You are on page 1of 8

MODELLING MIXED CONVECTION HEAT TRANSFER

AT INTERNAL BUILDING SURFACES

Ian Beausoleil-Morrison1
1
CANMET Energy Technology Centre, Natural Resources Canada, Ottawa Canada
PhD Student, Energy Simulation Research Unit, University of Strathclyde, Glasgow U.K.
(ibeausol@nrcan.gc.ca)

ABSTRACT the interactions between thermal and visual per-


The treatment of convective heat transfer at internal formance (e.g. occupants closing blinds in
building surfaces has a significant impact on the response to glare).
simulation of heat and air flow. Accurate In addition to widening the scope of BSim and inte-
approaches for the range of flow regimes experi- grating modelling methods, advanced models have
enced within buildings (buoyant flow adjacent to been developed for some of the important heat trans-
walls, buoyant plumes rising from radiators, fan- fer mechanisms:
driven flows, etc.) are required, as is the ability to • Conduction transfer function and finite-difference
select an appropriate method for the case at hand techniques are well-developed and widely utilized
and to adapt modelling to changes in the flow. for modelling transient heat transfer through the
A new approach—drawing upon previously pub- building fabric.
lished methods—has been developed for modelling • Ray-tracing approaches for view-factor calcula-
mixed convection within mechanically ventilated tion in conjunction with radiosity models are
rooms. It is applicable for rooms ventilated with often used for inter-surface radiation exchange.
ceiling mounted diffusers and is appropriate for
both heating and cooling. ESP-r simulations per- • Detailed ground-contact models have been cre-
formed with the mixed flow model indicate that the ated to consider the impact of time-varying
prediction of heating and cooling loads is highly ground temperatures and the transient heat storage
sensitive to the treatment of surface convection and of the surrounding soil.
that significant errors can result if an inappropriate BSim will continue to evolve towards more inte-
model is employed. The results also reveal that the grated and more highly resolved modelling
choice of convection algorithm can influence design approaches, driven by the need to address the com-
decisions drawn from a simulation-based analysis. plex nature of real-world design and analysis prob-
lems. One area where further refinement is neces-
INTRODUCTION sary is the treatment of convective heat transfer at
Building simulation (BSim) has evolved consider- internal building surfaces (e.g. walls, windows).
ably from its origins over three decades ago. The This is the topic of the current paper and a key ele-
early BSim tools were strictly thermal models, used ment in a research effort aimed at advancing the
to estimate building energy consumption and peak integrated modelling of heat and air flow in build-
heating and cooling loads. Although these early ings.
tools considered the thermal impact of user-pre- This paper outlines a number of issues pertaining to
scribed infiltration and ventilation rates, the flow of the modelling of internal surface convection, placing
air was not simulated, nor was the interdependency the significance of this heat flow path in context. A
of heat and air flow considered. new procedure for modelling mixed-flow convection
The scope of BSim has widened considerably in is then described. Following this, a series of ESP-r
recent years to consider physical processes which (ESRU 1997) simulation results are presented to
are not strictly thermal. As well, there has been a demonstrate the application of the new approach and
trend towards integrated modelling: to illustrate the sensitivity of BSim results to the
treatment of internal convection.
• The simulation of building loads and plant equip-
ment has been integrated. IMPORTANCE OF CONVECTION
• Macro-scale air flow models have been incorpo- Treatment of Surface Convection in BSim
rated to couple the simulation of heat and air flow A number of BSim programs treat surface convec-
and to analyze pollutant dispersion within build- tion as an explicit heat flow path, although many
ings. combine convection with inter-surface radiation,
• Illumination models have been coupled to enable modelling the two processes with a "film" coeffi-
the assessment of visual comfort and to consider cient and solving for some fictitious "operational"

