Norwegian University of Science and Technology Department of Petroleum Engineering and Applied Geophysics FRICTION FACTOR IN SMOOTH AND ROUGH GAS PIPELINES An Experimental Study Elling Sletfjerding A Dissertation for the Partial Fulfillment of the Requirements for the Degree of Doktor Ingenior Trondheim Jamuary, 1999Abstract Flow of high pressure natural gas in pipelines has been studied experimentally. Pipeline flow of natural gas is characterized by high Reynolds numbers due to the low viscosity and relatively high density of pressurized gas. Friction factor correlations for high Reynolds number flow in smooth and rough pipes were developed. To study the effect of wall roughness on pipe flow at high Reynolds numbers 8 test pipes with different wall roughness were fabricated. The wall roughness in 6 of the test pipes was varied by adding glass beads in an epoxy coating applied on the pipe wall. One test pipe was treated with a smooth epoxy coating and one was left untreated. The inner diameter of the test pipes was 150 mm, Measurements of the pressure drop in the pipes were made in a closed flow loop at line pressures of 25, 70, 95 and 120 bar. The Reynolds number of the flow was varied in the range 2-30 million. ‘The wall roughness of the test pipes was measured with a stylus instrument. Correlations between the directly measured wall roughness and the friction factor at fully rough flow conditions were presented. To characterize the wall roughness of the test pipes a parameter combining a measure of the roughness height (R,) and the texture of the wall roughness (H') was used. Due to the high Reynolds number of the flow, minute irregularities of the pipe wall had significant effect on the friction factor in the pipe. The measured wall roughness of the test pipes was in the range 1.4 < Ry <31 ym. ‘The flow experiments in test pipes was compared with data from operating pipelines in the North Sea. The offshore pipelines are coated with the same epoxy coating as used in the test pipes. The friction factor in coated offshore gas pipelines showed smooth behavior when the additional pressure drop due to welds were accounted for. The study of coated gas pipelines showed that the friction factor was significantly lower than predicted by standard correlations. iiiAcknowledgements I want to address my word of thanks to my supervisor, Professor Jon Steinar Gudmundsson, for giving me the opportunity to engage in this work. I have benefited from his continuous encouragement and guidance throughout my studies. With his enthusiasm and creative ideas he has helped me proceed in my work. Dr. Karl Sjoen in Statoil has given me support and valuable comments during my 3 1/2 years of studies. I am very grateful for his interest and for inviting to a close cooperation with Statoil. 1 am greatly indebted to Statoil for funding my research. I have appreciated Statoil’s support and their efforts to involve me in their gas transport activities. My visits to the gas transport control center at Byggnes have given me insight to the challenges of operating gas pipeline systems. ‘The experimental part of my work was done at Statoil’s K-Lab (Ki ‘Technology Laboratory). Warm thanks to Torleif Pedersen, Roar Haland, Asbjorn Erdal sto Metering and and the staff for their support and cordial welcome during my many visits. The experience I gained during my work at K-Lab has been fundamental both to me as a professional scientist and for my research. I would like to thank my fellow graduate students, the professors and the staff at the Department of Petroleum Engineering and Applied Geophysics. I have benefited from the interesting discussions with the scientists as well as the advice from the workshop at the department. Finally, I would like to thank my family. In particular, Charlotte and Magnus have with their joy encouraged me during these years. And that has made all the difference. vviContents Abstract iii Acknowledgements v Contents vii List of Tables xi List of Figures xiii Nomenclature xv 1 Introduction 1 1.1 Background 1 1.2 Scope of work... . eens 4 13 Organization 5 2 Theory and Literature Review 7 21 Pipe flow 7 22 Friction factor correlations... . an) 23° High Reynolds number flow... . nn) 24 Rough pipes... 2.2.0.0... beeen eee 19 25 Rongh surfaces... 0.0.2... vec e eee BD 2.6 Measurements of surface roughness re) 2.7 Roughness characterization . 4 viiContents 3. Flow Experiments 3.1 3.2 33. 34 Test pipes Coating Instrumentation ‘Measurements . . 3.4.1 Main test period 3.4.2 Second test period Results Error Analysis . 3.6.1 Measurement Uncertainty 3.6.2. Wall tap errors. 4 Pipeline data 41 4.2 43 Capacity tests . Modelling . . . Results 5 Roughness Measurements Bl 52 33. BA 55 Instrument... 2.0.0... 0 6 Measurements . Roughness of test pipes Ronghness of full-scale pipes Comparison with Nikuradse’s rou 6 Correlations 6.1 62 63 64 Rongh test pipes Smooth test pipe New correlations Pipeline data 7 Discussion igh pipes 35 35 36 37 40 40 41 AL 43 43 aT 103 103 107 108 ut 123Contents ix 8 Conclusions 129 References 131 A Published papers 139List of Tables 21 31 32 33 34 35 37 38 39 3.10 31 3.12 3.13 314 3.15 41 42 43 44 45 46 Nikuradse’s measuerements Inner diameter of test pipes Measured distance between pressure taps . Coating of test-sections . . Gas compositions in mol percent, Pipe 1 measurements... Pipe 2 measurements . . See Pipe 3 measurements . . . Pipe 4 measurements . . . Pipe 5 measurements ..... 2... Pipe 6 measurements . . Pipe 7 measurements . . . cee Pipe 8 measurements... oe eee eee Pipe 1 measurements, second test period... 2... 0c v ee eee eee Error estimate of AP measurements Uncertainty in variables, main test period... 0.02. eee Pipeline data in the study. cette eee Reported data, main test periods ©... 0... eee Reported gas compositions... 2... eee Data calculated for the study Calculated Reynolds numbers Sees Calculated outlet pressure 20... ee eee xi 32 74 74 acy cy 16 16List of Tables BL 52 53 5a 5B 5.6 61 62 63 Recommendations of cut-off wavelengths from DIN 4768 a 87 Cut-off and minimum resolved wavelength 87 Roughness of test pipes, measured week 51 1997 . . 87 Roughness of the coated test pipe, measured after the second flow test . 88 Roughness of coated full-scale pipes 88 Hurst exponent of roughness profiles ©... eee 88 Measured roughness of test pipes. 14 Constants in the curve fit of (1/f)'/? versus log(r/k) 14 Constants in the curve fit of k, versus the measured roughnes 14List of Figures 24 22 23 24 25 31 3.2 33 34 35 3.6 37 38 AL 52 53 54 55 5.6 37 Velocity profiles in pipe flow . . Viscous length scale in turbulent pipe flow . . . Coordinate system Nikuradse’s measurements in rough pipes Roughness parameters of test profiles... 