Energy and Buildings 43 (2011) 517528

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Simplied calculation method for design cooling loads in

underoor air distribution (UFAD) systems
Stefano Schiavon, Kwang Ho Lee , Fred Bauman, Tom Webster
Center for the Built Environment, University of California, 390 Wurster Hall, Berkeley, CA 94720-1839, USA

a r t i c l e

i n f o

Article history:
Received 20 July 2010
Received in revised form 12 October 2010
Accepted 18 October 2010
Keywords:
Underoor air distribution (UFAD)
Cooling load
HVAC sizing
Overhead air distribution (OH)
Mixing ventilation

a b s t r a c t
This paper describes the development of a simplied calculation method for design cooling loads in
underoor air distribution (UFAD) systems. The simplied design tool is able to account for key differences
between UFAD and traditional mixing overhead (OH) systems. These include: (1) difference between
design day cooling load proles, (2) impact of a thermally stratied environment for UFAD versus wellmixed for OH, and (3) impact of heat transfer (temperature gain) in underoor air supply plenums. The
new design tool allows the use of a familiar load calculation procedure for OH mixing systems as input to
the UFAD design tool. Based on 87 EnergyPlus simulations, four regression models have been developed
to transform the OH cooling load into the UFAD cooling load, and then to split this total load between the
supply plenum, zone (room), and return plenum. The regression models mainly depend on oor level,
and position (interior or perimeter) and orientation of the zone under analysis. Although considered in
the analysis, supply air temperature, window-to-wall ratio, internal heat gain, plenum conguration,
climate, presence of the carpet and structure type do not strongly inuence the developed models. The
results show that, generally, UFAD has a peak cooling load 19% higher than an overhead cooling load and
22% and 37% of the total zone UFAD cooling load goes to the supply plenum in the perimeter and interior,
respectively.
2010 Elsevier B.V. All rights reserved.

1. Introduction
Underoor air distribution (UFAD) is a mechanical ventilation
strategy in which the conditioned air is delivered to the zone from
a pressurized plenum through oor mounted diffusers. UFAD has
the potential to increase exibility to change ofce layout and to
reduce the oor-to-oor height [1]. UFAD systems may also provide
improved indoor air quality [2] and thermal comfort, as well as
energy savings [1].
1.1. UFAD cooling load calculation
The most commonly used cooling load calculations are based
on the assumption that a zone is well mixed, meaning its temperature is uniform throughout the zone [3]. This assumption is valid
over a wide range of conditions but cannot be directly applied to
stratied UFAD systems. The instantaneous cooling load is the rate
at which heat energy is convected to the zone air at a given point
in time. Computation of cooling load is complicated by the radiant
exchange between surfaces, furniture, partitions, and other mass in

Corresponding author. Tel.: +1 510 809 5163; fax: +1 510 643 5571.
E-mail address: kwhlee@berkeley.edu (K.H. Lee).
URL: http://www.cbe.berkeley.edu (K.H. Lee).
0378-7788/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2010.10.017

the zone. Most heat sources transfer energy by both convection and
radiation. Radiative heat transfer introduces a time dependency to
the process that is not easily quantied. Radiation is absorbed by
thermal mass in the zone and then later transferred by convection
to the space. This process creates a time lag and dampening effect.
The convective portion, on the other hand, is immediately transformed into cooling load [3]. The thermal storage effect is critical
in differentiating between instantaneous heat gain for a given space
and its cooling load at that moment. Accounting for the time delay
is a major challenge in cooling load calculation. The sum of all space
instantaneous heat gains at any given time does not necessarily (or
even frequently) equal the cooling load for the space at that same
time [4]. Thus, using only the heat gains and not taking into account
the thermal storage effect is not a proper way to calculate the design
airow rate.
The most common cooling airow design methods for UFAD systems [58] are based on the assumption that the design day peak
zone cooling load prole for UFAD and a well-mixed system is the
same. Schiavon et al. [9] showed that the presence of the raised
oor affects the cooling load proles, therefore it is possible that
the cooling load prole of UFAD is different from the cooling load
prole of a building ventilated by an overhead system (also called
mixing ventilation). Moreover, the presence of stratication affects
the heat exchange in the zone and this may also affect the cooling
load prole. In order to accurately simulate the UFAD system, two

518

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

Nomenclature
AHU
CL
GF
MF
OH
Pconf
RPF
SPF
TF
UCLR
UFAD
WWR
ZF

air handling unit

cooling load
ground oor
middle oor
overhead (mixing) air distribution system
plenum conguration
Return Plenum Fraction
Supply Plenum Fraction
top oor
UFAD Cooling Load Ratio
underoor air distribution system
window-to-wall ratio
Zone Fraction

Subscript
SP
supply plenum
Z
zone
RP
return plenum

important features should be captured by the software: heat gain

into the underoor supply plenum and the room air stratication.
Based on our review of available simulation tools, EnergyPlus is one
of only a few building simulation programs that can consider those
two key features of UFAD simulation and has been validated based
on extensive laboratory testing. It performs a full heat balance of
the underoor supply plenum and thus calculates the temperature
rise of the conditioned air inside the plenum during each time-step
[1012]. However, EnergyPlus is a complex software and the cost
and effort related to modeling a building for the cooling load calculation using EnergyPlus is, most of time, prohibitive for HVAC
designers. Therefore it is valuable for practitioners to have a simplied method that transforms the cooling load calculated for a
well-mixed system into a cooling load for UFAD without the need
for using EnergyPlus. The drawback to this method is imprecision
of the model and limitations imposed by the range of conditions
over which the model has been developed.
1.2. UFAD cooling load split
A distinguishing feature of any UFAD systems is the use of an
underoor plenum to deliver supply air through oor diffusers into
the conditioned space. Cool supply air owing through the underoor plenum is affected by the heat transfer from both the concrete
slab (in a multi-story building) and the raised oor panels. Field
measurements [13,14] and computer uid dynamic analyses [13]
showed that usually the air warms up signicantly, resulting in

