Vous êtes sur la page 1sur 13

IPENZ Transactions, Vol. 25, No.

1/CE, 1998

Hydraulic roughness of bored tunnels


Mark Stuart Pennington1, BScEng, MScEng, TM.IPENZ
Physical roughness data from various bored tunnels in Southern Africa were collected. An in-depth investigation into energy losses in conduits and the mechanisms associated therewith was made. The roughness data were analysed mathematically and linked to these energy loss mechanisms, to predict the expected hydraulic friction effects. Similar investigation was made on the inner surface of spun concrete pipes, and the results predicted using the methods developed accurately matched those obtained by physical testing of the pipes, giving confidence in the accuracy of the methods employed. Expected hydraulic loss coefficients for various surface types encountered in the tunnels sampled were calculated, and these resembled closely those values predicted in other such tunnels. A method for accurate determination of hydraulic friction loss coefficients from measurements of physical roughness data was developed. Keywords: water tunnels tunnel design unlined bored tunnels hydraulic resistance
1

Montgomery Watson, PO Box 9463, Hamilton

This paper, first received on 15 September 1997, was received in revised form on 6 April 1998.

1. Introduction
This paper presents the results obtained from a research project carried out at the University of Natal, Durban, South Africa during 1994 and 1995. The topic researched was the hydraulic roughness or resistance of unlined, bored water tunnels, and the research was sponsored by the Water Research Commission of South Africa. The expected hydraulic resistance is one of many factors which need to be known in the design of water tunnels. Knowing the resistance, the necessary diameter and/or difference in head required for a given design flow rate may be calculated. This is commonly done using the Darcy-Weisbach head loss equation
hf = flv 2 2 gd

.................................................(1)
head lost due to friction Darcy-Weisbach friction factor length of conduit mean velocity of fluid acceleration due to gravity diameter of conduit

where

hf f l v g d

= = = = = =

In terms of friction factor, f, the Colebrook-White equation (given in (2) below) has been shown to yield results consistent with those obtained experimentally.

k 1 2.52 .....................(2) = 2 log10 + f 3.71d Re f where Re is the pipe Reynolds number and k is the equivalent sand grain diameter from Nikuradses (1933) experiments on roughness in pipe flow.
For hydraulically rough or fully-developed flow at high Re (usually encountered in tunnels), equation (2) reduces to

IPENZ Transactions, Vol. 25, No. 1/CE, 1998

1 3.71d = 2 log10 ...................................( 3) k f

As an alternative to equation (1) with (2) or (3), Mannings formula is often used as an hydraulic resistance equation. The advantage offered by this is that Mannings n is independent of diameter and effectively depends only on surface roughness. For a closed conduit flowing full at high Re, Mannings n and Darcy-Weisbach f are related by

d6 f n= ..........................................(4) 4 8g In order to evaluate friction factor f from equation (3), the Nikuradse equivalent sand grain diameter k needs to be known (or estimated). The problem is that k is an equivalent dimension and does not relate specifically to any measurable surface roughness dimension other than the diameter of single-sized sand grains glued to the inside of a pipe (Nikuradse, 1933).

2. Measurements of surface roughness


The first step towards establishing this link was to obtain a large amount of sufficiently detailed raw data. For this purpose, a tunnel roughness measuring device, consisting of a movable laser scanner mounted on a track, was developed at the University of Natal. By picking up the reflection of an emitted laser beam from the tunnel wall in consecutive, discrete intervals, the wall roughness profile was digitised. Distance readings from the scanners one metre long track to the tunnel wall were taken every 0.5mm to an accuracy of 0.1mm, giving 2000 points per metre. In this way a two-dimensional profile of the roughness sampled at various locations along a tunnel wall could be plotted. A threedimensional picture of the surface roughness was obtained by taking parallel scans each a set distance apart, dependent on the degree of accuracy required. Examples of the two-dimensional plots obtained (to an exaggerated vertical scale) are shown in Figure 1. A more detailed description of the apparatus, its operation and its accuracy is given by Pennington (1995).

