Vous êtes sur la page 1sur 27

Flow Turbulence Combust (2013) 90:127

DOI 10.1007/s10494-012-9419-7

Explicit Friction Factor Accuracy and Computational


Efficiency for Turbulent Flow in Pipes

Herbert Keith Winning Tim Coole

Received: 19 June 2012 / Accepted: 4 October 2012 / Published online: 31 October 2012
Springer Science+Business Media Dordrecht 2012

Abstract The implicit ColebrookWhite equation is the accepted method for accu-
rately estimating the friction factor for turbulent flow in pipes. This study reviews
28 explicit equations for approximating the friction factor to integrate both the
accuracy to the implicit ColebrookWhite equation and the relative computational
efficiency of the explicit equations. A range of 901 Reynolds numbers were selected
for the review between Re 4 103 and 4 108 and 20 relative pipe roughness
values were selected between D 106 101 , thus producing a matrix of 18,020
points for each explicit equation, covering all the values to be encountered in
pipeline hydraulic analysis for turbulent flow. The accuracy of the estimation of the
friction factor using the explicit equations to the value obtained using the implicit
ColebrookWhite equation were calculated and reported as absolute, relative per-
centage and mean square errors. To determine the relative computational efficiency,
300 million friction factor calculations were performed using randomly generated
values for the Reynolds number and the relative pipe roughness values between the
limits specified for each of the explicit equations and compared to the time taken by
the ColebrookWhite equation. Finally, 2D and 3D contour models were generated
showing both the range and magnitude of the relative percentage accuracy across the
complete range of 18,020 points for each explicit equation.

Keywords Fluid mechanics Turbulent flow Friction factor Explicit equation


ColebrookWhite Computational efficiency

H. K. Winning (B)
CB&I, 40, Eastbourne Terrace, London, W2 6LG, UK
e-mail: kwinning@cbi.com

T. Coole
Dr. Buckinghamshire New University, Queen Alexandra Road, High Wycombe,
Buckinghamshire, HP11 2JZ, UK
e-mail: Tim.Coole@Bucks.ac.uk
2 Flow Turbulence Combust (2013) 90:127

Nomenclature

D Pipe inside diameter (mm)


fd Darcy friction factor (dimensionless)
ff Fanning friction factor (dimensionless)
Re Reynolds number (dimensionless)
Absolute Pipe Roughness (mm)

1 Introduction

Pressure loss in a pipeline is calculated using the DarcyWeisbach equation, which


requires the friction factor to be known; the friction factor being dependent on
both the relative roughness (absolute roughness divided by the internal diameter)
of the internal wall of the pipe and the Reynolds number which defines the flow
regime. The pipe is considered to be smooth when the projections of the surface are
completely submerged in the viscous laminar layer [1]. The thickness of the viscous
laminar layer is dependent on fluid properties; therefore the effect of the absolute
pipe roughness on the fluid is dependent on the magnitude of the Reynolds number.
Due to its applicability over a wide range of the Reynolds numbers and relative
pipe roughness values, the ColebrookWhite equation [2, 3] has become the accepted
standard for calculating the friction factor in pipes. This equation is a development of
the Prandtl [4] smooth pipe equation and the von Karman [5] rough pipe equation.
The ColebrookWhite equation is customarily given by Perry and Green [6] as:
 
1 1.256
 = 4 log +  (1)
ff 3.7D Re f f

However, the ColebrookWhite equation is implicit with respect to the friction


factor and therefore requires an iterative solution, such as using the Newton
Raphson method. Where numerous calculations are required such as for the hy-
draulic analysis of long distance pipelines transporting compressible fluids this will
have a significant effect on the time taken to obtain a solution, even for fast computer
based systems. To address this issue, a number of explicit equations for the estimation
of friction factor have been developed since the first explicit relationship proposed
by Moody [7, 8] in 1947. These equations vary in the accuracy of the approxi-
mation of the friction factor, determined largely by the complexity of the explicit
equation.

2 Comparative Studies

There have been a number of comparative reviews of explicit equations, usually as


part of proposing a new explicit equation. The equations reviewed in the comparative
studies discussed are identified in Table 5. The first of these reviews was carried out
in April 1985 by Gregory and Fogarasi [9]. The impact of this study on subsequent
research is evident from the books [10, 11] and papers [1216] citing their work,
frequently using the conclusions of their study as a starting point for new work.
Flow Turbulence Combust (2013) 90:127 3

