Available online at www.sciencedirect.

com

Building and Environment 38 (2003) 753 – 762

www.elsevier.com/locate/buildenv

Modelling !ow and heat transfer around a seated human body

by computational !uid dynamics
Dan N(rtoft S(rensen∗ , Lars K(llgaard Voigt
Department of Mechanical Engineering, Fluid Mechanics Section and International Centre for Indoor Environment and Energy,
Technical University of Denmark, Building 403, DK-2800 Kgs. Lyngby, Denmark
Received 22 October 2002; received in revised form 6 January 2003; accepted 31 January 2003

Abstract

A database (http://www.ie.dtu.dk/manikin) containing a detailed representation of the surface geometry of a seated female human body
was created from a surface scan of a thermal manikin (minus clothing and hair). The radiative heat transfer coe7cient and the natural
convection !ow around the manikin were calculated using computational !uid dynamics. Results were compared to published data and to
measurements by particle image velocimetry. The agreement was generally good and variations across the manikin well predicted. The
uniqueness of the present work is twofold: (1) the geometry of the computational manikin has all surface features of a human being; (2)
the geometry is an exact copy of an experimental thermal manikin, enabling detailed comparisons between calculations and experiments.
? 2003 Elsevier Science Ltd. All rights reserved.

Keywords: Human body; Thermal manikin; CFD; Radiation; Natural convection

1. Introduction The manikins were placed in a wind tunnel facing down-

An important goal in research on the indoor environment
is an enhanced understanding of the mechanisms responsi-
ble for human perception to di>erent exposures. Normally,
human subjects are used for tests concerning perception of
the indoor environment, while experiments with manikins,
equipped to resemble humans, may provide valuable in-
formation regarding velocities, temperature, pollutant con-
centration in the breathing zone, etc. As an alternative to
manikins and human subjects, computational !uid dynam-
ics (CFD) may be used to obtain detailed information about
the !ow Aeld and distributions of temperature and concen-
trations inside a room and particularly in the close vicinity
of a person.
In several studies, the !ow around geometries
resembling a standing human body have been calculated.
Brohus presents CFD calculations of three di>erent geo-
metrical models, constructed using rectangular shapes, and
compares the results to measurements on a standing thermal
manikin [1,2]. The complexity of the models ranges from
a simple rectangle to a manikin with legs, torso, and head.
∗ Corresponding author. Tel.: +45-4525-4300; fax: +45-4593-0663.
E-mail address: dns@mek.dtu.dk (D.N. S(rensen).
stream, with a contaminant source placed in front of them.
The results show that the contaminant concentration in
the breathing zone varies signiAcantly with the geometry
of the manikin, but for all models, the results are more
accurate than for the case without a manikin. Moreover,
they conclude that inclusion of legs in the simulation is
important when the source is located near the !oor at low
velocities.
The experimental set-up from [2] was used by Hyun
and Kleinstreuer [3] to calculate the !ow Aeld and concen-
tration distribution for a breathing manikin constructed to
resemble a human being with divided legs, cylindrical neck
and spherical head. A comparison between inhaled pollu-
tant concentration during constant inhalation and normal
(transient) breathing was made, and a signiAcant in!u-
ence from the breathing set-up was observed on the dose
concentrations, which are usually smaller for the transient
case.
In a number of studies, Murakami and co-workers use
CFD to calculate the !ow around a smooth shape resem-
bling the human body in standing posture. They investi-
gate the coupling between convective heat transfer from
the manikin and the buoyancy-driven !ow Aeld around the
manikin placed in a low-velocity (almost stagnant) displace-