SANDSTONE Chainage 3500 Ngoajane South


10

mm

5 0 -5 -10 0 100 200 300 400 500 600 700 800 900 1000

mm

GRANITE Chainage 3200 Emolweni


10

mm

5 0 -5 -10 0 100 200 300 400 500 600 700 800 900 1000

mm

SHOTCRETE Chainage 1600 Emolweni


10

mm

5 0 -5 -10 0 100 200 300 400 500 600 700 800 900 1000

mm

FIGURE 1: Physical roughness plots obtained from scanner.

Field trips to four different tunnels bored by Tunnel Boring Machine (TBM) were made. All of the tunnels were sampled, with samples being taken at 100m intervals. Within these tunnels, four different surface types were encountered and sampled. A summary of the sampling done is given in Table 1. TABLE 1: Summary of sampling - numbers of samples
In-situ Unlined Unlined Shotcrete TOTAL Tunnel Name concr lining granite sandstone Emolweni, Inanda-Wiggins Aqueduct 5 27 0 18 50 Clermont, Inanda-Wiggins Aqueduct 0 0 36 30 66 Ngoajane North Drive, Lesotho Highlands 0 0 108 0 108 Ngoajane South Drive, Lesotho Highlands 0 0 37 0 37 TOTAL 5 27 181 48 261

Differences in texture obtained from the various surface types are evident in Figure 1. In this figure any trends in the data have been removed using a linear least squares best fit, and are measured relative to the mean value which corresponds to the zero of the ordinate.

3. Statistical analysis of data


Having collected all these data, the challenge was then to extract something meaningful from them. In attempting to evaluate f, as given by equation (3), the phenomenon of flow at the boundary must be studied. Energy lost in boundary shear in a fluid flowing in fully developed turbulent flow past a rough boundary is due to the formation and subsequent dissipation of eddies. The amount of energy lost depends on at least two factors, these being the size of the eddies generated and the frequency at which they are shed. The size of eddy is directly dependent on the height of the roughness projection from which the eddy is spawned, and the frequency of eddy generation is dependent on the longitudinal spacing between consecutive roughness projections. Other factors such as shape and spatial distribution of the roughness elements will also have an effect on the amount of energy lost. Due to the random nature of the roughness pattern produced by TBM on rock, these other factors are assumed to have a common effect, with the major differences being due to the roughness height and spacing. It should, however, be noted that the spacing is a means of quantifying the number of eddies spawned (per unit length) and it is this (ie the number of eddies) which is directly linked to the amount of energy lost due to friction at the boundary. After identifying height and spacing as the two roughness characteristics having the dominant influence on energy loss due to boundary roughness in tunnels, meaningful analysis could be carried out on the physical roughness data collected. In this study, two different roughness height measurements were compared against each other. They were (i) the standard deviation, h , and (ii) the mean range, h. The spacing was measured by the mean wavelengths of the surface, obtained via the power spectrum of the data. It is also possible to represent physical roughness with a sinusoid of equivalent dimensions. These roughness descriptors are defined and described in detail in Appendix 1.

4. Linking physical roughness to hydraulic resistance


For flow at high values of Reynolds Number, Re, the flow essentially skims over the roughness crests, causing the formation of cylindrical eddies in the space between the tops of the roughness elements and the bed, with sizes varying according to the space available. In considering the energy lost to friction at the boundary, it may be seen that it is the net volume (or area in a two-dimensional representation) of eddies generated which is the causing mechanism. It is evident that this eddy volume is directly proportional to the heights of the roughness projections causing the turbulence. However, it would appear that the spacing of such projections has no bearing on the net eddy volume, provided Re is sufficiently high to cause wake interference flow (Morris, 1955). It is proposed that for surfaces of random roughness, the overall eddy generation is equivalent, by definition, to that over a sinusoid of equivalent dimensions. Note that hydraulic resistance in the context of wake interference flow is primarily by the mechanism of eddies being spawned in the hollows between

protrusions and being transported into the flow where their energy is dissipated through mixing and viscosity. Consider the sinusoidal boundary shown in Figure 2.

