357

Two-dimensional convection from heated wires

at low Reynolds numbers
By D. C . COLLIS AND M. J. WILLIAMS
Aeronautical Research Laboratories. Australian Defence Scientific Service

(Received 16 July 195s and in revised form 21 March 1959)

Measurements of heat transfer from circular wires placed normal to a horizontal

airstream have been made in the Reynolds number range 0.01 to 140. The
Nusselt number can be related to the Reynolds number and temperature loading
by an expression of the form

where the values of n, A and B (see table 3) depend on whether the Reynolds
number is above or below the value for which a vortex street exists in the wake
of the wire. This value of the Reynolds number (R = 44) is independent of the
intensity and scale of the stream turbulence. The theoretical heat transfer
relation based on the Oseen approximation is approached asymptotically as
R --f 0, provided free convection is negligible.
Free convection effects diminish rapidly with increasing Reynolds number so
that the orientation of the wire with respect to the vertical has a negligible in-
fluence on heat transfer except at very low velocities. For horizontal wires at
verylow Reynolds numbers, free convection is significant,when, roughly speaking,
the Reynolds number is less than the cube root of the Grashof number.

1. Introduction
Heat transfer from circular cylinders by forced convection has been widely
studied because of its engineering importance. This paper is concerned only
with the range of variables encompassed in instruments employing heated
cylindrical wires as sensing elements (Corrsin 1949). The most important
application of hot-wire instruments is probably in fluid velocity and turbulence
measurements. Unfortunately, the many advantages of the hot-wire anemo-
meter in such work are somewhat offset by large uncertainties in the absolute
magnitude of measured quantities. This deficiency is in part attributable to the
use of inaccurate heat-transfer relations. The experimental investigation
described below has yielded a new general relation governing heat-transfer by
two-dimensional forced convection in air which should assist in improving the
accuracy of hot-wire measurements.
Dimensional analysis of the equations for convection of heat by an incom-
pressible fluid (Goldstein 1938) shows that the dimensionless heat-transfer
 358 D.C. Collis and M.J. Williams
coefficient, or Nusselt number N , is a function of the flow parameters and fluid
properties as follows:
= f (R,G, g), (1)
where R is the Reynolds number, G the Grashof number, and cr is the Prandtl
number. This analysis assumes that the fluid properties are independent of
temperature and that it is a continuous medium. The former assumption is
rarely valid, and it is customary to take account of this by introducing a para-
meter representing the temperature loading of the heated body, e.g. TWIT, or
T,/T,, where T denotes temperature and the suffixes W , 00, m respectively
signify conditions at the body aurface, in the free stream, and the arithmetic
mean of these. I n the experiments reported here, some of the heated wires were
sufficiently small in diameter for the effects of the molecular nature of air to be
experienced. Molecular effects take the form of changed boundary conditions
as compared with continuum flow-there is a jump in temperature between the
surface of the wire and the gas adjacent to it, and the fluid slips or moves over
the surface with a finite velocity. I n rarefied gas dynamics the ratio of mean free
path to cylinder-diameter, i.e. the Knudsen number K , is recognized as the
parameter characterizing molecular effects. Thus, for the complete specification
of convective heat transfer from fine hot wires, a relation of the following type
is required:
N = f(R, G, fl, K , TmIT,). (2)
There is still insufficient knowledge of the subject to permit the formulation of
such a relation, and in any case it would be too unwieldy for practical purposes.
I n practice, free and forced convection are treated separately. As a result of
recent investigations (e.g. Collis & Williams 1954), free convection from wires in
air can be specified in the form
N = f(G,TmIT,) (3)
for Grashof numbers as small as It has been further reported (Beckers
et al. 1956) that free convection at Grashof numbers less than about unity is
independent of Prandtl number, so that equation (3) is valid for other fluids in
this case. Forced convection from hot wires in air has long been specified by
simple relations of the following type, or by its equivalent in dimensional
variables :
N = f(R,T,/T,). (4)
The results of the present measurements are also given in this form. The variation
of Prandtl number of air with temperature is small and can be combined with
other temperature loading effects. Explicit introduction of the parameter K
has been dispensed with, because molecular effects are small enough to be treated
as a correction to the continuum heat transfer coefficient. This correction,
described in Appendix A, is made on the assumption that temperature jump is
solely responsible for the observed reduction in the Nusselt number. Rough
estimates suggest that the effect of velocity slip would lie between and of the
jump effect in the present experiments.
At sufficiently low Reynolds numbers, forced and free convection interact in
a manner which has not yet been fully investigated. Ower & Johansen (1931)
and Cooper & Linton (see Ower 1949) have described this interaction, essentially
 Two-dimensional convection from heated wires 359
in a qualitative manner. I n the particular case of horizontal wires in a hori-
zontal air stream, sufficient data was obtained in the present series of experiments
to permit the formulation of a criterion for the onset of significant free convection
as the Reynolds number is decreased from values where forced convection pre-
dominates. Further, it has been clearly shown that at higher Reynolds numbers
beyond this small region of interaction, the effect of free convection is quite
negligible. This observation shows that the method of hot-wire anemometer
calibration which involves the measurement of heat transfer at zero Reynolds
number is erroneously based.
Since the primary aim of this work was to establish precise heat transfer laws
for forced convection at low Reynolds numbers the existing information on this
subject will now be examined in more detail.

