Applied Thermal Engineering 24 (2004) 1291–1300

www.elsevier.com/locate/apthermeng

A novel design for panel radiators

S.M.B. Beck *, S.C. Grinsted, S.G. Blakey, K. Worden
Department of Mechanical Engineering, The University of Sheffield, Mappin Street, Sheffield S1 3JD, UK
Received 12 September 2003; accepted 14 November 2003

Abstract
Previous work by the Sheffield group Building Services Eng. Res. Technol., 22 (3) (2001) 185 has shown
the importance of radiation heat transfer in the output from the panel radiators used for heating in most
buildings. The work described here continues this work by considering the enhancement of the heat transfer
from these emitters by the use of either one or two high emissivity sheets placed between the interior
surfaces of a double radiator. This is in contrast to conventional radiators which use fins to increase the
surface area. These fins add to the expense of fabricating the radiator and are very susceptible to fouling
from dust which reduces their heat output considerably.

2004 Elsevier Ltd. All rights reserved.

Keywords: Domestic radiators; Heat transfer; Radiation; Free convection

1. Introduction

Radiators are the most popular central heating emitters in the UK. Of the various designs
available, the steel panel radiator, usually equipped with convection fins to improve their heat
output, are common in domestic, business and industrial environments. In this type of device,
hot water is passed through the hollow radiator. As the radiator is hotter than the air sur-
rounding it, an amount of heat is transferred to the air and thus the water exits at a lower
temperature.
The amount of heat transfer varies primarily with the water temperature and the size and
design of the radiator. The first law of thermodynamics states that in steady state heat flow, all of
the energy put into a system must come out again, so those which are described as more efficient
cannot actually use less energy to provide the same amount of heat.

*
Corresponding author. Tel.: +44-114-222-7730; fax: +44-114-222-7890.
E-mail address: [email protected] (S.M.B. Beck).

1359-4311/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2003.11.026
1292 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300

Various solutions have been implemented that reduce the large amount of wall area taken up
by panel radiators. The ducted air systems used in the recent past and skirting board radiators
reduce the space lost to heating systems. Reducing the size of panel radiators would open up
better possibilities to the building engineer and architect as less wall space would be lost to these
devices. It must not be forgotten that both better insulation and the superior control of ventilation
have already reduced the requirements for heating in recent years.
Although radiators are known as radiators, most of their heat output is by natural convec-
tion. The steady flow energy equation for the air flow around a radiator states that the
rate of conductive heat transfer from the radiator is equal to the product of the flow rate of the
air, the specific heat capacity of air and the change in temperature of the air across the radi-
ator.
The rest of the heat transfer is due to thermal radiation. This is based on the fourth power of
the absolute temperature, and the emissivity of the surface, which is a measure of how closely it
approximates to a blackbody, which has an emissivity of 1.
By way of an example a 0.7 by 1.4 m single radiator has a heat output of 1800 W and a surface
area of just under 1 m2 per side [2]. The radiation heat transfer into the room from the panel
facing it is about 400 W. Therefore just under 1/4 of the heat transfer is due to radiation. If
account is taken of the side facing the wall [1], the proportion of heat transfer due to radiation
rises to 45%. For a double radiator, the convective heat transfer doubles, but the radiative heat
transfer remains the same.
The thermal performance of radiators is measured in accordance with EN 442 part 2 [3], which
specifies a standard test room subject to certain test conditions. Part of the standard states that the
emissivity of the paint used in the room is to be greater than 0.9. The actual output of an emitter
will, however, vary based on the installation and operating condition [4]. Peach [5] recommends
that comparison be made based on Ôspecific product heatÕ, the heat emitted per unit mass of
material. However, as stated previously by the authors, a heat output per wall area may actually
be a better measure for the customer [1].
Several aspects of radiator design affect their output. Some are based on their position and
operating points,

(a) The output of radiators can be slightly increased by decreasing their height above the ground
and by increasing their spacing from the wall [6].
(b) Decreasing the water flow rate through a radiator can lower the heat output [7,8].
(c) The attachment of fins to panel radiators increases the convection heat transfer.
(d) Different connection positions can affect the performance. The most common installation
being with both connectors at the bottom (BOE). However introducing the flow at the top
(TBOE) can improve the temperature distribution within the radiator and is used in the stan-
dard.
Others are based on the radiation from the radiator.
(e) Facing the wall adjacent to the radiator with insulated reflector can lower the heat loss
through the wall by 70% [9]. This will however lower the heat output from the radiator [1]
as the heated wall acts as another convecting surface.
(f) It is well known that fouling can dramatically lower the heat output from radiators that rely
on convection.
 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300 1293

(g) The use of metallic paint finishes can reduce the radiant component of radiator heat outputs
by up to 10% [5].