-1-
temperature. techniques—highly sensitive to the boundary condi-
For those programs that do treat internal surface tions supplied (assumed) by the user (e.g. Awbi
convection explicitly, most employ the so-called 1998).
well-stirred assumption. This treats the room air as The application of boundary conditions in BSim is
uniform and characterizes surface convection by a relatively straightforward. The model boundary is
convection coefficient (h c ) and the temperature dif- (typically) placed at the exterior of the building fab-
ference between the room air (T air ) and the solid ric: boundary conditions can be established in the
surface (T surface , also assumed to be of uniform tem- form of exterior conditions—dry-bulb temperature,
perature): wind velocity, etc.—drawn from an appropriate
weather-data file. However in modelling room air
q′′conv = h c ⋅ (T air − T surface ) (1)
flow with CFD, the model boundary is located
where q′′conv is the convective heat flux from the air within the building: the user must supply boundary
to the surface. conditions in the form of internal wall conditions
(surface temperatures or heat flow) and air flows
Some BSim programs employ time-invariant h c val- entering/leaving the room. The fundamental
ues (either user-prescribed or "hard-wired" in the dilemma is clear. A room does not exist in isolation:
program’s source code), although many recalculate wall temperatures and air flows through openings
h c for each surface each time step, using some user- are dynamic and dependent on the external weather
selected or fixed equation. Many h c equations are in conditions, states prevailing throughout the rest of
use, some are particular to buildings and are appro- the building, and the operation of plant equipment,
priate for specific flow regimes, while others are these in turn depending on conditions within the
general relations from the heat transfer literature. room.
Relevance to Thermal Modelling CFD researchers have begun to address this issue by
Given these simplified and varied approaches it is integrating dynamic fabric models and inter-surface
not surprising that the recently completed IEA radiation models into CFD codes (e.g. Holmes et al
BESTEST project (Judkoff and Neymark 1990). This allows room air flow to be calculated by
1995)—the objective of which was to systematically prescribing boundary conditions external to the
test and diagnose sources of disagreement between building or in adjoining spaces, rather than within
BSim programs—identified the modelling of surface the room.
convection to be one of the primary causes of dis-
agreement between programs. Integration of CFD into BSim
Negrão (see Clarke et al, 1995) extended this con-
Numerous researchers have examined the sensitivity cept by integrating a CFD code into ESP-r, the two
of BSim thermal predictions to the modelling of models operating in tandem, "handshaking" on a
internal convection (e.g. Waters 1980; Irving 1982; time-step basis. A thermal and (optionally) a net-
Spitler et al 1991; Clarke 1991; Fisher and Pedersen work air flow representation of the whole building
1997). Their work has demonstrated that predic- and plant is established in the BSim program while a
tions of energy demand and consumption can be CFD model is created for a single room. BSim
strongly influenced by the choice (made by program establishes the boundary conditions for the CFD
developer or user) of h c algorithm. Differences of model. Once the CFD solution converges, it passes
20-40% in energy predictions were noted by some thermal or air flow results to the BSim model, which
of these authors. uses the data to calculate the surface temperatures,
More importantly, the predicted benefits from design energy flows, and air flows throughout the building.
measures were, in some cases, found to be sensitive This process is repeated each time step. The reader
to the approach used to model internal surface con- is referred to Clarke and Beausoleil-Morrison (1997)
vection. As a result, the choice of h c algorithm for an overview of the "handshaking" mechanisms.
could affect the design decisions drawn from a sim- The power of the BSim-CFD integrated modelling
ulation-based analysis. These observations alone approach is clear. BSim has the potential to supply
provide ample motivation to improve the modelling realistic time-varying boundary conditions for CFD,
of internal surface convection within BSim. while CFD has the potential to predict the details of
Relevance to Air Flow Modelling flow and temperature fields within particular zones,
The significance of internal convection modelling is thus enabling flow visualisation, studies on pollutant
not limited to thermal simulations, however, but is dispersion, and thermal comfort assessments.
also of relevance in modelling indoor air motion. Success of the BSim-CFD approach, however, is
Computational Fluid Dynamics (CFD) has been critically dependent upon the treatment of the
widely and successfully applied in the prediction of physics at the model boundaries, the locations at
room air motion (e.g. Jones and Whittle 1992). which the BSim and CFD systems interact. Conse-
However, accuracy is—as with all modelling quently, any errors in the modelling of surface