0.2.0.0... ee eee ee Static pressure tap . Schematic view of a test pipe Schematic view of the K-Lab loop . Friction factor, all test pipes... 0.0... eee Friction factor pipes 1,234 2.00... 20002000 ee Friction factor pipes 8,5,6,7 Test of AP/Ax in pipes 1-4... neces Test of AP/Ax in pipes 5-8 Friction factor in gas pipelines . . Schematic view of the stylus instrument Roughness of test pipes, Ra values. . . Roughness of test pipes, Ry values Roughness of test pipes, R, values Roughness profiles, coated pipe and steel pipe . . . Roughness profiles, pipes 3 and 4 eee eee Roughness profiles, pipes Sand 5... 0.2 eee eee ee xiii 32 33 34 34 62 62 63 G4 65 66 67 68 7 89 89 90 OL 92 9358 59 5.10 bl 5.12 5.13 5.14 5.15 6.1 62 63 64 65 6.6 67 6.8 List of Figures Roughness profiles, pipes 6 and7 ........ 95 Structure function plot lo... eee 96 Structure function plot 2 97 Power spectral density plot... 00.000 98 Power spectral density plot 2.0... 0.000 ce eee eee 99 Roughness profiles in full-scale pipes 100 Structure function and power spectrum of full-scale pipes... .. 0... 101 Mean bead size versus measured roughness... 0... 2... + 102 Sand-grain roughness of test pipes 115 Correlations between (1/f)!/? and log(r/k) . 6 Correlations from Nikuradse’s work . . uz Correlation of k, versus measured roughness... 0... 20... 000% us Roughness function of the test pipes... 2... eee 9 Friction factor smooth coated pipe... 0... 0... 50% Gooco0 Ii) Roughness function and transition region . 121 Pipeline data corrected for the effect of welds .... <2... eee 22)Nomenclature Pipe cross sectional area Constants Pipe diameter Fractal dimension Diameter of pressure tap Pressure gradient Enthalpy Drag factor Moody friction factor, f = T/3p0? Fanning friction factor, fr = tw/4pU Friction factor at fully rough conditions Power Spectrum Acceleration of gravity Hurst exponent Roughness parameter Roughness height, Effective roughness Roughness length scales Sand-grain roughness Reduced roughness height, k/I* Length Depth of pressure tap ‘Tracing length Weld spacing Viscous length scale, I = v/u, xvNomenclature Gas mole weight mass flow rate Number of samples Constant Pressure Pressure difference Probability density Volumetric flow rate Heat transfer rate Reynolds stress Universal gas constant Inner radius of pipe Autocorrelation function Arithmetic mean roughness Reynolds number, Re = ©? ‘Transition Reynolds number Von Karman number, Re, Root-mean-square roughness ‘Mean peak-to-valley roughnes Structure function ‘Temperature Compressibility factor Profile height Axial velocity Mean (bulk) velocity Center-line velocity Axial fluctuating velocity Friction velocity, u. = /70/p Variance Radial fluctuating velocity Axial coordinate Distance between pressure taps Radial coordinate yf uDNomenclature Greek symbols a Pipeline inclination angle 8 Constant 5 weld height € Height of burr on wall tap ¢ Random number y Gas gravity © Von Karman constant 1 Dynamic viscosity vy Kinematic viscosity, v = 4/p Wavelength Ae Cut-off wavelength w — Radial frequency, w = 2r/A p Density p(€) Normalized autocorrelation o Standard deviation T Shear stress Wall shear stress Shift Mean square Subscripts and superscripts Mean value + Scaling by inner variables 12 Pipeline inlet and outlet conditionsxviiiChapter 1 Introduction Wall friction is the domination source of pressure drop in gas pipelines and the friction factor is a key parameter in determining the transport capacity of the pipelines. The thesis reports the work done during my graduate studies from 1995-98 to investigate the wall friction in gas pipelines. This introduction chapter describes the motivation for the work, the main objectives and the organization of the thesis, 1.1 Background Pipelines are the most viable means of transporting large amounts of gas from the wellhead to the consumer (Adewumi, 1997). The costs of pipeline transport are considerable, both in terms of investments in infrastructure and operational costs. To exploit effectively the transport capacity of the pipelines is of primary importance. ‘The steady state transport capacity of a gas pipeline can be calculated from flow models. ‘The capacity is the amount of gas transported at a given pressure drop and temperature. ‘A simple model for the transport capacity of a horizontal gas pipeline is (Katz and Lee, 12 1. Introduction 1990): (1.1) Equation 1.1 shows that the mass flow rate is related to the pressure difference between, the inlet and outlet [P? — P?], the friction factor f, the pipe geometry (L and D), the temperature and the gas properties. The friction factor is an empirical parameter used to describe the energy loss due to friction between a solid surface and a fluid. Equation 1.1 shows that the mass flow rate is proportional to the friction factor to the minus 1/2 power (rn ~ f~"/?). Thus, the capacity of a pipeline will be larger if the friction factor is low. Likewise the pressure drop (P} — P2) at a given flow rate will be smaller if the friction factor is low. It is well known in the gas industry (Katz and Lee, 1990) that the classical Cole>rook-White friction factor correlation (Colebrook, 1939) gives conservative estimates of the friction factor in a gas pipeline; the overestimation of the friction factor leads to an underestimation of the transport capacity of the pipeline. To operate a gas pipeline efficiently, an accurate knowledge of the capacity is needed. Therefore, good friction factor estimates are crucial. ‘The flow in offshore gas pipelines is characterized by large Reynolds numbers (Re ~ 10") due to the low gas viscosity and the relatively high density at typical operating pressures (100-180 bar). From the Colebrook-White correlation (Colebrook, 1939) it is found that minute irregularities on the pipe wall will have a significant effect on the friction factor at high Reynolds numbers. However, in the experimental work from which the Colebrook-White correlation was developed the maximum Reynolds number was 1 x 10°, one decade lower than values encountered in offshore gas pipelines. The aim of this work has been to investigate the influence of wall roughness on the friction factor at Reynolds numbers typical for gas pipelines. ‘To reduce the friction factor in gas pipelines internal coatings can be applied. Used since the 1950's (Kut, 1975), internal coatings have a pay-back time