This last fact shows how important radiation can be in the total heat output from panel
radiators. It was decided to investigate whether it was possible to increase the total heat transfer
from a radiator by using high emissivity sheets placed in the gap between two flat panel radiators.
These sheets should heat up by radiation and then transfer this heat to the air. This configuration
should be cheaper to make and easier to clean than conventional finned radiators.

2. Experiment

Two radiator setups were connected in series as shown in Fig. 1. It is appreciated that this
results in lower input temperatures for the second radiator. All of the radiators used in this study
were 600 mm high by 600 mm wide. The radiators were positioned at a height of 150 mm above
the floor in accordance with EN 442-2 [3]. Two single radiators were attached to the circuit using
flexible connections so that the distance between them could be varied. The water was heated in a
domestic hot water tank by an Immersion heater. The heater was connected to the mains via a
Variac, which allowed the power supplied to the boiler to be directly controlled. The boiler was
connected via the cistern to a three-speed pump, using a flow rate control loop. This ensured that
a constant flow rate was maintained throughout the system. This is shown in Fig. 1.
The flow rate was measured using a turbine flow meter connected to a computer. The pump was
used on its lowest setting which, in conjunction with the flow rate control pipe, limited the flow
rate to 1 l min
1 . This ensured a sensible temperature drop of around 5 C across each radiator.
The water temperature was measured using K-Type thermocouples attached to the surface of
the copper piping. As the outside of the pipe work was insulated, the piping was assumed to be at
the same temperature as the water. The thermocouples were attached to a computer for recording
data. The five main thermocouples used in the testing are also shown in Fig. 1.
2 Single panel Radiators T2 Double radiator with fins
T1

Radiator 1 Radiator 2
T3 T5
T4
Valve Flow Rate
control pipe
Tank with
immersion
heater
Flowmeter
Pump

Cistern

Fig. 1. Schematic of apparatus.

1294 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300

Insulation Insulation Insulation

High High
Emissivity Emissivity
Sheet Sheets

Radiator panels Radiator panels Radiator panels

2 separated panels Single sheet (mod1) Two sheets (mod 2)

Fig. 2. Radiator configurations.

T1 measured the temperature input into radiator 1.

T2 and T3 measured the temperature at the outlet of both panels. This also provided a means of
balancing the flow rate between the two single panels.
T4 measured both the overall output temperature of radiator 1 and also the input temperature for
radiator 2.
T5 measured the outlet temperature of radiator 2.
Two thermocouples were used to monitor other temperatures.
T6 was used to record the ambient temperature.
T7 was used to monitor the temperature of the black sheet when it was in place.

As only the heat transfer between the inside surfaces of the radiators was of interest in these
investigations, the outside surfaces of the panels were insulated to a thickness of 50 mm with
expanding foam insulation with a thermal conductivity of 0.04 W m
1 K
1 . This gave a heat loss
through the rear of the radiators of approximately 0.04 kW m
2 .
The three different configurations shown in Fig. 2, were tested using different spacing of panels
and sheets. The power output from each was recorded for a variety of operating conditions. The
experiments were run over several months. This allowed the effect of fouling on the heat transfer
from the conventional radiator with internal fins to be measured.

3. CFD modelling

In order to obtain a further insight into the air flow around the radiator configuration, CFD
simulations were carried out, using Gambit [10] and Fluent 5.1. Because the practical investiga-
tion contained no convection velocity modelling, the fluent model was created in an attempt to
analyse the flow between the radiator panels on their own and with the convective sheets in place,
assessing both the temperatures and velocities involved.
A 2-D model of a 4 · 3 m room as specified by the british standard (EN442-2) was created using
a non-uniform Gambit grid. The grid was created with this varying grid spacing in order to
 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300 1295

concentrate the main calculation on the areas of interest. The grid density is much higher around
the radiator itself, and also around the wall areas. This placed practical limitations on the grid and
was another reason why a three dimensional problem was not modelled. The grid contained
56,124 cells, 112,981 faces and 58,856 nodes.
The spacing difference results in a grid that is six times more concentrated around the radiator
sections and around the wall. This was done for two reasons

1. As the convection currents are expected to lead to high velocities around the wall sections, the
grid was also more concentrated there.
2. The boundary layers needed to be modelled reasonably well, so that a fairly small grid size was
required near the walls. Due to time considerations, the standard wall function was used even
though it is appreciated that the two layer model would provide substantially better results [1].