2. Review of data on forced convection at low Reynolds numbers

The results of the more important forced convection investigations have been
collected and correlated by McAdams (1954). The scatter of results is rather large,
but a correlation between N and R, satisfactory for many engineering purposes,
has been achieved for temperature loadings up to 1000°C by evaluating the
thermal conductivity k and the dynamic viscosity ,u of the gas at a temperature
halfway between the cylinder temperature and the ambient or free-stream tem-
perature, i.e. +(Tw+ Tm),and the density p at free stream temperature. I n the
Reynolds number range 0.1 to 250,000, McAdams gives co-ordinates of a recom-
mended mean curve, whilst for the smaller range 0.1 to 1000, he gives the equation
N = 0*32+0 ~ 4 3 R @ ~ ~ . (5)
Mean equations such as this smooth out the finer details, and therefore are
too crude to form the basis for accurate measuring devices, particularly those
which, like the turbulence hot wire, depend on the values of aN/aR at the
operating point. It is probable that a single careful investigation, covering
perhaps a more restricted range of the variables, would yield more satisfactory
heat-transfer relations.
Of the various sources quoted by McAdams, only two contribute extensive
data at Reynolds numbers in the range of operation of hot-wire instruments
(viz. R < 100)) namely King (1914) and Hilpert (1933). Hot-wire anemometry
is almost invariably based on an empirical relation obtained by King. If a wire
of diameter d , placed normal to an airstream of velocity V , and temperature T,,
is heated to a temperature T, then the heat transfer coefficient H W cm-l deg-1
as given by King's law is
H = A[1+ y(Tw - T m ) ] + B[1 + 6(T, - T m ) ] 1/( V d ), (6)
where A = 2.50 x l O - 4 ( 1 + 3 5 4 with d in cm, B = 1.012 x 10-2,,/d, and the tem-
perature coefficients when T, = 17 "C are y = 0.00114, 6 = 0*00008.
King's experiments were performed with a whirling arm and consequently
were subject to interference from draughts, both natural and induced. However,
more recent investigations in wind tunnels or air jets of low turbulence level
have been found to confirm King's law, but these investigations have generally
360 D.C. Collis an,d M . J . Williams
covered a very limited range of the variables, and it is doubtful whether the
results have been examined sufficiently critically. It will be shown later, in fact,
that King’s results involved some fairly large systematic errors, and that
because of this the results can be interpreted as being consistent with the proposed
new heat transfer relation which differs somewhat from equation (6).
The work of Hilpert is exceptional in that it extends over a very wide Reynolds
number range and exhibits a high degree of consistency throughout. Hilpert
expressed his results in the form of a power law

N = .[.(%):I”. (7)

where G and n are constant in specified ranges of R. The method of correlation

in this equation differs from that of McAdams in that the gas density as well as
the conductivity and viscosity are evaluated at the mean temperature,t
4(Tw+ Tm). Table 1 contains values of C and for the ranges of interest here.

R C 1E
1-4 0.891 0.330
4-40 0.821 0.385
40-4000 0.615 0.466
TABLE1

There are two reasons which possibly account for the fact that the results
have not been applied extensively in instrument work. First, the temperature
factor was based on a series of measurements in which the Reynolds number
varied only through a small range. It was therefore not shown that this factor
retains the same form in other ranges of Reynolds number, although some
support was derived by correlating some of King’s data at lower Reynolds
numbers. Secondly, the series of simple power laws is probably only an approxi-
mate representation of the data, which could lead to appreciable errors in
deriving the local slope of the heat-transfer curve. Although the experimental
results were published in tabular form, no alternative analysis seems to have
been attempted, and the discontinuities in the slope of the heat-transfer curve
which Hilpert noted at R = 4 and R = 40, have been generally ignored in
instrument work. The second of these irregularities obviously could arise from
the sudden change in nature of the flow, which occurs when vortices begin to
detach from the rear of the cylinder. The former, which was less well defined,
might be due to the formation of standing vortices-a more gradual process.
A theoretical solution to the forced convection problem for Reynolds numbers
in the range of 0.1 to 100 is extremely difficult, since neither the slow viscous flow
approach on the one hand, nor the use of boundary layer approximations on the
other, is allowable for most of the range. The one well-known solution for forced
t Allowance for this difference has not been made by McAdams (1954) in incorporating
Hilpert’s results in McAdams’ figure 10-7. The required adjustment amounts to an in-
crease of 133% in Reynolds number. A rather smaller adjustment to McAdams’ recom-
mended curve seems indicated.
 Two-dimensional convection f r o m heated wires 361
convection from wires, that due to King, avoids the di%culty by assuming
potential flow. King’s solution is somewhat tedious to evaluate, but can be
approximated closely by two equations of simpler form. Written in non-
dimensional terms these are?
2
N = ___ ( R c < 0.08),
log (2el-YalRa)
where y E = 0-577 ... is Euler’s constant, and

The second equation is identical in form with King’s experimental law, except
that i t does not include the effect of temperature on the physical properties of
the gas. Comparison with experimental data shows that it overestimates the
heat transfer by up to 40 %. Apart from this, the basic assumptions of the theory
have been held to be unsatisfactory (Goldstein 1938). The theory thus is of no
help in assessing the relative merits of experimental results.
A useful piece of theory (although of limited scope) is that based on the Oseen
approximation. This is of course strictly valid only in the limit as R --f 0. Cole
& Roshko ( 1 954) find that this approximation yields the solution
2 8
- = log --YE.
N Ra
Experimental work by the same authors failed to substantiate equation (10) due,
it was thought, to disturbance of the two-dimensional convection by three-
dimensional effects.
The investigation which is now described was provoked by observations that
hot-wire anemometer calibrations made under near-ideal conditions tended to
deviate systematically, and always in the same manner, from King’s law
(equation (6)). The nature of these deviations was such as to indicate a somewhat
different dependence of heat-transfer on velocity from that of equation (6).
A modification of King’s law is proposed in this paper following careful measure-
ments involving a wide range of air velocities, wire diameters and temperature
loading. The existence of at least one discontinuity in the heat transfer relation
as previously observed by Hilpert has been verified. The dependence on tem-
perature loading was found to agree in magnitude with Hilpert’s finding rather
than with that of King.

3. Description of measurements
3.1. Equipment
Heat transfer measurements in the speed range 6-140ft./sec were made with
electrically heated wires in a small closed-return wind-tunnel (test-section area
104 x 133in.) of conventional design. An excellent distribution of velocity and
a low turbulence level were obtained by using a single wire gauze screen in the
t The specific heat at constant volume has been replaced by that a t constant pressure,
which seems t o be more appropriate in this problem.
362 D . C. Collis and M . J . Williams
settling chamber followed by a 4 : 1 contraction of area. The longitudinal com-
ponent of turbulence was found to be 0.08 at 50ft.lsec. Air speed was deter-
mined from the pressure drop across the contraction, which was calibrated by
means of a Pitot-static tube placed at the position later occupied by the heated
wires. The temperature of the air in the tunnel was slightly sensitive to changes in
atmospheric conditions, and also changed several degrees during runs at the
higher speeds due to viscous dissipation. As heat-transfer measurements were
made for temperature loadings as small as 30 "C, a check on changes in ambient
temperature was thus essential. This was made by means of a rapid response,
platinum thermometer placed near to, but clear of, the wake of the heated wire.
From time to time an ambient temperature reading was obtained from the test
wire itself. Comparison of such readings with those of the monitoring thermo-
meter served as a check on changes in the test wire arising from strain or other
causes. Ambient temperature varied between 10 and 25°C during the experi-
ments.
For measurements a t lower air speeds, namely, from 0.08 to 1*42ft./sec,a
cylindrical duct fitted at the intake end with a bell mouth, honeycomb and gauze
was used. Air speed was measured by means of a Simmons' shielded anemo-
meter (Simmons 1949), which was originally calibrated on a whirling arm at the
National Physical Laboratory.? Flow conditions in this duct were much less
steady than in the wind-tunnel, and velocity measurements were less accurate.
Thus, in the experimental results presented later, there is a greater scatter
amongst data taken at speeds less than Gft./sec.

