You are on page 1of 15

VOLUME 15, NUMBER 1 HVAC&R RESEARCH JANUARY 2009

101
A Smart Mixed-Air Temperature Sensor
Adam Wichman James E. Braun, PhD
Student Member ASHRAE Fellow ASHRAE
Received December 29, 2007; accepted June 15, 2008
Accurate mixed-air temperature (MAT) measurements are notoriously difficult to obtain in
packaged air-conditioning equipment for small commercial applications because of space con-
straints and the use of small mixing chambers. However, MAT is important for control of econo-
mizers and in diagnostic algorithms for both economizers and vapor compression equipment.
This paper demonstrates that a single-point measurement of MAT can provide accurate results
when combined with a correlation for bias error that depends on damper control signal and dif-
ference between outdoor and return-air temperature. The correlation could be determined using
a self-calibration procedure with measurements of mixed, outdoor, return, and supply air tem-
peratures. A system was tested in a laboratory over a wide range of outdoor conditions and
damper positions, and the single-point measurement errors were reduced by about a factor of
five when using the bias correction. Furthermore, errors in outdoor air fraction (OAF) deter-
mined from temperature measurements were reduced by about a factor of four.
INTRODUCTION
Economizer systems typically use mixed-air temperature (MAT) measurements for feedback
control of the outdoor and return-air dampers. In addition, diagnostic methods for both econo-
mizer systems and air-conditioning equipment require accurate measurements of MAT. How-
ever, packaged air-conditioning equipment for small commercial applications typically have
small chambers for mixing outdoor and return air and can have very nonuniform temperature
and velocity distributions at the inlet to the evaporator. As a result, there can be significant bias
errors associated with employing single-point and averaging sensors (Avery 2002; Carling and
Isakson 1999; Robinson 1999). Furthermore, the mixing process can change significantly as the
position of the dampers changes with economizer operation. An array of at least four tempera-
ture sensors mounted symmetrically about the duct centerline may be necessary to achieve good
accuracy for MAT (Wichman 2007).
The requirement for employing averaging sensors adds significant cost and may not ensure
accuracy. Recently, a method was developed (Lee and Dexter 2005; Tan and Dexter 2005, 2006)
that corrects for the bias error associated with employing a single-point sensor. The method
employs fuzzy sensor fusion, whereby the results of single-point measurement are fused with
temperature distribution predictions from a CFD model for the mixing process. Although it was
demonstrated that this method can provide improved estimates of the MAT, it is not obvious how
the method can be generalized without requiring that CFD simulations be performed for any spe-
cific geometry of mixing box and sensor placement that may be encountered.
This paper develops and evaluates a method for estimating the bias error associated with a
single-point measurement. The method only requires single-point temperature measurements for
mixed, return, outdoor, and supply air that are already typically available for diagnostics and
Adam Wichman is a reactor engineer at PSEG Nuclear LLC, Hancocks Bridge, NJ. James E. Braun is a professor at
Purdue University, West Lafayette, IN.
2009, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in HVAC&R Research, Vol. 15,
No. 1, January 2009. For personal use only. Additional reproduction, distribution, or transmission in either print or digital form is not permitted
without ASHRAEs prior written permission.
102 HVAC&R RESEARCH
economizer control. The combination of a single-point measurement and bias prediction could
be integrated into a smart-MAT sensor to allow self calibration and dramatically improved accu-
racy. The proposed method was evaluated using data obtained for an air-side economizer and
mixing chamber integrated with a typical small commercial rooftop air conditioner.
SYSTEM DESCRIPTION, MEASUREMENTS, AND
MIXING PERFORMANCE
Experiments were performed on a five-ton rooftop air conditioner equipped with an econo-
mizer. The rooftop unit was set up inside environmental chambers to simulate indoor and outdoor
conditions. A diagram of the experimental setup is shown in Figure 1. Figure 2 shows a picture of
the mixing chamber and dampers depicting the directions for airflow and a filter that is just
upstream of the evaporator coil. The evaporator coil is located in very close proximity to the out-
door and return-air dampers, and the air intakes are not symmetrical, leading to very poor mixing.