-2-
convection will be propagated (perhaps amplified): implement an "adaptive" algorithm to allow convec-
if BSim supplies inaccurate boundary conditions for tion calculations to be responsive to local flow con-
CFD, CFD will calculate an incorrect temperature ditions. The BSim program would possess a suite of
and flow field for the zone; the erroneous results methods for calculating h c , each one appropriate for
passed from CFD to BSim will lead to errors in sur- a specific flow regime (e.g. buoyancy driven flow
face temperatures and energy flows throughout the over walls, forced flow at ceilings). The program
building, causing errors in the boundary conditions would select the appropriate approach for each inter-
supplied to CFD for the next time step. Clearly, an nal surface, on a time step basis, based upon the
accurate treatment of internal surface convection is configuration and the prevailing operational states.
critical to the BSim-CFD integrated approach. Such an approach is currently under development
Treatment of Surface Convection in BSim-CFD within ESP-r (Beausoleil-Morrison and Strachan
There are two basic options for modelling surface 1999).
convection in the BSim-CFD simulator: The ESP-r adaptive approach draws upon numerous
1) Have CFD calculate the air-to-surface heat trans- h c algorithms reported in the literature. However,
fer based on the CFD-predicted flow and temper- an important flow regime—mixed convection in
ature fields. which both mechanical (fan) and buoyant forces are
important—is not adequately addressed by existing
2) Have BSim calculate the heat transfer using h c algorithms. Consequently a new method has
empirical relations with surface-averaged and been developed, the subject of the next section.
zone-averaged temperatures.
The first option is the more general and desirable MIXED CONVECTION MODEL
approach: the calculations can respond to local flow
patterns and local—rather than surface-averaged— Numerous algorithms exist for establishing h c .
heat transfer can be predicted. Some are general in nature while the applicability of
others is restricted to specific building geometries
But, the prediction of surface convection remains and plant systems. Most are simple in form, often
problematic for CFD, principally because of the regressions of empirical data which give h c as a
nature of turbulence in room air flow and the related function of air and surface temperatures for a single
treatment of near-wall regions. The standard k − ε flow regime.
turbulence model with log-law wall functions
remains (by far) the most commonly employed The two methods that form the basis of the new
approach in the CFD modelling of room air flow, mixed convection model are reviewed in this sec-
although it has been well demonstrated that this can tion. As space does not permit treatment of other
lead to significant errors in surface convection pre- approaches, the reader is referred to Beausoleil-
dictions (Chen and Jiang 1992; Yuan et al 1994; Morrison and Strachan (1999) for descriptions of
Awbi 1998). other h c algorithms and further references.
Alternate turbulence models have been assessed and Alamdari and Hammond Correlations
improved results have been observed, but only in Alamdari and Hammond (1983) presented correla-
some cases, and usually at the expense of higher tions for buoyancy-driven convective heat transfer
compute requirements and/or stability (e.g. Chen for use in BSim programs. Correlations which cover
1995). laminar, transitional, and turbulent flow regimes for
the following three configurations are given:
Research is underway to develop new methods to
accurately resolve the wall heat transfer (e.g. Barp • Vertical surfaces.
and Moser 1998, Xu et al 1998). Indeed, one new • Stably-stratified horizontal surfaces (e.g. warm air
set of wall functions has been developed (Yuan et al above a cool floor).
1994), although their applicability is limited to • Buoyant flow from horizontal surfaces (e.g. cool
buoyancy-driven flow over vertical surfaces. These air above a warm floor).
research efforts may one day result in robust and
general methods to enable CFD to resolve wall heat Rather than conducting new experiments, they drew
transfer. However, until this time the second option upon data reported in the literature to develop their
outlined above remains the way forward: that is, for correlations, which are cast in a continuous form
BSim to calculate surface convection using empiri- suitable for implementation into BSim programs.
cal relations with surface-averaged and zone-aver- The relation for vertical surfaces, for example, is
aged temperatures. given by,
1/6
Need for Improved Convection Modelling  
1/4 6
The above discussion highlights the importance of   ∆T    1/3 
6
h c = 1. 5 ⋅ + 1. 23∆T
H     
(2)
improving the modelling of internal surface convec-  
tion. The way forward, it is proposed, is to  

-3-
where ∆T is the air-surface temperature difference In the case of the isothermal ceiling jet—the most
and H is the height of the vertical surface. applicable configuration in the context of BSim—
The correlations cover the full range of ∆T and Fisher found, interestingly, that the surface convec-
dimensions relevant to building applications. How- tion was independent of the inlet velocity of the ceil-
ever, there are limitations to their applicability: ing jet, but rather depended upon the jet’s volumetric
flow rate. He also found the buoyancy forces of the
• The correlations were generated from data cold jet to be negligible. The form of the
derived from experiments on isolated or free sur- correlations1, expressed in dimensionless parame-
faces, whereas air flow within rooms more closely ters, reflect these observations. The relation for
approximates flow within an enclosure. walls, for example, is given by,
• The correlations are not applicable for mechani-
Nuwalls = − 24. 8 + 0. 36 ⋅ Re0.8 (4)
cally driven jets as experienced in actively venti- e