undesirable and uncontrolled air temperature discharged at the

diffusers. This phenomenon is often referred to as thermal decay.
The amount of heat entering the underoor plenum directly
inuences the design cooling airow rate. Bauman et al. [15] by an
analysis of heat transfer pathways in real underoor plenums has
provided evidence that supports two widely observed thermal phenomena in UFAD systems: (1) part of the cooling load is removed
in the underoor supply plenum; and (2) temperature gain (thermal decay) in open underoor supply plenums is often larger than
expected. They estimated, based upon a simplied rst-law model,
that for typical multi-story building congurations (raised access
oor on structural slab with or without suspended ceiling), 3040%
of the total room cooling load in the interior zone is transferred
into the supply plenum and about 6070% is removed in the zone.
Based on the above mentioned results. Bauman et al. [8] developed
a practical design procedure for determining the required cooling
air ow rate in an interior zone of a UFAD system. In this earlier
study the cooling load split was xed, therefore it will be valuable
in the current study to investigate the variability of this split due
to various design and operating parameters.
The purpose of this study is to develop, by means of energy simulations with EnergyPlus, a simplied model for the transformation
of the cooling load calculated for a well-mixed OH system into a
cooling load for UFAD and a model for prediction of the split of the
UFAD cooling load between the supply plenum, zone (room), and
return plenum.
2. Methods
2.1. Indexes
For sizing the zone cooling airow rate the hourly peak design
day cooling load value is commonly used. Hereafter, the term
hourly peak design day cooling load is referred to simply as cooling load. Fig. 1 shows the models that are needed to transform the
cooling load calculated for a well mixed OH system into a UFAD
cooling load and to split the UFAD cooling load into the supply
plenum, the zone and the return plenum.
The indexes reported in Fig. 1 are dened below.
The UFAD Cooling Load Ratio (UCLR), dened in Eq. (1), is the
ratio of the cooling load calculated for UFAD to the cooling load
calculated for a well mixed OH system. UCLR equals 1 means that
the two cooling loads are the same. UCLR greater than 1 means that
the cooling load for UFAD is higher than the cooling load for OH,
for example, UCLR = 1.12 means that the UFAD cooling load is 12%
higher than the OH cooling load. UCLR lower than 1 means that
UFAD cooling load is lower than the OH cooling load.
UCLR =

CLUFAF
CLOH

(1)

Fig. 1. Schematic ow diagram of the models needed to transform the cooling load calculated for a well mixed OH system into a UFAD cooling load and to split the UFAD
cooling load into the supply plenum, zone and return plenum.

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

The cooling load for the OH system has to be calculated with

standard cooling load software (e.g., based on the Heat Balance
method or the RTS method [3]). The OH zone should: (a) have the
same oor to oor height as the UFAD zone; (b) have no underoor
plenum; and (c) have a return plenum that has a height equal to the
sum of the UFAD supply and return plenum heights. As shown in
the equations above, the plenums (both supply and return) should
be modelled separately from the room spaces for both OH and UFAD
systems, since the interaction between plenums and rooms turned
out to be important phenomena that should be taken into account.
Especially the heat gain of the conditioned air inside the underoor
plenum turned out to be very important and thus we recommend
to model the plenums separately.
The Supply Plenum Fraction (SPF), dened in Eq. (2), is the ratio
of the cooling load removed in the supply plenum to the total
UFAD cooling load. SPF may vary between 0 and 1. If, for example,
SPF = 0.35 this means that 35% of the UFAD cooling load is removed
in the supply plenum.
SPF =

CLUFAD
SP
CLUFAD

(2)

The Zone Fraction (ZF), dened in Eq. (3), is the ratio of the cooling load removed in the zone (room) to the total UFAD cooling load.
ZF may vary between 0 and 1. If, for example, ZF = 0.60 this means
that 60% of the UFAD cooling load is removed in the zone.
ZF =

CLUFAD
Z
CLUFAD

(3)

The Return Plenum Fraction (RPF), dened in Eq. (4), is the

ratio of the cooling load removed in the return plenum to the total
UFAD cooling load. RPF may vary between 0 and 1. For example, if
RPF = 0.05 this means that 5% of the UFAD cooling load is removed
in the return plenum.
RPF =

CLUFAD
RP
CLUFAD

(4)

From Eqs. (2)(4) it can be deduced that:

CLUFAD
+ CLUFAD
+ CLUFAD
= CLUFAD
Z
RP
SP
and
SLPF + ZF + RPF = 1

(5)

3. Energy simulation
With EnergyPlus it is possible to simulate the same building
with the same boundary conditions with either an overhead or a
UFAD system and then, for the UFAD building, calculate for each
zone the cooling load split between the supply plenum, the zone
and the return plenum. To develop the models listed in the previous paragraphs a sufciently large database has to be developed.
The database development was divided in two stages. In the rst,
a screening of nine possible variables, listed in the section simulated cases, affecting UCLR, SPF, ZF and RPF was performed. The
most inuential variables were selected and a balanced matrix (all
possible combinations of the levels of the chosen independent variables) of simulations was developed. In the second stage, the results
of the balance matrix were used to develop the models. In the following section, the input data for the energy simulation program
is discussed.1

1
The European standard 15265-2006 [16] recommends a format for reporting
the input data of an energy simulation. The following presentation of input data
complies with the guidance in the standards.

