2014 ASHRAE/IBPSA-USA

Building Simulation Conference

Atlanta, GA
September 10-12, 2014

INVERSE MODELING TO ESTIMATE THE EFFECTIVE

LEAKAGE AREA IN BUILDINGS
Te Qi1, Zheng ONeill2, and Godfried L Augenbroe1,
1
Georgia Institute of Technology, Atlanta, USA
2
The University of Alabama, Tuscaloosa, USA

ABSTRACT
Infiltration plays a significant role in the energy
consumption of buildings. Estimating infiltration rates
can provide insights into building energy consumption
patterns. Both Effective Leakage Area (ELA) and Air
Change rate per Hour (ACH) can be used to describe
the air infiltration rate in the building. This paper
presents an approach using the ELA as the major
parameter to model infiltration in existing buildings
The conventional way to estimate the aggregated ELA
value of a building is through a blower-door test, which
is rather expensive in practice. This study introduces a
lower cost alternative to estimate the ELA using inverse
modeling approach .

INTRODUCTION
Infiltration can lead to many negative effects in
commercial buildings in the U.S. Unexpected and
uncontrolled high air change rates can reduce the
thermal comfort and indoor air quality (IAQ), interfere
the operation of mechanical ventilation systems, and
cause moisture damage in building envelopes and mold
growth on internal surfaces and increase total building
energy consumption. Studies have shown that
infiltration is responsible for 33% of the total heating
energy use in U. S. office buildings (Emmerich 2005).
Therefore, controlling air leakage in buildings as part of
retrofits can achieve significant energy savings.
Building energy simulation tools can provide insights
into the infiltration impact on the total energy
consumption using a reasonable estimation of air
infiltration rates. There are two main ways to estimate
air infiltration rates: determining the Effective Leakage
Area (ELA) or determining the Air Change rate per
Hour (ACH). This paper proposes an inverse modeling
approach (also referred as calibration) to estimate the
ELA. This approach integrates a multi-zone airflow
model in CONTAM (CONTAM 2013) with a building
energy model in EnergyPlus (EnergyPlus 2013). We
first calculate dynamic infiltration rates in CONTAM
with different ELA estimations as inputs. Then we use
these calculated dynamic infiltration rates as inputs to

EnergyPlus. Monitored energy consumption data from

the building is then used as the target value for the
calibration. It is well known that the stochastic nature of
weather, occupant behavior, infiltration and the
physical aspects of building components have a strong
impact on building energy consumption. Unfortunately,
these uncertainties are usually ignored in a
deterministic simulation such as EnergyPlus. This paper
investigates the feasibility of replacing the blower-door
test with the inverse modeling approach through an
office building case study in the presence of parameter
uncertainties.

BACKGROUD
1. Blower-door test approach
The air leakage through the building envelope can be
measured with a pressurization test, which is commonly
called a blower-door test (ASHRAE 2009). In this
method, a large fan or blower is mounted in a door or
window inducing a roughly uniform large pressure
difference across the building shell (ASTM Standards
E779 and E1827; Canadian General Standards Board
(CGSB) Standard 149.10; ISO Standard 9972). The
airflow required to maintain this pressure difference is
the infiltration rate under this applied pressure
difference. Consequently, the result of a pressurization
test is combinations of different pressure differences
and corresponding airflow (infiltration) rate. Equation 1
shows the formula to calculate the total leakage area.

where

equivalent or effective air leakage area, cm2;

predicted airflow rate at
pressurization test data), m3/s;

(from curve fit to

air density, kg/m3;

reference pressure difference, Pa;
discharge coefficient.

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

323

Some common airtightness ratings include the effective

air leakage area at 4 Pa assuming
(Sherman
and Grimsrud 1980) and the effective air leakage area
at 10 Pa assuming
=0.611(CGSB standard 149.10).
2. Empirical model
When onsite blower-door test data are not available,
some empirical equations can be used to get a
reasonable estimate of ELA. One classical infiltration
model is the Sherman-Grimsrud model developed by
Sherman and Grimsrud (Sherman and Grimsrud 1980).
Equation 2 and Equation 3 show the mathematical
formulas. Table 1 and Table 2 show the value of stack
coefficient and wind coefficient used in these equations.
(ASHRAE 2009)

+
2

used to calculate the transient infiltration rate for a

given ELA.
Table 2 Wind Coefficient Cw

where

airflow rate, m3/s;

effective air leakage area, cm2;
stack coefficient, (L/s)2/(cm4K);
average indoor-outdoor temperature difference
for time interval of calculation, K;
wind coefficient, (L/s)2/[cm4 (m/s)2];
average wind speed measured at local weather
station for time interval of calculation, m/s.