This study compared the accuracy of 12 explicit friction factor equations to results
obtained using the ColebrookWhite equation. For each equation, Reynolds num-
bers in the range 4 103 to 108 and relative pipe roughness values between 108
to 5 102 were used. Thus, using a matrix of 6 Reynolds numbers and 8 relative
pipe roughness values for each equation generated a total of 48 data points for each
equation. In this study all the explicit equations were stated as being in terms of the
Fanning friction factor. During the review of this paper, it was found that the Wood
[17, Eq. 7] and the Serghides [18, Eqs. 20 and 21] were returning the Darcy friction
factor rather than the Fanning friction factor as stated, this being confirmed by
reference to the original sources. The Fanning friction factor is based on the hydraulic
radius, whereas the Darcy friction factor is based on the hydraulic diameter, thus
Darcy friction factor values are four times those of the Fanning friction factor.
Because of this error, their conclusion that Chen [19, Eq. 13] represented the last real
improvement is questionable. While Eq. 13 represented an improvement in accuracy
over the Moody [8, Eq. 5] by an order of magnitude of five, the Serghides [18, Eq. 20]
represented an improvement in accuracy over the Moody [8, Eq. 5] by an order of
magnitude of seven.
A second comparative review was carried out by Zigrang and Sylvester [20] in
June 1985, using the same boundary conditions as used by Serghides [18]: a matrix
of 70 values from 10 relative pipe roughness values from 4 105 to 5 102 and
7 Reynolds values from 2500 to 108 . This study was very similar to that of Gregory
and Fogarasi with respect to the explicit equations reviewed, with the exception of
dropping the Wood [17] equation and including the equation from Chen [21]; this
equation is not reviewed in the current study as the author stated that it was a
simplified equation and not expected to be of high accuracy, which was confirmed
by preliminary analysis. The recommendations were similar to those of Gregory
and Fogarasi, with the Chen [19, Eq. 13] being recommended as the simplest of the
most accurate equations, while the Serghides [18, Eq. 20] was identified as being the
most accurate overall. In addition, this study compared the equations complexity, as
defined by the number of algebraic notation calculator key strokes required to solve
the equation for Re = 103 and D = 0.001 [20].
Two further reviews were conducted in 2007 and 2008 by Goudar and Sonnad
[22, 23]. Both of these reviews only assessed the accuracy of the equations, with the
only new equations reviewed since 1985 being Manadilli [24, Eq. 23] in the 2007
review and Romeo et al. [25, Eq. 24] in the 2008 review. Both of these studies were
conducted using 20 relative pipe roughness values, ranging from 106 to 5 102
and 500 values of Reynolds numbers ranging from 4 103 to 108 , producing a matrix
of 10,000 points. This represents a significant increase in the size of the matrix
of values reviewed compared to the previous two studies. In the 2008 study [23],
the conclusion was that the Serghides [18, Eq. 20] was the most accurate of those
equations reviewedthe Serghides [18, Eq. 20] was not reviewed in the 2007 study.
In 2009, a further study was carried out by Yildirim, based on the same matrix
of values as used by Goudar and Sonnad. Yildirim does not make any definitive
recommendations, but concludes the study by summarising the data previously
presented in tabulated form, highlighting that the three most accurate equations are
Chen [19, Eq. 13], Sonnad and Goudar [14, Eq. 25] and Romeo et al. [25, Eq. 24].
Unfortunately neither of the Serghides equations were reviewed in this comparative
study.
4 Flow Turbulence Combust (2013) 90:127

In 2011, Brkic [26] performed a review of 20 equations, using 20 relative pipe


roughness values, ranging from 106 to 7.5 102 and 37 values of Reynolds
numbers ranging from 104 to 108 , producing a matrix of 740 points. In addition
to determining the accuracy, this study used the same method for estimating the
complexity of the equation as used by Zigrang and Sylvester [20], 26 years earlier.
This review introduced equations including Eck [27, Eq. 9], Rao and Kumar [28,
Eq. 26], Buzzelli [29, Eq. 27], Avci and Karagoz [30, Eq. 28] and Papaevangelou
et al. [31, Eq. 29]. This review did not offer any definite conclusions other than the
equations were all very accurate, with the exception of the equations: Round [32,
Eq. 14], Eck [27, Eq. 9], Moody [8, Eq. 5], Wood [17, Eq. 7] and Rao and Kumar
[28, Eq. 26].
In 2011 another review of 16 equations was conducted by Genic et al. [33],
performed over the same range of values of the Goudar and Sonnad and the Yildirim
studies with a very large matrix of 1 million points. This review introduced the
equations of: Altshul cited in [33, Eq. 6], Tsal [34, Eq. 22], Fang et al. [35, Eq. 32]
and two by Brkic [26, Eqs. 30 and 31]. This review, like the work of Romeo et al.
[25] in 2002, used the Model Selection Criterion (MSC) and the Akaike Information
Criterion (AIC) to perform a statistical comparison of the relative computational
efficiency. This review concluded with the recommendation of the use of the Zigrang
and Sylvester [36, Eq. 18].
The aim of this study is to provide a consistent review of the available explicit
equations, across a wide range of Reynolds numbers and relative pipe roughness
values, thus identifying the absolute and relative accuracy to the implicit Colebrook
White equation, the applicable range for each equation and the relative computa-
tional efficiency for each explicit equation.

3 Analysis

The explicit equations were all coded as Visual Basic functions in an Excel
spreadsheet. The equations were coded in terms of the friction factor (either Darcy
or Fanning) as stipulated in the cited equation. To ensure a uniform output for
the equations, where the friction factor was returned as a Fanning friction factor,
the result of the calculation was multiplied by four. Therefore, all the friction
factor functions return the Darcy friction factor. A range of Reynolds numbers and
relative pipe roughness values were selected to encompass the limits of the previous
comparative studies reviewed. A range of 901 Reynolds numbers was selected for the
review between Re 4  103 and 4 108 and 20 relative pipe roughness values
were selected between D 106 101 , thus producing a matrix of 18,020 points
for each calculation. The dimensions of the matrix were chosen to ensure that any
local variations were identified, and that sufficient data was collected to generate
smooth contour plots.

3.1 Accuracy to the Implicit ColebrookWhite Equation

For each of the explicit equations the minimum, maximum and mean absolute
error (Eq. 2) from the ColebrookWhite equation was calculated across the range
of Reynolds numbers and relative pipe roughness values studied. In addition the
Flow Turbulence Combust (2013) 90:127 5

minimum, maximum and mean relative percentage error (Eq. 3) was calculated and
finally the mean square error (MSE) (Eq. 4) was determined. The ColebrookWhite
values were determined to an accuracy of 109 for comparative purposes. These
results are in Table 1 and are ordered by the mean relative percentage error.

