Trip
Non-idle trips (weighted by number of trips)
Figure 5 - Humidity vs. Temperature for Phoenix Dataset
100
Q.
K
•5 80
m
^ 60
CD
I 20
CD
Q:
B a a
40"
...J...f.a°..D,
iSM.:i i!1"
oioHHaSaa °i§ll
III
o a a
a a
a B
60
70
80
90
100
110
Temperature (F) - Start of Trip
Non-idle trips
-------
Figure 6 - Heat Index
160-
LL, 140-
X
8
E 120-
^^
^ 100-
80 '
60
^
^^'
* ..— *
^
^
0
-Jr
J^
v*
****'** ^
^P
^f
^^
• *!
^^
*^
*"*
^^***
t
i _<
«*
/
#*
*
•»^^^
^^^
^
Rel Hum
80%
60%
40%
75 80 85 90 95 100 105
Temperature (F)
Note: Heat Index values based on shady conditions
Lines represent curve fit of tabular data
Figure 7 - Compressor-On vs. Heat Index
.8
6 61
b
.4
Q.
§ .2
O
0.0
60
70
80
90
100
110
Heat Index (F) - Start of Trip
Non-idle trips (weighted by number of trips)
R squared = 0.84
-------
Figure 8 - Compressor-On vs. Heat Index by Time of Day
1.0
•^ .8
o
CO
LL
c .6
CO
CO
CD
.4
0.0
60
70
80
90
100
Heat Index (F) - Start of Trip
Quadratic regression of non-idle trips
Weighted by trip duration
Period
Sunset-Sunrise
4 pm - Sunset
10 am -4 pm
Sunrise-10 am
110
Table 1 - Proposed "Raw" Demand Factor Equations
(Demand Factor = Constant + a* (Heat Index) + b*(Heat Index)2)
Period
Morning/Afternoon
Peak Sun
Daytime Combined
Night
All Combined
Constant
-2.930273
-5.307355
-4.101082
-1.257412
-3.631541
a
0.059110
0.113973
0.086382
0.006753
0.072465
b
-0.000213
-0.000521
-0.000367
0.000143
-0.000276
R2
0.54
0.17
0.43
0.52
0.44
-------
Table 2 - Proposed Demand Factor Equation Forms
Heat Index
65 & below
66
74
76
96
101
104
110 & above
Morning/Afternoon
Constant = 0
Daytime
Morning/ Afternoon
ii
All
ii
ii
Constant = 1
Peak Sun
Constant = 0
Daytime
ii
Peak Sun
ii
ii
All
Constant = 1
Night
Constant = 0
"
Night
"
"
All
"
Constant = 1
Figure 9 - Proposed Demand Factor Functions
65 70 75 80 85 90 95 100 105 110
Heat Index
Morning/Afternoon Peak Sun
•Night
-------
Figure 10 - Solar Radiation - Sunny and Cloudy Day (Fort Peck, MT)
1000
10 12 14 16 18 20 22
Hour of Day
-June 26 1997 June 24 1997
-------
Table 3 - Analysis of Variance (ANOVA) on Non-Idle Trip Dataset
with Vehicle as Random Effect
De
Source
Intercept
Average
Speed
Trip Duration
Soak Duration
Idle Fraction
Heat Index
Period of Day
Vehicle
Period *
Vehicle
Tests of Between-Subjects Effects
pendent Variable: Compressor Fraction (fraction of time engaged /all driving)
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Hypothesis
Error
Type III Sum
of Squares
.712
25.255
1 .607E-02
23.878
2.213E-02
23.878
4.694E-04
23.878
4.418E-03
23.878
2.687
23.878
.787
5.336
6.527
4.677
3.571
23.878
df
1
516.244
1
500
1
500
1
500
1
500
1
500
3
70.612
16
58.048
36
500
Mean Square
.712
4.892E-02
1 .607E-02
4.776E-02
2.213E-02
4.776E-02
4.694E-04
4.776E-02
4.418E-03
4.776E-02
2.687
4.776E-02
.262
7.556E-02
.408
8.056E-02
9.918E-02
4.776E-02
F
14.551
.337
.463
.010
.093
56.263
3.472
5.063
2.077
Sig.