{ } { }

where

normalized leakage area, dimensionless;

effective leakage area at 4 Pa (

), cm2;

gross floor area (within exterior walls), m2;

building height, m;
reference height of one-story building, 2.5m.
Table 1 Stack Coefficient Cs

Stack Coefficient

House Height
one
two
three
0.000145 0.00029 0.000435

3. Inverse Modeling Approach

The inverse modeling approach is motivated by the fact
that the blower-door test approach is too expensive and
the empirical approach is usually not accurate enough.
In the inverse modeling approach, a multi-zone model
which is able to estimate the transient infiltration rate
with given ELA is coupled with the energy model in a
dynamic simulation program. The multi-zone model is

A network airflow model idealizes a building as a

collection of zones modeled as nodes, such as rooms,
hallways, and duct junction, joined by flow paths
representing doors, windows, fans, ducts, etc.
The network model airflows are driven by the pressure
differences across the paths. Three are three types of
forces that drive the airflow through airflow paths:
wind, temperature differences (stack effect), and
mechanical devices.
This paper uses CONTAM to build the multi-zone
model. CONTAM is a multi-zone indoor air quality and
ventilation analysis computer program designed to
calculate airflows, contaminants concentrations and
personal exposures.

CASE STUDY PROCEDURES

This section shows the key procedures of the inverse
modeling process and details of one case study. The
five key steps of the inverse modeling approach
include:
1. Choose calibration period
2. Build a reliable energy model in EnergyPlus
3. Develop a multi-zone model using CONTAM
that can be used for different ELAs.
4. Export the infiltration rates calculated in
CONTAM into EnergyPlus for a range of ELA
values

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

324

5. Compare the simulation results with monitored

data and find the ELA that has the best fit to the
real data
The above procedure is a hand-crafted estimation
process which is good enough for the feasibility study
in this paper. Ideally, however, this should become a
full-blown optimization process to estimate ELA, in
other words, the EnergyPlus model should be able to
automatically import the infiltration rate calculated
from the CONTAM in each run based on a certain ELA
test value. This would allow to run the whole process in
an optimization loop until the best match is found, e.g.
by minimizing the sum of squared difference between
model predictions and real measurements.
At the current stage, due to the complexity of the
connection between CONTAM and EnergyPlus for
large and complex building models, the whole process
was not yet automated. Instead, a manual process is
being used. For example, in the case study in this paper,
10 discrete ELA inputs were defined in an interval from
10,905cm2 to 32,343cm2 to find a reasonable
approximation of the best estimation for the actual
ELA. In fact, this means that we test ten samples from a
uniformly distributed interval of ELA and perform a
simplified manually generated Monte Carlo simulation.
A case study is used to illustrate the proposed method.
The case study building, Building 101 in the Navy
Yard, is the temporary headquarters of the U.S.
Department of Energys Energy Efficient Building
Hub. The building, owned by the Philadelphia
Industrial Development Corporation (PIDC), has
become one of the nations most highly instrumented
commercial buildings1 . Figure 1 shows the west view
of building 101.

Because the infiltration rate has more impacts on the

energy consumption in heating season than cooling, we
chose a heating-dominated month as the simulation
period and compared the measured and estimated
heating energy consumption in that period. Because
wind speed and wind direction also have strong impacts
on the infiltration rate, a period of two weeks that
exhibit the most fluctuating wind velocity was chosen
to maximize the impact of infiltration on total heating
energy consumption as well as mitigate the interference
of other parameters. Based on the reasoning above, 1/01
to 1/14 in 2011 is chosen as the calibration period.
Figure 2 shows the wind speed from 1/01 to 1/14.