Absolute error = | ftrue festimated | (2)

 
ftrue festimated
Relative error = 100 (3)
ftrue

i=N
( ftrue festimated )2
Mean square error (MSE) = i=1
(4)
N
The explicit equations were also reviewed and grouped according to the number of
iterations of the ColebrookWhite function as defined by the number of calls made
to the log function, either to base 10 or the natural log. In addition, an adjectival
grade was given to each of the explicit equations, according to the size of the MSE:
Large: MSE 5 106 .
Medium: MSE 108 < 5 106 .
Small: MSE 1011 < 108 .
Very Small: MSE < 1011 .
These results are in Table 2, ordered by the mean square error.
Contour plots were produced (Appendix), which mapped the relative percentage
error (z value) to the log of the relative pipe roughness (x axis) and the log of
the Reynolds number (y axis). A description of the error distribution was defined
according to:
Upper Diagonal: where the magnitude of the MSE is medium or small and
the lowest errors are mainly confined to the upper diagonal (higher Reynolds
numbers and rougher pipe). These equations exhibit the greatest error in the
transitional zone, where Reynolds numbers are low and the relative roughness is
high.
Diagonal Ridge Error: where the magnitude of the MSE is medium or small and
the distribution of the highest errors is generally confined to the diagonal line,
from high Reynolds numbers to rough pipe.
Diagonal Ridge: where the magnitude of the MSE is small and the distribution of
the lowest errors is generally confined to the diagonal line, bisecting the values.
This is the inverse of the Diagonal Ridge Error.
Unclassified: where the error distribution does not conform to one of the
previous classifications and is generally spread across a significant range of the
matrix.
These different error distributions are identified in Fig. 1, for Churchill [43, Eq. 8],
Chen [19, Eq. 13], Romeo et al. [25, Eq. 24] and Moody [8, Eq. 5] respectively.
The magnitude of the error increases from dark blue through green, yellow and
orange to red, though the absolute values for the colour banding is dependent on
the magnitude of the MSE and therefore varies for each equation. It should be noted
6 Flow Turbulence Combust (2013) 90:127

Fig. 1 Error distribution classification

that the vertical scale for all the contour plots is the same to allow direct comparisons
to be made.
The previous studies have reported the accuracy of the explicit equations using a
variety of methods.
Relative percentage errorGregory and Fogarasi [9], Zigrang and Sylvester [20]
and Brkic [26]
Absolute errorGoudar and Sonnad [22, 23]
Relative percentage error and MSEGenic [33]
Relative percentage error, absolute error and MSEYildirim [16]
While the MSE provides a good indication to the overall accuracy of the explicit
equation, spikes within the data only become apparent with the maximum percentage
error. As can be seen from the contour plots, some of the explicit equations have very
high errors over a small range of input values thus producing spikes, which while
immediately obvious on the plots, might not be so readily identified in the tables and
certainly cannot be identified if both the MSE and maximum relative percentage
errors are not reported.
However, neither of these methods satisfactorily indicates the range of applicable
Reynolds numbers and relative pipe roughness values that produce an acceptable
accuracy for a given explicit equation. Using a combination of both 2D and 3D
contour plots the applicable range is easily indentified, even for the large matrix of
values being examined.

3.2 Computational Efficiency

Although the previous reviews which considered the complexity or efficiency for the
explicit equations did so by reference to either the number of algebraic notation
Flow Turbulence Combust (2013) 90:127 7

calculator key strokes or by statistical modelling, this study determined the relative
computational efficiency for the explicit equations by comparing the time taken to
perform the equations. It is believed that this provides a more practical method of
determining the relative computational efficiency.
The 100 million friction factor calculations were performed using randomly
generated values for the Reynolds number and the relative pipe roughness values
between the limits specified in the study, for each of the explicit equations. The total
time taken to perform the calculations was recorded and the relative computational
efficiency determined by dividing the time taken for the explicit equation to perform
the calculations by the time taken by the implicit ColebrookWhite equation. This
was performed three times and the mean values for the relative computational
efficiency calculated and ranked. These results are shown in Table 3.

3.3 Comparative Analysis

The explicit equations were ranked according to both the MSE (Rank A) and the
computational efficiency (Rank B); these rankings were then combined to produce
an overall ranking (Rank C). In order to produce a final ordinal ranking, the raw
combined rank was then ranked according to Rank A, as shown in Table 4.
As expected, those equations with the greater number of internal iterations
were generally more accurate and required a greater computational effort. While
the accuracy ranged from 1.75 104 to 5.62 1012 , the relative computational
efficiency only ranged from 0.488 to 0.716. These results are shown in Table 4.

4 Summary

As can be seen from Tables 1 and 2, the accuracy ranking differs slightly depending
on whether the equations are ordered by MSE or mean relative percentage error,
with the most significant changes being for the Churchill [41, Eq. 12], Brkic [26,
Eq. 30] and Rao and Kumar [28, Eq. 26], though generally the variations are small.
This study found that the Buzzelli [29, Eq. 27] is the most accurate overall, with an
MSE of 5.62 1012 , with the second most accurate being the Serghides Eq. 2 [18,
Eq. 20] with an MSE of 7.51 1012 .
The relative computational efficiency as shown in Table 3, demonstrates the
general correlation between accuracy and computational efficiency. Ignoring those
equations with large magnitudes of error as shown in Table 2, the most computa-
tionally efficient equations were those of Eck [27, Eq. 9], Churchill [43, Eq. 8] and
Swamee and Jain [42, Eq. 11]; these equations all produced medium magnitudes
of error. The equations with a small magnitude of error which were most compu-
tationally efficient were: Shacham [38, Eq. 15], Zigrang and Sylvester Eq. 11 [36,
Eq. 17] and Fang et al. [35, Eq. 32]. For the two most accurate equations; Serghides
Eq. 2 [18, Eq. 20] and Buzzelli [29, Eq. 27], the variation in the ranking of the
computational efficiency was not significant, these equations being ranked 20th and
27th respectively.
Table 4 combines the performance of the equations based on accuracy and
computational efficiency. The explicit equations with the lowest combined ranking
were: Sonnad and Goudar [14, Eq. 25], Serghides Eq. 3 [18, Eq. 21] and Serghides
8

