21 @ C. N.

Pannell et al - A new integration technique for flowmeters with chordal paths

A new integration technique for flowmeters

with chordal paths
C. N. PANNELL, W. A. B. EVANS and D. A. JACKSON

A class of volume flowmeter exists in which measurements that are essentially

line integrals of the velocity field along a series of parallel chords are combined
linearly to form an estimate for the mean velocity. The paper shows how a priori
knowledge of the functional forms of the velocity profile may be used in a
systematic manner to optimize the design of such an instrument.
Keywords: flowmeter, integration, chordal paths, velocity field

Introduction instruments of this type have so far been obtained

This work is concerned with the optimization of the from consideration of various tabulated schemes for
design of a particular class of volume flowmeters, N-point 1-D quadratures 7 i.e.
illustrated schematically in Figure 1. Measurements of
the velocity field (assumed steady or at least slowly _ f(x) g(x) dx = ~ wi f(xi) (3)
i=l
varying compared with the measurement time) are
made along a series of parallel chords symmetrically Here, the integral of some function f(x) multiplied by
placed in the flow. a known weight function g(x) over a standard interval
For the purpose of illustration, the flow channel is approximated by a sum of terms involving evalu-
will be considered to be circular in cross-section and ations of the function f at N points. Weight functions
completely filled. Each measurement conventionally which have been used are (i) g(x) = 1 (standard Gaus-
yields a quantity Vc(xi) proportional to the velocity sian quadrature) 2'3 and (ii) a scheme 4 based on Tche-
integrated along a chord. The mean velocity estimator bychev polynomials of the second kind 1 in which
VE is calculated from g ( x ) = % / 1 - x 2.
N
V[ = ~.t °)i V c ( x i ) (1)
i=1

where the set of weights {w~} and the set of perpen-

dicular spacings or abscissae {xi} have been deter-
mined beforehand. Indeed, as the {xi} represent the
physical positioning of various transducers they are in
all existing instruments (to our knowledge) fixed at the
mechanical design stage.
The major problem in the scheme of flow
measurement described here is that the sum repre-
sented by equation (1) is of course used only as an
approximation to the true mean velocity V defined by

/
x1 f X
~/= ;~ f~" V(r, O)rdOdr o
flo fg~ rdOdr

= --
it
'f2I: V(r, O) rdOdr (2)
x 2 = -x 3
x 1 =-xq

where V(r, O) is the velocity distribution and the flow /

channel is assumed to be of unit radius. The approxi- J
mation varies in accuracy as the functional form of the
velocity profile changes, as it will, for example, if the
Reynolds number or upstream conditions are altered. Figure 1 Schematic of 4-chord flowmeter cross-section
The {~i} and {xi} used in the design of all showing paths along which the mean velocity is
integrated to yield the quantities Vc(xi), i = 1,2, 3, 4.
The mean velocity is then determined from
The authors are with the University of Kent, Canterbury, Kent CT2 VE = ~=1 tai Vc(xi) where o~i and xi have been previ-
7NR, UK ously fixed.

0955-5986/90/040216-09 (~) 1990 Butterworth-Heinemann Ltd

Flow Meas. Instrum. Vol 1 July 1990 217

As to the question of accuracy, it is known that in D B

case (i) the relation expressed by equation (3) is exact
if
Y
f(x) = P.(x) (4)
where Pn(X) is a polynomial of order 2 N - 1 or less,
while in case (ii) the relation is exact if f(x) is of the
form
f(x) = 2 P.(x) 6) >
where P(.)(x) is a polynomial of order 2N - 1 or less.
The accuracy of the estimate of V, expressed by
equation (1), therefore rests on how well the line
integrals of the velocity along a chord, Vc(xi), are
approximated by functions of the form given by equ-
ations (4) or (5).
It appears, as we shall show, that in a wide A C
variety of flow situations the approximation of Vc(x~)
by equations (4) or (5) is not particularly good and that
a knowledge of the functional forms of the velocity II
profiles likely to be encountered may be used to
derive systematically a set of optimal weights and B,D
abscissae by the method of Harris and Evans5. These
weights and abscissae are in a sense 'tailored' to the
functional form of the velocity profile. They give
smaller errors having a controllable distribution for a
given number of measurement chords N, and in fact
allow the error in principle to be reduced to zero at N
arbitrary values of the parameters characterizin8 the
flow.
After a brief discussion of the realization of what
may be termed 'chord flowmeters' we will consider as
an example the case of axially symmetric flow in a
smooth pipe, in the laminar and fully developed
turbulent regimes. It is assumed that the functional
form of the velocity profile is adequately described
here by a single parameter, the Reynolds number.