0360-1323/03/$ - see front matter ? 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0360-1323(03)00027-1
754 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762
ment ventilated room, as well as in a horizontal !ow Aeld
(0:25 m=s), considering two levels of turbulence intensity
in the supplied air [4]. The calculations show the buoyant
plume and the distribution of surface temperature for the
three cases. The convective heat transfer coe7cient is cal-
culated and compares well with previous experiments. The
study is extended with cases of upward and downward !ow,
and a strong in!uence from the !ow direction is found, with
a much higher heat transfer coe7cient in the downward !ow
case than in any of the other cases. For the upward !ow
case, the heat transfer coe7cient is almost the same as in the
stagnant !ow case [5]. Also, the !ow, temperature, and heat
transfer coe7cient are considered for two manikins placed
closely together as well as for a single manikin in outdoor
settings (horizontal velocity, 2:5 m=s) [6]. At the higher
velocities, buoyancy e>ects are found to be very small.
Finally, models are included for thermo-physiological reg-
ulation internally in the body as well as moisture transport
from sweating. These are coupled with the CFD and radi-
ation models, resulting in a complete model for the energy
balance of humans [7].
The geometrical shapes of the manikins in the above stud-
ies are simpliAed to reduce the geometrical complexity, and
thus the computational requirements. The arms are com-
bined with the body, and all facial extremities are removed.
In most cases, the e>ects of these simpliAcations are negli-
gible. However, if detailed information is required in, say,
the breathing zone, the presence of mouth, nose, and ears
may be required. Furthermore, as will be evidenced in the
present study, the presence of arms, armpits, and a high res-
olution of the legs signiAcantly in!uences local heat transfer
properties.
The unique feature of the present study is that the compu-
tational manikin is an exact copy of the experimental ther-
mal manikin, enabling similar set-ups during calculations
and experiments. Furthermore, the geometrical resolution is
high and includes all relevant features of the human body,
except hair and clothes. The computational manikin pre-
sented here was Arst used in [8], to study the natural convec-
tion !ow around a human body. The manikin was also used
to study mixed convection !ow around a person (although
without hands) in a wind tunnel [9,10]. Furthermore, de-
tailed segment-to-segment view factors were calculated to
estimate the radiative heat exchange between di>erent body
parts of the manikin [11].
A di7cult task to accomplish experimentally is the sepa-
ration between heat removal by convection to the !ow Aeld
near the body, and radiation to the surfaces of the room, as
discussed by e.g. de Dear et al. [12]. This separation is eas-
ily accomplished by CFD and in the present work the distri-
bution of radiative heat transfer coe7cients over the entire
body was calculated. Furthermore, average values for di>er-
ent segments of the body (i.e. head, arms, chest, etc.) were
calculated and compared to published measurements. Also,
natural convection !ow around the manikin was calculated
and compared to measurements in a similar set-up. The cal-
culated heat transfer, integrated across individual segments,
agreed well with measurements, as did the velocity above
the head of the manikin. The latter was obtained by particle
image velocimetry (PIV).
The geometry of the manikin described in this article may
be downloaded from http://www.ie.dtu.dk/manikin, where
information concerning the Ale structure and format is given
as well. The availability of the presented geometry for down-
loading through the internet will hopefully motivate other
researchers to use the same geometry. In this way, much of
the uncertainty—namely the di>ering geometries—of dif-
ferent CFD studies may be eliminated.
2. Methods
2.1. Geometry of the manikin
The type of thermal manikin 1 chosen for this study
has been used in a number of experimental investigations
[13–15]. The manikin was placed in a seated position with
the arms hanging down by the side. The manikin was
undressed and without hair, since those details could not
be accurately resolved by the laser scanning measuring
technique used. Furthermore, modelling clothing and hair
correctly is a di7cult task.
A laser scanning technique 2 was used to extract the
geometrical details of a thermal manikin in a seated po-
sition. The geometry is described by a closed surface of
triangular elements with an accuracy of around 0:5 mm,
compared to the original geometry. The geometry is
shown in Fig. 1, which gives a front view of the com-
plete geometry as well as detailed pictures of the left
hand and the facial region. Three di>erent resolutions
of the surface are available with approximately 125.000,
250.000, and 500.000 cells for the coarse, medium and
Ane surface grids, respectively. The Ane surface reso-
lution is shown on the facial region in Fig. 1. Also, a
slightly modiAed version of the Ane surface grid was
divided into 16 body segments (Fig. 2), corresponding
approximately to the original division of the thermal
manikin.
Table 1 shows the areas of the individual body segments
of the scanned manikin. For comparison, the areas provided
by the manufacturer of the thermal manikin are shown as
well, and some di>erences can be observed. The di>erences
between individual segments may be explained by slightly
di>ering divisions of the manikin into body parts. However,
the di>erence in the whole-body surface area (0:12 m2 ) in-
dicates that an actual di>erence exists. Considering the to-
tal area for the segments comprising the legs (foot, leg,
and thigh), the di>erence is almost 0:1 m2 , which accounts
1 Manufactured by P.T. Teknik, Denmark.
2 Cascade Computing A/S, Denmark.
 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762 755

Table 1
Surface area of the individual segments of the manikina

Segment Area (CFD) Area (exp)