The economizer damper was configured to be manually controlled using a 210 V variable
DC voltage source to the motor. The motor had a DC voltage output indicating damper position
using the same 210 V scale as the input, where a 2 V input returns the damper minimum posi-
tion of about 5% outdoor air and the 10 V input returns a damper position of about 75% outdoor
air. Due to the flimsy construction of the return-air damper, significant leakage passed the return
damper when it was closed at the 10 V motor input, preventing the economizer system from
achieving an outdoor-air fraction (OAF) greater than 75%. For the purpose of characterizing
damper position, a normalized damper-control signal ( ) was determined as
, (1)
where DS is the damper control signal.
Figure 1. Experimental setup for economizer (not to scale).

D
DS 2V
8V
-------------------- =
VOLUME 15, NUMBER 1, JANUARY 2009 103
Four different measurement grids were set up in the unit with a total of 33 type T thermocou-
ples, each having a rated accuracy of 1.8R (1K). Nine thermocouples were arranged on grids
at both the outdoor and return-air inlets. Fifteen thermocouples were placed symmetrically in five
rows and three columns at the inlet to the evaporator coil to measure the MAT. Air-sampling
tubes used to measure dew point were attached to the center of both the outdoor and return-air
measurement grid. Ten sampling tubes were also fastened to the MAT grid, with two tubes in
each of the five rows that contained the temperature sensors. The sampling tubes were connected
to a dew-point hygrometer via an air-sampling pump. Figure 3 shows the mixed-air measurement
grid where T represents thermocouple measurements and D represents dew-point sampling tubes.
Figure 2. The economizer air-mixing chamber with arrows demonstrating direction of
airflow.
Figure 3. Mixed-air measurement grid.
104 HVAC&R RESEARCH
The temperature of the supply air after evaporator and fan was measured using nine thermo-
couples also mounted as a grid. Supply and outdoor airflows were also measured to determine
the outdoor-air fraction (OAF). The rooftop unit supply air was measured using standard noz-
zles, whereas an array of hot-wire anemometers was used to determine outdoor airflow rates.
Data were collected over the entire range of damper positions for both heating and cooling
operating conditions as summarized in Table 1. The first set of tests was for heating mode at
20F, 30F, and 40F outdoor-air temperature (OAT) and 70F return-air temperature (RAT) for
three different airflow rates. Through this testing, it was verified that the mixed-air conditions
were nearly independent of the supply airflow rate for a given damper position. As a result, it
was deemed unnecessary to conduct cooling-mode tests at different airflow rates, and a flow rate
of 2000 cfm (943.9 l/s) was used for tests in cooling mode.
One approach used to check the validity of the data was to compare OAF values determined
directly from airflow measurements to OAF data calculated from temperature data. It can be
shown using energy and mass balances on the mixing chamber within the economizer that the
OAF can be estimated from temperature measurements as follows:
(2)
where OAF is the ratio of outdoor airflow to the total supply airflow rate, OAT is outdoor-air
temperature and RAT is return-air temperature. OAF determined in this manner is often utilized
as a low-cost measure of ventilation flow.
Table 1. Economizer Test Matrix
Mode
Total Airflow,
cfm (L/s)
% OA
OAT,
F (C)
Outdoor RH,
%
RAT,
F (C)
Indoor RH,
%
Cooling 2000 (943.9) 575 55 (12.8) 4075 75 (23.9) 50
Cooling 2000 (943.9) 575 65 (18.3) 40100 75 (23.9) 50
Cooling 2000 (943.9) 575 75 (23.9) 40100 75 (23.9) 50
Cooling 2000 (943.9) 575 85 (29.4) 2075 75 (23.9) 50
Cooling 2000 (943.9) 575 95 (35.0) 2040 75 (23.9) 50
Cooling 2000 (943.9) 575 105 (40.6) dry 75 (23.9) 50
Cooling 2000 (943.9) 575 115 (46.1) dry 75 (23.9) 50
Heating 2000 (943.9) 575 20 (6.7) dry 70 (21.1) Dry
Heating 2000 (943.9) 575 30 (1.1) dry 70 (21.1) Dry
Heating 2000 (943.9) 575 40 (4.4) dry 70 (21.1) Dry
Heating 1600 (755.1) 575 20 (6.7) dry 70 (21.1) Dry
Heating 1600 (755.1) 575 30 (1.1) dry 70 (21.1) Dry
Heating 1600 (755.1) 575 40 (4.4) dry 70 (21.1) Dry
Heating 1200 (566.3) 575 20 (6.7) dry 70 (21.1) Dry
Heating 1200 (566.3) 575 30 (1.1) dry 70 (21.1) Dry
Heating 1200 (566.3) 575 40 (4.4) dry 70 (21.1) Dry
OAF
MAT RAT
OAT RAT
------------------------------- =
VOLUME 15, NUMBER 1, JANUARY 2009 105
Figure 4 shows calculated OAF as a function of the measured OAF for data obtained in this
work. For these results, all of the available thermocouple measurements were used to determine
average temperatures. Overall, the calculated OAFs were within 8% of the measured values over
the range of conditions considered. This is within the experimental uncertainty of the OAF val-
ues determined from temperature measurements. On average, the calculated OAF deviated from
the measured values by about 2%.