Realization of chord flowmeters b

Ultrasonic techniques
Figure 2 Deployment of beams in a "chord-type" ultra-
The following section describes two possible tech- sonic flowmeter. (a) Side view - x points out of paper.
niques, one acoustical and one optical, to measure the AB and CD are respective transmitter/receiver pairs,
quantity whose connecting paths make angles 8 with the pipe
axis. (b) Cross-sectional view. The chord marks the
Vc(x,) Jch V(xi' y)dy (6) projection of ultrasonic beams AB and CD onto a
cross-sectional plane.
i.e. the line integral of the axial component of the
velocity field along a chord parallel to the y axis.
The realization of the chord flowmeter using
measurements of transit-time in ultrasonic beams is Here, Vs is the velocity of sound in the fluid, assumed
described, for example, in references 2, 3, 4, 6 and 7. large compared with the flow velocity, and d is the
With reference to Figure 2, a series of ultrasonic distance A-B. Similarly,
pulses is projected along the paths shown between d
wideband transmitting and receiving transducers. The CStCD = 1 D (8)
beam paths are such that the projection of the up- V~ - ~ ~c V(r) cos 8 d LCD
stream/downstream beam pair onto a cross-sectional
plane is a chord. Typically, several beam pairs will be
deployed in the pipe, although one pair only is shown Thus
for clarity. d d
A pulse sent from A will arrive at B at time &tAB
m m

6t^B 6tco
later, where D
d
&tAB---- 1 1"B (7) V(r)dL^e + f~ V(r)d Lm).
d
V, + --~ J^ V(r) cos e d LAB (9)
218 C N. Pannell et al - A new integration technique for flowmeters with chordal paths

If V is a function of the radial co-ordinate only, i.e. is estimation given a set of quantities Vc(x3 representing
independent of the axial co-ordinate z, it may be seen line integrals of the velocity field along chords
that the right hand side of equation (9) reduces to x~ = constant. We will use as an illustration the case
2cotO /. +VTc~, of axially symmetric turbulent and laminar flows, the
d J-vT:~. V(xi, y)dy (10) functional form of which we assume to be adequately
described by a single parameter, the Reynolds
which is proportional to the line integral of the number. It should be noted that the method is not
velocity field across the chord x~ = constant. restricted to cases of axial symmetry but is applicable
to any case where the general functional form is
Optical techniques describable by a (hopefully) small number of para-
One possible realization of a scheme for measuring meters. It was pointed out in the first section that the
the quantity given by equation (6), which is currently use of N measurement chords allows us to have
under investigation at the University of Kent, is shown exactly zero error in the estimate of V for N specific
in Figure 3 and is based on the technique of transit values of the flow parameters. We have chosen all N
anemometry. Light from a laser diode is collimated values to be Reynolds numbers in an axially symmet-
and then passed through an anamorphic prism pair in ric flow, but we could, for example, have chosen
order to increase the ellipticity of the beam. A beam- some to be shape-determining factors indicating a
splitter is then used to produce a pair of parallel and deviation from symmetry. Functions of this kind are
highly elliptical beams. Seeding particles entrained in discussed, for example, in reference 8.
the flow are swept through the beam pair. Light
scattered from the seeding particles is collected by a The method
suitably designed concentrator and the autocorrelation Genera/
of the photodector output current yields information
about the mean fluid particle transit time between the We are attempting to replace the integral
beams and hence, under suitable conditions, the mean 1 f2~ 1
velocity. F/ = ~ jo Yo V(r, e) rdrde
It should be noted that this configuration is a
rather unusual application of transit anemometry in by a sum of terms of the form of equation (I). This
that conventional systems are designed to give high may be written in Cartesian coordinates as
spatial resolution while this system is designed to have
ideally zero resolution along the direction of the F/ = 1 f^ y V(x, y)dxdy (12)
beams. It

where A is the region inside the unit circle, which in

Optimization of weights ~, and absscissae xi in turn may be expressed as
equation (1)
Introduction = -- V(x, y)dy dx (13)
Tf 1 -