Table 1 Absolute, mean square and relative percentage error ordered by mean relative percentage error
Equation Absolute error MSE Relative percentage error
Min Max Mean Min Max Mean
(27) Buzzelli [29] 1.139E12 9.401E06 1.224E06 5.622E12 3.547E09 2.355E02 0.005
(20) Serghides Eq. 2 [18] 2.096E12 9.467E06 1.491E06 7.511E12 4.309E09 2.372E02 0.006
(18) Zigrang and Sylvester Eq. 12 [36] 2.096E12 3.051E05 4.027E06 7.067E11 4.309E09 1.305E01 0.019
(21) Serghides Eq. 3 [18] 2.095E12 4.697E05 6.668E06 1.886E10 4.308E09 3.680E01 0.037
(24) Romeo et al. [25] 2.566E09 9.877E05 2.528E05 1.330E09 2.730E05 1.226E01 0.057
(16) Barr [37] 4.175E11 2.588E04 2.315E05 1.945E09 3.180E07 5.533E01 0.063
(13) Chen [19] 3.493E09 1.537E04 3.388E05 2.032E09 2.902E05 3.387E01 0.097
(15) Shacham [38] 1.738E13 3.557E04 2.941E05 3.834E09 3.573E10 8.912E01 0.125
(17) Zigrang and Sylvester Eq. 11 [36] 1.098E12 1.866E04 2.790E05 3.101E09 2.258E09 9.921E01 0.141
(32) Fang et al. [35] 5.795E09 5.534E04 5.412E05 6.551E09 4.616E05 5.347E01 0.156
(25) Sonnad and Goudar [14] 1.285E07 3.867E04 5.222E05 1.053E08 1.344E03 9.688E01 0.174
(29) Papaevangelo et al. [31] 5.746E09 1.491E03 1.020E04 4.089E08 3.273E05 1.411E+00 0.230
(19) Haaland [39] 2.700E08 9.734E04 1.191E04 2.670E08 1.675E04 1.434E+00 0.373
(23) Manadilli [24] 6.875E09 2.484E03 1.561E04 1.277E07 3.636E05 2.710E+00 0.393
(10) Jain [40] 3.229E09 2.533E03 1.803E04 1.325E07 1.642E05 3.172E+00 0.452
(12) Churchill [41] 2.028E09 1.723E02 1.856E04 2.185E07 1.089E05 1.631E+01 0.475
(11) Swamee and Jain [42] 6.917E08 2.786E03 1.838E04 1.606E07 1.769E04 3.343E+00 0.478
(31) Brkic [26] 1.058E09 2.445E03 1.850E04 1.352E07 5.678E06 2.799E+00 0.479
(8) Churchill [43] 1.142E08 2.812E03 1.891E04 1.677E07 1.046E04 3.407E+00 0.492
(30) Brkic [26] 1.105E09 1.723E03 2.187E04 1.511E07 1.364E05 3.396E+00 0.721
(9) Eck [27] 6.901E08 3.506E03 3.899E04 4.640E07 1.265E04 8.209E+00 1.503
(28) Avci and Karagoz [30] 2.940E08 6.184E03 9.576E04 3.353E06 1.398E04 6.084E+00 1.716
(7) Wood [17] 2.265E07 1.128E02 1.536E03 5.913E06 1.859E03 2.825E+01 3.876
(14) Round [32] 1.573E07 1.336E02 2.260E03 1.835E05 8.530E04 1.315E+01 4.474
(5) Moody [8] 3.629E08 2.809E02 3.962E03 7.142E05 2.869E04 2.658E+01 6.098
(22) Tsal [34] 2.689E09 4.132E02 6.435E03 1.746E04 1.693E05 3.915E+01 8.894
(6) Altshul cited in [33] 4.564E08 4.132E02 6.684E03 1.751E04 2.187E04 3.972E+01 11.449
(26) Rao and Kumar [28] 5.086E06 3.412E02 3.469E03 5.159E05 2.838E02 8.548E+01 13.270
Flow Turbulence Combust (2013) 90:127
Flow Turbulence Combust (2013) 90:127 9

Table 2 Error magnitude and distribution, ordered by MSE


Equation MSE Magnitude Distribution
of error
(27) Buzzelli [29] 5.622E12 Very small Upper diagonal
(20) Serghides Eq. 2 [18] 7.511E12 Very small Upper diagonal
(18) Zigrang and Sylvester Eq. 12 [36] 7.067E11 Small Upper diagonal
(21) Serghides Eq. 3 [18] 1.886E10 Small Upper diagonal
(24) Romeo et al. [25] 1.330E09 Small Diagonal ridge
(16) Barr [37] 1.945E09 Small Upper diagonal
(13) Chen [19] 2.032E09 Small Diagonal ridge error
(17) Zigrang and Sylvester Eq 11. [36] 3.101E09 Small Upper diagonal
(15) Shacham [38] 3.834E09 Small Upper diagonal
(32) Fang et al. [35] 6.551E09 Small Unclassified
(25) Sonnad and Goudar [14] 1.053E08 Medium Upper diagonal
(19) Haaland [39] 2.670E08 Medium Upper diagonal
(29) Papaevangelo et al. [31] 4.089E08 Medium Unclassified
(23) Manadilli [24] 1.277E07 Medium Diagonal ridge error
(10) Jain [40] 1.325E07 Medium Upper diagonal
(31) Brkic [26] 1.352E07 Medium Upper diagonal
(30) Brkic [26] 1.511E07 Medium Upper diagonal
(11) Swamee and Jain [42] 1.606E07 Medium Upper diagonal
(8) Churchill [43] 1.677E07 Medium Upper diagonal
(12) Churchill [41] 2.185E07 Medium Upper diagonal
(9) Eck [27] 4.640E07 Medium Upper diagonal
(28) Avci and Karagoz [30] 3.353E06 Medium Unclassified
(7) Wood [17] 5.913E06 Large Unclassified
(14) Round [32] 1.835E05 Large Unclassified
(26) Rao and Kumar [28] 5.159E05 Large Upper diagonal
(5) Moody [8] 7.142E05 Large Unclassified
(22) Tsal [34] 1.746E04 Large Unclassified
(6) Altshul cited in [33] 1.751E04 Large Unclassified