lated buildings, but rather are restricted to flow


The Nusselt number is defined as Nu = h c V room 1/3
/k
regimes dominated by buoyant forces.
and the enclosure Reynolds number as
• The correlations are only applicable for configura- Ree = V̇ in /ν V room
1/3
, where V room is the room volume,
tions in which buoyancy is a result of temperature k is the thermal conductivity of air, V̇ in is the volu-
differences between the room air and room sur- metric air flow of the jet, and ν the kinematic viscos-
faces. They are not applicable, for instance, for ity of air.
the flow regime generated by a warm plume rising
These correlations were derived for the range of
from a radiator.
ventilation temperatures relevant for cooling
Fisher Correlations (10o C ≤ T in ≤ 25o C) and for a very large range of
Fisher (1995) performed experiments within a ventilation rates (3 ≤ ac/h ≤ 100; the data of Spitler
mechanically ventilated room-sized enclosure to et al 1991, acquired in the same experimental facil-
develop correlations for internal surface convection. ity, were used for the higher flow rates).
The experiments spanned a range of air flows and
Buoyant forces caused by temperature differences
ventilation-air temperatures.
between internal surfaces and the room air were
For the majority of the experiments the internal sur- very small in the isothermal ceiling-jet experiments.
faces were held at the same temperature, the so- Sixteen combinations of ventilation rate and temper-
called isothermal room. In one group of experi- ature were assessed. In all cases, the internal sur-
ments a single wall was chilled in order to examine faces were controlled to 30°C. The corresponding
the combined impact of buoyant forces against the mean room-air temperatures (not reported) can be
wall and mechanical effects (the non-isothermal estimated by performing a heat balance on the room:
room). Convection correlations for three classes of the average surface-air ∆T over the 16 experiments
flow were developed: was 2.4°C; the greatest difference was 3.8°C.
• Isothermal rooms with ceiling jets emanating Although the non-isothermal experiments examined
from radial ceiling diffusers. the combined impact of buoyant forces against the
• Non-isothermal rooms with ceiling jets emanating wall and mechanical effects, the range of tempera-
from radial ceiling diffusers. tures examined was narrow and the combination of
temperatures (T coldwall < T in < T hotwalls ) atypical.
• Isothermal rooms with free horizontal jets ema-
nating from wall air supplies. Fisher’s results represent a significant contribution
to the modelling of internal convection in BSim.
The room’s interior surfaces were covered by 53
However, as with all approaches, there are limiting
panels, each an independent resistance-heater. Heat
factors:
input to each panel was controlled to maintain the
desired surface temperature. Surface convection • The isothermal correlations are strictly applicable
was derived from these measurements by evaluating when the flow regime is dominated by a mechani-
surface energy balances for each of the 53 panels. cally driven jet, buoyancy caused by surface-air
By maintaining all surfaces at the same temperature temperature differences being negligible.
(in the isothermal cases), radiation exchange was • The non-isothermal correlations are not generally
minimized, thus reducing uncertainty in deriving the applicable to mixed flow, wherein buoyant forces
surface convection. adjacent to some surfaces are important (e.g. a
To minimize uncertainty the results were correlated window exposed to the outside) and may assist or
with the ventilation-air temperature (T in ), rather than oppose the mechanical forces.
the room-air temperature (T air ) as in Equation 1:
1
Fisher and Pedersen (1997) present an alternate
q′′conv
hc = (3) regression (in dimensional form) of the same experi-
(T surf − T in ) mental data.