Lfoot 0.0452 0.043

Rfoot 0.0456 0.041
Lleg 0.1081 0.089
Rleg 0.1083 0.089
Lthigh 0.1858 0.160
Rthigh 0.1894 0.165
Lhand 0.0340 0.038
Rhand 0.0315 0.037
Larm 0.0507 0.052
Rarm 0.0525 0.052
Lshoulder 0.0745 0.073
Rshoulder 0.0779 0.078
Pelvis 0.1744 0.182
Head 0.1206 0.100
Chest 0.1530 0.144
Back 0.1421 0.133

Whole body 1.594 1.476

a The Arst column of data shows the areas (m 2 ) calculated by the CFD
Fig. 1. View of the scanned manikin body, zoom of left hand and of
program, from the laser scanned surface. The second column shows the
the face. For the latter, the Ane-grid geometrical resolution of the surface
areas provided by the manikin manufacturer.
scan is shown.

2.2. Computational grid

The results presented in Section 3 were made in a room

Fig. 2. The division of the surface of the scanned manikin into segments.
The head segment comprise the neck, face, mouth and nostrils. The hand
segments include the thumbs.
for most of the missing surface. The reason for the large
variation for these segments is explained by the construction
of the knees of the thermal manikin, which includes an un-
heated !exible joint made up of cloth. This part of the leg is
not included during the processing of the experimental data
[16], whereas it is included in the surface of the scanned
surface and, correctly, in the CFD calculations.
of length, width and height of 2.95, 2.95 and 2:4 m, respec-
tively, and the manikin was positioned so that the tip of the
nose was centred between the walls (x=y=0), at a height of
z = 1:25 m (Fig. 3). In this way, the feet were 0:02 m above
the !oor. The computational grid used to resolve the volume
of the room was constructed as two separate regions (Fig. 3).
(1) A small rectangular volume, completely surrounding the
geometry of the scanned manikin. (2) An outer region cov-
ering the rest of the room. This division enabled the use of
an unstructured grid around the complex geometry of the
manikin, whereas a structured grid was used for the outer
region. The two regions were connected by so-called “arbi-
trary couples” [17], which ensures a correct interpolation of
the dependent variables across the interface. Furthermore,
local reAnement of the structured grid was made above the
manikin, where the buoyant plume was expected.
The rationale of the two-region set-up was that structured
grids (hexahedral cell topology) ensure higher accuracy for
a given number of grid points and that stretching towards the
side walls was easy to accomplish. However, the complex
geometry of the manikin was very di7cult to resolve by a
structured grid, using a reasonable number of grid points
[8]. For that reason an unstructured grid (tetrahedral cell
topology) was used for the volume surrounding the manikin.
To resolve the boundary layer around the manikin, 20 layers
of extruded triangular prisms were created at the surface of
the manikin with an initial height of 0:2 mm, and a growth
rate of 1.13 between layers (Fig. 4).
756 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762

Fig. 3. The computational domain divided into two regions. The inner
region is resolved by an unstructured grid and the outer region by a
structured grid.