3.2. Wire assemblies

I n some earlier work in this field the effect of finite wire length has been
largely eliminated by attaching potential taps to the heated wire, so that end
losses are not included in the measured heat transfer. This device is not practic-
able for the finer wire used in the present work. The procedure adopted was to
use very long wires and correct the results for the small residual transmission of
heat to the supports by the method given by Simmons & Beavan (1934). The
maximum correction necessary was 3.6 yo.
The heavier gauges of wire were mounted by soldering them between slender
steel prongs as in figure 1(a). Tension was applied subsequently by a screw
adjustment. The finest gauge wires which were prepared from Wollaston wire
could not be mounted under tension without risk of breakage by vibration. An
arrangement in which tension is maintained by air-drag was therefore used. An
assembly of this type is shown in figure 1 ( b ) . The Wollaston wire is soldered in
a slack loop trailing downstream of the prongs. The silver layer is then etched from
the centre of the loop and the platinum is straightened by manipulation of the
remaining silver supports. At the maximum speed at which the finest wire was
used, namely, 100ft./sec, the platinum was deformed into a catenary with the
tangent at either end inclined at about 5' to the normal to the air stream. This
deformation would have a negligible effect on the rate of heat transfer.
t I n using the Simmons' anemometer as a transfer device, no assumption about the
heat transfer law applicable to this instrument is involved.
 Two-dimensional convection from heated wires 363

wire

12 s.w.g. copper wire

$
P.V.C. slewing
Screw adiustmenf
of tension

Approx. half size

(a) 0.00535and 0oo090 cm diameter wires

- 26 s.w.g. spring
steel wire

c a

( b ) 0000295 cm diameter wire

FIGURE
1. Wire support geometry.

3.3. Properties of the heated wires

Wires of nominally pure platinum were employed for all of the measurements.
The dimensions of each wire and the temperature coefficient of electrical resist-
ance of the material are given in table 2. An optical interference method was used
to determine the diameters of the finer specimens of wire whilst a similar tech-
nique combinedwith theuse of slip gaugeswas usedfor the thickest wire. Accuracy
of the two methods was about 2 and 1 yo, respectively. A travelling micro-
scope was used for the length measurements. The temperature coefficients of
resistance were measured over the interval 0-lOO"C, using melting ice and
steam baths. The variation of this coefficient is indicative of impurities in the
platinum.
The length of each specimen was made as great as practicable to minimize
both the metallic conduction heat transfer and other three-dimensional effects.
Free convection measurements (Collis & Williams 1954) have shown that all
end effects (including metallic conduction) are small for 2/d 2 20,000: even at
364 D. C. Collis and M . J . Williams
very small values of the Nusselt number. As the Nusselt number increases the
hot air film diminishes in thickness, and rough calculations based on the Langmuir
concept of a concentric cylindrical film suggested that Zld > 1000 would be
adequate for these experiments. However, the concentric film concept is not
valid for forced convection, so that in the absence of a more reliable guide a
large margin of safety in the aspect ratio was considered desirable. The most
critical condition occurred with the finest gauge wire a t the lowest velocities.
The specimen with Z/d = 5370 was therefore used for all measurements at speeds
less than 6ft./sec. At higher speeds most of the measurements were made with
a shorter, and therefore stronger specimen in order to minimize errors arising
from strain and permanent deformation of the wire.

Diameter d Length 1 Aspect ratio Temp. coeff.

(cm) (cm) Ild of resistance
0.000295 0.872 2950 0.00370
0.000295 1.585 5370 0.00370
0~00090 7.80 8660 0.00380
0.00535 11.10 2070 0.00342
TABLE2

3.4. Measurement of heat-transfer coeficients

The wires under test were heated electrically and the power dissipated in the
wire was computed from measurements of the heating current and wire resist-
ance. The heat lost by convection is equated to the electrical power dissipation,
less losses to the supporting wires. The temperature loading was determined by
using the test wire as an electrical resistance thermometer. Temperatures on the
International Temperature Scale were obtained by first calculating temperatures
on the platinum scale using the measured temperature coefficients and then
applying a correction by the Callendar method (Kaye & Laby 1948),assuming the
difference coefficient 6 = 1.5. Resistance of the wire at ambient temperature
was determined by extrapolating to zero power dissipation. Ambient tem-
perature was measured on a mercury thermometer prior to a run and
corrected for rapid changes by reference to the platinum thermometer referred
to in $3.1.
Radiation losses were computed and found to be negligible throughout.

3.5. Eddy shedding observations

A hot wire anemometer was mounted in the wake of an unheated wire, of
0.00535 cm diameter, and a turbulence amplifier with a useful frequency range
up to about 70 kilocycles per second was used to detect eddies shed from the
upstream wire. Turbulence level and scale were varied by inserting square-mesh
wire grids into the working section upstream of the wires. Determinations of
the Reynolds number at which eddy shedding commenced was made for a range
of turbulence conditions.
 Two-dimensional convection f r o m heated wires 365

4. Results
The results discussed in $3 4.1 to 4.4 refer to horizontal wires in a horizontal
airstream normal to the wire. The effect of orientation is considered in 0 4.5.

4.1. Method of correlation

When formulating empirical relationships it is customary to attempt to
remove explicit dependence of the Nusselt number on temperature loading by
evaluating the physical properties of the fluid at some suitable temperature.
Choosing this temperature on the grounds of physical significance does not
always lead to a unique simple relationship between the variables. Here, the
system used by Hilpert has been followed (except where otherwise stated) as it
leads to a compact expression of the results in the Reynolds number range con-
cerned. Thus, the fluid properties-thermal conductivity, density and viscosity
-are evaluated at a temperature which is the arithmetic mean of the free stream
(or ambient) and cylinder temperatures. This system, it will be recalled, differs
from that used by McAdams, who evaluated density a t the free stream tem-
perature.
Values of the thermal conductivity of air were computed from a formula
given by Kannuluik & Carman (1951) which, for thermal conductivity in
electrical units, is
kt = 2-41 x 1O-y 1 + 0.00317t - 0.0000021t2), (11)
where kl W cm-l deg.-l is the thermal conductivity at temperature t "C. The
ratio of viscosity to density of air, i.e. the kinematic viscosity pip, was obtained
from tables in Goldstein (1938).

