Beruflich Dokumente
Kultur Dokumente
Jeff Stein is a senior engineer and Mark Hydeman is a principal at Taylor Engineering, LLC, Alameda, Calif.
γ ≡ – ln ( SCC ) (7) oped from tuning data on the left (surge) side and right
(normal) side of the peak efficiency point.
Fan efficiency can be accurately predicted as a function of Regardless of the equation order, care must be taken to
gamma. The recommended procedure is to break the function provide a continuous function through the critical gamma
into two parts: an equation for gammas in the “surge” region point.
and another for gammas in the “normal” region. For this
Figure 4 shows the accuracy of the characteristic curve
reason, it is important to get an accurate indication of the “crit-
fan model across 224 points, representing a wide range of fan
ical gamma,” which is the gamma that corresponds to the
speeds, pressures, and airflows. This tuning data came from a
system curve of highest fan efficiency. This can be done with
manufacturer’s selection program, and the results are
the manufacturer’s software by iterating on the airflow condi-
presented against this tuning data set. We used third-order
tions in the vicinity of the critical gamma.
polynomials to represent efficiency of gamma, with separate
Figure 3 shows the R-square regression statistic for vari- equations in the “surge” and “normal” regions (Equations 8
ous orders of polynomials to the example 66 in. plenum fan. and 9).
As demonstrated in Figure 3, a first-order polynomial (i.e.,
The characteristic curve model has been found to be accu-
straight line) is reasonably accurate. Higher order polynomials
rate for at least six types of fans: plenum, backward inclined,
provide a better fit, but they require more data to calibrate and
airfoil, mixed flow, propeller, and vane-axial with fixed
can produce undesirable results between calibration data
blades. This model does not apply to fans with variable pitch
points. A third-order regression appears to provide a good
blades or inlet vanes. Figure 5 shows sample gamma curves for
balance between calibration accuracy to the tuning data set
four types of fans. The curves are divided into the surge and
and rational function behavior between calibrating data
non-surge regions in order to illustrate the relationship
points. These equations are of the form:
between peak efficiency and the surge region.
2 3 Table 1 presents the characteristic curve fan model fit
η fan_left_of_peak_efficiency = S 0 + S 1 × γ + S 2 × γ + S 3 × γ
results across a range of manufacturers and fan types (plenum
(8) and housed, airfoil, and flat blade) and wheel diameters. This
table presents the coefficient of variation root mean square
2 3
η fan_right_of_peak_efficiency = N 0 + N 1 × γ + N 2 × γ + N 3 × γ error (CVRMSE) across the 57 fans in the database. Again,
(9) this represents third-order polynomial fits against tuning data
from manufacturer’s selection programs. Typically a fit of 1%
where S0…S3 and N0…N3 are regression coefficients devel- to 3% CVRMSE is excellent; a 5% fit is acceptable. Table 1
Figure 6 Manufacturer A 66 in. plenum fan efficiency map Figure 7 Phi as a function of efficiency.
using the characteristic curve fan model.
shows that the average CVRMSE is well within the excellent φ right_of_peak_efficiency
region. 2 3
(11)
= PN 0 + PN 1 × η + PN 2 × η + PN 3 × η
Figure 6 depicts the predicted fan efficiency from the
characteristic curve fan model for the example 66 in. plenum where PS0…PS3 and PN0…PN3 are regression coefficients
fan. The predicted efficiency is plotted on the Z-axis as a func- developed from tuning data on the left (surge) side and right
tion of the airflow (cfm, X-axis) and fan static pressure (H2O, (normal) side of the peak efficiency point.
Y-axis). The efficiency is computed between the minimum Figure 7 shows the model parameters and r-square statis-
and maximum characteristic system curves. When viewed tic for the example 66 in. plenum fan. Once φ is determined
from the top, this curve presents the same XY plane as the from Equations 10 and 11, Equation 2 can be used to calculate
manufacturer’s fan curve (Figure 1). The Z-axis goes from fan speed (N, rpm) from φ, airflow (CFM), and fan diameter
white at the highest efficiency to purple at the lowest effi- (D, in.).
ciency. It is interesting to note that fan efficiency falls more
rapidly with a change in airflow in the “normal” region than in DISCUSSION
the “surge” region.