In this section we show how the methods of reference

5 may be applied to the problem of mean velocity where the term inside parentheses which represents
the line intemegrai
aloral
along a chord from y = -
(a) Plan view
to y = + ~ / 1 - x ~ can be regarded as a function of
x:
Collimating Anamorphic
lens prism pair
/ ~ Beamsplitter

R is the Reynolds number and is added to the

argument list of V to show the explicit dependence of
Laser V on R. In an effort to make this paper reasonably
diode self-contained the appendices deal more fully with the
theory of 'tailored' quadratures (i.e. quadratures devel-
oped using a knowledge of the functional forms to be
integrated) including conditions for the existence of
solutions to the equations for the weights and abscis-
(b) Side view Light concentrator sae given by equations (1 7) and (18). However, in this
section a more pragmatic view will be taken, the
question of existence being regarded as answered in
the affirmative if solutions can be found numerically.
We can now apply the methods of reference (5)
Photodlode
to the integration of
Beam dump
Flow channel VCR) = ~7r - f(x; R) d x (15)
Figure 3 Optical technique for evaluating the fine in- As equation (15) is to be represented by a sum of N
tegral of velocity along a chord, Vc(xi). terms
Flow Meas. Instrum. %/ol 1 July 1990 219

N /'7
(16) V* = V ~ - (22)
i=1
we seek a solution of the set of 2N non-linear and u is the kinematic viscosity.
equations in the coi and xi The relation between V + and y+ is then usually
N expressed in terms of what may be described as a
~, ~f(x~; Ri) = F/(&) semi-empirical function 8 defined in a piecewise
i=1 manner according to the dominant physical process
0<x~<l j = 1. . . . 2N (17)
thus;
N even
N V+ = ~ y + ) (23)
(~if(xi; Rj)= V(Rj) (18)
i=1 0 < y+ < 5.5
j = 1. . . . 2N(xl =O),ON×~<I laminar film
N odd g y + ) = -3.04 + 5.01ogey + 5.5 < y+ < 30
buffer layer (24)
Where f is given by equation (14) and the 2 N values
of the Reynolds number are chosen within the speci- [5.5 + 2.5 Iogey + y+ > 30
fied range, Rmin =< Rj = < Rmax (18). turbulent core
The function f(xi; R) appearing in equation (15)
and (16) requires the integration of the velocity along This approach has been refined for our present
a chord x~ = constant. We found that in order to purposes in two ways:
evaluate this to sufficient accuracy for the system of
equations (17) or (18) to converge requires the devel- (1) by using the method of Rothfus and Monrad 1° to
opment of another tailored quadrature scheme of high introduce a correcting factor into g, thereby reduc-
order (a 28-point scheme was used). This was found ing the residual Reynolds number dependence of
necessary as the end-point behaviour of the function is the 'constants' in equation (24) to a negligible
such that even a 20-point Gaussian quadrature gave level.
only 2-3 places of decimals. (2) by replacing the piecewise function g by an ana-
In the case of an axially symmetrical flow profile, lytic function having the correct asymptotic beha-
equation (14) and the fact that V(x, y; R) is even in x, viour at both large and small y+. This function has
imply that f is even in x~. Thus in equations (17) and several parameters which have been determined
(18) the existence of a solution point at x~ implies a by least-squares fitting to the empirical data pre-
solution point at - x i . Having found a solution of sented in reference 10; not only is this new
equations (17) or (18), the following simple strategy is function a better fit to the empirical data which
adopted in order to achieve a symmetrical distribution has been accumulated, it is differentiable an arbit-
of chords: rary number of times. The existence of higher
order derivatives (g defined by equation 24 has a
(a) chords requiring abscissae xi > 0 are symmetrically discontinuous first derivative) enables the set of
placed and the corresponding weight w~ is halved. equations (17), (18) to be solved systematically
(b) Schemes involving an odd number of symmetric- using a multi-variable Newton iteration scheme 11.
ally placed chords will need one placed at x = 0.
In this case the corresponding weight is unaltered. Finally, our function, VT(r, R), is such that the true
mean velocity Q defined by equation (12) may be
found analytically as a function of the Reynolds
Form of function used number. This is important as it is needed in the
As mentioned in the introduction it is our intention to solution of equations (17) and (18).
illustrate the general method for the case of laminar/ Further details of the construction of our 'uni-
turbulent flow in a smooth pipe. Such a flow can versal velocity function' are given in Appendix B.
reasonably well be described by a single parameter, The two functions Vdr) for the laminar case and
the Reynolds number. For the laminar case, the func- VT(r, R) for the turbulent case have been combined
tional form is simply: into a single function using a third 'switching' function
s(n, R/Rc) which has the property
VL(r ) = 1 - r2 (19)
1 R<< Rc
when normalized to have unit maximum velocity, 5(n, R/Rc) = 0 R >> Rc (25)
For the turbulent case it is conventional to work
in terms of 'reduced' quantities for velocity and per- and the width of the transition region is defined by the
pendicular distance from the wall: parameter n.
V Here we took
V+ = (20)
V*
S(n, R/Rc) = (26)
y+ = yV* (21)
12
Where V* is the so-called friction velocity defined in with n = 16, Rc = 2000. The critical Reynolds number
terms of the mean velocity V and 'friction factor' Rc at which transition occurs was therefore fixed at
(reference 9) by 2000 and the value of n fixed the Width of the
 220 C. N. Pannell et al - A new integration technique for flowmeters with chordal paths