for the investments of1. Introduction 3 typically 3-5 years due to reduced operating costs (Singh and Samdal, 1988). In Norway 35 % of offshore generated power is used for gas export (OED, 1998), and internal coatings have been used successfully to reduce the costs of gas transport. In coated pipelines it is found that the calculated pressure drop from flow models is lower than the measured pressure drop. The main source of the difference is error in the friction factor correlations used. In this thesis an experimental study of the friction factor in coated test pipes is compared to data from gas pipelines in order to propose a more accurate friction factor correlation for coated pipelines. The topic of friction factors in pipes is still of major interest in the gas industry. Many studies have been made on this subject in gas transmission companies. However, few studies have been published in the open literature. One exception is the American Gas Association (AGA) report (Uhl, 1965) which has been the main reference on the subject of steady state flow in gas pipelines. Since the AGA report was published development in characterization of natural gases, pipe flow theory and measurement techniques have teken place, and a revisit of the topic of pressure drop in gas pipelines is due. ‘The AGA report (Uhl, 1965) used data from onshore gas pipelines in the United States. ‘The operating pressures of the pipelines in the AGA work was lower than 70 bar. Nodern subsea gas pipelines in Norway are operated at pressures up to 200 bar. The diameter of the subsea pipelines in Norway is typically 1 m (40 inches). The AGA work discussed the flow in pipelines of smaller diameters, maximum 0.75 m (30 inches). The intention of this work has been to improve the understanding of steady state flow in large diameter gas pipelines operated at high pressures.4 1, Introduction 1.2 Scope of work This work was aimed at reducing the uncertainty in friction factor calculations for gas pipelines. The main goal was to develop a friction factor correlation for the Reynolds number range experienced in high pressure gas pipelines. Reducing the uncertainty in the friction factor, the transport capacity of the long distance pipelines can be calculated more accurately. A main challenge was to investigate and understand the wall roughness influence on flow in pipes, especially in the high Reynolds number regime (Re > 1 x 10°). In the experimental work the roughness influence on the friction factor was systematically investigated for Reynolds numbers in the range 2 x 10° — 30 x 10°. With the use of modern wall roughness measurement techniques, development of correlations between the directly measured wall roughness and the friction factor was an objective of the study. Such correlations would allow predictions of the pressure drop in pipes from direct measurements of wall roughness.1, Introduction 5 1.3 Organization ‘The thesis is organized in seven main chapters. The contents of the chapters are summarized below. Theory and literature review. The chapter covers the theory and the relevant references used in this work. Two main topics are discussed; flow in pipes including smooth and rough pipes, and measurements and characterization of rough surfaces. Flow experiments. The high pressure flow measurements of natural gas in pipes are described. The experimental work, which aimed at measuring the friction factor in smooth and rough pipes, covered Reynolds numbers in the range 2 x 10° — 30 x 10°. Pipeline data. Data from operating gas pipelines is presented and analyzed. Roughness measurements. The measurements and analysis of wall roughness are discussed. The chapter includes measurements both on test pipes used in the experimental work and on full-scale pipes. Correlations. Friction factor correlations based on experimental data and field data are presented. Discussion. The results from the experimental work and the field data are compared and discussed. Conclusions. This final chapter summarizes the thesis and presents the final conclusions. Appendix. Two published papers are appended.Chapter 2 Theory and Literature Review The purpose of the chapter is to present the theory and the literature which is the basis for the the ‘The chapter is in two main parts. The first part concentrates on flow in pipes and reviews knowledge related to turbulent flow in pipes. The second part. focus on measuring and characterization of roughness and presents results from the literature which is appropriate for use in this thesis. 2.1 Pipe flow The work presented in the thesis concerns steady state flow in natural gas pipelines. The appropriate governing equations are the one-dimensional momentum and energy equations (Gregory et al., 1979). dP pod, dol?) _ Get egsina + Loo ate =0 (2a) ah gal 1dj TUG tasine+ T= 0 (2.2) Equations 2.1 and 2.2 can be derived from the Navier-Stokes equations (Haaland, 1975) or by force balance and the use of the first law of thermodynamies (Massey, 1990) 78 2. Theory and Literature Review From the momentum equation (Equation 2.1) it is seen that the pressure gradient dP/dr in a pipe is a result of three different mechanisms; gravity, friction between the moving fluid and the pipe wall and acceleration of the fluid, However, for normal operating conditions in a gas pipeline the acceleration term is small compared to the friction term. (Haaland, 1975) and can often be neglects The frictional pressure drop in a pipe is directly related to the wall shear stress of the moving fluid. When the flow in a pipe is fully developed (the cross-sectional velocity distribution does not change with distance <), the shear stress profile, r(y), can be found from force balance (Schlichting, 1979): (2.3) (2.4) From the wall shear stress the friction factor can be defined (Schlichting, 1979): Te fas To? (2.5) Schlichting used the term resistance coefficient (A) in his book and it is equal to Moody’s definition (Moody, 1944). Several other definitions are in use for the frietion factor. Among these are the Fanning friction factor fr = x85 1 fie. In the following the Moody friction factor is used and the subscript yy is omitted. In steady state horizontal and fully developed incompressible pipe flow the friction factor will directly give the relationship between the pressure drop and the flow rate: Le es The relation given in Equation 2.6 can be obtained by combining the expression for the wall shear stress (Equation 2.4) with the definition of the friction factor (Equation 2.5). Equation 2.6 is often called the