Figure 5 shows the impact of using a reduced number of MAT sensors for averaging on the
average error compared to using 15-point averaging. The results are presented as a function of the
difference between the outdoor and return-air temperatures, where each point resulted from aver-
aging the absolute values of the errors for all damper positions. The sensor numbers specified in
Figure 4. Calculated OAF as a function of measured OAF.
Figure 5. Average temperature errors between the best 1-, 2-, and 4-point combinations of
sensors and the 15-point average as a function of OAT RAT.
106 HVAC&R RESEARCH
the legend refer to the labels presented in Figure 3 and represent the best locations for one, two,
and four-point averaging sensors, as determined by the data. As would be expected, the average
differences between using a reduced number of sensors and the 15-point measurement decrease
nearly to zero as the difference between the OAT and RAT approaches zero. The use of a single
sensor in the center of the measurement grid produced a maximum error of about 4.2F (2.3C) at
the lowest OAT, and a root-mean-squared error (RMSE) of 2.7F (1.5C). The two-point combi-
nation of sensors had maximum and RMS errors of 2.7F (1.5C) and 0.8F (0.44C), whereas
the use of four sensors had maximum and RMS errors of 1.7F (0.94C) and 0.6F (0.33C).
Although the use of a single MAT sensor would not be adequate under many operating condi-
tions, the use of a single, centerline humidity sensor would work well for typical air-conditioning
conditions. Table 2 shows RMSE associated with using single or two-point averaging of
mixed-air dew-point sensors (see Figure 3 for sensor locations) as compared with 10-point aver-
aging. The evaluation was done for cooling-mode tests with outdoor air at 95F (35C) and 40%
RH, and indoor air at 75F (23.9C) and 50% RH for ten different damper positions ranging from
5%75% OAF. The use of individual sensorseither D
3
and D
8
provided good accuracy com-
pared to use of all ten dew-point sensors. Slightly better accuracy was achieved by averaging sen-
sors D
3
and D
8
.
SMART MIXED-AIR TEMPERATURE SENSOR
The goal is to correct a single-point MAT measurement to account for nonuniform temperature
distributions. The method presented here utilizes a combination of available temperature sensors
(return, outdoor, mixed, and supply air) to estimate the bias error in the MAT under conditions
where the compressor is not operating. When the evaporator is not providing cooling, an accurate
estimate of the MAT can be obtained by subtracting the fan temperature rise from a supply-air
temperature measured after the supply fan. Air within the supply-air duct is assumed to be well
mixed, since it has passed through the evaporator, fan, and ductwork. The fan-temperature rise
can be estimated under conditions where the MAT measurement is accurate, such as when the
outdoor-air damper is closed tight or when the OAT and RAT have similar values. Once the
fan-temperature rise is known, the bias error can be determined with the compressor OFF for any
damper position and outdoor condition as the difference between the single-point MAT and base-
line MAT (i.e., supply-air temperature corrected for fan-temperature rise). Data would be col-
lected for a range of damper positions and outdoor conditions in order to determine a model for
the bias error. In order to facilitate the training process, the damper could be sequenced through
the entire range of damper positions under various outdoor conditions that are expected to pro-
duce large errors (e.g., hot and cold). The single-point MAT, corrected for bias error, could then
be used during times when the compressor is operating.