FIGURE 2: Flow at sinusoidal boundary. Assume that the area between the bed and the tops of the roughness crests comprises eddies. For two different random surfaces whose equivalent sinusoids have the same amplitude, it is seen that the net eddy generation per unit length in the direction of flow is the same. Therefore it is argued that spacing has no effect on the amount of energy consumed in the formation and subsequent dissipation of eddies formed at the bed. If this is true, then the equivalent sand grain diameter, k, may be taken to be a function of the average or representative height of the surface roughness elements only. Other factors such as shape are disregarded due to the random nature of the tunnel roughness. Note that the above only applies to flow at sufficiently high values of Reynolds Number.

5. Relationship between k and h


LeCocq and Marin (1976) showed that sufficient accuracy is obtained if k is equated with average roughness height at the boundary, h. (No indication of how this h was calculated or measured was given in their paper). This, however, can only be true for high Re at which the flow exhibits wake interference flow, as outlined above. Using the values of h and h calculated for surfaces of known hydraulic resistance, this postulate was tested. The first surface used for this test was that occurring within spun concrete pipes. From the literature it was found that a representative value for Mannings n of such pipes is n = 0.010 (Pennington, 1995). It is, however, suspected that this is a lower bound for the roughness and that many spun concrete pipes would exhibit slightly higher n-values. The laser scanner was used to sample the physical roughnesses of such pipes, and from these measurements values of h and h were calculated. Using these, values of friction factor, f , were calculated using equation (2) with k = h or h, and compared with expected values. Since the diameters of pipes tested varied, it is more convenient to compare the corresponding calculated values for Mannings n obtained in this way. A summary of these values is given in Table 2. Table 2: Mean estimated Mannings n-values for concrete pipes. method A : Colebrook-White equation, k=h B : Colebrook-White equation, k=h Expected value from literature Mannings n 0.0117 0.0113 0.0100

From Table 2 it is evident that use of either h or h for k in the Colebrook-White equation (2) yields comparable values for n, and these are similar to the expected value, found in the literature. Heerman (1968) attempted to establish the link between physical roughness and hydraulic resistance. His methods were found to be inaccurate for the types of flow found in tunnels (Pennington, 1995), but some valuable information was available in his thesis. He included the physical roughness dimensions of the pipes of varying roughness configurations which he tested, providing a source of roughness data for which the corresponding hydraulic resistances have been measured. 6

The different roughnesses were made by casting plaster in the form of sinusoids of varying dimensions within 150mm outside diameter pipes. For each roughness, Heerman made head loss measurements in flowing air at various values of Re from which the friction factor, f, could be calculated. By analysis of generated sinusoidal roughness data corresponding exactly to the roughnesses tested, theoretical values for friction coefficients could be calculated (by the methods given above), and these were compared to the experimental results. Very good correlation of these was observed. In selecting the best k/h relationship, the sums of squares of the deviations of theoretical from measured values were calculated. From this it was found that using k= h yields very similar results to those obtained using k= h, with the former being marginally more accurate. It was also found that both of these yielded results of an acceptable degree of accuracy (Pennington, 1995), and therefore the postulate of k=h can be applied with confidence to physical roughness data taken from bored tunnels, when attempting to calculate the corresponding hydraulic resistance parameters.

6. Application to tunnel roughness data


Using both the standard deviation and the mean range as measures of roughness height, and by letting these heights be equivalent to k in equation (2), values of f and hence n for all sets of data taken in all tunnels sampled were calculated. These resulting n values are summarised in Figure 3 in which maxima and minima, together with the ranges one standard deviation either side of the mean values are shown.