The number of triangular elements used to describe

the geometry of the manikin (Section 2.1) was too high
to use as a surface grid for the CFD calculations. For-
tunately, the geometrical variations were small for most
of the surface (e.g. legs, chest, back), and the size of
the surface cells was determined from requirements of
the !ow and thermal resolution, rather than from the
geometrical requirements. Thus, a total of 23 000 trian-
gles were used to describe the surface of the computa-
tional manikin used in the CFD calculations. Most of
Fig. 4. Side view of the symmetry-plane of the manikin. The structured
these were placed around the face, ears and thumbs to and unstructured grid-regions are visible. The zoom at the surface of the
resolve the Ane geometrical features in these regions. computational manikin shows the layers of prism cells used to resolve
The full grid comprised slightly more than one million the boundary layer !ow.
cells. For the grid-dependency study (Section 3.3) an
even Aner grid was made, comprising more than three
million cells with 64 000 cells on the surface of the
manikin. was used with a second order scheme for the convective
terms. 3
2.3. Numerical methods In this study, we consider only surface-to-surface radia-
tion and in STAR-CD, this is modelled by a discrete beam
The commercial CFD code STAR-CD [17] was used approach, in which the surfaces of the calculation domain
to calculate the !ow and heat transfer. The !ow Aeld was are subdivided into adjoining patches. From each radiation
calculated by the three-dimensional and steady Reynolds patch, a user-deAned number of beams are emitted and traced
averaged Navier–Stokes (RANS) equations, combined through the !uid until another patch is reached, thereby
with continuity and energy equations, and closed with a 3 Denoted MARS (Monotone Advection and Reconstruction Scheme)
low Reynolds number k– model [18]. Standard values in [17]. A MUSCL (Monotone Upstream-Centered Schemes for Con-
were used for the turbulence model [17], and the turbulent servation Laws) type scheme with a TVD (total variation diminishing)
Prandtl number was Axed at 0.9. The SIMPLE algorithm limiter [19]. Run with a compression factor of 0.05.
 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762
creating a radiation-dependent pair of patches. The number
of beams between two such patches determines the view
factor between the patches. The radiative heat exchange be-
tween two patches is thus a function of the view factor be-
tween the two patches and depends on the emissivity and
average temperature of each of the two patches. All surfaces
are treated as grey with a constant emissivity.
The size of the radiation patches determines the resolu-
tion of the radiative heat exchange. For the walls, ceiling
and !oor, the length of the patch was set at 0:3 m. For the
surface of the manikin, each boundary face was deAned as a
patch. Because of the large number of boundary faces used
to resolve the details of the hands, the radiation patches for
the hands comprised several boundary faces to decrease the
number of radiation patches. From the above, a very high
resolution of the radiative heat transfer for the surface of the
manikin may be expected.
For the calculation of the radiative heat transfer coe7cient
without !ow (Section 3.1), only the energy equation was
solved with zero conductivity for the !ow, thereby isolating
the e>ect of radiative heat exchange between surfaces.
3. Results and discussion
In Section 3.1, the radiative heat transfer coe7cients are
calculated and compared to measurements from de Dear
et al. [12]. Section 3.2 compares calculated and measured ve-
locity Aelds for natural convection !ow around the manikin.
Finally, Section 3.3 discusses the accuracy of the numerical
calculations.
3.1. Radiative heat transfer coe9cients
The emissivity of the nude manikin was assumed to be
0.95 [12], whereas the outer walls, !oor and ceiling of the
room were given an emissivity of one. The temperature of
the room surfaces was Axed at 19◦ C, whereas the tempera-
ture of the surface of the computational manikin was 31◦ C.
For the calculations in this section, the main computing
time was spent during the calculation of the view factors,
whereas the actual radiative heat !ux calculations converged
in a few iterations due to the decoupling of the radiation
problem from the !ow problem. The resulting surface !uxes
were divided by the temperature di>erence between the com-
putational manikin and the outer walls (i.e. 12◦ C) to give
the radiative heat transfer coe7cient (in [W=(m2 K)]).
Fig. 5 shows front and back views of the calculated distri-
bution of the radiative heat transfer coe7cient and the vari-
ations are seen to be rather large. Notably, the sides of the
body and the inner sides of the arms have smaller radiative
heat transfer coe7cients than the front, back and outer sides
of the arms because of the proximity between the arms and
the body. The pelvic region has a very small radiative heat
transfer coe7cient because of the in!uence from the thighs.
Low values are found on the inner sides of the legs due to
757
Fig. 5. Distribution of radiative heat transfer coe7cient [W=(m2 K)].
The maximum value is 5:7 W=(m2 K) and the minimum value is
0:4 W=(m2 K).
the small distance between these. Furthermore, the head is
seen to in!uence the radiative heat transfer on the upper part
of the torso and, Anally, the lower side of the breasts has a
decreased radiative heat transfer.
The high resolution of the radiative heat transfer co-
e7cient in the present calculations has not previously
been obtained. To compare the data with measurements,
average values across each segment (Fig. 2) were calculated.
Table 2 compares the present calculations with measure-
ments of a seated thermal manikin from de Dear et al.
(Table 5 in [12]). The experimental approach used in [12]
is based on a heated thermal manikin, where each segment
is heated individually. To obtain the radiative heat transfer
coe7cient, two measurements for each body segment are
made: one with a low-radiating surface, and one with the
normally used surface. The average segment temperature is
kept the same for the two measurements, and the convective
heat transfer can thus be assumed to be constant. Therefore,
the heat !ux di>erence between the two measurements is
a direct measure of the radiative heat !ux, which is then
translated into the radiative heat transfer coe7cient.
The segmentation and posture of the manikin in the
present calculation di>ers from the segmentation and pos-
ture in the experimental work, and these di>erences prob-
ably account for most of the deviations found between
calculations and experiments.
From Table 2, a reasonable agreement is found for most
segments, though signiAcant deviations can be seen for the
head, arms, chest, and feet.
A major di>erence between the two set-ups is that the
experimental manikin is wigged with shoulder-length hair,
758 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762