Correcting for Temperature Rise across the Fan
The fan-temperature rise, , can be determined when the compressor is OFF during a
self-calibration mode under conditions where the single-point MAT measurement is known to
Table 2. Root-Mean-Squared Errors (RMSEs) of Single- and Two-Point Dew-Point
Measurements Compared to the Ten-Point Dew-Point Measurement
Dew Point Sensor(s) RMSE, F (C)
D
3
0.61 (0.34)
D
8
0.50 (0.28)
Two-point (D
3
and D
8
) 0.48 (0.27)
T
f an
VOLUME 15, NUMBER 1, JANUARY 2009 107
be accurate. One such condition can be achieved by closing the outdoor damper if the damper
seals well. In this case, the mixing process is eliminated and the mixed-air stream should have a
uniform temperature. The ability of the outdoor damper to close tightly can be evaluated by
comparing RAT to MAT, with the damper control signal set to its minimum value. If these tem-
peratures are significantly different, then it is necessary to collect MAT data when the outdoor
and return-air temperatures are close together. In either condition, the can be calculated as
the difference between the SAT and MAT or
, (3)
where SAT is supply-air temperature measured after the supply fan.
The rooftop unit incorporated a fixed-speed fan so that it was only necessary to determine a
single fan-temperature rise correction. In the economizer system tested in this study, the dampers
did not seal well, and 12 data points were collected for conditions with the outdoor and return-air
temperatures nearly the same. The differences between the SAT and a single-point MAT mea-
surement (MAT
1pt
) were calculated for each of these data points and then averaged to give an
estimate for of 1.19F (0.66C). The that was measured for these 12 data points var-
ied from 1.11F to 1.36F (0.62C to 0.75C). For systems with a variable-speed fan, it should be
possible to correlate fan-temperature rise with input control to the fan.
With the compressor turned off, the supply-air temperature (SAT) can be corrected for the
temperature rise across the fan to obtain a good baseline representation of the MAT as follows:
(4)
The baseline MAT can then be used to determine bias errors for the single-point mea-
surement.
The fan-temperature rise estimate was used to correct the measured SAT for all of the data
points collected in this study with the compressor OFF at different outdoor conditions and damper
positions as described in Table 1. Figure 6 shows differences between SAT
corrected
and the
T
f an
T
f an
SAT MAT =
T
f an
T
f an
MAT SAT
corrected
SAT T
f an
=
Figure 6. Corrected SAT compared to 15-point MAT (test matrix of Table 1).
108 HVAC&R RESEARCH
15-point average MAT (MAT
15pt
) as a function of the normalized damper position sorted accord-
ing to the difference between OAT and RAT. The overall RMS difference was calculated to be
0.87F (0.48C) for all of the experimental data. The most significant differences occur for low
OAT (cold ambient conditions) for damper signals within the middle of the control range. This
combination of conditions results in very nonuniform velocity and temperature distributions
within the mixing section but may be unrealistic since the damper would most likely be closer to
a minimum position for cold or hot outdoor conditions where economizer control would not be
employed. Furthermore, the primary goal for the smart sensor is to determine good estimates of
MAT under conditions where the air conditioner is running. For these conditions, the differences
between SAT
corrected
and MAT
15pt
are generally less than 1F for this system.
The SAT
corrected
may be a better measure of the actual MAT than the MAT
15pt
, because the
velocity profile is likely to be more uniform, and the fan promotes mixing. The differences in
Figure 6 are larger under conditions where nonuniform velocity distributions are expected to
have the greatest effects. It should be noted that in some applications, the supply-air velocity
profile could also be very nonuniform, depending on sensor location and duct transitions, bends,
etc., downstream of the fan. For the remainder of this paper, SAT
corrected
is referred to as the
baseline MAT, MAT
baseline
.
Effect of Damper Position on Mixed-Air Temperature Bias Error
In this study, the bias error in a single-point measurement of MAT was evaluated as
, (5)
where MAT
1pt
was measured at the centerline of the duct at the inlet to the evaporator filter.
Figure 7 shows the relationship between the single-point MAT
error
and normalized damper
control signal ( ) determined from the laboratory measurements. The data is organized into
groups based on the difference between OAT and RAT to demonstrate how the MAT
error
becomes larger as the difference between the OAT and RAT becomes larger. Also, the sign of
MAT
error
is positive when the OAT is less than the RAT and negative otherwise. The MAT
error
MAT
error
MAT
1pt
MAT
basel ine
=
Figure 7. Single-point MAT
error
as a function of
D
.