519

3.1. Building location and weather data

In the reference case, an ofce building located in Baltimore, MD
was modelled. The weather in Baltimore is characterized as mixed
and humid according to [17]. To calculate the cooling load prole
for each zone of the building, simulations were conducted for a total
of 87 cases for a summer design day. ASHRAE 0.4% summer design
conditions were assumed [18].
3.2. Description of the ofce building
A three-story prototype ofce building with a rectangular shape
(52.8 m 35.2 m) and aspect ratio of 1.5 was chosen for this study.
The oor plate size is 1858 m2 (total oor area is 5574 m2 ) and
each oor is composed of 4 perimeter zones, an interior zone and
a service core, which represent approximately 38%, 46% and 16% of
the oor area, respectively. The oor to oor height is 3.96 m and
the return plenum height is 0.6 m. The raised oor height is 0.4 m.
In the building ventilated by the overhead system the raised oor
is not present and the oor to oor height is equal to that served
by the UFAD system. In order to maintain the same zone height,
the return plenum height of the overhead building is equal to the
sum of the UFAD supply and return plenums, i.e., 1 m. Strip windows are evenly distributed (i.e., a ribbon window) in the walls.
The construction and thermal properties of the windows change in
each tested climate and they comply with ASHRAE 90.1 [17]. Three
building construction options are used as summarized in Table 2:
light, medium and heavy weight. The effect of thermal mass is taken
into account. Internal mass (e.g., furnishings) was simulated with
25 mm of wood with an area equal to 50% of the oor area. The wood
has a conductivity of 0.15 W/(m K), a density of 608 kg/m3 , a specic
heat of 1.63 kJ/(kg K) and a thermal mass of 25.35 kJ/(m2 K). Internal
mass increases the amount of surface area that can participate in
radiative and convective heat exchanges and it also adds thermal
mass to the zone. These two changes can affect the time response
of the zone cooling load in opposite directions [3]. In the building
with the UFAD system there is a raised oor. Its properties were
taken from a common, commercially available product. The raised
oor has a thickness of 0.025 m, a conductivity of 0.14 W/(m K), a
density of 1185 kg/m3 , and a specic heat of 0.669 kJ/(kg K).
3.3. Internal temperature, ventilation and inltration rate, and
HVAC system
From 5:00 till19:00 the cooling system controls the internal
air temperature to a cooling temperature setpoint of 23.9 C. The
inltration was assumed equal to 0.333 L/(s m2 ) (ow per exterior
surface area). The outdoor air ow rate was equal to 0.76 L/(s m2 ).
The outdoor air ow rate was provided from 5:00 till 19:00.
HVAC air distribution system models for overhead and UFAD
were similar. The overhead model employs variable air volume
(VAV) boxes with hot water reheat coils for all zones. In the UFAD
system the air is distributed through swirl diffusers in interior zones
and linear bar grille diffusers in the perimeter zones. The bar grilles
are ducted from the outlet side of fan coil units (FCU) served by
variable speed fans. The FCU shuts off when perimeter zone temperatures are in the heating-cooling dead band, between 21.1 and
23.9 C. In cooling mode the fan is on, and in the heating mode both
the fan and heating coil are on. An overhead VAV reheat system
serves the service core, although no loads are assigned to this zone.
Both systems are served by a single variable speed central station
air handling unit (AHU) including a return air economizer, chilled
water cooling coil, hot water heating coil and supply fan. A model
for AHU fan control is employed to simulate a static pressure reset
control strategy. The central plant consists of a central centrifugal
chiller with variable speed pumps and a two-speed cooling tower.

520

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

Fig. 2. Occupancy, lighting, equipment and HVAC schedules.

A gas red hot water boiler provides hot water to all heating coils.
Complete details of the system and plant designs can be found in
[19].

SPF, ZF and RPF was performed. The most inuential variables were
selected and a balanced matrix (all possible combinations) of simulations was developed. In the second stage, the results of the balance
matrix of simulation were used to develop the models.

3.4. Internal heat gains and occupancy

4.1. Screening
The occupants contribute to both sensible and latent heat loads
in the building. The activity level of the occupants was 1.3 met (1
met = 58.15 W/m2 ), and the total heat produced per occupant was
thus around 140 W. The balance between sensible and latent heat is
calculated by the software. Three peak occupancy levels have been
studied: 22.3; 11.2 and 8.4 m2 /person). There were no people in
the service core zone. In this paper, the fraction of the design value
is dened as the ratio of the actual number of occupants present
at their desks to the maximum number of occupants the room is
designed to accommodate. The occupants presence in the building
varied according to the schedule shown in Fig. 2. Three levels of
internal cooling loads (people, equipments and lighting) were simulated (22.1; 40.9, 58.1 W/m2 ). The equipment and lighting loads
follow the schedules shown in Fig. 2. Three levels of lighting loads
(lighting) were simulated (10.81; 11.8, 12.9 W/m2 ).
3.5. Simulation software
A robust building energy simulation program, EnergyPlus version 3.1.0.027, was used for the simulations. EnergyPlus is a
whole-building energy simulation program developed by DOE [11].
It calculates the thermal loads to be satised and predicts HVAC
system operation needed to full the required comfort conditions.
EnergyPlus was selected because it is a heat balance based simulation program and the heat balance method is the current industry
standard method for calculating space loads [3] and, to the authors
knowledge, it is one of only a few building simulation programs that
have the ability to explicitly [1012]. For more information about
validation of the EnergyPlus program, see [20].
4. Simulated cases
As mentioned above the work was divided into two stages. In
the rst one, a screening of nine possible variables affecting UCLR,

To determine the most signicant variables affecting UCLR, SPF,

ZF and RPF, 33 simulations were performed.
The inuence of the following parameters was investigated:
oor level, zone orientation, supply air temperature, windowto-wall ratio, internal heat gain, plenum conguration, climate,
presence of the carpet and structure type. For each simulation the
cooling load prole and the four indexes were calculated for the ve
zones (interior, east, north, west, south) of each oor level (ground,
middle, top oor). The supply air temperature is the air temperature supplied to the plenum. In this model, heat gain in the supply
ducts was not considered. Three levels of supply air temperature
were studied: 13.9 C, 15.6 C and 17.2 C. Three window-to-wall
ratios were investigated: 20%, 40% and 60%. Different WWRs are
achieved by varying the window height only. Three internal gain
levels were simulated. Seven different U.S. climates were investigated, which are summarized in Table 1. The inuence of the carpet
was also investigated. Three construction types were simulated,
light-, medium-, and heavy-weight. The construction characteristics are summarized in Table 2. For the properties of the materials
used, please refer to Table 18 of [3]. Two plenum congurations
were investigated: series (the AHU conditioned air goes from the