Table 2 pared to measurements by PIV, 4 and care was taken to

Radiative heat transfer coe7cients for di>erent body segmentsa ensure similar set-ups. In both cases, the manikin was nude,
Body segment Calculations Experiments Deviation unwigged and without breathing. The room geometry was
[W=(m2 K)] [W=(m2 K)] (%) identical to that described in Section 3.1.
The PIV technique requires particles in the measuring
Foot 5.36 4.2 24
Leg 5.12 5.4 −5 zone and therefore the !oor comprised perforated plates
Thigh 4.61 4.6 0 through which these particles were introduced. The entire
Hand 4.12 3.9 5 ceiling was covered with cloth through which the air!ow
Arm 4.30 5.2 −21 was exhausted. The set-up created a vertical !ow with an
Shoulder 4.63 4.8 −4
average velocity of approximately 0:02 m=s. To model these
Pelvis 5.01 4.8 4
Head 5.22 3.9 25 conditions in the CFD calculations, the !oor was deAned
Chest 4.73 3.4 28 as an inlet with supply velocity 0:02 m=s and a turbulence
Back 5.07 4.6 9 intensity and turbulent length scale of 30% and 0:004 m, re-
spectively. All of the ceiling was deAned as an outlet bound-
Whole body 4.83 4.5 7
ary condition.
a Calculated radiative heat transfer coe7cients for di>erent body seg- During experiments, the nude thermal manikin was oper-
ments of the computational manikin (=0:95, Twall =19◦ C, Tbody =31◦ C). ated at a surface temperature of 31◦ C, which was also used
Experimental values are from [12]. The deviation is relative to the calcu- for the CFD calculations. The set-up was placed in a large
lated values. The whole-body values are based on area-averaging of the
individual segment-values. When left and right side exist (e.g. left leg
laboratory facility and was not insulated. The wall temper-
and right leg), the average value is used for the radiative heat transfer ature of the set-up was monitored and was almost constant
coe7cient. at 19:75◦ C. In the CFD calculations, this temperature was
used for the supply and for the side walls. With regard to
radiation in the CFD model, both !oor and ceiling had a
temperature of 19:75◦ C.
whereas the computational manikin is bald. For the head It was assumed that the e>ects from the small through!ow
of the manikin, this probably explains the larger radiative velocity were small relative to the buoyancy-driven !ow
heat transfer coe7cient of the calculation compared to the created by the heated body, essentially resulting in a natural
measurements. This is also believed to explain the larger convection !ow. However, as a consequence of the through-
calculated value for the radiative heat transfer for the back !ow, the temperature away from the body was close to the
of the manikin. supply temperature, and for the subsequent calculations of
The position of the arms is di>erent in the two set-ups. heat transfer coe7cients the temperature di>erence between
For the computational manikin, the arms are hanging down the body and the supply was used, since this is close to the
by the sides, relatively close to the torso and thighs, whereas normally used temperature di>erence at a height of 0:6 m.
the experiments are carried out with arms that are bent, and For the experiments, the control system of the thermal
seem to be further away from the body (Fig. 2 in [12]). Thus, manikin varied the heat !ux for each segment of the manikin
the shadow e>ects from the torso and thighs are greater for to obtain the speciAed temperature, and the heat !ux for each
the calculations than for the measurements, explaining the segment was logged. It should be noted that the measured
lower radiative heat transfer for the calculations. heat !ux comprises the total (radiative and convective) heat
The largest di>erence between calculations and mea- loss.
surements is found for the chest. The only reasonable Since all boundaries have Axed temperatures in the CFD
explanation for this is that the legs are more bent in the ex- calculations, radiation to the outer surfaces as well as be-
perimental set-up than in the calculations, thereby moving tween segments was decoupled from the !ow Aeld and tem-
the thighs closer to the chest. Consequently, the shadow perature distribution of the interior domain. The complexity
from the thighs is increased. However, this can probably of the additional non-linearity introduced by radiation is thus
not account for the very large di>erence in radiative heat avoided. For this type of boundary condition, however, the
transfer. heat !ux from the manikin to the surrounding !ow is a part
The whole-body average of the radiative heat transfer co- of the solution, rather than being speciAed explicitly. Since
e7cient is larger for the present calculations than for the the buoyant !ow at a distance from the manikin is mainly
measurements in [12]. However, the calculated value of governed by the total power supplied to the !ow, the plume
4.83 W=(m2 K) is in very good agreement with the widely
accepted value of 4:7 W=(m2 K) [20].