Eq. 2 [18, Eq. 20]; these having decreasing magnitudes of error of medium, small and
very small respectively.
As 12 of the equations reviewed did not have stated applicable ranges, this study
has applied the same conditions to all the equations. While this obviously affects
the accuracy of the results claimed by the authors where an applicable range was
stated, it has been done to give a level comparison of the equations reviewed;
however, this is addressed by reference to the 2D and 3D contour plots which
clearly show the range of applicability for each equation. It should be noted that
the Rao and Kumar [28, Eq. 26] was developed to satisfy experimental data and
not to approximate the implicit ColebrookWhite equation and therefore shows to a
large extent the variation between the experimental data and the implicit Colebrook
White equation. However, it highlights the potential error that can be introduced
through the incorrect selection of an explicit method for approximating the friction
factor.
10 Flow Turbulence Combust (2013) 90:127

Table 3 Relative computational efficiency


Equation Relative computational efficiency Ranking
Run 1 Run 2 Run 3 Mean
(6) Altshul cited in [33] 0.4773 0.4766 0.4779 0.4773 1
(5) Moody [8] 0.4926 0.4869 0.4881 0.4892 2
(7) Wood [17] 0.5284 0.5226 0.5239 0.5250 3
(9) Eck [27] 0.5539 0.5634 0.5597 0.5590 4
(22) Tsal [34] 0.5692 0.5685 0.5648 0.5675 5
(14) Round [32] 0.5795 0.5787 0.5802 0.5795 6
(8) Churchill [43] 0.5846 0.5787 0.5853 0.5829 7
(11) Swamee and Jain [42] 0.5948 0.5889 0.5955 0.5931 8
(25) Sonnad and Goudar [14] 0.5999 0.5940 0.6058 0.5999 9
(10) Jain [40] 0.6050 0.5991 0.6007 0.6016 10
(12) Churchill [41] 0.5999 0.6042 0.6078 0.6040 11
(28) Avci and Karagoz [30] 0.6173 0.6063 0.6129 0.6122 12
(15) Shacham [38] 0.6224 0.6216 0.6232 0.6224 13
(19) Haaland [39] 0.6377 0.6267 0.6334 0.6326 14
(17) Zigrang and Sylvester Eq. 11 [36] 0.6326 0.6369 0.6334 0.6343 15
(32) Fang et al. [35] 0.6429 0.6369 0.6437 0.6411 16
(21) Serghides Eq. 3 [18] 0.6480 0.6420 0.6488 0.6463 17
(23) Manadilli [24] 0.6480 0.6471 0.6590 0.6514 18
(13) Chen [19] 0.6531 0.6522 0.6539 0.6531 19
(20) Serghides Eq 2 [18] 0.6633 0.6522 0.6590 0.6582 20
(16) Barr [37] 0.6633 0.6573 0.6641 0.6616 21
(31) Brkic [26] 0.6735 0.6726 0.6743 0.6735 22
(18) Zigrang and Sylvester Eq. 12 [36] 0.6786 0.6675 0.6743 0.6735 23
(26) Rao and Kumar [28] 0.6837 0.6828 0.6897 0.6854 24
(27) Buzzelli [29] 0.6939 0.6879 0.6948 0.6922 25
(30) Brkic [26] 0.6939 0.6879 0.6999 0.6939 26
(24) Romeo et al. [25] 0.7144 0.7083 0.7102 0.7110 27
(29) Papaevangelo et al. [31] 0.7399 0.7390 0.7409 0.7399 28

5 Conclusions

This study has identified the most accurate explicit equations for approximating the
values obtained by the implicit ColebrookWhite equation, as well as determining
the relative computational efficiency for each explicit equation. By the innovative
use of 2D and 3D contour modelling over an extensive range of values, this study has
also identified the applicable range for each explicit equation in a way that enables
the reader to quickly assimilate the information from the vast number of underlying
calculations.
Any explicit equation can only approximate the value as obtained by the implicit
ColebrookWhite equation. Therefore, the value of an explicit equation to approxi-
mate the implicit ColebrookWhite equation is a function of both the accuracy and
computational efficiency and a review of one without the other is of questionable
value. As can be seen from Table 4, the most appropriate equations are not the most
accurate, but rather those that are of a higher order of accuracy while being more
efficient.
While determining the friction factor in the region of 1 % or better is considered
acceptable from a practical viewpoint, by referring to the tables and contour plots
Table 4 Explicit equations ordered by accuracy and computational efficiency
Equation Internal Accuracy Computational efficiency Combined ranking
Iterations MSE Rank A Relative effort Rank B A+B Rank C
(25) Sonnad and Goudar [14] 2 1.053E08 11 0.5999 9 20 1
(21) Serghides Eq. 3 [18] 3 1.886E10 4 0.6463 17 21 2
(20) Serghides Eq. 2 [18] 3 7.511E12 2 0.6582 20 22 3
(15) Shacham [38] 2 3.834E09 9 0.6224 13 22 4
(17) Zigrang and Sylvester Eq. 11 [36] 2 3.101E09 8 0.6343 15 23 5
(10) Jain [40] 1 1.325E07 15 0.6016 10 25 6
(9) Eck [27] 1 4.640E07 21 0.5590 4 25 7
(27) Buzzelli [29] 3 5.622E12 1 0.6922 25 25 8
(18) Zigrang and Sylvester Eq. 12 [36] 3 7.067E11 3 0.6735 23 26 9
Flow Turbulence Combust (2013) 90:127

(13) Chen [19] 2 2.032E09 7 0.6531 19 26 10


(32) Fang et al. [35] 1 6.551E09 10 0.6411 16 26 11
(19) Haaland [39] 1 2.670E08 12 0.6326 14 26 12
(11) Swamee and Jain [42] 1 1.606E07 18 0.5931 8 26 13
(8) Churchill [43] 1 1.677E07 19 0.5829 7 26 14
(7) Wood [17] 0 5.913E06 23 0.5250 3 26 15
(16) Barr [37] 2 1.945E09 6 0.6616 21 27 16
(5) Moody [8] 0 7.142E05 26 0.4892 2 28 17
(6) Altshul cited in [33] 0 1.751E04 28 0.4773 1 29 18
(14) Round [32] 1 1.835E05 24 0.5795 6 30 19
(12) Churchill [41] 1 2.185E07 20 0.6040 11 31 20
(24) Romeo et al. [25] 3 1.330E09 5 0.7110 27 32 21
(23) Manadilli [24] 1 1.277E07 14 0.6514 18 32 22
(22) Tsal [34] 0 1.746E04 27 0.5675 5 32 23
(28) Avci and Karagoz [30] 2 3.353E06 22 0.6122 12 34 24
(31) Brkic [26] 4 1.352E07 16 0.6735 22 38 25
(29) Papaevangelo et al. [31] 2 4.089E08 13 0.7399 28 41 26
(30) Brkic [26] 4 1.511E07 17 0.6939 26 43 27
(26) Rao and Kumar [28] 2 5.159E05 25 0.6854 24 49 28
11
12

Table 5 Reviewed explicit equations


Year Author and reference Explicit friction factor equation Applicable range Notes Eq.