Efforts were made to use reasonable expansion ratios. Three grids were built to match the three
different modifications tested practically. An overall separation distance of the two panels of 9 cm
was decided on as a sensible compromise between power and practicality. To further mimic the
experiments, the sheets were spaced equidistance between the panels.
There were no flow boundaries as the flow was driven by free convection, heated at the radi-
ator, and cooled around the room as in [1]. The boundary conditions used were

• Heated surfaces: walls at 70 C.

• Room walls: at 20 C in accordance with EN 442 part 2 [3].
• Sheets between radiator panels: walls at 50 C. This temperature was measured experimentally.

The k–e turbulence model was used for all of the modelling. Previous work [1] had showed that
the two layer wall model would give better results, but due to time constraints, the standard wall
function was used throughout.

4. Neural network analysis

In order to make the results easier to interpret and to reduce the uncertainty and scatter from
the experiments, it was decided to use what is in effect a complicated, multi variant method of
curve fitting, the neural network.
The neural network paradigm used for this study was the multi-layer perceptron (MLP). For
the sake of completeness, a brief description of the MLP is given here; for a more detailed dis-
cussion, the reader is referred to Bishop [11].
The networks used for this study were designed and trained using the package MLP [12]. The
MLP is simply a collection of connected processing elements called nodes or neurons arranged
together in layers. A set of signal values pass into the input layer nodes, progress forward through
the network hidden layers and the result finally emerges through the output layer. Every node is
connected to each node in the preceding layer through a connection of a given weight and also
to nodes in the following layers. Signals pass through the node as follows: a weighted sum is
1296 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300

performed at each node of all the signals from the preceding layer, giving the excitation of the
node. This is then passed through a non-linear activation junction to emerge as the output of the
node to the next layer. Various choices for the function are possible; the hyperbolic tangent
function f ðxÞ ¼ tanhðxÞ was used here.
The first stage of using a network to model an input–output system is to establish the
appropriate values for the connection weights. This is the training or learning phase. The type of
training adopted here is a form of supervised learning and makes use of a set of network inputs for
which the desired network outputs are known. At each training step, a set of inputs are passed
forward through the network, yielding trial outputs which are then compared to the desired
outputs. If the comparison error is considered small enough, the weights are not adjusted. If,
however, a significant error is obtained, the error is passed backwards through the net and a
training algorithm uses the error to adjust the connection weights. The algorithm used in this work
is the backpropagation algorithm.
Once the comparison error is reduced to an acceptable level over the whole training set, the
training phase ends and the network is established.
Once the network has been trained, it can then be tested with another dataset. Finally, a set of
equations can be produced from the weights so that interpolation between results can be pro-
duced.
For this work there was only a single output, the heat output in Watts. The inputs were:

1. The configuration (Fig. 2) which was given a number based on the configuration (finned, no,
one or two sheets).
2. The input temperature (in C).
3. The gap (in cm).

The height and width of the radiator could also be used as an input parameter, but all of the
radiators were 600 mm square.

5. Results and discussion

The power and input temperatures are shown for all three configurations and a series of dif-
ferent spacing in Fig. 3. For easy comparison, the heat output with an inlet temperature of 70 C
is shown in Table 1.
One effect to note is that there are two sets of results for the double radiator, labelled min and
max. These refer to the reference radiator at the beginning and at the end of the experiment. They
show the effect of fouling on this radiator. It can be seen that the heat output of the radiator
dropped by 20% in the four months that the experiment was running. Cleaning these fins is a
difficult task and the panels proposed here will be far easier to keep dust free than these fins.
It will also be seen that the heat output for the different configurations vary very little for
different panel spacing. This indicates that in this flow regime the heat transfer is independent of
the gap. The only exception to this is the solid grey circles which is a 5 cm gap with a single sheet
in the middle where the heat transfer is up to 10% lower than that for a larger gap.
 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300 1297

1000
900
3cm

Power output (Watts)

800 5cm
7cm
700 9cm
11cm
600
5cm mod1
500 7cm mod1
9cm mod1
400 11cm mod1
5cm mod2
300
7cm mod2
200 9cm mod2
11cm mod2
100 Double radiator - Min
Double radiator - Max
0
55 60 65 70 75 80
Input Temperature (° C)

Fig. 3. Input temperature and output heat for various configurations.