transition such that in the turbulent regime there was It should be noted that if one imposes some
approximately 1% of the residual laminar form at criterion on the error distribution in a given range of
R = 2500 while in the laminar regime there was 1% Reynolds number, such as minimizing the maximum
of the turbulent form at R = 1500. It was felt that even error, a more even distribution of errors will result.
if this approach is difficult to justify on purely physical Thus although with an N-chord measurement scheme
grounds (the switching function does not, for example, one can zero the error at N values of Reynolds
display hysteresis at the transition) it was more realistic number (or, in general, N values of flow-describing
than imposing a sharp transition at a particular Rey- parameters), this will not result in the best error
nolds number as it caters for a wider variety of performance if the full range of flow velocities is likely
functional forms of mean velocity profile. to be encountered. The values of the Reynolds num-
The function used to describe the full laminar and bers shown on the various figures are close to the
turbulent flow regimes was optimum in the sense of minimizing the maximum
error, but we do not claim that they are exactly
V(r, R) = 5(16, R/2OOO)VL(r)
- + [I - 5(16, R/2OOO)]VT(r, R)
(27)

Results
In order to obtain the optimized weights and abscissae 0
it is necessary to solve either equations (17) or (18)
with the function f(x; R) defined by equation (14), and L -2
the function V(x,y; R) defined by equation (27)
with r = %/ x ~ + y" .
We have chosen to illustrate the situation by -4
considering the cases of two, three, four and five o~
chord schemes (N = 2, 3, 4, 5). For each value of N, -6
three cases are considered:
(1) The use of Gaussian quadrature -8
(2) The use of a quadrature based on Tchebychev
polynomials of the 2nd kind.
(3) The 'tailored' quadrature obtained by solution of -I0
10 3 10 4 10 5 10 6 1o 7
equations (17) or (18). In all cases the percentage
errors were calculated from Reynolds number