D
VOLUME 15, NUMBER 1, JANUARY 2009 109
has an extremely nonlinear relationship with due to the damper geometry and sensor place-
ment. The errors are very large at low values of , because the single temperature sensor was
exposed mostly to return air. When is around 0.3, the MAT
error
is significantly less, because
the return damper was closed to a point where the single sensor experienced a good mixture of
outdoor and return air. The MAT
error
is higher for larger values of , because the sensor
encountered mostly outdoor air. It is interesting to note that the errors for the single-point sensor
are still significant for the two limiting cases where the return damper is closed and the outdoor
damper is closed. This is because these dampers do not seal very well. Therefore, at large differ-
ences in return and outdoor temperature, the difference between maximum and minimum tem-
peratures within the cross-section of the mixed-air section can be significant. Furthermore, the
location of the single sensor at the center of the mixing chamber is a poor representation for
these two extreme damper positions.
Correlating Mixed-Air Temperature Bias Error
The results of Figure 7 indicate that there is a highly nonlinear dependence of MAT
error
on
damper control signal when a single-point measurement is employed. Simple analytical func-
tions cannot capture the nonlinearity. However, for any given damper position, the bias error has
a nearly linear dependence on the difference between the outdoor and return-air temperature. As
a result, a simple but effective approach is to utilize separate linear bias models for different
ranges of damper control signal.
To evaluate this approach, the economizer data was divided into ten equally sized bins
according to normalized damper position from 0 to 1. Separate linear correlations were then cre-
ated for each bin. Figures 8 and 9 show typical results for two of the damper bins. In general, the
single-point MAT error correlates reasonably well with damper position and the difference
between OAT and RAT.

D
Figure 8. Single-point MAT
error
as a function of the difference between OAT and RAT for

D
in the range of 00.1.
110 HVAC&R RESEARCH
The resulting correlations were used to predict the MAT
error
, which was then used to cor-
rect the single-point MAT. Figure 10 compares corrected single-point and baseline values for
MAT for all of the data points collected in this study (as described in Table 1). The corrected
MAT agrees very well with the baseline values, with an overall RMS error of 0.57F
(0.32C). This corresponds to a factor of about 5 reduction in error, compared to using an
uncorrected single-point measurement.
Impact on Estimating Outdoor-Air Fraction (OAF)
The corrected single-point MAT can provide significant improvement in estimates of OAF
that are determined from temperature measurements using Equation 2. Figure 11 shows compar-
isons between OAF calculated with the uncorrected single-point MAT and the baseline MAT. In
this case, the RMS difference between the sets of data for OAF is 0.091. The errors are signifi-
cantly greater at values of OAF less than about 0.3. Figure 12 compares OAF calculated with the
corrected MATs to OAFs determined with baseline MAT values. The use of the 10-bin bias cor-
relation resulted in an RMS error of only 0.025, which is nearly a factor of four improvement
compared with using the uncorrected measurements.
TRAINING WITH LIMITED DATA
The results presented in the previous section were developed using a full set of temperature
data with OATs that ranged from 20F to 115F (6.7C to 46.1C). If this system were imple-
mented in the field, it could take at least six months for the building system to encounter such a
large range of conditions. The case studies presented in this section consider the effectiveness of
the correlations in extrapolating when trained with limited data from winter and summer periods.
Winter Training Data
For this case, data points for OATs ranging from 20F to 40F (6.7C to 4.4C) were used to
develop correlations for predicting MAT
error
. Using just this data, there were several data points
Figure 9. Single-point MAT
error
as a function of the difference between OAT and RAT for

D
in the range of 0.040.5.
VOLUME 15, NUMBER 1, JANUARY 2009 111
for each bin of normalized damper control signal. This correlation was then applied to all of the
data, and the RMS errors were calculated for the corrected MAT and the calculated OAF. Fig-
ures 13 and 14 compare MAT and OAF values determined from single-point, corrected mea-
surements with the baseline values for all of the data. The RMS errors for MAT and OAF are
1.1F (0.61C) and 0.06, respectively. This is a significant improvement over the uncorrected
single-point measurement but was not as effective at correcting the MAT as the correlation
developed using all of the economizer data.
Figure 10. Corrected single-point MAT compared to the MAT
baseline
using the 10-bin
correlations (test matrix of Table 1).
Figure 11. OAF calculated with uncorrected single-point MAT, compared to OAF deter-
mined with the baseline MAT.
112 HVAC&R RESEARCH
Summer Training Data
The correlation for this case was developed using OAT data in the range of 85F to 105F
(29.4C to 40.6C). This data set did not cover the entire normalized damper control signal range
as well as for the winter case. Several of the bins only had two data points. As a result, the correla-
tion developed using this data yielded a very poor correction when applied to data outside of the
range used for training. Figures 15 and 16 show results for MAT and OAF over the entire range of
economizer data for correlations trained with the limited summer data. The RMS errors calculated
for the MATs and the OAFs are 8.2F (4.6C) and 0.23, respectively, which are significantly
Figure 12. OAF calculated with the 10-bin corrected, single-point MAT, compared to
OAF calculated with the baseline MAT.