4.2. Correlation of forced Convection data

The range of Knudsen number K , involved in these measurements is too small
to attempt an accurate evaluation of the effect of temperature jump on heat
transfer. However, in order to reduce all of the data to comparable conditions
a correction was applied to reduce the measured heat-transfer coefficients to
continuum values. Figure 2 has been drawn to exhibit the magnitude of the
correction necessitated by temperature jump. A low-temperature loading was
chosen for this illustration to avoid complications arising from temperature
dependence of the fluid properties. The derivation of a correction for temperature
jump based on kinetic theory is given in Appendix A. It is assumed that the mea-
sured rate of heat transfer is that which would occur in a continuum under the
temperature difference which exists in the gas outside the region of discontinuity.
The continuum Nusselt number Nc is then related approximately to the apparent
or measured value N, in the following way for low temperature loadings:
1 1 -
2K.
NM NC

For high-temperature loadings, allowance must be made for the temperature

dependence of mean free path and thermal conductivity as shown in Appendix A.
366 D. C . Collis and M . J . Williams
Applying the correction enables all data pertaining solely to forced convection
to be consolidated into a single curve for each temperature loading. This is
demonstrated on a logarithmic scale for a low- and a high-temperature loading
by means of figure 3. The series of points which diverge sharply from the upper
forced convection curve a t low Reynolds numbers are associated with heat
transfer due to mixed free and forced convection. The conditions under which
free convection becomes important are discussed in a later paragraph. The

3
2/R
FIUURE
2. Effect of temperature jump on heat transfer.

evaluation of fluid properties at the arithmetic mean temperature T, clearly does

not entirely eliminate the temperature loading TWITwor T,/Tw as a necessary
parameter. Nevertheless, the residual effect of temperature loading is small and
is essentially independent of Reynolds number. The effect amounts to roughly
7 % increase in Nusselt number for an increase in temperature difference from
30 to 320 "C. Thus, for the range of variables covered in this work, the dimension-
less heat-transfer coefficient, temperature loading and mass flow can be corre-
lated by an equation of the form
 Two-dimensional convection from heated wires 367
where g and f are functions to be determined. A satisfactory form for the tem-
perature function is

It should be noted that this function is somewhat dependent on the values chosen
for the thermal conductivity, and various sources differ somewhat in their values
at elevated temperatures. I n figure 4 the effectiveness of relations (13) and (14)
in condensing the whole body of forced convection data into a single curve is
demonstrated.

log,, R
FIGURE3. Variation of continuum Nusselt number with Reynolds
number showing residual effect of temperature loading.

log,, R
FIGURE
4. Nusselt number uniquely correlated with Reynolds
number and temperature loading.
368 D . G. Collis and M . J. Williams
The next step is to find a convenient analytical approximation to the flow
function f ( R ) . Since it is usually accepted that the data conform to King’s
equation (with modified constants),it will be instructive to make the comparison,
remembering that the investigation was undertaken because of suspected
departures from King’s law. Figure 5 shows clearly that the supposed linear

FIGURE
5. Demonstration of the inadequacy of the heat
transfer relation N = A -tB JR.

relation between N and JR is no more than a rough approximation at Reynolds

numbers less than about 44. Thus, the constants of King’s law as determined by
experiment must be somewhat arbitrary, because of the latitude available in
approximating the curves of figure 5 by straight lines. I n particular the slope
of such lines is very variable, which explains a great deal of the inconsistency
common in turbulence measurements by hot wire. At Reynolds numbers greater
than 44 or thereabouts, eddies are shed from the rear of the cylinder. The change
in the velocity distribution, which occurs abruptly, gives rise to the marked
change of slope of the heat-loss curves which can be seen in figure 5 , a t about
 Two-dimensional convection from heuted wires 369
,/R = 6-6. The function f ( R )thus has some sort of discontinuity at this critical
Reynolds number. By trial and error it was found that the experimental data
were in good agreement with the equation
T -0.17
N(e) =A+BRn

if the constants A , B, n have the values given in table 3 in the Reynolds number
ranges prescribed there. Most of the data shown in figure 4 are replotted in
figure 6, showing how the 0.45 power law is satisfied in the range specified in
table 3. The change following the onset of eddy shedding is clearly defined. At

07
7

06
6

02
3

Buoyancy
2 04
7 4

5
-
hE
t)

03
3
7
6
T T

02
2

01

0
1

8" C 06
6
0s
8
1.o
10
R0'45

FIGURE
6 . Heat transfer data correlated by 0.45 power law
in the range 0.02 < R < 44.
24 Fluid Mech. 6
370 D. C . Collis and M . J . WiEliams
R = 0.02, equation (15), with appropriate constants, overestimates the Nusselt
number by 2 or 3 yoand the error increases rapidly if Ghe same relation is used at
smaller Reynolds numbers.

0*02<R<44 44<R<140
?a 0.45 0.51
A 0-24 0
B 0.66 0.48
TABLE3

4.3. Forced convection at very low Reynolds numbers

The theoretical solution for two-dimensional forced convection from cylinders
obtained by Cole & Roshko (1954) using the Oseen method, is believed to be
correct in the limit as R -+ 0, and should not be greatly in error for R 4 1.
A comparison of the experimental results at low Reynolds numbers, with the
Oseen solution, thus provides a check on the general soundness of the measure-
ments and in particular of their approach to two-dimensionality .This comparison
is made in figure 7 for large and small temperature loadings. The straight (broken)
line represents the Oseen solution (equation (10)).Some data depart noticeably

log,, €2
FIGURE
7. Comparison with Oseen solution.

from the forced convection curve because of buoyancy effects, but the smooth
curve through the bulk of the points asymptotes to the Oseen solution as R -+ 0.
This curve deviates from the theoretical one by about 5 yo at R = 0.01, and
diverges steadily as the Reynolds number increases, as would be expected. The
discrepancy is thus too great for the Oseen solution to have much value as a
working rule for forced convection at low Reynolds numbers. However, a relation
 Two-dimensional convection from heated wires 371
of similar form can be used to describe the experimental results for R < 0.5
with an error less than the experimental scatter:

This equation agrees with equation (15)to within 2 yoin the range 0.02 < R < 0.5,
and is in better agreement with experiment at R < 0.02 than is equation (15).
Equation (16) will almost certainly be a good approximation for two-dimensional
heat transfer a t any lower Reynolds numbers which are attainable under con-
tinuum conditions (or nearly such) and in the absence of significant buoyancy
effects.
4.4. Mixed free and forced convection
As we have noted, certain series of points diverge quite sharply from the
forced convection curve near the lower end of the Reynolds number range. This
divergence occurs because the components of velocity, induced in the heated