Table 1
Climates investigated.
Denomination

City

Climate according to ASHRAE 90.1

HTX
LVNV
SFCA
BMD
SWA
COIL
MMN

Houston (TX)
Las Vegas (NV)
San Francisco (CA)
Baltimore (MD)
Seattle (WA)
Chicago (IL)
Minneapolis (MN)

2A
3B
3C
4A
4C
5A
6A

Hot and Humid

Warm and Dry
Warm and Marine
Mixed and Humid
Mixed and Marine
Cool and Humid
Cold and Humid

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

521

Table 2
Characteristics for the construction for the light-, medium-, and heavy-weight construction class, in parenthesis the overall U-factors (W/m2 K).
Construction class

Exterior wall

Ceiling/oor

Partition

Roof

Ground oor slab

Light-weight (LW)

0.8 mm steel siding,

50 mm insulation
board, 50 mm air space,
19 mm gypsum (0.49)
100 mm brick, 50 mm
insulation board,
50 mm air space,
19 mm gypsum (0.47)
100 mm brick, 200 mm
of heavyweight
concrete, 50 mm
insulation board,
50 mm air space,
19 mm gypsum (0.45)

100 mm of lightweight
concrete, ceiling air
space, acoustic tile
(1.45)
100 mm of
heavyweight concrete,
ceiling air space,
acoustic tile (1.82)
200 mm of
heavyweight concrete,
ceiling air space,
acoustic tile (1.66)

19 mm gypsum, 50 mm
air space, 19 mm
gypsum (2.59)

100 mm of lightweight
concrete, ceiling air
space, acoustic tile
(1.45)
100 mm of
heavyweight concrete,
ceiling air space,
acoustic tile (1.82)
200 mm of
heavyweight concrete,
ceiling air space,
acoustic tile (1.66)

100 mm of lightweight
concrete, ceiling air
space, Heavy insulation
(0.27)
100 mm of
heavyweight concrete,
ceiling air space, Heavy
insulation (0.28)
200 mm of
heavyweight concrete,
ceiling air space, Heavy
insulation (0.28)

Medium-weight (MW)

Heavy-weight (HW)

interior to the perimeter), and parallel (the conditioned air is independently supplied to the perimeter and interior plenums).
Table 3 summarizes the simulated cases for the variable screening. As a reference case the following parameters were chosen:
supply air temperature equal to 17.2 C, internal heat gain equal to
22.6 W/m2 , structure type light-weight, WWR equal to 40%, series
plenum conguration, carpet present, and climate for Baltimore,
MD. The reference case for UFAD is case 2 and for OH it is case 1.
4.2. Balanced matrix
From the variable screening process (see related section in the
results) it was shown that the most important variables are the
oor level and zone orientation. The structure type, supply air temperature, and internal heat gain showed a small but not negligible
inuence on the indexes. A full balanced matrix of these parameters was developed and simulated. Three supply air temperatures
(13.9 C, 15.6 C and 17.2 C), three structure types (light, medium,
heavy), three internal heat gain levels (22.6 W/m2 , 40.9 W/m2 ,
58.1 W/m2 ), two ventilation principles (UFAD and OH) were studied, therefore 3 3 3 2 = 54 simulations were carried out. The
other variables were held constant at; WWR = 40, Plenum Conguration = Series, Carpet = Yes, City = Baltimore, MD.
4.3. Statistical analysis
The data distributions are reported as frequency histograms and
as box-plots when more than one variable is plotted. A box-plot
is a way of graphically summarizing a data distribution. In a boxplot the thick horizontal line in the box shows the median. The
bottom and top of the box show the 25th and 75th percentiles,
respectively. The horizontal line joined to the box by the dashed
line shows either the maximum or 1.5 times the interquartile range
of the data, whichever is smaller. Points beyond those lines may be
considered as outliers and they are plotted as circles in the boxplot graphs. The interquartile range is the difference between the
25th and 75th percentiles [21]. The normal distribution of the data
was tested with the ShapiroWilk normality test [22]. Correlation
between variables is reported with Spearmans rank coefcient if
the variable does not have a normal distribution and with the Pearson correlation if it has a normal distribution.
Regression models were selected based on R-squared adjusted
values and authors judgment of the maximum number of useful
explanatory variables. R-squared, the coefcient of determination
of the regression line, is dened as the proportion of the total sample variability explained by the regression model. Adding irrelevant
predictor variables to the regression equation often increase Rsquared, to compensate for this, R-squared adjusted can be used.
R-squared adjusted is the value of R-squared adjusted down for a

19 mm gypsum, 50 mm
air space, 19 mm
gypsum (2.59)
19 mm gypsum,
200 mm heavyweight
concrete, 19 mm
gypsum (2.94)

higher number of variables in the model. The statistical analysis

was performed with R version 2.10.1 [23].
5. Results and discussion
The results and discussion section is divided into two parts. After
a short introduction, where an example of cooling load proles
for overhead and for UFAD is shown, the results for the screening
analysis, which was used to select the variables for the balanced
matrix runs, are reported. In the second part, the results of the balanced matrix and the development of the regression equations are
described.
Fig. 3 shows the design day cooling load proles for overhead
(mixing) and UFAD systems for the ve zones of the middle oor
for the reference cases (Cases 1 and 2 in Table 3). The cooling load
proles for the two systems are different. The main reason for the
difference between the overhead system and UFAD is the thermal
behaviour of the raised oor [9]. The HVAC system is operating
from 5:00 to 19:00. The amount of energy generated in the building
or entering the building is the same for the two systems but the
amount that the HVAC system has to remove is different because,
for the OH system, part of the energy is accumulated in the slab
during the day and is released at night when the system is off. In
the UFAD system the presence of the raised oor reduces the ability
of the slab to accumulate heat, thereby impacting the cooling load,
as described by Schiavon et al. [9].
5.1. Screening
The raw data for the screening cannot be reported here
due to space limitations. They are publicly available at

Fig. 3. Cooling load proles for overhead (mixing) and UFAD systems for the ve
zones of the middle oor for the Case 1 and Case 2 of Table 3.