Darcy-Weisbach equation and was first presented by the French engineer Henri Darey who did experiments with pipe flow of water (Massey, 1990).2. Theory and Literature Review 9 The flow in a pipe appears in two physically different regimes. As first investigated by Reynolds (1883) this fundamental difference between the so-called laminar and the turbulent flow was documented. In pipes the flow is laminar for Reynolds numbers below 2000 — 3000 (Schlichting, 1979). For higher Reynolds numbers the flow is turbulent. Laminar flow in pipes is characterized by a parabolic velocity profile. Figure 2.1 shows the mean velocity profiles in laminar and turbulent pipe flow. In turbulent flow the velocity profile is much flatter resulting in larger velocity gradients and shear stress at the wall. The pressure gradient in laminar pipe flow is proportional to the mean velocity. However in turbulent flow the pressure gradient is proportional the mean velocity to a power close to or equal to two. The turbulent velocity profile changes with Reynolds number. The profile gets more "plug”-like as the Reynolds number increases, and the region close to the wall where large velocity gradients are found becomes more narrow. Traditionally, pipe flow is viewed as a special case of a boundary layer where the radius of the pipe is the boundary layer thickness. Close to the wall in smooth pipe flow there is a thin region, the viscous sub-layer, where the viscous actions are dominant and the velocity profile is linear. A velocity scale for this wall region is the friction velocity defined in terms of the wall shear stress and the fluid density: me y= 27) Eliminating the pressure gradient from Equation 2.6 the friction factor can be expressed in tems of the friction velocity and the mean velocity: £ (28) A length scale for the small scale turbulent motion close to the wall is the viscous length scale defined in terms of the friction velocity and the kinematic viscosity: yet (2.9) The viscous length scale decreases as the Reynolds number increases. In a pipe flow the radius of the pipe (the boundary layer thickness) is a measure of the large scale turbulent10 2. Theory and Literature Review motion, Because the radius of the pipe is constant the size difference between the small scales (the viscous length scale) and the large scales increase with Reynolds number. From Equations 2.8 and 2.9 the ratio l,./D (a measure of the spread in length scales) can be expressed with the Von Karman number Re, or in terms of the friction factor: Loy 1_1f D- uD ie wy (2.0) Employing the Blasius friction factor correlation (Equation 2.11) it is found that 1,/D = 5.03Re~/’. In Figure 2.2 the ratio l,/D is plotted using the Zagarola friction factor correlation for smooth pipes (Equation 2.31). For a gas pipeline with a diameter of 1 mand a Reynolds number of 20 x 10° it is found that the viscous length scale is approximately 1.6 um assuming that the flow is hydraulically smooth. ‘The existence of very small scale motion at high Reynolds numbers makes it difficult both to do reliable measurements of turbulent structure and numerical calculations of the turbulent flow. In addition if the wall roughness has a significant influence on the flow (when the wall roughness is comparable in size with the viscous length scale) direct numerical simulations of the flow even in a simple geometry as a pipe is still not. possible. In a recent paper Patel (1998) discusses the difficulties of doing computational fluid dynamics (CFD) in high Reynolds number flow over rough surfaces. Because of these difficulties much of the analysis of high Reynolds number pipe flow still is based on classical empirical work on friction factors and the velocity profiles near solid walls, The next sections will discuss classical and more recent knowledge in this area of turbulent flow. 2.2 Friction factor correlations Blasius (Schlichting, 1979) formulated the resistance formula for smooth pipes in 1911. ‘This formulation (Equation 2.11) was based on Reynolds law of similarity and expressed2. Theory and Literature Review pee the friction factor as an exponential function of the Reynolds number. f= oats (2.11) The relation was based on experiments which were available at the time and the constant was made to fit those. Later it has been shown not to be accurate above Re = 100000 (Schlichting, 1979) Another empirical correlation relevant to the gas industry was the Weymouth frietion factor correlation (Equation 2.12) for gas pipelines (Katz and Lee, 1990). ‘The paper by Weymouth was presented in 1912. 0.00941 f=—pe (2.12) ‘The Weymouth correlation did not include viscosity nor wall roughness and is therefore expected not to be very accurate. ‘The further treatment of the problem of pressure drop in turbulent pipe flow was strongly related to developments in boundary layer theory (Schlichting, 1979). With new experimental works analyzed with the scaling arguments and dimensional analysis of boundary layer theory, new types of correlations were developed in the 1930's. The rest of this chapter will concentrate on these results. It can be shown (Tennekes and Lumley, 1987) that for turbulent pipe flow the appropriate mean flow equation is: = + duo (2.13) Equation 2.13 is derived from the Navier-Stokes equations assuming that the velocity components can be decomposed into a mean flow (time mean) and fluctuating parts (Reynolds decomposition) (Tennekes and Lumley, 1987). The first term in Equation 2.13 represents the so-called Reynolds stress or the turbulent stress, the second term is the viscous stress. The y-coordinate is measured from the wall and outwards. Figure 2.3, shows the coordinate system used.12 2. Theory and Literature Review For laminar flow the Reynolds stress is zero and the mean flow equation can be integrated dP 1 ie = ely) (2.14) in this case is at the center line Uma = ae a, ae Wimax = r®. The friction factor can be found directly from Equation 2.6: (2.15) on = For turbulent flow, the analysis is more complex. Introducing the definition of the friction velocity u2 “/ = —1425, Equation 2.13 can be written (Tennekes and Lumley, 1987). Ttot = 2.16) 5 (2.16) Equation 2.16 shows how the total stress 71. (the sum of Reynolds stress and viscous stress) varies linearly in pipe flow. Direct integration of the momentum equation in turbulent flow is not possible because of the Reynolds stresses which are generally unknown. However in boundary layer theory effort has been made to develop a dimensionally correct form of the friction factor correlation. Together