Forced convection

1 1 I I
01 02 03 04 5
dRMc

FIGURE
8. Interaction of free and forced convection.
24-2
372 D . C. Collis and M . J . Williams
fluid adjacent to the cylinder by buoyancy forces, become comparable in magni-
tude with components of the forced draught. Figures 3,4,6 and 7 show that the
points of divergence depend markedly on cylinder diameter. By plotting the
very low Reynolds number data on an enlarged scale as in figure 8, a temperature
dependence is also observed. I n this figure the method of correlation has been
modified to the extent that the gas density embodied in the Reynolds number is
evaluated at ambient temperature after the practice of McAdams (1954). This
allows the temperature function of equation (15) to be discarded for the particular

n
~ogmG,
FIGURE
9. Criterion for onset of buoyancy effect.

small range of Reynolds numbers involved-a fact which will prove convenient
presently (Appendix B). The data pertain to one wire diameter only, but the
buoyancy effect sets in at different Reynolds numbers for each temperature.
It is also clearly shown that, as the Reynolds number is further reduced, the
heat-transfer rate passes through a shallow minimum before reaching the free
convection value a t zero Reynolds number. This curious phenomenon was
apparently first observed by Ower & Johansen (1931) and later confirmed by
others. However, except for special cases, the conditions governing the occur-
rence of mixed free and forced convection do not appear to have been quanti-
tatively determined. Some information on this question can now be derived.
Since free convection and forced convection in a particular fluid (a = const.)
depend primarily on the Grashof and Reynolds numbers, respectively, a criterion
for mixed flow involving only these two parameters should exist. This is con-
 Two-dimensional convection from heated wires 373
sistent qualitatively with the observed dependence on diameter and temperature.
As a characteristic point of the mixed flow rbgime, we have chosen the condition
for which the Nusselt number of mixed convection equals that for pure free con-
vection at the same Grashof number. This point has the practical significance
that it defines the lowest Reynolds number at which the hot wire in question can
be used as an anemometer without ambiguity. Furthermore, the trend of the
data with increasing Reynolds numbers indicates that the effect of buoyancy
on the forced convection rapidly becomes negligible. I n figure 9, the values of
Reynolds and Grashof numbers, associated with these points of equal Nusselt
number, have been plotted. Points pertaining to the finest wire could only be
obtained by extrapolation. Giving less weight to these points, the conclusion
was reached that buoyancy effects are small provided?

R, > Gk. (17)

The suffixes denote evaluation of the fluid properties a t ambient temperature,
but the conclusion, being very approximate, is not sensitive to the temperature
dependence of these properties.
Another criterion based on a greater volume of data is obtained in Appendix B,
and is probably more accurate for a limited range of Reynolds number. It is
derived by assuming that there is no deviation from the pure forced convection
relation (equation (16)) due to buoyancy effect until the Nusselt number reaches
that due to free convection alone. The details of this analysis are given in
Appendix B. The criterion derived for Reynolds numbers less than about 0.1
is that
€2, = 1*85#%39

The straight line corresponding to zero temperature loading (Tm/Tm = 1) is

shown in figure 9. The overall trend of the experimental points is followed, but
in general the Reynolds number predicted is too high for a given Grashof number
because of the assumption of no-deviation from the forced convection relation
(16). The groups of points belonging to two of the three wires show additional
dependence on temperature in good agreement with equation (1 8). Apart from
the numerical factor it seems that equation (18) provides a fair criterion for the
onset of buoyancy effects. I n particular, it follows that

where Vmin is the lowest velocity which can be measured without ambiguity by
a hot wire of large aspect ratio. It is interesting to note that the lower limit of
usefulness of the hot wire is very little dependent on diameter. This discussion
applies, of course, only to wires of large aspect ratio, where the heat flow is two-
dimensional. For sufficiently small aspect-ratio wires three-dimensional heat
-f It has recently been pointed out by Dr J. J. Mahony that this conclusion could be
derived by the method of Appendix B if the heat transfer relations used are the Oseen
solution (10) and Mahony’s asymptotic formula for free convection at very low Grashof
numbers (Mahony 1956).
374 D. C. Collis and M . J . Williams
transfer may modify the heat flux a t Reynolds numbers higher than those
defined by equation (18).
Since the preceding discussion is based on that segment of the experimental
results for which errors were greatest, the quantitative aspects should be
regarded as being essentially exploratory in nature, and the results treated with
appropriate reserve.
4.5. Effect of orientation
The orientation with respect to the vertical, of the hot wire and of the plane
of the airstream, is important in free convection and consequently has a bearing
on the interaction of free and forced convection. As already stated, the preceding
discussion ($9 4.1 to 4.4) refers only to horizontal wires in a horizontal stream.
It has been shown (Collis & Williams 1954) that in free convection two-dimen-
sional heat transfer from vertical wires at very low Grashof numbers ( <
is not attained even for aspect ratios as large as 20,000. It is certain therefore
that the aspect ratio would be significant in the case of mixed free and forced
convection from vertical wires, for all practical values of the aspect ratio. The
limited accuracy of the equipment available for very low Reynolds number
observations did not justify a detailed study of this complicated phenomenon.
However, it has been demonstrated previously (Ower 1949; Simmons 1949) that
the region of interaction between buoyancy effects and forced convection extends
over a much smaller velocity range for vertical wires than it does for horizontal
wires. It may therefore be predicted that the forced convection heat-transfer
coefficients for vertical and horizontal wires will not differ sensibly for all
Reynolds numbers substantially greater than the value defined by equation (18).
Some measurements which bear out this conclusion have been published by
Simmons (1949), and further evidence is presented in figure 10 of this paper.
Heat transfer measurements from a single wire of large aspect ratio (l/d = 5400)
were made for both horizontal and vertical orientations. No significant difference
was found within the Reynolds number range (about 0.25 to 4) which was
covered. A systematic discrepancy between the two orientations, rising to
18 yo at the highest Reynolds number is discernible, but is within the limit of
reproducibility imposed by changes in the wire due to strain and annealing. The
effect of orienting the airstream in other than a horizontal plane has not been
examined a t all. It is clear that with a vertical stream the mechanism of the
interaction between free and forced convection will differ from that occurring in
a horizontal stream, since buoyancy forces will be acting in a line parallel to the
free stream instead of normal to it. The ambiguity in Nusselt number is unlikely
to be present in the case of an upward directed stream, but will undoubtedly
exist for a stream directed downwards.