Figure 4 Percentage error as a function of Reynolds

error = 100 x = (28) Number for 2-chord schemes: (A) Gaussian, (B) Tche-
V
bychev, (C) tailored.
where {toi}, {xi} are the sets of weights and
abscissae appropriate to the scheme and V (the
true mean velocity) was obtained by substituting
equation (27) into equation (12) and integrating
analytically.
1.5
The results are shown in Figures 4-7. In all cases A
both the Tchebychev and tailored quadratures give
essentially zero error for the laminar region, as ex- 1
pected. However, in the 2-chord measurement
scheme, the use of Gaussian quadrature results in a
peak error of approximately - 7 . 5 % . The Tchebychev k 0.5
quadrature produces a peak error of nearly 4% at a
Reynolds number of =3000. It can be seen that the
tailored quadratures result in minimum errors; it can ~ o
also be seen that the error is indeed zero at two values
of the Reynolds number. Graphs of error against chord
spacing for what is essentially a single-chord instru- -0.5
ment are also given in reference 6.
In the 3-, 4-and 5-chord schemes we see that the
peak error produced by the Tchebychev quadrature is M
in fact worse than the Gaussian. In all cases the 10 3 10 4 10 S 10 6 10 7
tailored quadratures produce smaller and more evenly
Reynolds number
distributed errors, over four orders of magnitude of
Reynolds number, being 0.1% or less for the 4-and Figure 5 Percentage error/Reynolds Number for three-
5-chord schemes. chord schemes.
 Flow Meas. Instrum. Vol 1 July 1990 221

0.8 Table1 We~ht and abscisue for three measurement

schemesdiscu,medin the text
0.6 n Weight Abscissa

A Gauss~n
0.q 2 1.0000000000 0.5773502691
0.8888888888 0.00000000(X)
L.
0.2 3 0.555555 5 5 5 5 0.7745966692
gJ

0.6521451548 0.3399810435
u
0J 4 0.3478548451 0.8611363115
#. 0.5688888889 0.0000000000
5 0.4786286704 0.5384693101
-0.2
0.2369268850 0.9061798459
Tchebychev
-0,q
2 0.9068996821 0.5000000000
0.7853981634 0.0000000000
-0.6 i s i iltl|l i i * IIIHt I I IlJil.] I * iJllll
3 0.5553603673 0.7071067812
10 3 10 zi 10 5 10 6 10 7
0.5975664329 0.30901 69944
Reynolds n u m b e r
4 0.3693163661 0.80901 69944
Figure 6 Percentage error/Reynolds Number for four- 0.5235987756 0.0000000000
chord schemes. 5 0.4534498411 0.5000000000
0.2617993878 0.86602 54038

Tailored toveloci~profiles
0.5 2 0.8695461811 0.4781968014

0.~,

0.3
/ 2

4
0.9303956329
0.5035368650
0.74408267521
0.22731061851
0.0000000000
0.7793809274
0.40673462160
0.88927588657
0.81119524020 0.0000000000
5 0.46573406517 0.68603131156
t,.
0.2 0.12434916330 0.94742033235
B

0.1

situation may be used to derive a set of weights and

abscissae for the positioning of the sensors in chord-
type volume flowmeters. These weights and abscissae,
derived by the methods of Evans et al, are significantly
-0.1
different from those used in the design of conventional
volume flowmeters, and are optimized in the sense of
i = * J =lt*l i i t i lllll i t i 1 ,HII L i i , J=l
giving minimum error in the estimation of the mean
103 10 q 10 5 10 6 10 7 velocity, over a wide range of flows.
Reynolds n u m b e r
Acknowledgements
Figure 7 Percentage error/Reynolds Number for five-
chord schemes. CNP would like to acknowledge the financial support
of Shell Exploration and Research (UK) Ltd during the
period in which this work was carried out.
WABE is grateful for several past discussions on
optimal as there are almost certainly several local quadrature topics with Dr C. G. Harris, Dr A. Genz,
minima. The weights and abscissae themselves are Prof P. Rabinowitz, Prof P. R. Graves-Morris, Dr C.
shown in Table 1, correct to ten places of decimals. It Woodcock, Ms R. D. Malakar and Dr C. Isenberg, that
can be seen that the quantities found by solving were directly, or indirectly, relevant to the methods
equations (17) and (18) are significantly different from herein used.
the other two cases, within the context of engineering
tolerances.
Appendix A: 'Tailored' quadratures

Conclusiom We give here a brief account of the mathematical

techniques that underlie the method used to obtain the
It has been shown that a knowledge of the range of optimal choice of abscissae (chord positions) and
functional forms likely to be encountered in a flow weights for flow measurements described in this
222 C N. Pannell et al - A new integration technique for flowmeters with chordal paths

paper. These techniques of 'tailored' quadratures are P[2N-1](U).