522

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

Table 3
Simulated cases for the variable screening.
Case

Ventilation system

Supply air
temperature [ C]

Internal heat load

[W/m2 ]

Structure type

Window-to-Wall
Ratio [%]

Plenum conguration

Carpet

City

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33

OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD
OH
UFAD

17.2
17.2
15.6
15.6
13.9
13.9
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2
17.2

22.6
22.6
22.6
22.6
22.6
22.6
40.9
40.9
58.1
58.1
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6
22.6

Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Medium
Medium
Heavy
Heavy
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light
Light

40
40
40
40
40
40
40
40
40
40
40
40
40
40
20
20
60
60
40
40
40
40
40
40
40
40
40
40
40
40
40
40
40

Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Parallel
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series
Series

Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes

BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
BMD
HTX
HTX
LVNV
LVNV
SFCA
SFCA
SWA
SWA
CIL
CIL
MMN
MMN

http://tinyurl.com/RawDataUCLR. In Fig. 4 the box-plots of

the UFAD Cooling Load Ratio (UCLR) for the screening simulations
are shown. In the box-plots, UCLR is plotted versus supply air
temperature, internal heat gain, structure type, oor level, zone,
window-to-wall ratio, plenum conguration, presence of the
carpet, and climate. In the same way the SPF and ZF are reported
in Figs. 5 and 6, respectively. In all comparisons, Cases 1 and 2 of
Table 3 were chosen as the reference cases.
The aim of the screening is to determine which parameters have
the most signicant inuence on the developed indexes, and therefore should be included in the balanced matrix. From Fig. 4 it can
be deduced that the oor level and the zone are the parameters
that mostly affect UCLR. The oor and the zone are also the parameters that mostly affect SPF (Fig. 5) and ZF (Fig. 6). Structure type,
supply air temperature and internal heat gain slightly affect SPF.
From the analysis of the RPF was deduced that the oor level is the
single parameter that mostly affects RPF. From the screening it can
be concluded that the most important factors are the oor level
and the zone orientation. Structure type, supply air temperature,
and internal heat gain have a small but not insignicant inuence
on the indexes. The WWR, plenum conguration, presence of the
carpet and climate do not signicantly affect UCLR, SPF, ZF, and RPF.
5.2. Balanced matrix
The raw data are publicly available at http://tinyurl.com/
RawDataUCLR. The summary statistics of all the balanced matrix
simulations are reported in Table 4. UCLR has a median value equal
to 1.19, this means that generally UFAD has a 19% higher peak cooling load than OH. On average, one third of the UFAD cooling load
goes into the supply plenum and the other two thirds to the zone.
The results are consistent with the predictions of Bauman et al. [15].

The results for RPZ are strongly skewed, i.e., the mean is different
from the median. From Fig. 9 it can be deduced that this is due to
the fact that the value of RPF is high only in the top oor where the
inuence of the solar radiation to the roof affects the cooling load
distribution.
5.3. UFAD Cooling Load Ratio (UCLR)
A frequency histogram of the UCLR is shown in Fig. 7. The
ShapiroWilk normality test (W = 0.9955, p-value = 0.29) and histogram demonstrate that UCLR has a normal distribution. The
correlation table (Pearson correlation coefcient r) showed that
UCLR is not correlated with the supply air temperature (r = 0.09)
and only slightly correlated with the internal heat gain (r = 0.16).
The box-plots of UCLR versus the independent variables (supply
air temperature, internal heat gain, structure type, oor level and
zone) are shown in Fig. 7.
From Fig. 7 it can be deduced that the supply air temperature,
internal heat gain and structure type have little effect on UCLR
and thus should not be taken into account in a regression model.
The oor level and zone type have a strong inuence on UCLR
(see Fig. 7). Using these results, a multi-variable linear model was
developed. The best model was selected based on the R-squared
adjusted method. To keep the model as simple as possible the minimum number of explanatory variables should be used. Using all
the available variables leads to an R-squared adjusted equal to 0.87.
Removing the structure type, internal heat gain and the supply air
temperature from the model simplies the model without excessively reducing the value of R-squared adjusted. The interaction
effects between the zone and the oor is signicant (p < 0.001) but
has little inuence on R-squared adjusted thus it has also been
excluded. The chosen model takes into account two variables as

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

523

Fig. 4. Box-plots of the UFAD Cooling Load Ratio (UCLR) for the screening simulations. In the box-plots UCLR is plotted versus supply air temperature, internal heat gain,
structure type, oor level, zone, window-to-wall ratio, plenum conguration, presence of the carpet, and climate.

shown in Eq. (6). C1 is a function of the oor level (ground oor,

middle oor and top oor). C2 is a function of the orientation (east,
south, west, north and interior zone).
UCLR = C0 + C1 + C2

(6)

with: C0 = 0.9528; C1 = 0 if oor is the ground oor; C1 = 0.1572 if

oor is between oors; C1 = 0.2379 if the oor is the top oor; C2 = 0
if the zone is north oriented; C2 = 0.1739 if the zone is east oriented;
C2 = 0.0999 if the zone is south oriented; C2 = 0.1349 if the zone is
west oriented; C2 = 0.0802 if the zone is the interior zone.
From the ANOVA analysis of the multi-variable regression model
was deduced that all the variables are signicant (p < 0.001) and
that Adjusted R-squared is equal to 0.81. Visual evaluation of the
plot of residuals versus tted values and of the normal probability
plot showed that the residuals have a normal distribution with a
mean of zero and that they have the same variance for each tted value. This means that the hypotheses of the linear regression
model are met and thus, the model is valid.