with empirical resulis these scaling argument for the velocity profile in turbulent boundary layers is the basis for the most friction factor correlations. Traditionally, the velocity profile in pipe flow is divided in two regions (Tennekes and Lum‘ey, 1987). In the core region, the flow is not directly influenced by the effect of the wall. The roughness of the wall and the viscosity has negligible effect on the flow compared to the turbulent mixing. The velocity profile depends on y,r and u, only. Presented as a velocity defect law (Tennekes and Lumley, 1987), this leads to (U.1, is the center-line velocity): UW) Ver _ py Swat — Ay ean) The velocity defect law was first presented as a postulate by Von Karman (Hinze, 1975)2, Theory and Literature Review 13 Clese to the wall in the inner region the effect of the viscosity and wall roughness dominates, and the velocity profile depends on y,v,u, and k;, where kj were length scales describing the roughness of the wall (height, length etc.). The variables are non-dimensionalized with the viscous length scale £. The velocity profile in the inner region is called the law of the wall: Uy) _ gy Ky =O pig) = verbs) (2.18) ‘The law of the wall was first presented by Prandtl as a result of flat plate boundary layer analyses (Hinze, 1975) Millikan (1938) assumed that the two regions should overlap in an intermediate region. By matching the two scaling relations (Equations 2.17 and 2.18) using their gradient, Millikan derived the functional form of the law of the wall and the velocity defect law: FO Dany, + 640k) (219) UW)—Uar Ay UU hind) +c, (2.20) ‘The constant « is traditionally called the Von Karman constant. Combining Equations 2.19 and 2.20 the relation between the center line velocity and the friction velocity is obtained (Schlichting, 1979): Ver 1 a zin(Re VF) + C1 + Calis) (2.21) Integrating the velocity defect law (Equation 2.20), the relation between the center line velocity and the mean velocity is (Schlichting, 1979): OT _ Ue, ue Combining this with Equation 2.21 a friction law was presented (Schlichting, 1979). 2 i For a smooth pipe (hi = 0), this relation was called the Pranddl law of flow in smooth +05 (2.22) : In(Re Vf) +O; + Ca(hex) (2.23) pipes. The Prandtl law of friction was first derived by Von Karman and Prandtl based on Von Karman’s similarity theory and the mixing length hypothesis (Schlichting, 1979).14 2. Theory and Literature Review In two extensive experimental works Nikuradse (1932) and (1933) performed experiments measuring velocity profiles and pressure drop in smooth and rough pipes. Using the data from the smooth pipe measurements the constants in Prandtl’s law of friction were adjusted (Schlichting, 1979). iS [f= 20b0tRe VP) ~08 = 20109 ME (2.24) ‘The Prandtl law of friction in smooth pipes showed "excellent (Schlichting, 1979) agreement. with Nikuradse’s experiments which reached a maximum Reynolds number of 34 x 10° In the investigation of flow in rough pipes Nikuradse found that the friction factor was dependent on the ratio of the roughness length scale (diameter of sand grains) to the viscous length scale. Three different regions were defined depending on the ratio j=: 10< 4 <5, f= f(Re) vi 2.5< pte <70, f= f(Re,ks/r) Pte 3. 70< fe, f= f(ks/r) jae In the first region, the roughness of the wall was imbedded in the viscous sub-layer near the wall so no effect of the roughness was felt by the outer flow. The flow was similar in character to smooth pipe flow. In the third region the pressure drop was dominated by the turbulence created by the roughness at the wall and the friction factor was funetion only of the wall roughness. This region was called fully rough pipe flow. The second region was a transition region between the two extremes where both the Reynolds number and the roughness length scales affected the friction factor. AA friction factor formula for the fully rough case was derived by Von Karman (Schlichting, 1979} in a similar manner as shown above for smooth pipe flow. The derivation relies on. the assumption that the flow in the outer layer is independent of the wall roughness and2. Theory and Literature Review 15 the ow in the wall layer is dominated by the wall roughness. Nikuradse (1933) presented a friction factor correlation in rough pipes based on Von Karman’s theoretical work i f Colebrook (1939) presented a friction factor correlation which was valid for all three 2log(--) + 1.74 = 2log(7 aur) (2.25) regions in pipe flow with rough walls. He stated that there were two length scales which would govern the friction factor in rough pipes; the viscous length scale 2 and the roughness length scale k and proposed that a linear combination of the two length scal would describe their joint influence on the flow. For a smooth flow k would equal zero. For a fully rough flow k > & and & would be the dominating length scale. Letting y: = Cik + C22 where C; and C2 are constants, the law of the wall may be written: Uy) uy, ) (2.26) Following a similar derivation as done for the Prandtl law of flow in smooth pipes, the Colebrook-White correlation was presented (Colebrook, 1939): vj ‘The derivation is reviewed in a recent paper by Matthew (1994). The constants in —2log( (2.27) Equation 2.27 was chosen such that in the limit k + 0 the Prandtl law of flow for smooth pipes (Equation 2.24) was obtained, and when ;#4- is large the Nikuradse’s sand grain roughness law (Equation 2.25) was obtained. ‘The Colebrook-White correlation has been widely used in engineering ever since it was published (Katz and Lee, 1990) (Massey, 1990). Together with the graphical chart (based on the Colebrook-White equation) presented by Moody (1944), this correlation has been the main tool for pressure drop calculations in pipes for over 60 years. However, in gas industry the correlation is known to give conservative estimates of the pressure drop in gas pipelines (Katz and Lee, 1990)16 2. Theory and Literature Review It is important to be aware of the underlying assumptions which were the basis for the friction factor correlations. First of all the two layer model for the velocity profile and the overlap of these layers in an intermediate layer was the basis. In addition it was assumed in the derivations (Schlichting, 1979) that the law of the wall (or the velocity defect law) was valid in the whole pipe cross section. The law of the wall is strictly not valid outside the intermediate layer. However, when correlating the theoretical result with Nikuradse’s measurements an accuracy of -+3% was obtained with the Prandtl law of friction for smocth pipes with Reynolds numbers from 10" — 3.2 x 10° (Zagarola, 1996) which was certainly an improvement compared to the empirical Blasius correlation. For many years the main problem with the Prandtl law of friction in smooth pipes and the Colebrook-White correlation has been that they are implicit equations and must be solved by iteration. Several explicit equations has been presented over the years to simplify the calculation of the friction factor. One such correlation was presented by Haaland (1983): k 3.75D tiny (2.28) Sal F y+ ( This correlation is especially suited for gas pipelines if n = 3. With n = 1, the equation is a good approximation to the Colebrook-White correlation (Equation 2.27). 