.000
.562
.496
.921
.761
.000
.020
.000
.000
a 3.197E-03 MS(VEH) + 2.718E-04 MS(PERIOD * VEH) + .997 MS(Error)
b MS(Error)
c .541 MS(PERIOD * VEH) + .459 MS(Error)
d .638 MS(PERIOD * VEH) + .362 MS(Error)
-------
Figure 11 - Compressor-On vs. Trip Duration
Compressor-On Fraction
o ->•->•
o ho 4*. a> bo b ho
^""* x s
-------
Figure 13 - Proposed Base Market Penetration Estimates
72- 74 76 78 80 82 84 86 88 90 92 94 96 98 00+
Model Year
•LDV's LDTs
Figure 14 - Proposed Rate of Unrepaired Malfunctions
nacp/0
1 3 5 7 9 11 13 15 17 19 21 23
- R13te
-------
Table 4 - Proposed Rate of A/C Malfunction
Vehicle Age
(years)
1
2
3
4
5
6
7
8
9-25
Consumers
Reports*
<2%
2 - 5 %
2 - 5 %
2-5%
5 - 9.3 %
5 - 9.3 %
9.3 - 14.8 %
9.3 - 14.8 %
n/a
Proposed
Estimates
0.5 %
2.0 %
3.5%
5.0%
6.5 %
8.0 %
9.5 %
11.0%
12.5 %
1997 Automobile Purchase Issue
-------
Appendix A - Heat Index Equations
Source: Meisner and Graves, "Apparent Temperature", Weatherwise, August 1985
This set of equations computes heat index under "mild" and "severe" sultriness. Mild sultriness
indicates conditions under which thermal equilibrium can be achieved with reduced clothing
thickness. Severe sultriness indicates conditions for which reductions in the skin's resistance to
heat and moisture flow are required to achieve thermal equilibrium. If the required clothing
thickness is less than zero for the "mild" equations, the "severe" equations are used to calculate
heat index. This set of equations is based on an adult wearing trousers and a short-sleeved shirt,
walking in the shade at 3.1 mph, standard sea level air pressure, wind speed of 5.6 mph and a
vapor pressure of 1.6 kPa.
Inputs: TF = Temperature (°F), RH = Relative Humidity (%)
Variable
TC
ES
E
"Mild Sultriness":
HER
ERA
QV
EZA
HR
ARA
AZA
Q2U
QJ
K
L
F
RF
Equation
(TF - 32)*(5/9)
6.11*10A (7.567*TC)/(239.7+TC)
0.01*RH*ES
4.18+0.36*TC
1/(17.4+HER)
180*(143-0.00112*TC-0.0168*0.1*E)
0.060606/EHC
3.35+0.049*TC
1/(11.6+HR)
0.060606/CHC
((TB-TC)+(PB-PINF)*ERA/(ZS-EZA))/(RS+ERA)
(Q.QV-(1-0.84)*Q2U)/0.84
(.0387+ARA)+(0.0521+AZA)/0.124-((37-TC)+(5.65-
0.1*E)/0.124)/QJ
((0.0387+ARA)*(0.0521+AZA)) - ((37-
TC)*(0.0521+AZA)+(5.65-0.1*E)*ARA)/QJ)/R
K*K-4*L
0.5*(-K+ SQR(F))
Comment
Temperature in Celsius
Saturation Vapor Pressure
Relative Vapor Pressure
if <0use "Severe"
-------
Variable
DF
Wl
W2
W3
W4
W5
W6
W7
Heat Index
"Severe Sultriness":
HC
HR
RA
ZA
QU
=£ZS
R3
C
-------
Appendix B - Peer Review Comments and Response
Subsequent to publication of the draft version of this report in January 1998, the
document was put out for stakeholder review. Formal peer review comments were also solicited
from two independent sources. No comments were received through the stakeholder review
process, and thus the peer review comments represent the only external feedback received on this
report. This section contains these peer review comments in full and our response to these
comments. The author of the first set of peer review comments requested confidentiality, and
hence is not identified here.
Commentor 1 (EPA responses inserted as italicized text)
The comment text is verbatim from the commentor, minus editorial corrections to maintain the
confidentiality of the commentor.
General
The report documents a valuable initial step towards accurate estimation of exhaust emissions
resulting from auto a/c operation. The approach is appropriate for a first step: focus scarce data
collection resources on measuring emissions at extreme ambient conditions, and make a series of
assumptions to estimate emissions at other conditions. It is a valuable contribution to the
published literature because it presents and analyzes a unique set of data from instrumented
vehicles.
Based on our review we offer a variety of suggestions, which fall into three major categories:
- evaluate changes in some analytical assumptions
- try to extract more information from Phoenix data
- plan to accomodate or acquire additional data
By making these suggestions, we do not wish to detract from the value of the report and its
usefulness as an initial step. It is far better than nothing, yields results that provide a credible
basis for policy, and provides valuable insights into operational aspects of mobile a/c systems
that have never been published before.