Figure 2 Wind speed profile from Jan/01 to Jan/14

Since we want to compare the predicted energy
consumption from the energy model with
measurements, it is necessary to use the actual weather
data in the energy simulation process. For this purpose,
this study uses actual meteorological year (AMY)
weather data from Weather Analytics.
(http://www.weatheranalytics.com/)
2. Build energy model in EnergyPlus
A detailed energy model was created in Design Builder
(Design Builder 2005). Figure 3 shows the building
model in Design Builder. However, to estimate ELA,
the model is simplified to make the complexity of the
model manageable without losing the accuracy of the
final simulation result. Figure 4 and 5 shows the zoning
before and after the simplification.

Figure 1 West-view of building 101

1.

Choose calibration period

(http://www.eebhub.org/projects-list/navy-yard-buildin
g-101)

Figure 3 Detailed Energy Model

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

325

Figure 4 Detailed First Floor plan

Figure 6 Basement Floor plan in CONTAM

Figure 5 Simplified First Floor plan

In EnergyPlus, one way to calculate the zone
infiltration rate is through the Zone Infiltration: Design
Flow Rate object and the underlying equation 4:
(

)(

+ |
+

)[
|+

Here
is the CONTAM simulation result

(2
33 values), we set

, A = 1, B = C = D = 0, so that the EnergyPlus

simulation can take the CONTAM result as input.
3.

Develop a multi-zone model using CONTAM

with different ELA
To calculate the dynamic infiltration rates, CONTAM is
used to create the multi-zone model for the case study
building. To build the multi-zone model in CONTAM,
one needs to zone the building reasonably to the level
that it is not too complicated while not affecting the
accuracy of results. Second, we calculate the wind
pressure coefficient on each faade. Third, we estimate
the ELA of each leakage component based on empirical
data. In this case, the ELA of each leakage component
on each facade is aggregated over the whole faade of
the zone as this dramatically reduces the computation
load of the pre-treatment of the infiltration rate results
from CONTAM. Figure 6 illustrates the air flow
network model in CONTAM. Figure 7 shows an
example of input in CONTAM.

Figure 7 Input example of an air leakage component

To calculate the wind pressure coefficient of each
facade, we utilize the web-based Cp calculator
(http://cpgenerator.com/). Urban context plays a very
important role in wind pressure profile calculation.
From Building101s urban layout, we find this area is
actually more like a suburban area, as there is no
building around the building within a distance of three
times of the height of building 101. Therefore, the
obstacle effect of surrounding buildings was ignored in
this study. Figure 8 shows the simplified facade profile.

Figure 8 Simplified Facade Profile

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

326

4. Estimate ELA variability

Estimating ELA variability in all leakage components is
an important part of the whole process. In this paper,
we use the window effective leakage area from
Miscellaneous Commercial and Institutional Building
Airtightness Data in CONTAM library, as shown in
Table 3 to predict the ELA of the whole facade.

Table 4 Estimated ELA of each Facade

Table 3 Window Leakage Rate in CONTAM Library

For the estimation of the ELA of windows, we average

the maximum and minimum ELA value and find that
the mean equals 2.6cm2/m2. For the estimation of the
ELA of roofs and above grade walls, we refer to the
suggested value in SSPC90.1 Envelope subcommittee
for component infiltration rate (Gowri et al. 2009)
5. Determine which ELA is the best estimate
The indicator used to determine the best estimate of
ELA in this study is Mean Square Error (MSE).
In statistics, MSE is one way of quantifying the
difference between values implied by an estimator and
the true values of the quantity being estimated. MSE is
a risk function, corresponding to the expected value of
the squared error loss or quadratic loss. MSE measures
the average of the squares of the "errors." The error is
the amount by which the value implied by the estimator
differs from the quantity to be estimated. The formula
of MSE is as equation 5:

Figure 9 Boiler Gas Consumption (Measured vs.

Predicted)
In the case study building, a blower-door test
(Camroden Associates, Inc 2012) was conducted as
well, which we used to compare with our results.
Table 5 Measurement from Blower Door Test

is a vector of n predictions
is the vector of the true values
In this case study,
is the real measured data of boiler
gas consumption, is the predicted boiler
consumption from the EnergyPlus model.