1 
106 3
1947 Moody [8] f f = 1.375 103 1 + 2 104 D + Re Re 4 103 and 108 1a, 1b, 1c, 1f & 1e (5)

D 0 102 1a, 1b, 1c, 1e & 1f

0.25
68
1952 Altshul cited in [33] fd = 0.11 Re+ D 4 (6)
A1
0.225 0.4
1966 Wood [17] fd = 0.094 D + 0.53 D + 88 D Re Re 4 103 and 5 107 1a, 1c, 1f & 2 (7)

D 105 4 102
0.134
where A1 = 1.62 D
7 0.9 

1973 Churchill [43] 1 = 2 log 3.7D + Re 1c, 1e, 1f & 4 (8)
fd

15
1973 Eck [27] 1 = 2 log 3.715D + Re 4 (9)
fd

0.9 

1976 Jain [40] 1 = 2.28 4 log D + 29.843
Re Re 5 103 and 107 1a, 1b, 1c, 1e & 1f (10)
ff

D 4 105 5 102

0.9 
6.97


1976 Swamee and Jain [42] 1 = 4 log Re + 3.7D Re 5 103 and 108 1a, 1c, 1e & 1f (11)
ff

D 106 5 102
 1
12

12
8 1
1977 Churchill [41] ff = 2 Re + 3 1a, 1e & 4 (12)
(A2 +A3 ) 2
 0.9 16
16
7
where A2 = 2.457 ln Re + 0.27 D A3 = 37530
Re

5.0452
1979 Chen [19] 1 = 4 log 3.7065D Re log A4 Re 4 103 and 4 108 1a, 1b, 1d, 1e & 1f (13)
ff
1.1098

0.8981
(/D) 7.149
where A4 = 2.8257 + Re D 107 5 102
Flow Turbulence Combust (2013) 90:127
Table 5 (continued)
Year Author and reference Explicit friction factor equation Applicable range Notes Eq.

Re
1980 Round [32] 1 = 3.6 log Re 4 103 and 108 1a, 1c & 1f (14)
ff 0.135Re( D )+6.5

D 0 5 102


5.02 14.5
1980 Shacham [38] 1 = 4 log 3.7D Re log 3.7D + Re Re 4 103 and 4 108 1b (15)
ff


4.518 log Re
7

1981 Barr [37] 1 = 2 log 3.7D + 0.52
1d, 1e & 4 (16)
fd Re 1+ Re29 ( D )0.7

5.02
1982 Zigrang and Sylvester Eq. 11 [36] 1 = 4 log 3.7D Re log A5 Re 4 103 and 108 1a, 1b & 1d (17)
ff
13
where A5 = 3.7D + Re D 4 105 5 102

Flow Turbulence Combust (2013) 90:127

5.02
1982 Zigrang and Sylvester Eq. 12 [36] 1 = 4 log 3.7D Re log A6 Re 4 103 and 108 1a, 1d, 1e, 1f & 3 (18)
ff
13 5.02
where A5 = 3.7D + Re A6 = 3.7D Re log A5 D 4 105 5 102
1.11 
6.9
1983 Haaland [39] 1 = 3.6 log Re + 3.7D Re 4 103 and 108 1a, 1b, 1c, 1e & 1f (19)
ff

D 106 5 102
2
2
(A8 A7 )
1984 Serghides Eq. 2 [18] fd = A 7 A9 2A8 +A7 1a, 1b, 1d, 2 & 4 (20)

12
where A7 = 2 log 3.7D + Re

7
A8 = 2 log 3.7D + 2.51A
Re

8
A9 = 2 log 3.7D + 2.51A
Re
2
2
7 4.781)
1984 Serghides Eq. 3 [18] fd = 4.781 A(A 8 2A7 +4.781
1a, 1d, 2 & 4 (21)

12
where A7 = 2 log 3.7D + Re

7
A8 = 2 log 3.7D + 2.51A
Re
13
Table 5 (continued)
14

Year Author and reference Explicit friction factor equation Applicable range Notes Eq.

0.25
68
1989 Tsal [34] A = 0.11 Re +D Re 4 103 and 108 (22)

If A 0.018 then fd = A D 0 5 102

If A < 0.018 then fd = 0.0028 + 0.85A


95
96.82


1997 Manadilli [24] 1 = 2 log 3.7D + Re0.983 Re Re 5.235 103 and 108 1c, 1e & 1f (23)
fd

D 0 5 102

5.0272
2002 Romeo et al. [25] 1 = 2 log 3.7065D Re Re 3 103 and 1.5 108 1d, 1e & 1f (24)
fd

4.567
log 3.827D Re D 0 5 102


0.9924
log 7.7918D

0.9345 
5.3326
+ 208.815+Re

0.4587Re
2006 Sonnad and Goudar [14] 1 = 0.8686 ln S(S/S+1)
Re 4 103 and 108 1c & 1e (25)
fd

where S = 0.1240 D Re + ln (0.4587Re) D 106 5 102
 
(2 D )1

2007 Rao and Kumar [28] 1 = 2 log 0.444+0.135Re 4 (26)


fd
Re

2
0.33 ln Re
6.5
where = 1 0.55e


B 2
B1 +2 log Re
2008 Buzzelli [29] 1 = B1 4 (27)
fd 1+ 2.18
B 2

0.774 ln(Re)]1.41
where B1 = [

1+1.32 D

B2 = 3.7D Re + 2.51 B1
Flow Turbulence Combust (2013) 90:127
Table 5 (continued)
Year Author and reference Explicit friction factor equation Applicable range Notes Eq.
6.4
2009 Avci and Karagoz [30] fd =  4 (28)