Below we itemize suggestions by page and paragraph number in the draft report. But first we
will provide a general statement of the three types of suggestions.
1. Evaluate alternative analytical assumptions
a) It is likely that the incremental emissions due to a/c operation vary with engine
rpm. Presumably this effect is captured in the assumption that the effect of
compressor engagement on overall engine load is linear, at a given rpm and load,
thus accounting for the fact that engine efficiency is worse at idle. In other words,
we assume that this compressor runtime data will be used in conjunction with
measured data on emissions with compressor on and off, at each point in the
driving (test) cycle.
Response: An aggregate model such as MOBILE6 doesn 't explicitly address
-------
RPM. These effects are assumed to be captured on an average basis through the
use of representative driving cycles
b) Heat index was assumed to be a good indicator of humidity effects, but the
Phoenix data are too anecdotal to confirm that hypothesis. On theoretical
grounds, we would not expect it to be valid. It appears from this data that the
statistical significance of runtime dependence on heat index is not significantly
different from its dependence on outdoor temperature. As a measure of human
comfort, it might predict a driverOs decision to turn on the a/c unit. However it is
unlikely to have a significant effect on compressor runtime fraction. Runtime will
depend on the sum of the sensible and latent heat load. While the compressor is
on, the temperature of the evaporator coil is approximately 35 F on most cars.
The sensible load depends on the difference between the evaporator surface and
indoor air temperatures. Latent load depends on the difference between the
absolute (not relative) humidities of indoor air and that of 100% humid air at the
evaporator surface temperature (about 35 jF). In the absence of indoor humidity
data, the total latent load might be calculated directly from an estimate of the rate
at which outdoor air infiltrates into the car. Assuming the infiltration rates are
constant between Phoenix and other climates, the ratio could be used to
extrapolate Phoenix results to Houston and other climates. The amount of air
leaking into the car would have to be assumed.
Response: The demand factor developed for MOBILE6 attempts to account for
both driver behavior and vehicle behavior, and we think heat index is the most
appropriate method for extrapolation into high humidity conditions. It is beyond
the scope ofMOBILE6, and available data, to separate these effects out,
particularly given the complexity involved in accurately modeling the operation of
the compressor, which will vary from vehicle to vehicle. Thus, we believe the
deterministic approach relating ambient conditions to overall compressor runtime
continues to be the most appropriate approach.
c) Solar load is probably irrelevant to your methodology, because most cars that
automatically control compartment air temperature do so using reheat. So do
drivers who prefer more moderate temperature air blowing on them, or who use
high evaporator fan speeds to ensure that air is distributed to throughout the
vehicle. In virtually all cars the compressor runs whenever the evaporator is not
frosted. Since many drivers control indoor air temperature by running the heater,
they simply set the heater to a lower level on sunny days. Since emissions depend
only on the total sensible plus latent loads, the composition of sensible loads
(solar plus reheat) will not matter. The only case in which solar loads would
increase emissions is when drivers control indoor temperature solely by
decreasing evaporator fan speed. In the future, introduction of automatically-
controlled variable-displacement compressors may change this, so now is the time
to start understanding solar loads.
-------
Response: Solar load is very relevant in terms of initial cabin temperature, which
will effect whether driver turns the A/C on at the start of a trip. As noted, it will
be less of an issue once the vehicle is cooled down initially.
d) The Phoenix data set contains no data on air leakage into the car, since it is under
the control of the driver who sets it between zero and 100% fresh air intake to a/c
system (about 250-350 cfm max, depending on vehicle). Actually infiltration is
not zero even at idle, since even the "recirc" setting must prevent carbon dioxide
levels from reaching reaching dangerous levels when the car is full of occupants.
Air leakage is a strong determinant latent loads, but contributes only a fraction of
sensible loads.
Response: Air infiltration isn 't something that can be realistically accounted
within the scope of'the MO BILE 6 analysis
e) Humidity affects compressor runtime only indirectly, so the reliance on relatively
dry Phoenix data may not be as shaky as it may first appear. The effect of
humidity is filtered twice. The first effect on compressor runtime is the latent
load caused by the humid outdoor air infiltrating into the car; this increases load
and therefore compressor runtime. However at the same time the a/c system
efficiency increases because the water condensing on the evaporator surface raises
the refrigerant evaporating temperature. That will partially cancel the effect of
adding the latent loads. Therefore a rough estimate (based on a crude estimate of
infiltration loads, which vary with vehicle and average speed and fresh air intake
settings) should be good enough for now. It will probably be necessary to ask
auto manufacturers for an estimate of leakage rates as a function of vehicle speed
and ventilation settings, unless you can extract an average from the Phoenix data
(see below).