CASE STUDY RESULT AND DISCUSSION

Following the procedures described above, the best
estimate of the ELA is listed in Table 4, where the total
building ELA is 23,437cm2 at the pressure difference of
4 Pa. Figure 9 shows the comparison of measured and
predicted boiler gas consumption with this estimated
ELA.

Table 5 lists the measurement results from the blowerdoor test. Based on the whole building depressurization
test conducted by Camroden Associates, Inc, we
derived the power law of the whole building:

23

From equation 6, we then calculated that the air leakage

rate of the whole building is 2.7 m3/s, 0.36 ACH at the
pressure difference of 4 Pa, and ELA is 10,483cm2 at
the pressure difference of at 4 Pa, which is less than
half of the value predicted by the inverse model.
However, due to the uncertainty of the energy model

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

327

and multi-zone model, this result is not a conclusive

adoption or rejection of the new inverse approach as a
predictor of the ELA of the building. There are some
reasons that uncertainty could explain the discrepancy.
One of them is that the energy model contains many
sources of uncertainty. Although the input of the energy
model is based on the building specification, there
could still be big discrepancies between the modeled
and real usage and operation scenario, especially
relevant to the HVAC system modeling.
Simple Uncertainty Analysis
The following rudimentary uncertainty analysis is about
estimating the possible ELA distribution as a result of
the uncertainty from the CONTAM model. For this
purpose we assume that the actual infiltration rate
ranges from 0.8V to 1.2V, where V is the deterministic
result from the CONTAM model. Based on this 20%
infiltration range, we can find the estimated ELA range
due to this variation, and get the uncertainty distribution
of the ELA as a result of CONTAM modeling error. As
shown in last section, the best estimate of the ELA is
2.25m3/s for the whole building. Table 6 gives the
infiltration rate range from a baseline calculated with
ELA =23437cm2 at the pressure difference of 4 Pa.
Table 6 Infiltration deviation from baseline

not lie in this range this is probably because we only

consider the uncertainty of the CONTAM model. With
other parameters uncertainty taken into account, the
ELA=10,483cm2 could possibly be within the range.

Figure 10 ELA distribution as a result of CONTAM

model uncertainty
Until now we have only considered the calibration of
one parameter. In the case we do multi-parameter
calibration (as shown in Figure 11 for schedule and
building material parameters) we may get better
calibration results for the estimated ELA. However this
requires that more detailed measured data (not just
hourly energy consumption) should be available to
make this work.

In this example, the ELA value will be between 1.8m3/s

and 2.7m3/s. Through a calculation, the corresponding
infiltration rates with these ELA are 18,240 cm2 and
29,403 cm2 respectively. Table 7 lists ELA margin with
CONTAM model uncertainty.
Table 7 ELA margin under CONTAM model
uncertainty
Figure 11 Multi-variable analysis diagram

With the assumption that uncertainty to calculate the

infiltration rate is 20% due to CONTAM model
uncertainty, we conclude that the ELA could lie
between (18,240cm2, 29,403cm2), as shown in Figure
10, which shows the possible ELA distribution as a
result of CONTAM modeling uncertainty. Although the
ELA value 10,483cm2 from the blower-door test does

CONCLUSION AND FUTURE WORK

The purpose of this research is to develop a new
approach to estimate the effective leakage area using
the inverse modeling approach as an alternative to the
blower-door test. From the comparison with simulation
results, the best estimate of the total building ELA from