4.6. Effect of turbulence

I n the low-turbulence airstream of the wind-tunnel the onset of periodic vortex
shedding was just detectable at R = 46. A small increase in Reynolds number
(about 5 yo)caused the intensity of the velocity fluctuation to rise sharply to a
level which remained roughly constant as the Reynolds number was further
 Two-dimensional convection from heated wires 375
increased. The kink in the heat transfer curve was estimated to occur a t R = 44,
which agrees well with the start of eddy shedding.
Square-mesh turbulence grids of mesh size +,Q and g i n . were placed in turn
across the stream loin. ahead of the 0.002in. (0.00535 cm) wire. The turbulence
level was thus increased in steps (Batchelor & Townsend 1948) whilst main-
taining the scale of small eddies (the microscale A) about the same. Observations
of the vortex street became steadily more difficult but vortex shedding still

RQ46

FIGURE
10. Effect of orientation on heat transfer.

commenced close to R = 46. The frequency of vortex shedding became increas-

ingly uncertain, due apparently to frequency modulation arising from the
fluctuating stream velocity, and to lateral displacements of the wake as a whole.
However, no significant change in the Strouhal number was observed. Since
the maximum effect of turbulence on the detachment of the vortex layer from
the cylinder might be expected to occur with high intensity and small scale
turbulence, the gin. mesh was installed a distance of 20 mesh lengths upstream.
Again no significant effect was observed.
It is perhaps of some interest to note that the vortex shedding frequency in
these observations varied from 27 to 68kilocycles per second.
376 D. C. Collis and M . J . Williams

5. Further discussion and comparison with earlier work

The experiments were designed to provide information pertaining to two-
dimensional heat transfer. The results show that the effect of aspect ratio is
certainly small, since agreement within the limits of error is obtained between
wires of aspect ratio varying from 2070 to 8660. Again, the agreement with the
Oseen solution for two-dimensional convection at very low Reynolds numbers
(9 4.3) is indicative that three-dimensional effects are small even at the lowest
Reynolds numbers involved, although this situation cannot, of course, continue
to indefinitely small Reynolds numbers.
A comparison with representative results taken from King (1914) and Hilpert
( 1 933) is made in figure 11. The first impression is of haphazard scatter, but closer

FIGURE
11. Comparison with previous forced convection investigations.

study shows that the differences between results are very largely systematic.
Hilpert’s work shows a high degree of self-consistency, and while exhibiting
similar trends to the authors’ curve the mean slope of a curve fitted to Hilpert’s
results would be somewhat less. Thus, while there is good agreement in values of
N in the vicinity of R = 100, Hilpert’s heat transfer is some 8 yogreater at R = 2,
and at high values of R Hilpert’s curve falls below the authors’ extrapolated
curve. The change of slope of the N - R relationship due to establishment of a
vortex street is discernible at about the same point (R+ 40)as the bend in the
authors’ curve. The latter does not exhibit a gradual bend at R = 4 as found by
Hilpert, but the occurrence of this bend was deduced from an insubstantial
number of data. Where the vortex street exists, the plot of Hilpert’s results
 Two-dimensional convection from heated wires 377
shows an irregularity involving at least one point of inflexion in the range
40 < R < 500 which is not reflected in his constants of table 1. Unfortunately,
the authors’ results are too sparse in this range to clarify the matter. The correct
relation for this range therefore remains in some doubt. Hilpert’s data below the
eddy-shedding Reynolds numbers conform well to a law involving the 0.45 power
of the Reynolds number as shown in figure 12. However, the constants differ
from those given in tabIe 3, for equation (15), particularly the constant. A .

3
R0’46

12. Comparison of Hilpert’s data with equation (15); R < 44.

FIGURE

The data taken from King and plotted in figure 11 represent results for the
lightest and heaviest wire gauges used by that author, two moderate temperature
loadings being selected in each case. Near R = 1, heat transfer from the thicker
wire is undoubtedly augmented slightly by the effect of buoyancy (see $4.4).
Apart from this it is remarkable that each series of points (of given diameter and
temperature loading) is spaced a constant distance in the ordinate direction from
each other and from the authors’ curve. These discrepancies are evidently due
to some deficiency in the measurement of the heat-transfer coefficient rather than
of the Reynolds number. Regardless of the precise nature of this deficiency, it
may be concluded that King’s heat-transfer coefficients show the same form of
dependence on the Reynolds number in the range 0.1 < R < 40 as the work
described here.
378 D. C. Collis and M . J . Williams
With regard to the effect of temperature loading, the authors’ results con-
stitute the only extensive body of consistent data available. The development
of a method of correlating the heat-transfer coefficient with temperature loading
was described in 0 4.2 (equations (13), (14)). The method satisfactorily accounts
for the few measurements taken by Hilpert a t temperatures other than 100 “C.
This can be seen from figure 11 where representative points at temperature
loadings up to 1024 “C are included. The effectiveness of equations (13) and (14)
in dealing with King’s results cannot be determined due to the uncertainties in
those results. King’s own analysis showed deviations of the temperature coeffi-
cients y and S of equation (6) of up to 50 % from their mean value. For R < 44,
equation (15) can be reduced to a dimensional form equivalent to King’s equa-
tion (6), viz.
H = A’[l+y(T,-T,)]+B’[l +S(TW-T!)] (Vd)0.45, ( 20)
where A’ and B‘ embody physical properties of the fluid and T, = 290 O R . No
significance attaches to differences between the values of A‘ and B’ and the
constants A and B of King’s equation because of the changed power of Vd in
the second term, but it is of interest to compare the representative values of y
and 6. The authors’ values given in table 4 together with those of King are
calculated assuming the following variation of conductivity and viscosity with
temperature : 0.80

The larger values found here have some significance in practical hot wire anemo-
metry, as discussed by Collis (1956).