Figure 13. Corrected single-point MAT correlation trained using winter data, compared
to the MAT
baseline
.
VOLUME 15, NUMBER 1, JANUARY 2009 113
worse than the results for the uncorrected single-point MAT. The bias error correction should not
be applied to conditions that are significantly different than those used for training. However, in a
practical application, the correlation could be continually updated, and the algorithm would prob-
ably not be required to extrapolate much beyond the range of data used for training.
CONCLUSIONS
This paper demonstrates that the bias error associated with using a single-point sensor for
MAT can be correlated with damper control signal and the difference between the outdoor and
Figure 14. OAF calculated with the single-point MAT correlation trained with winter
data, compared to OAF calculated with the baseline MAT.
Figure 15. Corrected single-point MAT correlation trained using summer data, compared
to the MAT
baseline
.
114 HVAC&R RESEARCH
return-air temperatures. A correlation for the bias error can be learned when the compressor is
off, using measurements of mixed, supply, return, and outdoor air temperature. When trained
with data covering a wide range of outdoor conditions and damper positions, the correction to
the single-point measurement provided a factor of five improvement in the accuracy of MAT.
Furthermore, estimates of OAF determined from temperature measurements were improved by
nearly a factor of four. However, it was found that a model for the bias determined from limited
data may not extrapolate well to other operating conditions.
In this study, only a constant-air-volume (CAV) system was considered. This is representa-
tive of packaged air conditioners used in small commercial buildings within the U.S. that utilize
ON/OFF control of compressors for capacity control. However, it is anticipated that the method
could be adapted to consider variable-air-volume (VAV) systems used in larger commercial
buildings. For this case, it should be possible to learn a correlation for fan temperature rise in
terms of input control signal to the fan. This is a topic for future work.
It is envisioned that the bias correction model would be self calibrating. During initial calibra-
tion or recalibration periods, the damper could be sequenced through its entire range of operation
with the compressor OFF. Data could be collected for each damper position, and linear correlations
for the dependence on the difference between OAT and RAT could be updated recursively using
the new data. Alternatively, the data could be stored in two-dimensional bins of damper position
and difference in OAT and RAT. Individual entries within the two-dimensional table could be
updated using simple averaging or averaging that weighs recent data more heavily (e.g., exponen-
tially weighted averaging). Interpolation can then be used for predicting the single-point measure-
ment bias after sufficient data has been collected in the range where the prediction is needed.
The method presented in this paper is primarily intended for single-point MAT measure-
ments. However, it could also be used in systems where multiple sensors are averaged to repre-
sent the MAT.
Figure 16. OAF calculated with single-point MAT correlation trained with summer data
compared to the OAF calculated with the baseline MAT.
VOLUME 15, NUMBER 1, JANUARY 2009 115
REFERENCES
Avery, G. 2002. Do averaging sensors average? ASHRAE Journal 44(12):4243.
Carling, P., and P. Isakson. 1999. Temperature measurement accuracy in an air-handling unit mixing box.
Proceedings of the 3rd International Symposium on HVAC, ISHVAC 99, Shenzhen, China.
Robinson, K.D. 1999. Mixing effectiveness of AHU combination mixing/filter box with and without fil-
ters. ASHRAE Transactions 105(1):8895.
Wichman, A. 2007. Evaluation of fault detection and diagnosis methods for refrigeration equipment and
air-side economizers. Masters thesis, Ray W. Herrick Laboratories, School of Mechanical Engineering,
Purdue University, West Lafayette, IN.
Lee, P.S., and A.L. Dexter. 2005. A fuzzy sensor for measuring the mixed air temperature in air-handling
units. Measurement 37(1):8393.
Tan, H., and A.L. Dexter. 2005. Improving the accuracy of sensors in building automation systems. Pro-
ceedings of the 16th IFAC World Congress, Prague, Czech Republic.
Tan, H., and A.L. Dexter. 2006. Automated commissioning of a cooling coil using a smart mixed-air tem-
perature sensor. Proceedings of the 7th International Conference on System Simulation in Buildings
(SSB 2006), Paper P14, Liege, Belgium.

You might also like