3.2. Natural convection :ow
Natural convection !ow around the computational
manikin was considered. The CFD calculations were com-
R
4 These unpublished experiments were carried out by Oktay Ozcan from
the Department of Aeronautics and Astronautics, Technical University
of Istanbul, while he was a visiting professor at the Department of
Mechanical Engineering, Fluid Mechanics Section, Technical University
of Denmark. His contribution is highly appreciated.
 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762
Fig. 6. Distributions of vertical velocity (m=s) (top) and temperature
(◦ C) (bottom). Sideview in symmetry-plane (left) and frontview (right)
in x = −0:12 m (centred near top of head).
above the manikin depends strongly on a correct prediction
of the heat !ux.
Before proceeding to the comparison between the CFD
calculations and the measurements, a few calculated results
are shown to clarify the !ow and heat transfer around the
computational manikin.
Fig. 6 shows the predicted distributions of vertical ve-
locity and temperature near the computational manikin. The
759
Fig. 7. Natural convection heat transfer coe7cient [W=(m2 K)].
rising plume is of signiAcant strength, with a maximum ver-
tical velocity above 0:5 m=s. The legs result in a wide plume
along the symmetry-plane (side-view), whereas the plume is
narrow in the front-view plane. Recirculating zones are seen
above the head and the shoulders and inside these zones,
the temperature becomes relatively high. The build-up of
momentum and thermal boundary layers can be observed,
notably along the back.
The complexity of the heat transfer is further illustrated
by the calculated distribution of the natural convection heat
transfer coe7cient around the computational manikin in
Fig. 7, where a signiAcant variation is seen. Locally, values
below one are found inside the recirculating zones at the
upper part of the head and shoulders, whereas values above
ten are found at the tip of the Angers and at the lower side
of the ears and nose.
The CFD predictions are compared to measurements in
Table 3, which contains average values of the total heat
!ux for each segment as well as the whole-body average.
Values are given for the CFD calculations as well as for
the experiments and, because the experiments include the
combined e>ect of radiation and convection, these are the
values provided in the table for the CFD calculations as
well. Furthermore, the table contains the deviation between
the measured and calculated heat transfer. The Anal column
shows the convective heat transfer coe7cient from the CFD
predictions (i.e. without the contribution from radiation).
Generally, the calculated and measured total heat !uxes
are in excellent agreement, with deviations below 10%. The
largest deviations appear for the legs, which can proba-
bly be explained by the fact that the areas for the seg-
ments comprising the legs di>ered signiAcantly between
the experimental and the computational manikin (see the
760 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762

Table 3
Total heat !ux and calculated convective heat transfer coe7cient from each body segmenta

Segment Total heat !ux Conv. coe>.

CFD Experiments Deviation CFD
(W=m2 ) (W=m2 ) (%) [W=(m2 K)]

Foot 113.0 101.7 11 4.66 (4.64)

Leg 92.0 116.7 −21 3.04 (3.03)
Thigh 87.8 98.8 −11 3.18 (3.19)
Hand 97.0 101.2 −4 4.50 (4.49)
Arm 91.5 99.1 −8 3.82 (3.76)
Shoulder 82.7 85.9 −4 2.71 (2.70)
Pelvis 88.1 90.5 −3 2.80 (2.82)
Head 99.8 104.0 1 3.62 (3.59)
Chest 80.2 78.5 2 2.38 (2.44)
Back 82.4 76.9 7 2.23 (2.23)

Whole body 89.67 88.26 2 3.13 (3.13)

a Calculated and measured total (convection and radiation) !ux from each body segment as well as the calculated convective heat transfer coe7cient
(Twall = Tinlet = 19:75◦ C). The heat transfer coe7cient is based on the di>erence between whole-body temperature and Twall . The whole-body values
are based on area-averaging of the individual segment-values. When left and right side exist, the average value is used. The column with the convective
heat transfer coe7cient contains the values for the Ane-grid solution in parentheses (see Section 3.3). Fig. 2 shows the various segments and Table 1
gives the segment areas.