2.4
ln(Re)ln 1+0.01Re D 1+10 D
0.24790.0000947(7log Re)4
2010 Papaevangelo et al. [31] fd =
2 Re 104 and 107 (29)

log 3.615D + 7.366
0.9142
Re

D 105 103

2
2011 Brkic [26] fd = 2 log 100.4343 + 3.71D 4 (30)

where = ln Re

1.1Re
1.816 ln ln(1+1.1Re)

2

Flow Turbulence Combust (2013) 90:127

2011 Brkic [26] fd = 2 log 2.18


Re + 3.71D 4 (31)

where = ln Re

1.1Re
1.816 ln ln(1+1.1Re)
 2
60.525 56.291
1.1007
2011 Fang et al. [35] fd = 1.613 ln 0.234 D + Re 3 103 and 108 (32)
Re1.1105 Re1.0712

D 0 5 102

1. Equation reviewed by:


a. Gregory and Fogarasi [9]
b. Zigrang and Sylvester [20]
c. Goudar and Sonnad [22]
d. Goudar and Sonnad [23]
e. Yildirim [16]
f. Genic [33]
2. In the Gregory and Fogarasi [9] paper it was incorrectly identified as returning the Fanning friction factor, rather than the Darcy friction factor
3. This equation was incorrectly stated in the Gregory and Fogarasi [9] paper, where the equation was given as:
 
1 = 4 log 3.7D 5.02 log A6
ff

4. No applicable ranges specified in the original paper


15
16 Flow Turbulence Combust (2013) 90:127

within this study it is possible to select the most suitable explicit equation based
on the accuracy required, relative pipe roughness, predicted flow regime, and the
number of calculations to be performed taking into account the inherent uncertainty
in estimating some of the input parameters (Table 5).

Acknowledgement The authors wish to acknowledge the importance of the comparative study
conducted by Gregory and Fogarasi which has provided the basis for much of the subsequent
research in this area and is still widely cited.

Appendix: 2D and 3D Contour Plots

Fig. 2 Moody [8, Eq. 5]

Fig. 3 Altshul cited in [33, Eq. 6]


Flow Turbulence Combust (2013) 90:127 17

Fig. 4 Wood [17, Eq. 7]

Fig. 5 Churchill [43, Eq. 8]

Fig. 6 Eck [27, Eq. 9]


18 Flow Turbulence Combust (2013) 90:127

Fig. 7 Jain [40, Eq. 10]

Fig. 8 Swamee and Jain [42, Eq. 11]

Fig. 9 Churchill [41, Eq. 12]


Flow Turbulence Combust (2013) 90:127 19

Fig. 10 Chen [19, Eq. 13]

Fig. 11 Round [32, Eq. 14]

Fig. 12 Shacham [38, Eq. 15]


20 Flow Turbulence Combust (2013) 90:127

Fig. 13 Barr [37, Eq. 16]

Fig. 14 Zigrang and Sylvester Eq. 11 [36, Eq. 17]

Fig. 15 Zigrang and Sylvester Eq. 12 [36, Eq. 18]


Flow Turbulence Combust (2013) 90:127 21

Fig. 16 Haaland [39, Eq. 19]

Fig. 17 Serghides Eq. 2 [18, Eq. 20]

Fig. 18 Serghides Eq. 3 [18, Eq. 21]


22 Flow Turbulence Combust (2013) 90:127

Fig. 19 Tsal [34, Eq. 22]

Fig. 20 Manadilli [24, Eq. 23]

Fig. 21 Romeo et al. [25, Eq. 24]


Flow Turbulence Combust (2013) 90:127 23

Fig. 22 Sonnad and Goudar [14, Eq. 25]

Fig. 23 Rao and Kumar [28, Eq. 26]

Fig. 24 Buzzelli [29, Eq. 27]


24 Flow Turbulence Combust (2013) 90:127

Fig. 25 Avci and Karagoz [30, Eq. 28]

Fig. 26 Papaevangelou et al. [31, Eq. 29]

Fig. 27 Brkic [26, Eq. 30]


Flow Turbulence Combust (2013) 90:127 25

Fig. 28 Brkic [26, Eq. 31]

Fig. 29 Fang et al. [35, Eq. 32]

References

1. Farshad, F., Rieke, H., Garber, J.: New developments in surface roughness measurements,
characterization, and modeling fluid flow in pipe. J. Petrol. Sci. Eng. 29, 139150 (2001)
2. Colebrook, C.F.: Turbulent flow in pipes, with particular reference to the transition region
between the smooth and rough pipe laws. J. Inst. Civ. Eng. 11, 133156 (1939)
3. Colebrook, C.F., White, C.M.: Experiments with fluid friction in roughened pipes. Proc. R. Soc.
Lond. Ser. A, Math. Phys. Sci. 161, 367381 (1937)
4. Prandtl, L.: Essentials of Fluid Dynamics. Hafner, New York (1952)
5. Von Karman, T.: The fundamentals of the statistical theory of turbulence. J. Aeronaut. Sci. 4,
131138 (1934)
6. Perry, R.H., Green, D.W.: Perrys Chemical Engineers Handbook, 7th edn. McGraw-Hill, Lon-
don (1997)
7. Moody, L.F.: Friction factors for pipe flow. Trans. Am. Soc. Mech. Eng. 66, 671684 (1944)
8. Moody, L.F.: An approximate formula for pipe friction factors. Trans. Am. Soc. Mech. Eng. 69,
10051006 (1947)
9. Gregory, G.A., Fogarasi, M.: Alternative to standard friction factor equation. Oil Gas J. 83, 120
127 (1985)
26 Flow Turbulence Combust (2013) 90:127