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

328

the inverse modeling approach is the 23,437cm2 at the

pressure difference of 4Pa, while the result from the
blower door test is 10,483cm2 at the pressure difference
of 4Pa. Because of the insufficient information of the
building and the uncertainty of the input parameters, the
study has not led to a definite statement whether the
proposed calibration of the ELA with energy
consumption data can replace a blower-door test to get
an equally valid or even better ELA estimate, but it
might be feasible.
As mentioned above, we set the uncertainty boundary
of the ELA in the CONTAM model as plus/minus 20%;
however, due to the simplification of the multi-zone
model and HVAC system network, this uncertainty
boundary could be actually greater than 20%.
In addition, other uncertainty sources such as plug load
power density and schedule, which will greatly affect
the boiler gas consumption, are not considered in this
study. With those parameters taken into consideration,
the predicted mean value of the ELA (shown in Figure
10) could be shifted to the left and the measured ELA
could be possibly fall into the range of predicted ELA.
Future work includes an automation of the calibration
process and a calibration under uncertainty.
1) Automation of the calibration
This presented work to estimate ELA includes
significant amount of manual work, which could be
avoided if we could couple CONTAM and EnergyPlus
to make them automatically run simultaneously. With
this automation process, we could conduct a full
optimization process with multiple variables or
propagate multiple parameter uncertainty to get a more
reliable result.
As an alternative of the CONTAM, the nodal airflow
network in EnergyPlus could also be employed as a
multi-zone model to calculate the dynamic infiltration
rate. The main reason why this module is not used is
because Energyplus 7.2 was the latest version when this
research was conducted, which does not have the
fuctionality to deal with variable air volume (VAV)
system. Decoupling the energy model and the airflow
network model has the advantage of increasing the
accuracy of the result, adding flexibility of the input
work, as well as facilitating the debugging process.
The newest released Energyplus 8.1.0 has the function
of integrating the VAV system in the airflow network
module. With the latest Energyplus, we could
accomplish the automation process and eliminate the
uncertainty introduced by simplifying the HVAC
system in CONTAM model.
2) Calibration under uncertainty

Since other uncertain parameters, such as Cp of each

faade, also affect the accuracy of the result, it is more
convincing to repeat the procedure under the influence
of those uncertain parameters.
There are basically three different levels to do the
uncertainty analysis.
First level is to create an ELA band to see how many
hours the hourly monitor data will lie in the output from
the energy model, then we can calculate the confidence
that the ELA is within this band. The second level is
that we can do the prior uncertainty quantification for
dominant parameters, for example, ELA of each
leakage component, and Cp of each facade, based on
the empirical data from expert judgment, which could
be derived from a pool of sources (experiments,
surveys, expert knowledge, industry standards, etc).
After we get the distribution of these parameters, we
can run the Monte Carlo simulation with the
distribution of these parameters and get a distribution of
the ELA estimate. For each run, the indicator of the best
estimate is the Mean Square Error (MSE) of monitor
data and model data. The third level is using a
high-level mathematical Bayesian calibration method
(Y. Heo 2012). Detailed discussion could be found in
(Qi 2012).

ACKNOWLEDGMENT
This work is partially funded by Energy Efficient
Buildings Hub, sponsored by the Department of Energy
under Award Number DE-EE0004261.

REFERENCES
ASHRAE
(2009).
Fundamentals.
Chapter16 American Society of Heating,
Camroden Associates, Inc 2012. Report on the
building envelope air tightness and pressure
relationship mapping, report submitted to
Philadelphia Industrial Development Corporation
CONTAM
http://www.bfrl.nist.gov/IAQanalysis/CONTAM/
Design Builder 2005
http://www.designbuilder.co.uk/
Emmerich, S.J., Persily, A.K. 2005. Impact of
Infiltration on Heating and Cooling Loads in U.S.
Office Buildings. AIVC Conference.
EnergyPlus
http://apps1.eere.energy.gov/buildings/energyplus/
Gowri, K., D. Winiarski, and R. Jarnagin. (2009).
Infiltration modeling guidelines for commercial
building energy analysis. PNNL #18898. Richland,
WA: Pacific Northwest National Laboratory.
M.C. Kennedy, A. OHagan, (2001). Bayesian
calibration of computer models, Journal of the

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

329

Royal Statistical Society, Series B 63 (2001) 425

464
Sherman, M.H. and Grimsrud, D.T., 1980,
Infiltration-Pressurization Correlation: Simplified
Physical Modeling, ASHRAE Transactions, Vol.
86(2), pp. 778-803
T. Qi, (2012) Inverse modeling to predict effective
leakage area, Master thesis, Georgia Institute of
Technology
Y. Heo, R Choudhary, G.Augenbroe,(2012)
Calibration of building energy models for retrofit
analysis under uncertainty, Journal of the Energy
and Savings, 47(2012)550-560

2014 ASHRAE (www.ashrae.org). For personal use only. Reproduction, distribution, or transmission
in either print or digital form is not permitted without ASHRAEs prior written permission.

330