-4-
• The room was cooled by the supply air in all buoyant or mechanically driven? This question can
experiments: forced-air heating systems were not be answered (qualitatively) by examining the surface
examined2. convection predicted by the two approaches. Alam-
• All experiments were carried out in a single room dari and Hammond (Equation 2) predicts surface
of constant dimensions, so the influence of room convection to be ∼8W /m 2 ; Fisher (Equation 4) pre-
aspect ratio is unknown. dicts ∼7W /m 2 . As the predictions are of the same
order, both buoyant and mechanical forces are con-
Mixed Flow: A Common Flow Regime
sidered significant. Since the forces are assisting,
Two algorithms that BSim programs can use to cal-
the surface convection should be higher than either
culate h c on a time-step basis have been described.
method alone predicts. In this case, flow is mixed
One (Alamdari and Hammond) is for purely buoyant
and the forces are assisting.
flow (buoyancy caused by ∆T between room air and
internal surfaces). The other (Fisher isothermal) is But during morning start-up (ie. recovery from night
for purely mechanically driven jets. However, in setback), the temperature of the supply air is much
mechanically ventilated rooms both forces will, in warmer (∼40°C) while the wall-air ∆T is lower
general, be present, and both can be significant. In (∼3°C). In this situation, Alamdari and Hammond
some cases the mechanical and buoyant forces will gives ∼6W /m 2 whereas Fisher gives ∼15W /m 2 . So
assist (act in same direction) while in others they in this situation the mechanical forces are dominant.
will oppose (act in opposite directions) or act trans- And on a sunny and relatively warm (-3°C) day the
versely (act in perpendicular directions). Neither the heating system supplies air just above the room air
Alamdari and Hammond nor the Fisher approach temperature and the wall-air ∆T is only ∼2°C. In
can fully characterize the convective regime in these this situation, Alamdari and Hammond gives
mixed flow cases. ∼3. 5W /m 2 whereas Fisher gives ∼1W /m 2 . In this
In addition, it is difficult (usually impossible) to pre- situation buoyant forces are dominant.
determine whether a configuration will be domi- The examination of other mechanically ventilated
nated by buoyant forces or mechanical forces. This building configurations in different climates would,
is best illustrated by example3. A well-insulated of course, lead to different observations. Although in
office with large glazing area is heated by a con- general it would be seen that for some thermal and
stant-volume forced-air system delivering 6 ac/h operational states, buoyant forces are dominant,
through ceiling-mounted diffusers. On a relatively while for others, mechanical forces are dominant,
cold day (-20°C) a supply air temperature of ∼30°C while yet for others both are important.
is adequate to heat the office, the heating load being
New Mixed Flow Correlation
offset by solar gains and gains from lights, occu-
A new model that blends the Alamdari and Ham-
pants, and office equipment. The internal surface of
mond and Fisher algorithms is proposed for the gen-
a wall facing the outdoors is about 4°C colder than
eral case of mixed flow. By considering the buoyant
the averaged room air temperature.
and forced correlations as asymptotic solutions, the
The warm jet emanating from the diffuser spreads following 3rd order sum gives h c when the buoyant
across the ceiling towards the walls (it adheres to the and mechanical forces are assisting or acting trans-
ceiling rather than dropping due to viscous and versely (modelled after Churchill and Usagi’s, 1972,
buoyant forces). The jet cools as it flows down the general expression for correlating rates of heat trans-
outside-facing wall. Velocity is relatively low by fer):
this point due to the jet’s spread. As the surface of 1/3
the wall is colder than the surrounding room air, air h c,mixed,assisting = (h c,Fisher )3 + (h c, A&H )3  (5)
 
adjacent to the wall contracts (becomes more dense)
and sinks due to gravity: buoyancy assists the When the buoyant and mechanical forces are oppos-
mechanically driven jet, both effects forcing flow ing h c is taken as the greater of the two predictions:
down the wall.
h c,mixed,opposed = max  h c,Fisher , h c, A&H  (6)
Both mechanical and buoyant forces drive the air  
down the wall, but is the flow predominantly
This model reproduces identically the Alamdari and
2
As buoyancy of the cold jet did not influence sur- Hammond result when forced effects are unimpor-
face convection, a heating system using the same type tant and reproduces identically the Fisher result
of diffuser should in fact generate a substantially simi- when buoyant effects are insignificant. When both
lar flow field. Therefore, it is felt that Fisher’s correla-
tions are equally applicable to room heating when
effects are important and are assisting, Equation 5
there are negligible surface-air temperature differ- results in a greater h c than either method alone.
ences. Equation 6 ensures that h c will not be lower than
3
The temperature data cited in the example were either method predicts.
acquired from an ESP-r simulation.