very general and potentially useful in numerous other
fields, which is why we believe it worthwhile to From (A.2) it is evident that we here are considering
devote this Appendix to their explanation. They con- integrands wih symmetric end-point behaviours at +1.
stitute a development over the years of the work of The weights and abscissae, for this case, will have the
Harris and Evanss, where the Gaussian Quadrature symmetry of Gaussian weights and abscissae, i.e.
method was extended to cater for end-point singular symmetric abscissae, +xi, will have equal weights, oJ~.
(but integrable) behaviours such as x- ~/2, In (x ) etc, It should be mentioned also that the quadrature-defin-
and linear combinations thereof. ing functional form (A.2) is also a generalization of
The 'general' quadrature problem is concerned that considered in reference 5 inasmuch as the non-
with finding numerical approximations to polynomial functions, s.(l )% are multiplied by the
polynomials, P~( )'s of higher than zero order, Mj. In
reference 5, only multiplicative constants, i.e. Mj = 0,
f(x)dx = ~ (ui f(xi) (A.1) were considered. On numerous occasions, in practice,
i=1
using high-order multiplicative polynomials almost in-
where the domain [a,b] can be finite, semi-finite variably leads to remarkable improvements in
[0, ~], or infinite [ - % 0o]. In nearly all cases, by accuracy (several more significant figures for the same
means of some straightforward transformation (which, value of N -- typically, we use N values <30). Note
of necessity, will be non-linear for semi-infinite and that we have 2N of these in total, so that in
infinite domains), the integration can be mapped to
one over a finite domain, which is normally chosen as
f(u).du = ~ c0i.f(xi),
[-1,1]. Most published tabulated quadrature weights I i=I
and abscissae pertain to this domain, i.e. are con-
cerned with we obtain 2N coupled (in general nonlinear) equ-
ations in the 2N unknowns (N~oi's and nx/s), by
equating the coefficients of these arbitrary polynomial
f(u).du = ~, (.Di.f(xi). coefficients on both sides. We have developed a
1 i=1
multi-variable Newton-Raphson program to solve
Published quadrature tables are always based on such equations for a wide choice of nonpolynomial
the optimal integration of some defined form of the and usually assymptotically-singular (but integrable!)
integrand function, f(x); e.g. the well-known Gaussian functions that frequently recur in Physical Applica-
Quadrature is based on polynomials, the N-point rule tions. Further, the program is readily (and has often
being, in fact, exact for integrating any polynomial of been) generalized to incorporate novel nonpolynomial
order ( 2 N - 1 ) or less over finite domains (for an functions as the need for them arises.
authoritative discourse on quadratures the reader is Of course, there are mathematical questions con-
referred to the text of Davis and Rabinowitz 12. As was cerning the uniqueness, existence of solutions (that are
emphasized in reference 5, many integrands of interest real and with all abscissae lying within the integration
are inherently not polynomial-like, and for these Gaus- domain, etc) to the above-mentioned nonlinear quad-
sian quadrature is not very suitable. The preferred rature-defining equations. These questions can be
approach here is to compute a quadrature that is satisfactorily answered for a wide class of defining
eminently suitable for the integrand function(s) we are functional forms that we have found useful (but not
dealing with, i.e. to manufacture a quadrature that is all!). More specifically, if it can be proven that the
'tailored' to the integrand, rather than depend on quadrature-defining functional form is composed of a
finding suitable published (dare we say 'off-the-peg') linear set of functions that constitute a Tchebychev
quadratures. By this we mean we must study the system of functions (see Karlin and Studden 13) over the
integrand functional form, in particular its assymptotic integration domain, then, by a well-known theorem TM
behaviour at the end-points of the domain (here we these properties follow. With quadratures of the type
are implying that, if the function has singular behavi- in equation (A.2), the weights and abscissae do display
ours at points internal to the domain, then the domain Gaussian symmetry, (~)i, +_x~). In this case, a sufficient
must first be split into sub-domains so that all singular condition for the existence of real weights and Positive
behaviours appear at domain end-points). Then we abscissae of magnitude less than unity may be shown
choose a suitable quadrature defining functional form, to depend on whether the even parts, {si(1-
viz: x) + s (x + 1)}, of the linear set of functions that enter
f(U) = PI2N-1-2.Zi(Mi+I)I(U) the quadrature-defining functiona form, constitute a
Tchebychev system over the domain [0, 1]. It is
+ ~,{PMj(1 -- U).Sj(1 - u) intended to report more fully on the mathematical
J technicalities involved here in a future publication Is.
+ PMi(U + 1).Sj(U + 1)} (A.2) In the gas flow application that is the main topic
of this paper, it is paramount to get the best Possible
where the si(1 + u)'s are the non-polynomial set of answer with the least number of chords (abscissae). As
functions, singular or otherwise, that are chosen to the integrancl function, f(x; R), defined, for symmetric
mimic the possible nonpolynomial assymptotic beha- flows, by equation (14) viz.
viour of our integrand function(s) at the domain end- f .v-T-:~
points. The above form is clearly a generalization of f(x; R) = Jo V(x, y; R)dy
Gaussian quadrature, for, if we have no sj(1 4-u)'s,
then the defining functional form is that of the N-point differs in behaviour at the two end-points (correspond-
Gaussian Quadrature, viz ing to the pipe centre and pipe wall) of the integrals
FlowYvfeas. Instrum. Vol 1 July 1990 223