Y 6
King 0.00114 0.00008
Collis & Williams 0.00164 0.00025
TABLE4

I n the lower end of the Reynolds number range (R of order l),comparison must
be made with the work of Cole & Roshko (1954). Reference has already been
made to the fact that good agreement has been found with the Oseen solution
(equation (10)) obtained by these authors. Cole & Roshko did not find such
agreement, but their experimental accuracy may have suffered greatly through
use of nominal values of wire diameter and of temperature coefficient of resist-
ance. Again, although very fine wires involving f i u d s e n numbers of up to Q were
used, no account was taken of ‘rarefaction’ effects. According to equation (12)
these might exceed 30 % of the measured heat transfer coefficient. In view of
this, a direct comparison with the authors’ results is not possible.
Some interesting indirect evidence consistent with the 0.45 power of R or V
in equations (15) and (20))respectively, is found in measurements by Newman
(1951) of the sensitivity of hot wire anemometers to yaw. It was originally
accepted that a hot wire responded to the normal component of velocity V sin $,
where $ is the angle of yaw, so that according to equation (6) the heat-transfer
 Two-dimensional convection from heated wires 379
coefficient varied linearly with ,/( V sin +). Newman reached the conclusion,
however, that the wire responded to Vi(sin 9+)0.457. I n the light of equation (20)
and a close examination of Newman’s data, it seems probable that the wire
actually responds to the 0.45 power of the normal component of velocity, i.e.
( V sin +)045.

Conclusions
1. Based on experiments in the Reynolds number range 0.01 to 140, a new
relation for two-dimensional forced convection from cylinders normal to an
airstream has been established. The law has the form
T -0.17
N(E) =A+BP,

where the values of n, A , B depend on whether the Reynolds number is above or

below the value for which a vortex street exists in the wake of the cylinder (see
table 3). Fluid properties k,p and p used in computing N and R are evaluated at
mean film temperature T,, and T, is in the normal range of room temperatures.
2. The extensively used relation of King was based on measurements made
at Reynolds numbers below this critical value. I n that region the new relation
differs from King’s in that the Reynolds number enters to the 0.45 power rather
than the 0.5 power, and the effect of temperature loading is found to be signi-
ficantly larger. These differences, particularly the former, appear to arise mainly
in the analysis of the results rather than in the measurements themselves.
3. Hilpert’s measurements at low Reynolds numbers (R < 44) are satisfied
by equation (15), except for the value of the constant A . Hilpert’s temperature
function differs in form from that in equation (15), but no data available are
sufficiently accurate to discriminate between them. For R > 44,there is a dis-
crepancy between Hilpert’s results and those of the authors which cannot be
resolved without further ,investigation.
4. The theoretical heat transfer relation based on the Oseen approximation
(equation (10))is approached asymptotically as R -+ 0, provided free-convection
or aspect-ratio effects do not intervene. A relation of similar form (equation (16))
satisfactorily describes two-dimensional forced convection for R < 0.5.
5. Free convection effects die out quickly with increasing Reynolds number
(at least for horizontal airstreams), so that except at very low velocities, the
orientation of the wire with respect to the vertical has negligible influence on
the heat transfer coefficient.
6. A rough criterion for the onset of buoyancy effects has been derived for
horizontal wires (equation ( 18)). This gives quantitative expression to earlier
observations that the minimum speed V, which can be measured with a hot-
wire anemometer increases with temperature loading, and also shows that Vmin
varies only as the $-power of the wire diameter.
7. Molecular effects reduce the heat transfer from fine wires below the con-
tinuum value (equation (15))by an amount which can be estimated by assuming
that the temperature differential is reduced by the ‘temperature jump ’ calculated
from kinetic theory.
380 D . C. Collis and M . J . Williams
8. The intensity and scale of the turbulence in the airstream does not affect
the Reynolds number a t which the vortex street develops in the wake of fine
wires. Thus the change in constants of equation (15)likewise occurs at the same
Reynolds number (R = 44) regardless of the stream turbulence.

The authors are indebted to the Chief Scientist, Australian Defence Scientific
Service, Department of Supply, Melbourne, for permission to publish this paper.

Appendix A
Effect of temperature jump on the heat-transfer coeflicient
It has been established (Kennard 1938) that when thermal conduction takes
place between a rarefied gas and a bounding wall, there is a discontinuity in
temperature at the wall. If the wall temperature is T,, and T, is what the tem-
perature of the gas would be if the temperature gradient along the outward-
drawn normal, aTpr, continued right up to the wall, then the discontinuity,
or temperature jump, is given by
aT
TJV -Ts = -[--.
ar
The constant [ has dimensions of length and is known as the temperature jump
distance.
Two assumptions are made in calculating the continuum value of the heat-
transfer coefficient from the values measured in the presence of temperature
jump: ( a )the measured rate of heat transfer is the same as would take place from
a cylinder of the same diameter, at a temperature T,, immersed in a perfectly
continuous gas; ( b ) the correction can be made directly to the measured value
of the heat-transfer coefficient, which is of course an average of the local value
taken over the circumference of the cylinder. Kennard derives an expression
for [ in terms of the properties of the gas and the surface:

where a is the accommodation coefficient of the surface for the particular gas,
k is the thermal conductivity of the gas, y = c,/c, is the ratio of specific heats of
the gas a t constant pressure and constant volume, p is the viscosity of the gas,
and L is the mean free path of gas molecules. Here L is defined by the relation
,u = cpVL, in which V = 4(8RT/n), where R is the gas constant per unit mass, p is
the densityof the gas, andc is a constant such that 0.491 < c < 0.499. Rearranging
and putting 4c = 2, we get
[ = - -2y
- 12-a
L,
y+lr a

where (T = ,uc,/k is the Prandtl number.

For air, y = 1.4 and (T = 0.72, and for platinum in air a + 0.9, so that, with
little error,
i3T
A0 = T S r - T , = - 2 L - . (A 4 )
ar
 Two-dimensional convection from heated wires 381
If a wire is maintained a t a temperature loading 8, = T, - T,, and q the heat
loss per unit area is measured, then the heat-transfer coefficient h,, is calculated
from the relation
q = h,OW, (A 5 )
and on the basis of assumptions (a)and ( b ) above it follows that the continuum
heat-transfer coefficient h, is given by
q = h,e,. (A 6)
Combining (A 5) and (A 6), we obtain

From the conduction equation, we have

and evaluating the quantities in equation (A 4)a t temperature T,, we also have

By eliminating q and 3Tlar from equations (A 6) to (A 9), it can be shown that

(A 10a)

which may be written alternatively as

(A l o b )

I n figure 13, 2Llk has been plotted against T - To,using values of k and L taken
from Kannuluik & Carman (1951) and Kennard (1938). The correct value of
l/h,, must be obtained by successive approximation, starting with T, = !&.
Actually the maximum correction to the present results amounted to 11 yo,
giving a jump of about 30" at a wire temperature of 300 "C. I n this case the error
in h, arising from evaluating 2Llk at q7'instead of T,is less than 4yo.
I n order to determine the continuum Nusselt number Nc from h,, the thermal
conductivity at the mean film temperature q,&= T,+&8, must be substituted
in the expression

I n the example quoted above, the adjustment to k, arising from the diminished
temperature loading amounts to - 3 yo,so that the continuum Nusselt number
(calculated from (A 11))is 13 Yo greater than the measured value.
At low temperature loadings although the jump correction AB/B, may be large,
the absolute temperatures T,,, T,andT, arevery nearly equal. If 2Llk isevaluated
at T, then
N --.h.9d (A 12)
c- k
382 D. C. Collis and M . J . Williams
The measured Nusselt number
hwd 7
N,, = k
and substituting in (A l o b ) we have the simple non-dimensional formula

where K = L / d = Knudsen number.