Response: Humidity is relied on for the heat-index calculation, primarily to
account for changes in driver behavior rather than compressor behavior
f) Fortunately your data were taken close to the fall equinox, so day length was
approximately 12 hours. If you attempt further studies of solar loads, or
calculations thereof, don't forget that day lengths in the US can exceed 15 hours
near the summer solstice
2. Extract more information from Phoenix data
a) The effect of solar load should be most noticeable on short trips dominated by the
"pulldown" period. On longer trips, solar is probably irrelevant due to use of the
car heater to control air temperature. Solar load (by increasing heat stored in the
passenger compartment's thermal mass) will determine the length of time needed
to cool the air to comfortable levels. Elevated indoor temperatures translate
directly into higher loads and to less-than-proportional increases in compressor
torque, because of slightly higher a/c system efficiencies. This may explain the
-------
results shown in Figure 1: compressor runtime is high for trips having high %
idle, because most of these are short trips with little cycling, in which pulldown of
"soak" loads are a large fraction of the total. If true, the same behavior might be
less noticeable in cloudier climates because the initial temperature of the vehicle's
thermal mass should be lower.
Response: To address this, we analyzed whether period of the day was a
significant variable for trips longer than 10 minutes; it was. However, it dropped
below 95 percent significance for trips above 15 minutes, which could be an
indication of the small sample size for trips of this length across the four periods.
In general, the average trip in MOBILE6 is estimated to be under 10 minutes.
Thus, the "pulldown " effect would predominate the usage ofA/C under average
MOBILE6 conditions.
b) Another hypothesis that might be tested concerns the slight decrease in
compressor runtime on trips having low % idle: most of these trips were at high
speed, so the a/c operated more efficiently due to the ram air effect cooling the
condenser; the effect of greater infiltration at high speed was probably offset by
the driver's use of reheat, since the cooling capacity of the a/c system at high rpm
probably still exceeded the loads, as might be indicated by the cycling rate. Such
an effect might have been hard to see in humid-climate data.
Response: The decrease was relatively minor, but was not statistically significant
due to small sample size. Because all trips were grouped together for the
calculation of air conditioning demand factors, this effect would be represented in
the results.
c) Try to extract an estimate of the contribution of latent loads from the Phoenix
data. Focusing on the longer trips, try to estimate the ratio of sensible to latent
loads using the method described in l(b) above. That is, assume that sensible
loads are proportional to (Toutdoor - 75 F) and that latent loads are proportional to
the difference between absolute humidities of indoor and outdoor air.
Response: Data isn 't available on indoor air humidity to perform this
calculation.
3. Accomodate and acquire additional data
a) Intensive testing of mobile a/c systems is underway at the Air Conditioning &
Refrigeration Center, University of Illinios at Urbana-Champaign. That will
provide compressor power and torque as a function of rpm and indoor temperature
and humidity, and outdoor ambient temperature. It should be published within the
coming year, so it would be wise to modify your models in anticipation of such
data.
b) The next time cars are instrumented by EPA for testing, perhaps the estimate of
-------
latent loads could be improved by measuring indoor humidities and temperatures.
Indoor humidities should be quite stable while a/c cycling rates are high; but
indoor humidity could increase during long off-cycles. Also it might be useful to
monitor fan and ventilation settings (fresh air vs recirculation), and the use of the
heater (on/off at least). All these driver-controlled variables affect a/c compressor
power, and measuring them for the purposes of making estimates is difficult.
Now that a/c emissions are monitored by EPA, it is possible that future a/c
systems will be controlled automatically to reduce emissions.
c) Ultimately it would be helpful to know how compressor rpm relates to vehicle
speed, on an aggregate basis. That would facilitate linkage of the University of
Illinois a/c test results to various driving cycles.
d) Nearly all today's cars cycle the compressor to keep the evaporator free of frost,
and use the car's heater or the evaporator fan to modulate a/c capacity. Both are
inefficient. Advanced systems using variable-displacement compressors can run
the compressor continuously, operate the evaporator at (higher, above-freezing)
temperatures that increase a/c system efficiency, with zero reheat. The overall
reduction in sensible load due to elimination of reheat, plus the higher operating
efficiency, may lead to substantial reductions in emissions. Automatic control of
ventilation is another future technology that might significantly reduce both
sensible and latent loads.