-5-
The mixed flow model has been incorporated into Canadian construction.
ESP-r, adding to the program’s existing convection Each zone is conditioned with a constant-volume
capabilities. Equation 5 is applied for all floors and forced-air mechanical system whose supply-air tem-
ceilings, because on these surfaces buoyant forces perature varies from 13°C to 43°C in response to
always act in a transverse direction to the jet result- loads. During occupied hours (5h00 to 20h00 week-
ing from radial ceiling diffusers. For walls, a test is days) the system delivers 60 L/s of outdoor air to
performed each time step to determine whether the each zone. The building is heated to 22°C, with an
wall-air ∆T results in a buoyant force that assists or 18°C setback during unoccupied hours. The cooling
opposes the mechanically driven jet, and corre- setpoint is 24°C while the building is allowed to free
spondingly a decision made on whether to apply float during unoccupied periods in the summer. At 6
Equation 5 or 6. ac/h, the system is sized to meeting the peak heating
A simulation with the mixed flow model was per- load but is undersized for cooling.
formed on the office previously described for the Impact of Model on Load Predictions
month of March using Ottawa weather data. Figure Three annual simulations—identical except for the
1 plots q′′mixed , q′′Fisher , and q′′A&H for the outside fac- treatment of internal convection—were performed.
ing wall. The surface convection heat flux is plotted The Alamdari and Hammond correlations were used
against the surface-air ∆T to best illustrate the in the first simulation (ESP-r’s default approach),
impact of the mixed flow model. Fisher’s correlations applied in the second, while the
mixed flow model was utilized in the third4. The
Fisher
Alamdari and Hammond
annual heating and cooling loads (normalized by
15.0 mixed flow model floor area) are given in the following table.

annual loads
h c algorithm heating cooling
convection at wall (W/m )
2

10.0
B
(MJ/m 2 ) (MJ/m 2 )
Alamdari & Hammond 271 207
Fisher 265 243
Mixed Flow Model 295 247
A
5.0
The mixed flow model predicts significantly higher
heating loads than either Alamdari and Hammond
(9% higher) or Fisher (11% higher). It also predicts
substantially higher cooling loads than Alamdari and
0.0
0 1 2 3 o
4 5 6 Hammond (19% higher) but only slightly more
| Twall − Troom air | ( C)
(<2%) than Fisher.
Figure 1: Constant Volume Heating
Impact on Thermal Comfort Predictions
The Alamdari and Hammond model correlates well
Clearly, the choice of h c algorithm has a significant
to ∆T , as expected (see Equation 2). However, the
impact on the prediction of annual heating and cool-
Fisher correlation does not, as it responds to T in , a
ing loads (and thus energy consumption). Another
function of the room’s heating load. At higher ∆T ,
(perhaps more) significant implication of algorithm
Alamdari and Hammond tends to dominate, but at
choice can be seen by examining Figure 2, which
lower ∆T , where buoyant forces are small, q′′mixed
plots the air temperature in the west zone on July 5,
approaches q′′Fisher (see region A in Figure 1). For a
a day with very high cooling loads.
large number of data points q′′mixed is greater than
both q′′Fisher and q′′A&H , indicating that buoyant and The system was not sized to meet the peak cooling
forced effects are often both important (see B in Fig- loads, a valid design decision in a climate with a
ure 1). short cooling season; significant capital cost savings
can be realized by sizing equipment to maintain the
setpoint temperature through the majority of the
APPLICATION OF NEW MODEL cooling season, but allowing temperatures to rise on
Description of Test Case the most severe days. In such a case a designer
A two-zone (150 m 2 floor area per zone) ESP-r might use BSim to assess whether thermal comfort
model representing one storey of a shallow floor- will be unduly compromised by the undersizing.
plate office building was created. The building,
4
located in Ottawa, has a north-south alignment and ESP-r’s adaptive convection algorithm does not
apply the Fisher or mixed flow equations when the
is moderately glazed (35% of external wall area), all forced-air system is inoperative; rather it switches to
windows facing east or west. The fabric assemblies, Alamdari and Hammond to more closely approximate
insulation levels, and internal gains are typical of the convective regime.

-6-
40 Alamdari &
Mixed Model
design Hammond
35
peak
option cooling load cooling load
temperature in west zone ( C) 30
(MJ/m 2 ) (MJ/m 2 )
base design 51.8 58.5
o

25
9 ac/h 52.5 (+1%) 61.6 (+5%)
20
10°C SAT 52.4 (+1%) 60.8 (+4%)
occupied period
VAV 52.5 (+1%) 63.1 (+8%)
15 overhangs 41.2 (-20%) 48.7 (-17%)
night purge 42.8 (-17%) 51.3 (-12%)
10
Alamdari and Hammond