5.4. Supply Plenum Fraction (SPF)

The frequency histogram of the Supply Plenum Fraction (SPF)
is shown in Fig. 8. The ShapiroWilk normality test (W = 0.9122,
p-value < 0.001) and the histogram showed that SPF does not have
a normal distribution, even if its deviation is not large. The correlation table (Spearmans rank r) showed that SPF is very weakly
correlated with the supply air temperature (r = 0.21) and with the
internal heat gain (r = 0.16). The box-plots of the SPF versus the
independent variables (supply air temperature, internal heat gain,
structure type, oor level and zone) are also shown in Fig. 8.
From Fig. 8 it can be deduced that the zone has a strong inuence
on SPF; in particular there is a signicant difference between interior and perimeter zones. The data was divided in two categories
(interior and perimeter), the analysis showed that the perimeter
zones have a median value equal to 0.22 and a narrow spread,
implying that there is not much variability, the interior zone has a
wider spread with a median value equal to 0.37. The narrow spread
of SPF in the perimeter zone means that the perimeter zone is an
effective variable explaining the amount of the cooling load going

Table 4
Mean, standard deviation, maximum and minimum values for the investigated parameters.
Parameter

Mean

Standard Deviation

Median

Minimum

Maximum

UFAD Cooling Load Ratio (UCLR)

Supply Plenum Fraction (SPF)
Zone Fraction (ZF)
Return Plenum Fraction (RPF)

1.18
0.28
0.62
0.10

0.13
0.11
0.16
0.15

1.19
0.27
0.63
0.02

0.82
0.07
0.25
0.00

1.55
0.72
0.93
0.51

524

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

Fig. 5. Box-plots of the Supply Plenum Fraction (SPF) for the screening simulations. In the box-plots SPF is plotted versus supply air temperature, internal heat gain, structure
type, oor level, zone, window-to-wall ratio, plenum conguration, presence of the carpet, and climate.

to the supply plenum and that we can expect that percentage is

varying within a small range. The wide spread of SPF in the interior
zone means that other factors inuence SPF.
A multi-variable linear model was developed for the plenum
fraction. The analysis of the R-squared adjusted values for the most
signicant combinations of variables showed that internal heat
gain, structure type, and supply air temperature do not contribute
enough to explain the variability of SPF and therefore, were not
included in the model. Fig. 8 shows that the inuence of the four
perimeter zones on SPF all behave in a similar way, therefore it is
possible to lump them into a single variable called zone lumped
that it can take only the value: interior and perimeter. The effectiveness of this new variable can also be seen by comparing the
R-squared adjusted for the zone (0.35) and the zone lumped (0.33);
they are almost the same, and thus, it is more convenient to use
the lumped zone variable. The two-variable model (lumped zone,
oor plus interaction effect) has the same R-squared adjusted as
the model with all ve variables. Thus, the simplest is chosen. The
model has the following structure:
SPF = C0 + C1 + C2 + C3

(7)

From the ANOVA analysis of the multi-variable regression model

it can be seen that all the variables are signicant (p < 0.001) and
that Adjusted R-squared is equal to 0.80.
The plot of residuals versus tted values used to check the validity of the hypothesis of constant variance required for the linear
model, showed that the hypothesis is violated (heteroscedastic).
The same problem was revealed in the plot of the square root of
the standardized residuals versus the tted values. It was evident

that the variability in the standardized residuals tends to increase;

therefore the model was not valid. Ignoring non-constant variance when it exists invalidates all inferential tools like p-value,
condence intervals and prediction intervals [24]. To overcome
the problem of non-constant variance a power transformation of
the response variable (SPF) was applied. The BoxCox method
[25] was used to nd the exponent of the transformation. The
results showed that applying the square root the dependent variable would remove the problem. The new equations will have the
following structure:(8)(SPF)0.5 = C0 + C1 + C2 + C3 with: C1 is a function of the zone type (interior or perimeter); C2 is a function of the
oor level (ground oor, between oor, and top oors); C3 is a function of the combination of oor level and zone type; C0 = 0.6179;
C1 = 0 if the zone is a interior zone; C1 = 0.2095 if the zone is a
perimeter; C2 = 0 if oor is the ground oor; C2 = 0.1242 if oor is
between oors; C2 = 0.0896 if the oor is the top oor; C3 = 0 if
the zone is interior zone or the oor is the ground oor; C3 = 0.0396
if the zone is a perimeter zone and the oor is between oors;
C3 = 0.1642 if the zone is a perimeter zone and the oor is the top
oor.
From the ANOVA table of the multi-variables regression model it
can be seen that all the variables are signicant (p < 0.001) and that
adjusted R-squared is equal to 0.79. Visual evaluation of the residuals versus the tted values showed that the heteroscedasticity
problem was solved. A visual evaluation of the normal probability
plot showed that the residuals have a mild non-normal distribution, and mild non-normality, which can safely be ignored [24].
This means that the hypotheses of the linear regression model are
met and thus the model is valid.

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

525

Fig. 6. Box-plots of the Zone Fraction (ZF) for the screening simulations. In the box-plots ZF is plotted versus supply air temperature, internal heat gain, structure type, oor
level, zone, window-to-wall ratio, plenum conguration, presence of the carpet, and climate.

Fig. 7. Frequency distribution of the UFAD Cooling Load Ratio (UCLR) and box-plots for UCLR versus supply air temperature, internal heat gain, structure type, oor level and
zone.

526

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

Fig. 8. Frequency distribution of the UFAD Supply Plenum Fraction (SPF) and box-plots for SPF versus supply air temperature, internal heat gain, structure type, oor level
and zone.