I 1
? 100 200

T-To
FIQURE
13. Variation of 2Llk with temperature for air at atmospheric pressure.

Appendix Criterion for the onset of buoyancy effects

This is derived for horizontal wires on the assumption that there is no devia-
tion from the forced convection relation for low Reynolds numbers, equation (16),
until the Reynolds number is reduced to the point where the forced convection
Nusselt number equals that due to free convection alone. The experimental
results of this paper show that this is substantially true.
I n the very low Reynolds number range where buoyancy effects are likely to
be significant, the forced convection relation (see 5 4.3) is

I n the range 10-10 < 0, c lo", free convection data of Collis & Williams (1954)
have been re-examined in the light of Mahony's (1956) theoretical work. It was
found that a semi-logarithmic relation similar to equation (B 1) fits the experi-
mental results very satisfactorily:
N-l = 0.88 - 0.43 logloG,. (B2 )
 Two-dimensional convection from heated wires 383
I n this relation the temperature loading has been eliminated by the method of
correlation employed, previously discussed at some length by the authors (1954).
Equating the Nusselt number in equations (B 1) and (B 2) yields a criterion of
the type required,
- i.e.
0.17
log R = 0.39($) log G, 1- 1.07 - 0.80

This is not a convenient equation to deal with, but fortunately a simpler approxi-
mate result can be obtained. In the range < R < 0.2 the temperature
function of equation (B 1) may be omitted if McAdams's correlation (see $ 2 )
is used (see note below). If RM, denotes Reynolds number evaluated in the
manner of McAdams, then in lieu of equation (B 3) we obtain the condition
log RMc= 0.39 log C, + 0.27, (B 4)
or Raft = 1.85 GE3'. (B 5)
If the viscosity-temperature relation be taken as p cc the relation con-
necting R, and R,,, is
RMc = R ,

Equation (B 5) can therefore be written as

(3
brn

-0-76
(B 6)

whence

where V,, is the lowest velocity that can be measured without ambiguity by a
hot-wire anemometer and the constant depends on the flow conditions ( p ,p,
.)''2

Note on McAdams correlation

A simple power law can be fitted locally to any part of the log N 05 log R curve
by determining the local slope p . Thus

p ) - @ l 7 = const. x RP.

A relation connecting R and RMc may be written a5

and equation (B 9) may be written

N = const. x R%,

It can be seen that, provided p lies near 0-17, the temperature function is nearly
annulled.
The slope p may be determined by graphical methods or by differentiation of
the empirical law which describes the particular range of interest. Upon differ-
entiating (B 1) and obtaining p , it is found that in the range 10-2 < R < 10-1
McAdams system will correlate data to within 1 yo at temperatures up to
300 "c.
 384 D.C. Collis and M . J . Williams

REFERENCES
BATCHELOR, G . K. & TOWNSEND, A. A. 1948 Decay of isotropic turbulence in the initial
period. Proc. Roy. SOC. A, 193, 539-58.
BECKERS,H. L., TER HAAR,L. W., TJOAN,LIE TIAM, MERK,H. J., PRINS, J. A. &
SCHENK,J. 1956 Heat transfer a t very low Grashof and Reynolds numbers. Appl.
Sci. Res. A, 6, 82.
COLE, J. & ROSHKO, A. 1954 Heat transfer from wires a t Reynolds numbers in the Oseen
range. Proc. Heat Transfer and Fluid Mechanics Inst. Univ. of Calif. Berkeley, Calg.
COLLIS, D. C. 1956 Forced convection of heat from cylinders at low Reynolds numbers.
J . Aero. Sci. 23, 697-8 (Readers’ Forum).
COLLIS,D. C. & WILLIAMS, M. J. 1954 Free convection of heat from fine wires. A.R.L.
Aero. note 140.
CORRSIN, S. 1949 Extended applications of the hot-wire anemometer. N.A.C.A., T . N .
1864.
GOLDSTEIN, S . (Editor). 1938 Modern Developments in Fluid Dynamics. Vols. I and 11,
Chaps. I and XIV. Oxford.
HILPERT,R. 1933 Warmeabgage von geheizten Drahten und Rohren im Luftstrom.
ForschArb. IngTVes. 4, 215-24.
KANNULUIK, W. G. & CARMAN,E. H. 1951 The temperature dependence of the thermal
conductivity of air. Aust. J . Sci. Res. 4, 305-14.
KAYE,G. W. C. & LABY,T. H. 1948 Tables of Physical and Chemical Constants, tenth
edition. Longmans, Green and Co.
KENNARD, E. H. 1938 liinetic Theory of aases. New York and London: McGraw-Hill
Book Co. Inc.
KING, L. V. 1914 On the convection of heat from small cylinders in a stream of fluid.
Determination of convection constants of small platinum wires with application to
hot-wire anemometry. Phil. Trans. A, 214, 373-432.
MAHONY, J. J. 1956 Heat transfer a t small Grashof numbers. Proc. Roy. SOC.A, 238,
4 12-23.
MCADAMS, W. H. 1954 Heat Transmission, third edition, Chap. X. McGraw-Hill Book
Co. Inc.
NEWMAN, B. G. 1951 Some contributions t o the study of the turbulent boundary-layer
near separation. Australian Council for Aeronautics. Report ACA-53.
OWER,E. & JOHANSEN, F. C. 1931 On a determination of the pitot static tube factor at
low Reynolds numbers with special reference to the measurement of low air speeds.
R & M 1437, British A.R.C., Appendix IV.
OWER,E. 1949 The Measurement of Airflow, third edition, Chap. X. Chapman and Hall
Led.
SIMMONS, L. F. G. & BEAVAN, J. A. 1934 Hot wire type of instrument for recording gusts.
R &. M 1615, British A.R.C.
SIXMONS, L. F. G. 1949 A shielded hot wire anemometer for low speeds. J . Sci. Instrum.
26, 407-11.