5
mixed flow model The mixed flow model responds to changes in the
flow regime (air change rates and supply air temper-
0
0 4 8 12 16 20 24
ature); higher cooling loads are predicted for the first
time (hours) three design options, a result of the increased cool-
Figure 2: Mixed Flow Model ing capacity and increased surface convection. In
In this case, the ESP-r results indicate that although contrast, the Alamdari and Hammond approach is
the setpoint temperature is maintained most of the not capable of responding to these changes in the
time, there are a few problematic days. On July 5, flow regime. Consequently, cooling load predictions
for example, the system is unable to maintain the are only slightly higher (∼1%) and due entirely to
setpoint temperature, particularly in the afternoon the fact that the cooling system was able to extract
when solar gains reach their peak. The temperature more energy because of its higher capacity.
of the zone drifts upwards, reaching a maximum just In contrast, the Alamdari and Hammond correlations
after 16h00 (refer to Figure 2). When the Alamdari predicted greater savings with overhangs and night
and Hammond correlations are used, a peak zone purging. These measures reduced cooling loads
temperature of 27.9°C is predicted: this might be substantially with both convection methods, but the
acceptable to the designer. However, when the lower h c produced in the Alamdari and Hammond
mixed flow model is employed, a peak zone temper- runs overpredicted the savings.
ature of 31.2°C is predicted; this would be deemed
unacceptable, leading the designer to alter the archi- CONCLUSIONS
tectural and/or mechanical features of the building.
More advanced and refined methods are required for
Impact on Assessment of Design Options modelling internal surface convection within BSim.
The designer might explore a number of options to The way forward, it is believed, is for BSim pro-
address the overheating problem, including: grams to adapt convection calculations to local flow
• Increasing cooling capacity of system by 50% by conditions, an approach that requires BSim pro-
increasing flow rate from 6 to 9 ac/h. grams to be populated with h c methods appropriate
• Increasing cooling capacity ∼25% by lowering for various flow regimes. To this end, a new
minimum supply air temperature from 13°C to approach has been developed for calculating h c for
10°C. mixed convection within mechanically ventilated
rooms, building upon Alamdari and Hammond’s
• Changing system to VAV with a constant supply
(1983) work on buoyancy-driven flow and Fisher’s
temperature of 13°C, a minimum flow of 6 ac/h,
(1995) work on forced flow. The new method is
and a maximum flow of 9 ac/h, effectively
applicable for rooms ventilated with ceiling
increasing cooling capacity by 50%.
mounted diffusers and is appropriate for both heat-
• Reducing solar gains by adding window over- ing and cooling. It is suitable when the convective
hangs. regime is dominated by a mechanically driven jet,
• Pre-cooling the building by night purging with 6 when it is dominated by buoyant forces resulting
ac/h with 100% outdoor air. from surface-air temperature differences, and when
both effects are important.
Each of these design options was simulated twice:
first with the Alamdari and Hammond correlations In a series of ESP-r simulations of a mechanically
and then with the mixed flow model. All measures ventilated office building, the mixed flow model pre-
reduced the peak zone temperatures, with varying dicted significantly higher heating loads (9% higher
degrees of success, and all had an influence on cool- than Alamdari and Hammond; 11% higher than
ing loads, as shown in the following table (loads for Fisher), indicating the importance of both buoyant
month of July). The numbers in parentheses indi- and forced effects. Therefore, simulating configura-
cate the difference relative to the base design. tions like this with a convection approach that