that we need, we generalized the above prescription,

as outlined in the general discussion of the method,
Q(R) --
V max
above, so as to work with asymmetric tailored quad-
ratures catering for different behaviours of the two end and applied the transformation
points, for optimal efficacy. Further, in this applica- V + --~ Q(R) V + = V +'
tion, the defining functional form is not of the form in
equation (A.2) in that it does not possess arbitrary y+ --, y+/Q(R) = y+' (B.1)
linear coefficients, but instead the integrand function all the experimental points fell into a single curve,
depends on R, the Reynolds Number. Thus to arrive at
2 N equations to solve for the N weights and N V +' = g(y*'). (B.2)
abscissae for the N-point formula would give an exact Instead of the usual piecewise function, we used
answer for the flow as calculated using the fitted the function
flow-velocity function shown in Figure 8 (see also
equation (B.3) Appendix B). Although, here, there is g(y*') = bl Io8,(1 + a l y * ' ) + b2(1 - e -a2y'')
no question of the quadrature-defining form being + b3(1 - e-a=Y÷')2 (B.3)
Tchebychev (it is not a linear set of functions!), suffice
to say that, for all our choices of Reynolds Numbers, which has the correct asymptotic behaviour at ex-
we did find real solutions with all the abscissae lying tremes of y+', i.e.
within the integration domain [0, 1].
g(y+' .-, O) ~- (blal + b2a2)y +, i.e. linear (B.4)

Appendix B: Details of the 'universal' velocity ~ y + ' ~ oo) == bl lose at + bl Ioge y * ' , (B.5)
function for turbulent flow used in the paper i.e. logarithmic
In the section above on form of function used, we and used a least-squares fit with a weight function
mentioned that the method of Rothfus and Monrad
had been employed to refine the usual expression for 1
oJt(y+ ') = (B.6)
the universal velocity profile (equation (24)) and that a (IOgl0Y* 90.3
new form of fitting function had been proposed. The to fit the function in the intermediate region. As can
details are given here. be seen from Figure 8, the function is an extremely
Rothfus and Monrad collected together a large good fit while containing the same number of 'free'
amount of experimental flow data and plotted them in parameters as equation (24).
reduced co-ordinates, where it was apparent that a The results of the fit were:
residual dependence of the 'constants' of equations
(24) precluded fitting by a universal (i.e. Reynolds al = 4.517368 x 10 -2
number independent) function. They observed,
a2 = 3.534997 x 10 -2
however, that if one were to take a suitable weakly
monotonic function of the Reynolds number (they bl = 2.572035
used the ratio of mean to maximum velocity):
b2 = 1.944606 x 101

b3 = 1.073174 x 101 .
Velocity f u n c t i o n - fit to data
A suitable empirical form for the friction factor
35
appearing in the section on form of function used
w a s ~6