2.3 High Reynolds number flow The friction factor correlations discussed in the last section have had success in a wide range of applications. It has been believed that the friction correlations could be extrapolated to infinite high Reynolds numbers (Schlichting, 1979). However, this statement has been shown to be somewhat optimistic (Purushothaman, 1993). In addition there have been doubts on the accuracy of the measurements of Nikuradse (Zagarola, 1996). Of significant importance both in turbulence theory and in industrial application where2. Theory and Literature Review 17 high Reynolds numbers are found was the discussion on the Reynolds number depeadence of the law of the wall (Hinze, 1962) (Purushothaman, 1993) (Tennekes, 1968}. The asymptotic matching procedure presented by Millikan (1938) resulted in the logarithmic law of the wall (Equation 2.19) where the constants 1 and C; were supposed to be independent of the Reynolds number. Hinze (1962) presented a reevaluation of the experimental results and concluded that the these constant were funetions of the Reynolds number. ‘Tennekees (1968) presented a second order theory for turbulent pipe flow to incorporate the Reynolds number dependence. He concluded that the leading constant in the law of the wall should be of the form 1 ~ C Re, "/, which meant that there would be a constant value only in the limit Re + oo. From the derivations in the last section it is evident that if the constants in the law of the wall is dependent on the Reynolds number the friction factor correlations will be directly influenced. Even more important, the Reynolds number dependence of the constants is an anomaly in the classical boundary layer theory and therefore questions the theoretical fundament for the friction factor correlations. Recently a new approach to incorporate the Reynolds number dependence of the velocity profiles in turbulent pipe flow was presented in two papers by Barenblat (1993) and Barenblatt and Prostokishin (1993). Starting from a power-law type of scaling Barenblatt (1993) introduced the concept of incomplete similarity. In principle Barenblatt stated that in the intermediate region the velocity profile (or the velocity gradient) depends on both viscosity (Reynolds number) and 4. This incomplete similarity assumption led to a velocity profile in terms of a power-law. (y) ot (2.29) uy v where both C and n were functions of the Reynolds number. Barenblatt proposed that /(2in Re) and C = “4488, skin friction relation for smooth pipes resulting the18 2. Theory and Literature Review scaling law was presented (Barenblatt and Prostokishin, 1993): f=8/ure (2.30) 3(V3 + 5n) YO) = Sen. + m+n) The friction relation fitted very well with the experimental results of Nikuradse (1932) in the high Reynolds number range (Purushothaman, 1993) and is a promising theory for the explanation of the Reynolds number dependence of the wall laws. The most recent and a very interesting work on friction factor relations in pipe flow has been done in the new experimental facility at Princeton University (Superpipe). Measurements of velocity profiles and pressure drop were presented for Reynolds numbers in the range 31 x 10° — 35 x 10° (Zagarola, 1996). These measurements are unique to date and were very thoroughly done in a specially designed high pressure facility. Zagarola presented two new friction correlations for smooth pipe flow. One was essentially similar to the Prandtl law of friction in smooth pipes (Equation 2.24) with adjustments made to the constant because of the new experiments. z 1.889 log(Re /f) — 0.3577 (2.31) vi At Re = 3.5 x 10" this relation give a 5.5 % larger friction factor than the Prandtl law of flow in smooth pipes (Equation 2.24). Zagarola claimed that this correlation predicts the friction factor for a smooth pipe within 11.2% in the Reynolds number range 100 x 10% — 35 x 10°. Zagarola also presented a friction relation taking into account: the shape of the near wall velocity profile: 1.872 log(Re y/f) — 0.2555 — es (2.32) Zagarola stated that this correlation will predict the friction factor in a smooth pipe within 41.2% in the Reynold number range 31 x 10% — 35 x 108. The difference between Equation 2.31 and 2.32 is apparent for low Reynolds numbers only (Zagarola, 1996). vi2. Theory and Literature Review 19 ‘There is still disagreement between researchers on the use of a power-law or a logarithmic velocity profile in turbulent pipe flow. Barenblatt et al. (1997 and 1998) and Zagarola et al. (1997) use the same experimental data to support their different views. This discussion is not followed up in this work because the present experimental work does not include measurements of velocity profiles. However the implications of this discussion on the type of friction factor correlations used should be noted. 