-7-
considers only buoyant or only forced effects would 1121-1128.
significantly underpredict heating loads and heating Clarke J.A. (1991), Internal Convective Heat Trans-
energy consumption. The mixed flow model fer Coefficients: A Sensitivity Study, Report to
resulted in substantially higher cooling loads than ETSU, University of Strathclyde, Glasgow U.K.
Alamdari and Hammond (19% higher) but only Clarke J.A., Dempster W.M. and Negrão C. (1995),
slightly more (<2%) than Fisher, indicating that the ‘The Implementation of a Computational Fluid
convection regime is dominated by mechanical Dynamics Algorithm within the ESP-r System’,
effects when the system is cooling (due to low sur- Proc. Building Simulation ’95, IBPSA, 166-175.
face-air ∆T ). Therefore, cooling loads in cases like Clarke J.A. and Beausoleil-Morrison I. (1997), ‘The
this could be accurately predicted with a convection Integration of Computational Fluid Dynamics into
approach that considers only forced effects, although Building Simulation’, IBPSA News, 9 (2).
an approach that considers only buoyancy effects ESRU (1997), The ESP-r System for Building
would lead to significant errors. Energy Simulations: User Guide Version 9 Series,
Additionally, an improper choice of h c algorithm ESRU Manual U96/1, University of Strathclyde,
could lead to inappropriate design decisions. Simu- Glasgow UK.
lations of the office using the Alamdari and Ham- Fisher D.E. (1995), An Experimental Investigation
mond approach indicated that peak temperatures of Mixed Convection Heat Transfer in a Rectangu-
were borderline acceptable on the most severe cool- lar Enclosure, PhD Thesis, University of Illinois,
ing days; whereas, the mixed flow model clearly Urbana USA.
showed that comfort conditions could not be main- Fisher D.E. and Pedersen C.O. (1997), ‘Convective
tained. Simulations of a number of design measures Heat Transfer in Building Energy and Thermal
aimed at mitigating the overheating problem illus- Load Calculations’, ASHRAE Transactions, 103
trated that the mixed flow model can respond to (2) 137-148.
changes in the flow regime resulting from mechani- Irving S.J. (1982), ‘Energy Program Validation:
cal system alterations, but the buoyancy-only Conclusions of IEA Annex 1’, Computer Aided
approach cannot. The impacts of the five design Design, 14 (1) 33-38.
measures assessed were found to be sensitive to the Jones P.J. and Whittle G.E. (1992), ‘Computational
convection method utilized, demonstrating that the Fluid Dynamics for Building Air Flow Predic-
choice of h c algorithm can influence design deci- tion--Current Status and Capabilities’, Building
sions drawn from a simulation-based analysis. and Environment, 27 (3) 321-338.
Judkoff R. and Neymark J. (1995), International
REFERENCES Energy Agency Building Energy Simulation Test
Alamdari F. and Hammond G.P. (1983), ‘Improved (BESTEST) and Diagnostic Method, IEA Energy
Data Correlations for Buoyancy-Driven Convec- Conservation in Buildings and Community Sys-
tion in Rooms’, Building Services Engineering tems Programme Annex 21 Subtask C and IEA
Research and Technology, 4 (3) 106-112. Solar Heating and Cooling Programme Task 12
Awbi H.B. (1998), ‘Calculation of Convective Heat Subtask B.
Transfer Coefficients of Room Surfaces for Natu- Holmes M.J., Lam J.K-W, Ruddick K.G., and Whit-
ral Convection’, Energy and Buildings, 28 tle G.E. (1990), ‘Computation of Conduction,
219-227. Convection, and Radiation in the Perimeter Zone
Barp S. and Moser A. (1998), ‘Heat Transfer Under of an Office Space’, Proc. ROOMVENT ’90, Oslo
Thermally Stratified Conditions at Horizontal Sur- Norway.
faces in Buildings’, Proc. ROOMVENT ’98, (2) Spitler J.D., Pedersen C.O., and Fisher D.E. (1991),
103-110, Stockholm Sweden. ‘Interior Convective Heat Transfer in Buildings
Beausoleil-Morrison I. and Strachan P. (1999), ‘On with Large Ventilative Flow Rates’, ASHRAE
the Significance of Modelling Internal Surface Transactions, 97 505-515.
Convection in Dynamic Whole-Building Simula- Waters J.R. (1980), ‘The Experimental Verification
tion Programs’, accepted for ASHRAE Transac- of a Computerised Thermal Model for Buildings’,
tions. Building Services Engineering Research & Tech-
Chen Q. and Jiang Z. (1992), ‘Significant Questions nology, (1) 76-82.
in Predicting Room Air Motion’, ASHRAE Trans- Xu W., Chen Q., and Nieuwstadt F.T.M. (1998), ‘A
actions, 98 (1) 929-939. New One-Equation Turbulence Model for Near-
Chen Q (1995), ‘Comparison of Different k − ε wall Natural Convection’, Int. J. Heat Mass Trans-
Models for Indoor Air Flow Computations’, fer, 41 3161-3176.
Numerical Heat Transfer, B(28) 353-369. Yuan X., Moser A., and Suter P. (1994), ‘Wall Func-
Churchill S.W. and Usagi R. (1972), ‘A General tions for Numerical Simulation of Turbulent Natu-
Expression for the Correlation of Rates of Trans- ral Convection’, Proc. 10th International Heat
fer and Other Phenomena’, AIChE Journal, 18 (6) Transfer Conf., Brighton UK, 191-196.

-8-

You might also like