10. Guo, B., Lyons, W.C., Ghalambor, A.: Petroleum Production Engineering: A Computer-
Assisted Approach. Gulf Professional, Oxford (2007)
11. Takcs, G.: Gas Lift Manual Tulsa. Pennwell Books, Oklahoma (2004)
12. zger, M., Yldrm, G.: Determining turbulent flow friction coefficient using adaptive neuro-
fuzzy computing technique. Adv. Eng. Software 40, 281287 (2009)
13. Schroeder, D.W. Jr.: A tutorial on pipe flow equations. In: PSIG Annual Meeting Bonita Springs.
Pipeline Simulation Interest Group, Florida (2010)
14. Sonnad, J.R., Goudar, C.T.: Turbulent flow friction factor calculation using a mathematically
exact alternative to the ColebrookWhite equation. J. Hydraul. Div. Am. Soc. Civ. Eng. 132,
863867 (2006)
15. Sonnad, J.R., Goudar, C.T.: Constraints for using Lambert W function-based explicit Colebrook
White equation. J. Hydraul. Eng. Am. Soc. Civ. Eng. 130, 929931 (2004)
16. Yldrm, G.: Computer-based analysis of explicit approximations to the implicit Colebrook
White equation in turbulent flow friction factor calculation. Adv. Eng. Softw. 40, 11831190
(2009)
17. Wood, D.J.: An Explicit Friction Factor Relationship, vol. 60. Civil Engineering American
Society of Civil Engineers (1966)
18. Serghides, T.K.: Estimate friction factor accurately. Chem. Eng. 91, 6364 (1984)
19. Chen, N.H.: An explicit equation for friction factor in pipes. Ind. Eng. Chem. Fund. 18, 296
(1979)
20. Zigrang, D.J., Sylvester, N.D.: A review of explicit friction factor equations. J. Energ. Resour.
Tech. 107, 280283 (1985)
21. Chen, J.J.J.: A simple explicit formula for the estimation of pipe friction factor. Proced. Inst. Civ.
Eng. 77, 4955 (1984)
22. Goudar, C.T., Sonnad, J.R.: Explicit friction factor in correlation ifor turbulent flow in rough
pipe. Hydrocarb. Process 86, 103105 (2007)
23. Goudar, C.T., Sonnad, J.R.: Comparision of the iterative approximations to the colebrook-white
equation. Hydrocarb. Process 87, 7983 (2008)
24. Manadilli, G.: Replace implicit equations with signomial functions. Chem. Eng. 104, 129 (1997)
25. Romeo, E., Royo, C., Monzn, A.: Improved explicit equations for estimation of the friction
factor in rough and smooth pipes. Chem. Eng. J. 86, 369374 (2002)
26. Brkic, D.: Review of explicit approximations to the Colebrook relation for flow friction. J. Petrol.
Sci. Eng. 77, 3448 (2011)
27. Eck, B.: Technische Stromungslehre. Springer, New York (1973)
28. Rao, A.R., Kumar, B.: Friction factor for turbulent pipe flow. Division of Mechanical Sciences,
Civil Engineering Indian Institute of Science Bangalore, India ID Code 9587 (2007)
29. Buzzelli, D.: Calculating friction in one step. Mach. Des. 80, 5455 (2008)
30. Avci, A., Karagoz, I.: A novel explicit equation for friction factor in smooth and rough pipes. J.
Fluids Eng. 131, 061203 (2009)
31. Papaevangelou, G., Evangelides, C., Tzimopoulos, C.: A New Explicit Relation for the Friction
Factor Coefficient in the DarcyWeisbach Equation, pp. 166172. Protection and Restoration
of the Environment Corfu, Greece: University of Ioannina Greece and Stevens Institute of
Technology New Jersey (2010)
32. Round, G.F.: An explicit approximation for the friction factorReynolds number relation for
rough and smooth pipes. Can. J. Chem. Eng. 58, 122 (1980)
33. Genic, S., Arandjelovic, I., Kolendic, P., Jaric, M., Budimir, N., Genic, V.: A review of explicit
approximations of Colebrooks equation. FME Transactions 39, 6771 (2011)
34. Tsal, R.J.: AltshulTsal friction factor equation. Heat-Piping-Air Cond. 8, 3045 (1989)
35. Fang, X., Xu, Y., Zhou, Z.: New correlations of single-phase friction factor for turbulent pipe flow
and evaluation of existing single-phase friction factor correlations. Nucl. Eng. Des. 241, 897902
(2011)
36. Zigrang, D.J., Sylvester, N.D.: Explicit approximations to the solution of Colebrooks friction
factor equation. AIChE J. 28, 514515 (1982)
37. Barr, D.I.H.: Solutions of the Colebrook-white function for resistance to uniform turbulent flow.
Proc. Inst. Civ. Eng. Part 2: Res. Theor. 71, 529535 (1981)
38. Shacham, M.: Comments on: An explicit equation for friction factor in pipe. Ind. Eng. Chem.
Fund. 19, 228229 (1980)
39. Haaland, S.E.: Simple and explicit formulas for the friction factor in turbulent pipe flow. J. Fluids
Eng. 105, 8990 (1983)
Flow Turbulence Combust (2013) 90:127 27

40. Jain, A.K.: Accurate explicit equation for friction factor. J. Hydraul. Div. Am. Soc. Civ. Eng. 102,
674677 (1976)
41. Churchill, S.W.: Friction factor equation spans all fluid flow regimes. Chem. Eng. J. 91, 9192
(1977)
42. Swamee, D.K., Jain, A.K.: Explicit equations for pipe flow problems. J. Hydraul. Div. Am. Soc.
Civ. Eng. 102, 657664 (1976)
43. Churchill, S.W.: Empirical expressions for the shear stress in turbulent flow in commercial pipe.
AIChE J. 19, 375376 (1973)