Table 1
Radiator output at an input temperature of 70 C
Configuration Heat output (W) Compared to double Compared to double Compared to empty
max (%) min (%) gap (%)
Double: max 950 100 125 226
Double: min 760 80 100 181
Two panels 670 71 88 160
One panel 580 61 76 138
Empty gap 420 44 55 100

It will be seen from Table 1 that, compared to an empty gap, a single sheet raises the heat
output by almost 40% and that two sheets raise the output by 60%. This is because the sheets heat
up and act as additional convecting surfaces which increase the heat output from the radiator. The
reason for this can be clearly seen from the results of the CFD modelling, shown in Fig. 4, which
shows the flow between the heated panels.
Two effects will be seen here which explain why the heat output is increased

1. With two sheets, there are three times as many heated surfaces than with no sheets, Even if each
surface only adds a small amount of heat to the air, their sheer number will add to the heat
output.
2. The peak velocities near the original panels are increased by the buoyancy effect of the added
panels. This increased velocity will decrease the thickness of the boundary layer on the heated
panel and will thus increase the heat transfer to the air.

This implies that if the fins fitted to conventional radiators were offset relative to each other,
then the radiation would heat up the opposing fins which would increase the heat output of the
radiator somewhat.
Shown in Fig. 5 are the results from the neural network analysis of the data. It will be seen that
the scatter has now been removed from the data. It will also be seen that some of the results,
1298 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300

Fig. 4. Velocity profiles with no, one and two sheets in the gap.

1000
900 3cm
5cm
Power output (Watts)

800
7cm
700 9cm
11cm
600 5cm mod1
500 7cm mod1
9cm mod1
400 11cm mod1
5cm mod2
300 7cm mod2
200 9cm mod2
11cm mod2
100 Double radiator - Min
Double radiator - Max
0
55 60 65 70 75 80
Input Temperature (°C)

Fig. 5. Heat output predicted from the neural network.

(particularly the double-max) have a decreasing gradient, this is due to the fact that firstly the
perceptron has no physics in it and that secondly the random selection of a number to describe the
geometry is not the best way to input this to the neural network.
 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300 1299

However, now that the neural network has been taught, it can be used to extract more data
about the system as well being a powerful interpolation tool.

6. Conclusions

It will be seen that two separated sheets of a high emissivity placed between the panels can
produce between 71% and 88% of the heat output of a finned double radiator. This is because they
heat up by radiation and then transfer this heat to the air by free convection. In effect, the surface
area that can transfer heat to the air is increased.
The finned radiator itself showed a 20% reduction in heat output between the beginning and
end of the experiments. The alternate design of radiator with sheets will be less susceptible to this
fouling than the finned ones, but will be far easier to clean. The new radiator design will be easier
and cheaper to make than conventional finned ones, though the heat output from finned radiators
could be increased by offsetting the fins relative to each other to make better use of the radiation.
The neural network analysis of this type of problem is still in its infancy, and the inputs have to
be optimised to enable the results to be more reliable.
Future work includes improving the CFD model to include radiation and a better wall model
and to use a variety of different sized radiators to allow the neural network analysis to be more
general.

Acknowledgements

The authors would like to thank Mr. M. Nettleship for his help with the experimental work.
They also acknowledge Fluent Europe for the use of their software.

References

[1] S.B.M. Beck, S.G. Blakey, M.C. Chung, The effect of wall emissivity on radiator heat output, Building Services
Engineering Research and Technology 22 (3) (2001) 185–194.
[2] http://www.myson.co.uk/Products01.html. If this is unavailable, any radiator manufacturerÕs catalogue will supply
this type of data.
[3] EN 442, Radiators and Convectors, Part 2. BSI, 1997.
[4] M. Guillaume, M. Gengoux, LÕemission des radiateurs et des convecteurs, Batiment International/Building
Research and Practice 16 (6) (1983) 366–370.
[5] S. Peach, Radiators and other convectors, Journal of Institute of Heating Ventilation Engineers 39 (2) (1972) 239–
253.
[6] K.J. Colthorpe, The performance of radiators and convectors, The Heating and Ventilating Research Council,
technical note no. 3, 1959.
[7] K. Waters, R.H. Fine, The performance of radiators and convectors using medium temperature hot water, The
Heating and Ventilating Research Association, laboratory report no. 59, 1969.
[8] I.C. Ward, Domestic radiators: performance at lower mass flow rates and lower temperature differentials than
those specified in standard performance tests, Building Services Engineering Research and Technology 12 (3) (1991)
87–94.
1300 S.M.B. Beck et al. / Applied Thermal Engineering 24 (2004) 1291–1300

[9] D.J. Harris, Use of metallic foils as radiation barriers to reduce heat losses from buildings, Applied Energy 52
(1995) 331–339.
[10] Fluent Incorporated, GAMBIT: Modelling Guide, Fluent Incorporated, Lebanon, NH, USA, 1999.
[11] C.M. Bishop, Neural Networks for Pattern Recognition, Cambridge University Press, 1998.
[12] K. Worden, MLP 3.4––A Users Manual (obtainable via k.worden@sheffield.ac.uk), 1996.