5.5. Return Plenum Fraction (RPF)

The frequency histogram of the Return Plenum Fraction (RPF)
is shown in Fig. 9. The ShapiroWilk normality test (W = 0.709,
p-value < 0.001) and the histogram showed that RPF has a strong
non-normal distribution. The correlation table (Spearmans rank)
showed that RPF is not correlated with the supply air temperature
(r = 0.11) and with the internal heat gain (r = 0.01). The box-plots
of RPF versus the independent variables (supply air temperature,
internal heat gain, structure type, oor level and zone) are also
shown in Fig. 9. From Fig. 9 it can been seen that RPF does not
vary much and the most signicant parameter is the oor level.
In particular, there is a clear difference between the top oor and
other oors, most of the variance is generated in the top oor. The
mean for the top oor is 0.30 (it has a normal distribution) and for
the other oors, the median is 0.01.
The Return Plenum Fraction model is summarized in Eq. (9). C1
is a function of the oor level (ground oor, middle oor and top
oor).
RPF = C1 X1

(9)

where: X1 = oor level. There are three possible levels: ground oor,
middle oor and top oor; C1 = 0.01 if oor level is the ground oor;
C1 = 0.01 if oor level is middle oor; C1 = 0.30 if the oor level is
the top oor.
5.6. Zone Fraction (ZF)
The frequency histogram of the Zone Fraction (ZF) is shown
in Fig. 10. The ShapiroWilk normality test (W = 0.971, pvalue < 0.001) and the histogram showed that ZF does not have a
normal distribution, even if its deviation is not strong. The correlation table (Spearmans rank) showed that ZF is very weakly
correlated with the supply air temperature (r = 0.10) and with the
internal heat gain (r = 0.18). The box-plots of ZF versus the independent variables (supply air temperature, internal heat gain, structure
type, oor level and zone) are also shown in Fig. 10. From Fig. 10

it can be deduced that the zone and oor level have the strongest
inuence on ZF; in particular, there is a strong inuence on the
interior zone. It is not necessary to develop a model for the prediction of the Zone Fraction because the sum of ZF, SPF and RPF has
to be equal to 1. The models for SPF and RPF have been developed
already, thus ZF = 1-SPF-RPF.
5.7. Summary of models
In this section the developed models are summarized.
The UFAD Cooling Load Ratio (UCLR) model is summarized in
Eq. (10).
UCLR = 0.9528 + C1 + C2

(10)

where: C1 is a function of the oor level (ground oor, middle oor

and top oor); C2 is a function of the orientation (east, south, west,
north and interior zone); X1 = oor level. There are three possible
levels: ground oor, middle oor and top oor; C1 = 0 if oor is the
ground oor; C1 = 0.1572 if oor is between oors; C1 = 0.2379 if the
oor is the top oor; C2 = 0 if the zone is north oriented; C2 = 0.1739
if the zone is east oriented; C2 = 0.0999 if the zone is south oriented;
C2 = 0.1349 if the zone is west oriented; C2 = 0.0802 if the zone is
the interior zone.
The Supply Plenum Fraction model is summarized in Eq. (11).
SPF = (0.6179 + C1 + C2 + C3 )2

(11)

where: C1 is a function of the zone type (interior or perimeter);

C2 is a function of the oor level (ground oor, between oor, and
top oors); C3 is a function of the combination of oor level and
zone type; C1 = 0 if the zone is a interior zone; C1 = 0.2095 if the
zone is a perimeter; C2 = 0 if oor is the ground oor; C2 = 0.1242
if oor is between oors; C2 = 0.0896 if the oor is the top oor;
C3 = 0 if the zone is interior zone or the oor is the ground oor;
C3 = 0.0396 if the zone is a perimeter zone and the oor is between
oors; C3 = 0.1642 if the zone is a perimeter zone and the oor is
the top oor.

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

527

Fig. 9. Frequency distribution of the UFAD Return Plenum Fraction (RPF) and box-plots for RPF versus supply air temperature, internal heat gain, structure type, oor level
and zone.

5.8. Example

The Zone Fraction model is summarized in Eq. (12).

ZF = 1 SPF RPF

(12)

The Return Plenum Fraction model is summarized in Eq. (13).

RPF = C1

(13)

where: C1 is a function of the oor level (ground oor, between

oor, and top oors); C1 = 0.01 if oor level is the ground oor;
C1 = 0.01 if oor level is middle oor; C1 = 0.30 if the oor level is
the top oor.

UCLR, SPF, ZF and RPF depend on the zone of the building and the oor level. In this example, UCLR, SPF, ZF and RPF
have been calculated for a perimeter zone with a south orientation located in a middle oor. The calculated indices are as
follows:
UCLR = 0.9528 + 0.1572 + 0.0999 = 1.21;
SPF = (0.6179 0.2095 + 0.1242 + 0.0396)2 = 0.337;
RPF = 0.01;
ZF = 1 0.337 0.01 = 0.653.

Fig. 10. Frequency distribution of the UFAD Zone Fraction (ZF) and box-plots for ZF versus the supply air temperature, the internal heat gain, the structure type, the oor
level and the zone.

528

S. Schiavon et al. / Energy and Buildings 43 (2011) 517528

5.9. Limitations of the study

Acknowledgments

The main limitations of this study are related to the selection

of the cases to be simulated and to the variables investigated.
The inuence on UCLR, SPF, ZF, and RPF of the following variables has been evaluated: building oor level, building zone, supply
air temperature, window-to-wall ratio, internal heat gain, plenum
conguration, climate, presence of the carpet and structure type.
We did not test the inuence of the different types of raised oor
construction, room height, insulation level, raised oor height and
building aspect ratio. We only considered ve zone orientations,
i.e., other possible orientations such as south-west corner are not
taken into account. Only UFAD diffuser types present in EnergyPlus
were studied. The interaction with a secondary cooling system, e.g.,
radiant panels, was not investigated. Solar shading has not been
applied because it is common in the design stage not to consider
shading for the design day calculation, and external shading is not
a common technology applied in U.S. commercial buildings. The
roof assembly does not comply with ASHRAE 90.1 [17], lower RPF
should be considered in the top oor if a better insulated roof is
used. The applicability of the developed models is limited to the
tested conditions.

For the help with energy simulations, the authors would like to
thank Tyler Hoyt, Assistant Specialist, and Jingjuan Feng, Graduate
Student Researcher; both are with the Center for the Built Environment. The present work was supported by the California Energy
Commission (CEC) Public Interest Energy Research (PIER) Buildings
Program and the Center for the Built Environment, University of
California, Berkeley.