2.4 Rough pipes In engineering most pipes cannot be considered ideally smooth at high Reynolds numbers (Schii therefore been of significant interes chting, 1979). The investigations of Nikuradse (1933) on flow in rough pipes has to engineers. Nikuradse used sand-grains and * Japanese lacquer” to vary the surface roughness of the pipes. The sand was ordinary building sand, sifted with sieves to get a narrow distribution of sizes. As an example grains of 0.8 mm were obtain by sifting sand with sieves of diameter 0.78 mm and 0.82:nm. In addition "several hundred” sand-grains were measured with a thickness gauge to verify that the arithmetical average was 0.8mm. ‘The test pipes in vertical position were filled with thin lacquer, emptied and left to dry for 30 minutes. Then the pipes were filled with sand of a certain size and then emptied. The pipes were then dried for two to three weeks, filled with lacquer again and dried for another two or three weeks. This last layer of lacquer formed a “direct coating” on the wall, and Nikuradse claimed tha unchanged” (Nikuradse, 1933). : The original form and size of the grains remained In his work Nikuradse used sand-grains and pipes of different diameter to vary the dimensionless parameter £. Table 2.1 lists the different values of pipe diameters, sand-grain diameter and the £-parameter used in Nikuradse’s work. Figure 2.4 shows the friction factor data from Nikuradse’s measurements. The symbols indicate different configuration of pipe diameter and sand-grains which has the same £. According to the20 2. Theory and Literature Review theory of dynamical and geometrical similarity the flow behavior should be the same for different choices of pipe diameter and sand-grains as long the parameter £ was kept constant. Nikuradse’s results verified this theory (Figure 2.4). However the results in Nikuradse’s measurements have been questioned by several authors. In (Zagarola, 1996) a critical review of the smooth pipe measurement by Nikuradse (1932) can be found. The experimental set-up for Nikuradse’s rough pipe experiments were similar to the smooth pipe set-up. Zagarola (1996) noted several inconsistencies in Nikuradse’s experimental set-up and work. Grigson (1984) showed how problems in defining the origin for the logarithmic velocity profile makes it difficult to determine the Von Karman constant from Nikuradse’s measurements. Nevertheless Nikuradse’s sand-grain roughness experiments have been the main reference on flow in rough pipes in over 60 years. In engineering it is common practice (Schlichting, 1979) to quote the equivalent sand grain roughness as a measure of the hydraulic roughness of a surface or a pipe. In this procedure Equation 2.25 is used to "back-track” the equivalent sand-grain roughness from measured friction factors. Charts and tables of equivalent sand grain roughness for various materials are given in handbooks, for example (Blevins, 1992). Several authors (for example (Robertson et al., 1968) (Powe and Townes, 1973)) have investigated turbulent flow in rough pipes with focus on turbulent structure, In the review article ”Rough-wall turbulent boundary layers” Raupach et al. (1991) review the work on flow over rough walls from various disciplines. It is shown in this review that there is strong support for the hypothesis of wall similarity. That is, at high Reynolds numbers the structure of the boundary layer outside the viscous sublayer would have the same structure as a smooth wall boundary layer. Raupach et al. also reviews ‘measurements on flow over roughened surfaces (roughness elements, ribs, meshes ete.).. ‘These types of measurements will not be discussed further here. Warburton (1974) performed measurements in graphite tubes of different, wall roughness. Using compressed air the Reynolds number in the tubes were varied from2. Theory and Literature Review 21 3 x 10*—2.5 x 10°, In his work Warburton used the mean peak-to-valley height measured with a profile meter as a measure of the surface roughness. He claimed that the friction factor in graphit tubes may be predicted from direct measurements of surface roughness. However, no information on the tracing length of his roughness measurements is given which makes it difficult to compare his results with the present work. In the gas industry the American Gas Association report "Steady Flow in Gas Pipelines” (Uhl, 1965) has been a main reference for pipeline hydraulics. In the report results from measurements on operating pipelines and from experiments were reported and friction factor correlations for gas pipelines were presented. The experimental work was from the Monograph "Flow of Natural Gas Through Experimental Pipelines and Transmissions Lines”, (Smith et al., 1956). For smooth pipe flow a correlation similar in form to the Prandtl smooth pipe law (Equation 2.24) was presented: Vz F,2log(Re Vf) — 0.9 = Fy2log (2) (233) The second constant in Equation 2.33 is slightly different from the constant in Equation 2.24. In addition a ”semi-qualitative” drag factor F; which takes into account effects bends, valves and welds was introduced. Guidelines for choosing the drag factor Fy was given in the report. For the fully rough regime Uhl et al. (1965) recommended the Nikuradse correlation (Equation 2.25). However instead of the sand-grain roughness ,, the effective or operative roughness was used: @ + 1.74 = 2log(7.41 (2.34) B R) The effective roughness ke accounted for welds, fittings and bends in addition to frictional effects of wall roughness. These two equations (2.33 and 2.34) presented by Uhl et al. are commonly called the AGA friction factor correlation. Uhl et al. (1965) argued that the Colebrook-White equation modeled the transition between smooth and rough flow badly. With the support of their data from pipelines and the experimental data of Smith et al. (1956) they proposed that an abrupt transition22 ‘Theory and Literature Review would be more aceurate for steel gas pipelines. Practically this means that the largest value of the friction factor obtained from the smooth (Equation 2.33) and rough (Equation 2.34) friction factor correlation was used, i.e. f = max(fe, fr). ‘The transition point between the smooth and rough flow was found from equality between the smooth law (Equation 2.33) and the rough law (Equation 2.34). 8 vie ‘The transition Reynolds number Rey was a function of the rough friction factor f,, the Rey = (raney (2.35) transmission factor Fy and the equivalent roughness ke. 2.5 Rough surfaces An excellent reference for studying measurement and characterization of rough surfaces is the book "Rough Surfaces” by T.R. Thomas (Thomas, 1982). The book includes numerous examples of applications in engineering where the surface roughness of a solid surface is an important parameter. This short literature review is based on Thomas’ book and will focus on characterization of surface roughness. It is meant to give a background to the roughness measures used in this work rather than giving a complete review of the subject of surface roughness. Measurement techniques will only be discussed briefly. The appearance of a surface i strongly influenced by its formative process (Thomas, 1982). In some cases, often under very controlled conditions, the formative processes may create surfaces that have Gaussian (Thomas, 1982) or fractal (Barabasi and Stanley, 1995) height distributions. However, in a steel surface particular fingerprints from milling, turning and honing is found. The resulting height distribution is often