Response: We will take these comments into account for future testing of air
conditioning activity.
4. Detailed comments
a) p 2 para 3: Data on compressor power as a function of rpm and evaporating and
condensing temperature may be available from compressor manufacturers'
calorimeter testing. It will tell you nothing about load (cycling rates etc) but
would be better than nothing. Temperature differences between the refrigerant
and air, indoors and outdoors, probably vary little with climate and do not vary
much among cars.
Response: Modeling an effect as a function of rpm would requires far more
resolution than MOBILE6 provides; in addition, activity information of
evaporating and condensing temperature isn 't readily available, so it could not be
easily implemented in an aggregate emission model.
b) p 2 para 3: If in the future you want to improve predictions of compressor runtime
as a function of climate and driving cycle, you will need data on cycling rate as
well as compressor runtime. Transient losses and latent loads may depend
strongly on the cycling rate (e.g. whether 50% runtime results from 10 cycles of 6
min duration or 100 cycles of 0.6 min duration).
Response: The Phoenix data only provided the total time the compressor was on
per trip, instead of real-time data. Future testing may be able to address this
-------
point, since real-time compressor function will be a target variable.
c) p. 2 para 3: Your method appears to assume that compressor runtime is
independent of rpm. It also acknowledges that this is probably untrue. We agree
that it may be important to account for this because emissions vary so greatly with
rpm.
d) p 4 para c: Mohinder Bhatti of Delphi Thermal Systems published an SAE paper
in 1997 which assumed that idling accounted for 15% of compressor runtime.
The Phoenix data appear to violate this assumption. You might want to ask him
for the source of his data.
Response: the Phoenix data indicates that trips with 100 percent idle have
relatively low compressor-on time, but this doesn 't necessarily contradict Dr.
Bhatti's findings. The compressor fraction was significantly higher for trips with
very high (but less than 100 percent) percentages of idle. The all-idle trips in
Phoenix may not be representative of idle operation in general -for example, they
may reflect very short key-on events. If all idle operation could be separated
from the Phoenix dataset (including the idle events on non-idle trips), it is likely
the overall compressor-on time would be high.
e) p 4 para 5: We doubt that humidity will account for half of total loads, as it might
on the test procedure. But it raises an interesting possibility for measuring indoor
compartment humidity (and perhaps measuring condensate removed) on the test,
and using it to calculate infiltration rates as a function of ventilation setting (fresh
air vs recirculation). If reheat is not used on the EPA test, the total sensible load
seen by the a/c is the sum of conduction and infiltration and solar. By controlling
the solar load and varying wind tunnel speed, EPA would have a unique
opportunity to gather data on load composition.
f) p. 5 para 1: Comparing noontime humidity to daily maximum temperature mixes
apples and oranges. If it is 85 F and 40% humidity at noon, then the temperature
peaks at 95 at 4pm, the relative humidity falls below 30% while the absolute
humidity remains constant. It is the absolute humidity that determines latent load.
Air at 40% relative humidity contains 40% more water at 95 than at 85 F.
Response: this comparison was made simply to illustrate that humidity levels in
other parts of the U.S. can be significantly higher than those observed in Phoenix.
g) p 5 para 2 and 3: Fig 6 is not much different from Fig 2. Since humidity is so low
in Phoenix, it is not clear that one can conclude from this data that the strong
correlation between compressor runtime and heat index is better than a method
based on the correlation with outdoor temperature, adjusted for absolute
humidities.
Response: The purpose of using heat index as the predictor variable instead of
-------
temperature is to develop a methodology to account for humidity as well as
temperature, rather than to find the best fit for the Phoenix data. The heat index
levels in Phoenix are driven almost exclusively by temperature, since very high
humidity levels are required to affect the index beyond temperature.
h) p 5 para 3: It may be true that using heat index is better than ignoring humidity,
but we suggest that you try to track the physics better by using absolute humidities
to estimate latent loads.
Response: Absolute humidity was was found to not be a particularly good
predictor of compressor fraction, as shown in Figure 3.
i) p. 8, para 1: Assuming a sunny day is probably not a problem for most cases, since
reheat will make up for loads blocked by clouds.
j) p 8 para 1: Radiative heat loading is very complex, and some of the suggested
assumptions are problematic. For example even with 100% cloud cover, solar
collectors work very well collecting the diffuse component of solar radiation.