6. Conclusions
The main conclusions of the development of a simplied calculation method for design cooling loads for underoor air distribution
(UFAD) systems are summarized below.
The design day cooling load proles for overhead and UFAD systems are different. In order to capture this difference the UFAD
Cooling Load Ratio (UCLR) has been introduced. UCLR is the ratio
of the cooling load calculated for UFAD to the cooling load calculated for a well mixed system (e.g., overhead). Generally, UFAD
has a peak cooling load that is 19% higher than the overhead
cooling load. A model to predict UCLR has been developed. UCLR
mainly depends on the oor level, zone type (interior, perimeter),
and orientation of the perimeter zone under analysis.
In a UFAD system the total cooling load is divided between the
supply plenum, the zone and the return plenum. Regression models to predict the split have been developed. The part of the
cooling load that goes to the supply plenum mainly depends on
the oor level, and the position and orientation of the zone under
analysis. For all cases considered, median values of the total cooling load going to the supply plenum were found to be 22% in the
perimeter zone and 37% in the interior zone. Except for the top
oor the Return Plenum Fraction is small.
Supply air temperature, window-to-wall ratio, internal heat gain,
plenum conguration, climate, presence of the carpet and structure type do not strongly inuence the difference in cooling
load between overhead and UFAD and the split of the cooling
load between the supply air plenum, the zone and the return
plenum.
The developed regression models allow the use of a familiar load
calculation procedure for mixing systems as input to the tool for
the UFAD cooling load calculation.

References
[1] F. Bauman, Underoor air distribution (UFAD) design guide, American society
of heating, Refrigerating and Air-Conditioning Engineers, 2003.
[2] Z. Jiang, Q. Chen, Air distribution effectiveness with stratied air distribution
systems, Final Report, ASHRAE Research Project RP-1373, Atlanta, US, 2009.
[3] ASHRAE Fundamentals Handbook, Chapter 18, American Society of Heating,
Refrigerating and Air-Conditioning Engineers, Inc., 2009.
[4] J.D. Spitler, Load calculation applications manual, ASHRAE (2009).
[5] F. Sodec, R. Craig, Underoor air supply system: guidelines for the mechanical
engineer, Report No. 3787A, Krantz GmbH & Co., Aachen, Germany, 1991.
[6] K. Loudermilk, Underoor air distribution for ofce applications, ASHRAE
Transactions 105 (1) (1999) 605613.
[7] YORK International, Convection Enhanced Ventilation Technical Manual, York,
PA, US, 1999.
[8] F. Bauman, T. Webster, C. Benedek, Cooling airow design calculations for
UFAD, ASHRAE Journal (October) (2007) 3644.
[9] S. Schiavon, K.H. Lee, F. Bauman, T. Webster, Inuence of raised oor on zone
design cooling load in commercial buildings, Energy and Buildings 42 (5) (2010)
11821191, doi:10.1016/j.enbuild.2010.02.009, http://escholarship.org/uc/
item/2bv611dt.
[10] Q.A. Liu, P.L. Linden, The EnergyPlus UFAD module, in: Proceedings of the Third
National Conference of IBPSA-USA, Berkeley, California, US, 2008, pp. 2328.
[11] EnergyPlus, Engineering Reference, U.S. Department of Energy, Energy
Efciency and Renewable Energy, Ofce of Building Technologies, 2010.
www.energyplus.gov.
[12] T. Webster, F. Bauman, F. Buhl, A. Daly, Modeling of underoor air distribution (UFAD) systems, in: Proceedings of the Third National Conference of
IBPSA-USA, Berkeley, California, US, 2008, http://www.ibpsa.us/simbuild2008/
technical sessions/SB08-DOC-TS11-1-Webster.pdf.
[13] H. Jin, F. Bauman, T. Webster, Testing and modeling of underoor air supply
plenums, ASHRAE Transactions 112 (2) (2006).
[14] F. Bauman, S. Schiavon, T. Webster, K.H. Lee, Cooling load design
tool for UFAD systems, ASHRAE Journal (September) (2010) 6271,
http://escholarship.org/uc/item/7hh1t2z4.
[15] F. Bauman, H. Jin, T. Webster, Heat transfer pathways in underoor air distribution (UFAD) systems, ASHRAE Transactions 112 (2) (2006).
[16] CEN (European Committee for Standardization), EN 15265-2006, Thermal Performance of Buildings Calculation of Energy Use for Space Heating and Cooling
General Criteria and Validation Procedures, European Committee for Standardization, Brussels, Belgium, 2006.
[17] ASHRAE, ASHRAE Standard 90.1-2004, Energy standard for buildings except
low rise residential buildings, American Society of Heating, Refrigerating and
Air-Conditioning Engineers, Inc., 2004.
[18] ASHRAE Fundamentals Handbook, Chapter 14, American Society of Heating,
Refrigerating and Air-Conditioning Engineers, Inc., 2009.
[19] P. Linden, J.K. Yu, T. Webster, F. Bauman, K.H. Lee, S. Schiavon, A.
Daly, Simulation of Energy Performance of Underoor Air Distribution
(UFAD) Systems, Building Energy Research Grant (BERG) Program, 2009.
http://www.escholarship.org/uc/item/7gr8r3d3.
[20] EnergyPlus, Testing and Validation, http://apps1.eere.energy.gov/buildings/
energyplus/testing.cfm.
[21] M.J. Crawley, Statistics: An Introduction Using R, Wiley, 2005.
[22] S.S. Shapiro, M.B. Wilk, An analysis of variance test for normality (complete
samples), Biometrika 52 (34) (1965) 591611.
[23] R: www.r-project.org.
[24] J. Faraway, Practical Regression and Anova Using R, 2002. http://cran.rproject.org/doc/contrib/Faraway-PRA.pdf.
[25] G.E.P. Box, D.R. Cox, An analysis of transformations, Journal of the Royal Statistical Society, Series B (Methodological) 26 (2) (1964) 211252.