1 1
3O f = '° (B.7)
5.74 2

÷
25
and from this and the definition of Reynolds number
>:
~_ 20 R 2rpV
_o
u = U
(B.8)
>
"o 15 where rp = pipe radius, we get
u
"O U
e,. 10
V, = h(R) (B.9)
2 rp

where
5
R
h(R) = (B.IO)
0 '
10-1 100 101 102 103 10q 105 106

Reduced distance from wail, Y+ Similarly, for y+ we have

Figure 8 Fit of universal velocity curve (equation (B.3)) 1( r)
y ÷ = -~- 1 - - - h(~ (B.11)
to data. rp
224 C. N. Pannell et al - A new integration technique for flowmeters with chordal paths

Finally, our equation VT(r, R) was obtained by sub- flowmeters. In: B. E. Jones (Ed), Instrument science and
stituting from equations (20), (21), (BI),(B9), (BIO) and technology, Vol 3 (1983)
(B11) into equation (B2), setting .rp to unity and 4 British Patent No 1 503852: Fluid flow measurement
normalizing so that the on-axis veloctiy is also unity. system for pipes. Filed by Westinghouse Electric Corp.
The result is (1978)
5 Harris, C. G. and Evans, W. A. B. Extension of numerical
g{ h(R) (1 - r)
quadrature formulae to cater for end point singular
behaviours over finite intervals. Intern.J. ComputerMath.
Section B 6 (1977) 219
VT(r'R)={ } g2Q(R)
h(R, (B.12) 6 Baker, R. C. and Thompson, "E. J. A two-beam ultrasonic
phase-shift flowmeter. Proc. Conf. on Fluid Flow
Measurement, NEL, Glasgow (April 1975)
The function Q(R) was obtained by fitting a poly- 7 Nolan, M. E., O'Hair, I. G. and Teyuandier, R. The
measurement of high pressure natural gas flows using the
nomial in log [ log R] to data presented in reference 9.
four path ultrasonic flowmeter developed by British Gas.
Proc. Int. Syrup. on Fluid Flow Measurement, Washing-
ton D.C., USA. (16-19 Nov 1986)
Table 2 Reynolds numbers at which error is zero for tail- 8 Salami, L. A. Errors in the velocity-area method of
orecl quadratures of Figures 4 - 7 measuring asymmetric flows in circular pipes. Proc. Int.
Conf. on Modern Developments in Flow Measurements,
Number UKAEA, Harwell (21-28 Sept 1971) paper 6.5
of chords R 9 Massey, B. S. The mechanics of fluids. Van Nostrand
(1968)
1.0 x 10 3, 7.0 x 10 3 10 Rothfus, R. R. and Monrad, C. C. Correlation of turbul-
1.0 x 10 3, 3.4 x 10 3, 2.5 x 10 4 ent velocities for tubes and parallel plates. Indust. and
1.0 x 10 3, 2.9 x 10 3, 8.0 x 10 3, 4.0 x 10 4 Engng. Chem.47 (6)(1955) 1144
1.0 x 103 , 3.0x 103 , 5 . 7 x 103 , 2 . 0 x 104, 11 Press, W. H., Flannery, B. P., Teukolsky, S. A. and
1.0 x 103 Vetterling, W. T. Numerical recipes. Cambridge Univer-
sity Press (1988) ISBN 0-521-30811-9.
12 Davis, P. J. and Rabinowitz, P. Methods of numerical
integration (2nd Edn). Academic Press, Orlando, Florida
(1984)
References 13 Karlin, S. and Studden, W. J. Tchebycheff systems with
applications in analysis and statistics. Wiley (Inter-
1 Abramowitz, M. and Stegun, I. A. Handbook of mathe- science) New York (1966)
matical functions. Dover, 9th Edn, ISBN 0-486-61272-4. 14 Polya, G. and Szego, G. 'Aufgaben und Lehrsatze aus
2 Fisher, S. G. and Spink, P. G. Ultrasonics as a standard der Analysis'. Vol. II, Springer (1925) 48.
for volumetric flow measurement. Proc. Int. Conf. on 15 Evans, W. A. B. Manuscript in preparation (1990)
Modern Developments in Flow Measurement. UKAEA, 16 Miller, D. S. Internal flow systems. BHRA Fluid Engineer-
Harwell (21-28 Sept 1971) paper 3.4 ing Series, Vol. 5. Pub. BHRA Fluid Eng. (1986) ISBN
3 Brain T. J. S. and Scott, R. W. W. Survey of pipeline 0-900983-787