And nighttime radiative cooling reduces need for a/c, especially on clear nights
when the surface of the car is radiating to the cold blackness of space.
Response: this suggests that the treatment of 100 percent cloud cover as
"nightime " may be overstating the effect of cloud cover (which is very small as it
is). Because solar radiation data (or even reliable cloud cover data) was not
available for Phoenix, we don't have the ability to separate "direct" and
"diffuse " radiation.
k) p 8 para 3: Soak duration is likely to have a large effect on short trips having little
cycling, because of the effect on pulldown load. After indoor temperature
stabilizes, effects of soak have vanished.
1) Fig 1: Clarify that the first bar is deleted from the data set.
m) Fig 2: It is clear that each data point represents the time-weighted average runtime
of all data points in that temperature bin. But is the least squares fit weighted too?
If so, explain how. Do the averages include trips taken with the a/c off for the
whole time? Is there any systematic dependence of errors and scatter on long vs
short trips or those having higher/lower cycling rates?
Response: The demand factor curves were generated from data aggregated over
every trip by heat index level. This approach accounts for the disparity in trip
length implicitly, by calculating compressor fraction as the total time the
compressor is engaged divided by total trip time at a given heat index. This
approach also reduces the influence of random effects (such as vehicle), since the
aggregate trips for a given heat index are comprised of several vehicles.
n) Fig 6: Instead of using a parabola, why not select an exponential or other curve
that naturally has a horizontal asymptote?
-------
o) Fig 10: All curves should approach a horizontal asymptote. It would help if data
points were shown along with the curve fits. Try re-fitting with exponential curve
to capture asymptote?
Response: While the quadratic approach with truncation may not be the most
elegant, the terms are significant, and another approach would be unlikely to
yield results much different from the current approach.
p) Fig 11: Lack of dependence on speed suggests that ram air flow into condenser
improves a/c system efficiency. Load reductions due to wind cooling the solar-
heated metal, and load increases due to increased infiltration are probably
neutralized by reheat.
q) Table 1: Explain how this relates to Fig. 7. Also show formulae or explain better
how the numbers were calculated.
5. Questions
a) p 2 para 1: is the 850 W/m2 solar load applied to a car in a wind tunnel or is the
vehicle stationary? Stationary test will overestimate effect of solar, since car
surface will be hotter than while driving.
Response: The vehicle is stationary, but a variable speed fan provides cooling
and presumably produces conditions more representative of the road
p 2 para 3: Compressor ontime fraction should be defined carefully to clarify whether the
denominator is total trip time, or only that fraction of the trip during which the a/c system was
switched on.
Response: Compressor time is the fraction of compressor time divided by total
trip time, regardless of whether the A/C was on or not.
Commentor 2:
John Warner, PhD
Center for Statistical Consultation and Research
University of Michigan
Verbatim comments not available electronically. Hard copies available upon request. Comment
and responses are summarized below:
Comment: Given that the analysis is focused on the population of the Phoenix dataset as
a whole, rather than making inferences about idividual drivers/vehicles, vehicle should be
included as a random effect. Ignoring random effects may lead to unrealistically small
standard errors and unrealistically low p-values.
Response: The analysis for determining the significance of individual variables was
redone, adding vehicle as a random variable. The fundamental conclusions (as predictor
-------
variables, only heat index and period are significant) are unchanged. The AN OVA
results presented in Table 3 have been updated to reflect this change.
Comment: Main effects should be tested individually, or alternately it should be made
clear what approach was taken to determine the significance of individual effects.
Response: The methodology for performing the Analysis of Variance on main effects has
been added to the report.
Comment: There are likely approaches to modeling demand factor which are better than
a quadratic fit, given the level of truncation required with the quadratic curves to maintain
results between 0 and 1. Examples include a non-linear least-squares with random effects
(which a quadratic form would adderss); and "ideal" component would be to give higher
weights to trip lengths and for compressor fractions close to zero or one.
Response: The demand factor curves were generated from data aggregated over every
trip by heat index level. This approach accounts for the disparity in trip length implicitly,
by calculating compressor fraction as the total time the compressor is engaged divided by
total trip time at a given heat index. This approach also reduces the influence of random
effects (such as vehicle), since the aggregate trips for a given heat index are comprised of
several vehicles. While the quadratic approach with truncation may not be the most
elegant, the terms are significant, and another approach would be unlikely to yield
results much different from the current approach.
------- |