7. Values of Mannings n from literature


As part of the study, an extensive literature survey was conducted for information pertaining to hydraulic resistance in conduits, in particular bored water tunnels. From the limited amount of literature which was found on the resistance in bored tunnels, the following summary was compiled: Unlined: Stutsman (1988) : estimated from measurements Stutsman (1988) : calculated from head loss HDTC (1988) report LeCocq and Marin (1976) : from measurements Metcalf (in Pegram & Pennington, 1995) : from head losses Other: HDTC (1988) report - shotcrete linings Metcalf (in Pegram & Pennington, 1995) - shotcrete HDTC (1988) report - cast in situ concrete Metcalf (in Pegram & Pennington, 1995) - in situ concrete n = 0.016 n = 0.015 n = 0.012 n = 0.012 n = 0.0159 n = 0.0154 n = 0.016 n = 0.016 n = 0.0161

From Figure 3 it is evident that, contrary to suggestions found in the literature, shotcrete is rougher than unlined rock. This ties in well with what has been observed from a physical inspection of tunnels and what may be evident from Figure 1. However, having said this, it is accepted that the number of shotcreted surfaces sampled is relatively small. Furthermore, shotcrete roughness may vary considerably with particular material constituents, and with the method of application.

8. Recommended values for Mannings n in bored tunnels


The following emerge as the recommended values for Mannings n for use in the design of tunnels bored by machine: cast in situ concrete lining unlined sandstone unlined granite n = 0.0119 0.0009 n = 0.0154 0.0010 n = 0.0157 0.0008

shotcrete

n = 0.0161 0.0011

Manning's n by Variance for different surfaces


0.02

0.019

0.018

0.017

n
0.016 0.015 0.014 0.013 sandstone1 sandstone2 sandstone3 granite shotcrete1 shotcrete2

Surface

Manning's n by Mean Range for different surfaces


0.02

0.019

0.018

0.017

n
0.016 0.015 0.014 0.013 sandstone1 sandstone2 sandstone3 granite shotcrete1 shotcrete2

Surface

FIGURE 3: Values of Mannings n for tunnels. The agreement between these values derived from physical measurements and those found in the literature is encouraging. In consideration of the expected micro-roughness of bored tunnels, it is suggested that the above values for n be taken as representative. It should, however, be emphasised here that these n values only apply to the surface texture, and do not incorporate macro roughness effects (diameter / alignment changes, steps and other irregularities), as do commonly occur in bored tunnels (for example, when TBM cutters are changed). In deciding on the applicability of the results of this investigation to shotcrete roughness, it is suggested that, at the least, visual comparison of the surfaces concerned is made. This is made possible by the inclusion of photographs of every sampled section (shotcrete and unlined) being included in the references, Pennington (1995) and Pegram & Pennington (1996). In addition, raw data files (in ASCII format) of all physical roughness data sets used are included on diskette in the document by Pennington (1995).

9. References
Colebrook, CF & White, CM (1937), Experiments with Fluid Friction in Roughened Pipes, Proceedings, Royal Society of London, Vol.161, p.367-381. Heerman, DF (1968), Characterization of Hydraulic Roughness, Thesis submitted in partial fulfilment of requirements for PhD, Colorado State University, Colorado. Highlands Delivery Tunnel Consultants (1988), Delivery Tunnel Design Contract TCTA- 01, Technical Memorandum H2, Tunnel Roughness Report, Unpublished. LeCocq, R & Marin, G (1976), Evaluation des Pertes de Charges des Galleries DAmenee DEau Forees au Tunnelier et Non-Revetues, Translation by J Capell (1994), Keeve Steyn Inc., Unpublished. Manning, R (1889), On the Flow of Water in Open Channels and Pipes, Transactions, Institution of Civil Engineers of Ireland, Vol.20, p.161-207. Nikuradse, J (1933), Stromungsgesetze in Rauhen Rohren, Verein Deutscher Ingenieure, Forschungsheft, Vol.361. Morris, HM (1955), Flow in Rough Conduits, Transactions, American Society of Civil Engineers, Vol.120, p.373-410. Pegram, GGS & Pennington, MS (1996), A Method for Estimating the Hydraulic Roughness of Unlined Bored Tunnels, Water Research Commission (WRC) Report No 579/1/96, Johannesburg, South Africa. ISBN No. 1 86845 219 0. Pennington, MS (1995), Hydraulic Roughness of Bored Tunnels, Thesis submitted in fulfilment of requirements for MScEng, Department of Civil Engineering, University of Natal, Durban, South Africa. Stutsman, RD (1988), TBM Tunnel Friction Factors for the Kerckhoff 2 Project, Water Power 87, Brian W Clowes ed., p1710-1725.