discussion in Section 2.1). Also, the whole-body average

heat !ux agrees well. It was not possible to separate the ra-
diative and convective heat !ux in the present experiments,
but the excellent agreement of the total heat !ux justiAes the
CFD predictions of the natural convection !ow.
From Tables 2 and 3, it can be seen that the convective
and radiative heat loss from the body governs around 40%
and 60% of the total heat loss, respectively. This agrees well
with the Andings in [7], where 43% of the heat was lost
due to convection and 57% due to radiation (disregarding
the heat loss due to evaporation and respiration). Further-
more, de Dear et al. [12] reports an average natural convec- Fig. 8. Contours of vertical velocity (m=s), frontview in x = −0:12 m
tion heat transfer coe7cient of 3:3 W=(m2 K) for a seated (centred near top of head). Left: measured, right: calculated.
manikin, a value in close agreement with the present value
of 3:13 W=(m2 K). are explained by the large surfaces in the upward direction,
Compared to the widely accepted expression for the nat- which thus enables the build-up of a thick thermal boundary
ural convective heat transfer coe7cient [21], hc = 2:38(ts − layer. Although the head resides inside the thermal plume,
tw )0:25 W=(m2 K), which results in hc =4:36 W=(m2 K) for a the accelerated !ow around the head results in a large heat
temperature di>erence of 11:25◦ C, the whole-body convec- transfer coe7cient for that segment.
tive heat transfer coe7cient, predicted in the present study, To further validate the CFD calculations, these are com-
was signiAcantly smaller. However, the variation in reported pared to PIV measurements in Fig. 8, which shows the ver-
natural convection heat transfer coe7cients for humans is tical velocity in a vertical plane x =−0:120 mm (front view,
large (see discussions in [5,12]), and the present Andings approximately centred above the head). The agreement is
are within the reported values, although in the lower end. satisfactory, although a slightly higher velocity is apparent
Considering the natural convection heat transfer coe7- in the calculations.
cient in the last column of Table 3, there is a large vari-
ation. The hands and feet have very large convective heat 3.3. Numerical accuracy
transfer coe7cients because these peripheral segments are
mainly exposed to the cold rising air!ow which have not The numerical accuracy of the CFD calculations was en-
been heated by other parts of the body. This argument also sured in di>erent ways according to the procedures described
explains the rather large value for the arms, whereas the in [22]. Below, the di>erent means are discussed.
shoulders (with upper arms) are close to the heated body,
and thus are in!uenced by the thermal plume from the lower • As described in Section 2.3, the accuracy of the ra-
parts of the body. The very low values for the chest and back diation calculations depends on the number of beams
 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762
emitted from each radiation patch. The in!uence of the
segment-averaged radiative heat transfer coe7cient on
the number of beams was investigated for 100, 400, 1600
and 6400 beams, and a signiAcant change was found from
100 to 400 beams, and from 400 to 1600 beams, whereas
the change was very small from 1600 to 6400 beams.
The results in Sections 3.1 and 3.2 were therefore made
with 6400 beams emitted from each radiation patch.
• To assess the in!uence from the number of grid points
on the natural convection !ow, the calculations in Sec-
tion 3.2 were made using two di>erent grid resolutions
(see description of the grids in Section 2.2). The results
presented are for the coarse grid, except for the natural
convection heat transfer coe7cient in Table 3, which in-
cludes results for both grid resolutions. The di>erences in
the table are small, justifying the solution presented being
almost independent of the computational grid.
• For the calculations in Section 3.2, the use of a low
Reynolds number turbulence model requires a large num-
ber of grid points normal to the surface and, furthermore,
the centre of the computational cells closest to the surface
should be at a non-dimensional distance of y+ ¡ 1. 5 For
the calculations presented y+ was below one everywhere,
with an average value of 0.25. As described in Section
2.2, the near-surface grid comprised 20 layers of extruded
prism-cells, which was deemed to be su7cient. Thus, the
basic requirement of the turbulence model was fulAlled.
4. Conclusions
This study presented the detailed geometry of the surface
of a seated female, identical to a thermal manikin that is often
used for laboratory experiments. The geometry was used in
CFD calculations that were compared to measurements of
the heat transfer and !ow around the manikin. The study is
unique in two ways: (1) the high degree of detail for the