In addition to the characteristic curve fan model, which As displayed in Table 1 and Figure 4, the characteristic
predicts fan efficiency, a second model was developed for curve fan model is quite accurate at predicting fan efficiency
predicting fan speed from airflow and fan static pressure. This from manufacturer’s data. The model presented in the
model, referred to as the Phi model, is used to simulate the ASHRAE Secondary System Toolkit provides similar results.
performance of fans when they are riding the fan curve (e.g., As stated in the objective, the real issue is how to take these
at fixed speed). With variable-speed driven fans, this occurs models and apply them to design and analysis. This includes
when the fan reaches its minimum speed for motor cooling. a number of subtopics:
The Phi model is derived from the Transys program, as 1. The need to model the other components in the fan system.
reported in the ASHRAE Secondary Toolkit (Clark 1985; This includes the motor, physical drive (belts, coupling, or
Brandemuehl et al. 1993). This model produces the dimen- gears), and variable-speed drive (if applicable).
sionless flow coefficient, φ (Equation 2), from fan efficiency, 2. The accuracy of the fan system model compared to field-
as predicted by the characteristic curve fan model. measured data.
φ left_of_peak_efficiency 3. A methodology to apply the model to new construction,
2 3
(10) where the airflow and pressure demand need to be esti-
= PS 0 + PS 1 × η + PS 2 × η + PS 3 × η mated.
sizes within a model line using other perfect fan laws. Under Manufacturer A tested the 54 in. fan and extrapolated the
ANSI/ASHRAE Standard 51-1999 (ANSI/AMCA Standard performance to the 60 to 73 in. sizes. Using the fan laws, one
210-99), manufacturers are not required to test all fan sizes. can exactly duplicate the curves from 60 in. through 73 in.
According to the standard, test information on a single fan may using the 54 in. fan data.
be used to extrapolate the performance of larger fans that are Figure 9 also shows one airfoil fan and a forward curve
geometrically similar using the fan laws (ASHRAE 2000). fan. These fans have different curve shapes than the plenum
Figure 9 shows curves for several fans, including five sizes of fans.
Manufacturer A’s plenum airfoil fans. The 54 to 73 in. diam- Figure 10 shows the highest efficiency (efficiency at crit-
eter plenum airfoil fans have virtually identical curve shapes, ical gamma) for a number of housed airfoil fans from two
just shifted along the x-axis (gamma). The 49 in. version has manufacturers as a function of wheel diameter. By reviewing
a different peak efficiency and curve shape. This suggests that the step changes in the peak efficiency data as a function of fan
354 ASHRAE Transactions: Symposia
diameter, it is clear from this figure which fans the manufac- model works for systems with fixed-speed fans and fans with
turers tested and which they extrapolated. For example, both variable-speed drives, but it will not work for fans with inlet
Manufacturer A and Manufacturer B tested their 30 in. fans. vanes or variable-pitch blades. Those challenges are left up to
Manufacturer A then extrapolated the 30 in. data all the way future researchers.
up to 73 in. (the variability in the peak efficiency of the Manu- Also left to future researchers is the development of
facturer A 30 in. to 73 in. fans is due to rounding and sampling generalized fan models based on the characteristic curve fan
error). Manufacturer B only extrapolated the 30 in. up to 36 in., model. The techniques described thus far require tuning data
then they tested the 40 in.and extrapolated that all the way to specific to each fan to be evaluated. However, there are clear
73 in. Manufacturer A’s 30 in. is more efficient than Manufac- patterns between gamma curves for fans of the same type (see
turer B’s 30 in. but not more efficient than the Manufacturer Figure 9) A single gamma curve could be used to represent all
B’s 40 in. Had Manufacturer A tested a 40 in. (or larger) fan, fans of a certain type (housed airfoil, plenum airfoil, plenum
they might have found that it had higher efficiency than flat blade, etc.). This curve could then be translated along the
equally sized Manufacturer B’s fans. gamma axis using the perfect fan laws and along the efficiency
To use the characteristic curve fan model in a design axis as a function of diameter. Figure 10 shows such a function
context, we suggest the following process: for housed airfoil fans.