10

10. Appendix 1: Details of statistical analysis of roughness data


10.1 Standard deviation
In order to obtain a number representative of the average height of roughness projection for each data set, the variance of the data was calculated. The square root of this, the standard deviation , is a linear measure of the average deviation of the data from the mean. This deviation may be linked to amplitude by considering a sinusoid, the variance of which is given by 2 = var[a.sin(2x / )] = =
2 1 [a.sin(2x / )] dx 0

a2 2 That is, standard deviation and amplitude are linked by


= a 2

The crest-to-trough height of a sinusoid is equal to twice the amplitude, so the average height of roughness elements obtained via the standard deviation, h, may be given by h = 2a = 2.83 This measure of roughness height, h, is easily calculated from the data sets.

10.2 Mean wavelength


In order to extract from the physical roughness data some measure of average or representative wavelength associated with the roughness, frequency analysis was used. The power spectrum of a set of data, obtained via the Fourier Transform, shows how the variance of the data is distributed with frequency. For each of the data sets obtained, the sample spectrum, as given by equation (6) below, was calculated and plotted.

C XX ( ) = N
where

t= N

x e
t

N 1

2 j 2 t

...........................( 6)

CXX() N t

= = = =

sample spectrum number of data points length of series sampling interval

It is worth noting that the integral of the sample spectrum gives the variance of the data set. In order to find the longitudinal spacing representative of the roughness, we developed the idea of the centroidal frequency, C, which is found by calculating the centroid of the sample spectrum. This is done using:
C = 1 2

. C ( ). ................................(7)
i XX i i =1

which yields the mean wavelength C as: C = 1 C

The above treatment indicates how a set of tunnel roughness data consisting of 2000 points may be represented by one (or both) of two parameters, these being the average roughness height and the centroidal wavelength, or spacing. It should be stressed that the height measure, h, obtained via the variance of a data set is particularly susceptible to outliers or trends in the physical data, because of

11

the squaring of the variation of the individual deviations from the mean. As an alternative to h, another representative roughness projection height was sought.

12

10.3 Mean range


Having already found the representative interval between roughness projections of a rough surface (i.e. C), it was possible to calculate the variation in height within intervals of this length. This sampled variation, called the range ri , is given in a specific interval (xi , xi+c ) by the relation

ri = max i [ xi , xi + c ] min i [ xi , xi +c ]
with i ranging from 1 to N-C. The mean range h obtained for the data set as a whole is then given by
h = 1 N c
N c i =1

This measure of average or representative roughness height is a far more robust estimator than h. It requires more computation, but this is not a problem with modern computing power. In wellbehaved data sets (i.e. no trends or outliers), the values of h and h are very similar, in which case no greater accuracy is afforded by either one. As would be expected of data of the sort being dealt with here, the value of C has a marked influence on the resulting value of h. This is shown in Figure A1. In this figure, various values for the mean wavelength, C , were arbitrarily chosen and plotted against the corresponding resulting values of h.
Effect of Wavelength on Mean range
9

mean range (mm)

0 0 50 100 150 200 250 300 350 400 450 500

wavelength (mm)

FIGURE A1: Effect of C on h. Figure A1 shows the dependence of h on C , and emphasises the importance of maintaining a consistent method for the computation of C. Even though power spectra of certain data sets may exhibit certain dominant frequencies at which the variance is concentrated, calculation of C via the mean range as outlined above will be, on the whole, representative of the particular variance distribution. Also evident from this figure is that h tends to be proportional to the square root of C.

10.4 Equivalent sinusoid


Knowing both the representative wavelength (spacing) and height (twice the amplitude), a set of roughness data may be represented by an equivalent sinusoid of wavelength C and amplitude equal to either h/2 or h/2.

13

Vous aimerez peut-être aussi