manikin used in the calculations; and (2) the fact that we
have access to identical detailed geometries numerically and
experimentally.
The radiation heat transfer coe7cient was predicted and
compared to published data [12] for various body parts,
as well as to whole-body data [21]. The agreement with
the whole-body data is good, and di>ering posture probably
explains the variations for the individual body parts.
Measurements of the natural convection !ow around the
experimental, thermal manikin were compared to CFD cal-
culations of an identical set-up and posture. Comparing mea-
sured and calculated heat !uxes for the individual body parts,
the agreement was found to be excellent and, furthermore,
PIV measurements of the velocity above the head of the
manikin were well predicted.
5 y + = u Ty =, where Ty is the distance normal to the surface, 
 1 1
the kinematic viscosity, and u = (w =)1=2 is the friction velocity,  is
the density and w the wall shear stress.
761
The surface geometry is available through the internet and
we hope that other researchers will take advantage of this
opportunity to use identical geometries for di>erent studies.
Acknowledgements
This work was supported by the Danish Technical Re-
search Council (STVF) as part of the research programme
of the International Centre for Indoor Environment and
Energy.
References
[1] Brohus H, Nielsen PV. CFD models of persons evaluated by
full-scale wind channel experiments. In: Proceedings of Roomvent
96, Yokohama, Japan, vol. 2, 1996. p. 137– 44.
[2] Brohus H. Personal exposure to contaminant sources in ventilated
rooms. PhD thesis, Aalborg University, Denmark, 1997.
[3] Hyun S, Kleinstreuer C. Numerical simulation of mixed convection
heat and mass transfer in a human inhalation test chamber.
International Journal of Heat and Mass Transfer 2001;44(12):
2247–60.
[4] Murakami S, Kato S, Zeng J. Development of a computational
thermal manikin—CFD analysis of thermal environment around
human body. In: Proceedings Tsinghua-HVAC-95, Beijing, 1995.
p. 349 –54.
[5] Murakami S, Kato S, Zeng J. Flow and temperature Aelds around
human body with various room air distribution, part 1—CFD
study on computational thermal manikin. ASHRAE Transactions
1997;103(1):3–15.
[6] Murakami S, Zeng J, Hayashi T. CFD analysis of wind environment
around a human body. Journal of Wind Engineering and Industrial
Aerodynamics 1999;83:393–408.
[7] Murakami S, Kato S, Zeng J. Combined simulation of air!ow,
radiation and moisture transport for heat release from a human body.
Building and Environment 2000;35:489–500.
[8] Voigt LK. Navier–Stokes simulations of air!ow in rooms and
around a human body. PhD thesis, Technical University of Denmark,
Denmark, 2001.
[9] Topp C, Nielsen PV, SHrensen DN. Application of computer
simulated persons in indoor environmental modeling. ASHRAE
Transactions 2002;108(2):1084–9.
[10] Topp C. In!uence of geometry of a computer simulated
person on contaminant distribution and personal exposure. In:
ROOMVENT2002, Copenhagen, Denmark, 2002. p. 265 –8.
[11] SHrensen DN. Radiation between segments of the seated human body.
In: ROOMVENT2002, Copenhagen, Denmark, 2002, p. 317–20.
[12] de Dear RJ, Arens E, Hui Z, Oguro M. Convective and radiative
heat transfer coe7cients for individual human body segments.
International Journal of Biometeorology 1997;40(3):141–56.
[13] Melikov A, Zhou H. Comparison of methods for determining
equivalent temperature under well-designed conditions. In: Proceed-
ings of the Sixth International Conference FLORENCE ATA,
Firenze, Italy, 1999.
[14] Tanabe S, Arens EA, Bauman FS, Zhang H, Madsen TL. Evaluating
thermal environments by using a thermal manikin with controlled
skin surface temperature. ASHRAE Transactions 1994;100(1):
39–48.
[15] Melikov A, Cermak R, Mayer M. Personalized ventilation:
evaluation of di>erent air terminal devices. Energy and Buildings
2002;34(8):829–36.
762 D.N. S-rensen, L.K. Voigt / Building and Environment 38 (2003) 753 – 762
[16] Trans P. Private communication. Regarding construction of thermal
manikin, 2002.
[17] STAR-CD. Methodology, STAR-CD Version 3.15. Computational
Dynamics Limited, 2001.
[18] Lien FS, Chen WL, Leschziner MA. Low-Reynolds-number
eddy-viscosity modelling based on non-linear stress–strain/vorticity
relations. In: Proceedings Third Symposium on Engineering
Turbulence Modelling and Measurements, Crete, Greece,
1996.
[19] van Leer B. Towards the ultimate conservative di>erence scheme,
V: a second-order sequel to Godunov’s method. Journal of Comput-
ational Physics 1979;32:101–36.
[20] ASHRAE. Thermal comfort. In: ASHRAE handbook of
fundamentals. Atlanta: ASHRAE, 2001 [chapter 8].
[21] Fanger PO. Thermal comfort. Copenhagen: Danish Technical Press,
1970.
[22] SHrensen DN, Nielsen PV. Quality control of computational !uid
dynamics in indoor environments. Indoor Air 2003;13(1):2–17.