1. Develop a simulation model of the facility.
2. Export the hourly demand for fan airflow cfm. ACKNOWLEDGMENTS
3. Bin the data by hours spent at increments of airflow (10 to The authors would like to acknowledge the input and
20 bins should suffice). work of other members of our research team, including Cathy
Higgins of the New Buildings Institute, Steve Taylor of Taylor
4. Develop a system curve that represents the coincident pres-
Engineering, Erik Kolderup and Tianzhen Hong of Eley Asso-
sure at each fan airflow. Note: this may actually be a family
ciates, Lynn Qualman from SBW Consulting, Inc., and Roger
of curves representing issues such as supply pressure reset
Lippman from New Horizon Technologies. They would also
control and the fixed pressure overhead for individual fans
like to recognize the contributions of our technical advisory
run alone or in parallel.
team members. Finally, special thanks to the numerous build-
5. Evaluate the performance of alternate fans across the bin ing engineers and property managers at these sites for putting
data using the system curves to develop a coincident pres- up with our intrusions at their buildings and for their signifi-
sure. cant assistance in our work.
This, of course, is moot if software developers incorpo-
rate the characteristic curve fan model in their programs REFERENCES
directly for a parametric analysis.
AMCA. 1990. AMCA Publication 203-90, Field perfor-
CONCLUSIONS mance measurement of fan systems, 0203X90A-S.
Arlington Heights, Ill.: The Air Movement and Control
As shown in this paper, it is possible to accurately predict Association International, Inc.
manufacturers’ fan performance using the characteristic curve
ASHRAE. 1999. ANSI/ASHRAE Standard 51-1999 (ANSI/
fan model. However, it is a challenge to predict field perfor-
AMCA Standard 210-99), Laboratory Methods of Test-
mance due primarily to inaccuracies in instrumentation,
ing Fans for Aerodynamic Performance Rating. Atlanta:
system effects due to field conditions, and inaccuracies in the
American Society of Heating, Refrigerating and Air-
manufacturers’ reported performance data (due mostly from
Conditioning Engineers, Inc.
their extrapolation of test data). From a design perspective,
some of these issues are moot, as biases in instrumentation, ASHRAE. 2000. 2000 ASHRAE Handbook—HVAC Systems
measurement, and data reporting will tend to cancel out in a and Equipment, Chapter 18, Fans. Atlanta: American
comparative analysis of design alternatives. Society of Heating, Refrigerating and Air-Conditioning
To serve as a design tool, a predictive fan model should be Engineers, Inc.
developed to predict brake horsepower from airflow and fan Brandemuehl, M.J., S. Gabel, and I. Andresen. 1993. HVAC
static pressure. These are the inputs that are typically provided 2 Toolkit: Algorithms and Subroutines for Secondary
in a simulation tool. The model should also have discrete HVAC System Energy Calculations. Atlanta: American
submodels for the separate fan system components so that Society of Heating, Refrigerating and Air-Conditioning
analysis can be done on the impact of design alternates for Engineers, Inc.
each of those components. CALMAC. 2003. Data from the non-residential new-con-
Overall, the characteristic curve fan model meets all of struction database available from the California Mea-
these requirements. It is accurate at predicting manufacturers’ surement Advisory Council’s web site, <http://
data, relatively easy to tune, and could easily be incorporated www.calmac.org/>.
into existing simulation tools. The authors have successfully Clark, D.R. 1985. HVACSIM+ building systems and equip-
employed it in Visual Basic code. There are limitations; this ment simulation program: Reference Manual. NBSIR