more complex and non-Gaussian. A steel pipe as used in natural gas pipelines is made up of a rolled flat plate of a given thickness which is pressed into an O-shape and welded longitudinally. We should expect that the rolling of the plate will have a significant effect on the appearance of the surface,2. Theory and Literature Review 23 The rolling will flatten the peaks in the original surface but leave the valleys untouched. However at very small length scale the surface properties will be independent of the 1g (Thomas, 1982). mackii The surface of a coated steel pipe is quite different from a steel pipe. The coating is applied after sand-blasting of the steel and with a film thickness suitable to cover all the peaks in the steel surface. We can expect that the coating creates a surface that is smoother than the steel surface and has a different character. In his study of the drag losses of painted ship hulls Grigson (1992) describes a good quality painted surface as *imperfectly smooth”. Such a surface is expected to have streamlined roughness clements in contrast to a sand-blasted surface which has a more *peaky” structure. As described in (Barabasi and Stanley, 1995) a surface may appear smooth or rough depending on the observation point. A steel surface may look very smooth and flat from distance but as we approach the surface more texture and details are seen and the surface will appear more and more rough. This means that the length scale of observation or the length scale at which the surface roughness is measured is very important for the result. 2.6 Measurements of surface roughness ‘The stylus instrument is the most commonly used instrument for measuring surface rouginess, National standards on surface roughness are defined in terms of the performance of stylus instruments (Thomas, 1982). Nevertheless, a variety of other techniques may be used to measure the roughness of a surface. After the stylus instrument, the most used instrument is the scanning electron microscope which theoretically has a much better resolution than the stylus instrument. Reference is made to Thomas’ (1982) book for a thorough discussion of other measurement techniques. In principle the stylus instrument works like a gramophone pick-up, a stylus traverses a surface and records the irregularities. Figure 5.1 shows a schematic view of the24 2. Theory and Literature Review equipment used in this study. The vertical range of the stylus instrument is limited by the cynamic range of the pick up. The vertical resolution is often so good that the real limitation is the background noise during measurements. Horizontally, the resolution is limited by the stylus dimensions (tip radius). The horizontal range is limited by the maximum tracing length of the pick-up (Thomas, 1982) ‘The stylus instrument produce comparative measurements (Sander, 1991). Comparable results are only obtained if the measurements are done under identical conditions. One should not expect the stylus instrument to give an exact representation of a surface. An obvious example is that deep valleys in the surface will not be accurately resolved with a stylus instrument (Thomas, 1982) because of the finite size of the stylus tip. In this work profile measurements are used to describe the roughness of the surfaces. A necessary assumption is that the surface is homogenous. A measured profile is a two-dimensional sample of a three-dimensional surface. There exists ways of measuring the roughness of a surface in three dimensions (Thomas, 1982), however these methods are more costly and not suitable for field measurements. Such methods will not be discussed in this work 2.7 Roughness characterization The simplest way to describe the roughness of a surface is by a single height parameter. Assuming that the surface consists of regular repeating features (sand grains of constant, diameter) a single height parameter (grain diameter) would give a good characterization of the roughness of the surface (Thomas, 1982). One such parameter which is commonly used in surface metrology is the peak-to-valley height.. ‘The peak-to-valley measure used in this study is the R., which is defined as the distance between the highest peak and the lowest valley averaged over 5 samples. The parameter is standardized in DIN 4768. By taking the mean over 5 samples the influence of2. Theory and Literature Review 25 exceptional peaks (or valley) in the profile is diminished. In addition the values given of R, in this study are ensemble means over several runs which again will reduce the influence of single extreme peaks. In addition to the peak-to-valley measure which is a pure vertical parameter, measures of the spread around the mean line are often used (Thomas, 1982). The most common such measure is the arithmetical mean roughness R,. In addition to the Ra, the root-mean-square roughness R, is used in this thesis. The definition of the two parameters are (Thomas, 1982): weg L ph L Re |d: (2.36) 2 ae In the definitions the profile height z is the departure from the mean line and L is the R, # sample length. Note that for a zero mean profile the Ry is identical with the standard deviation of the profile The parameters may also be found from the probability density function of profile heights (2): (2.37) 2p(z)dr The probabi (Bendat and Piersol, 1986) were o is the standard deviation. Assuming that the surface density function for a Gaussian distribution is p(z) = J exp( 3) has 2 Gaussian height distribution the arithmetical mean roughness, Rs can be written (Thomas, 1982): 0 0.80 = 0.8Ry (2.38) Even though the parameter R, is perhaps the most used roughness parameter it is a vertical parameter and it gives little information on profile shape (Sander, 1991) in the26 2. Theory and Literature Review horizontal direction. As an example the two profiles in Figure 2.5 will produce the same value of R, even though the profile a appear more rough” (the example is taken from ‘Thomas’ (1982) book). The Ry parameter is also a true amplitude parameter but will put more weight to peaks and valleys in the profile. Referring again to Figure 2.5 it is seen that Ry capture the removal of peaks better than Ra. The variation of the surface roughness in the horizontal direction is poorly described by R,,Ry and R,. To describe the variations in the horizontal direction there are many possible choices. According to (Thomas, 1982) the different methods mostly reduce to cither measures of the peak density (peaks per unit length) or the zero-crossing density (crossings of the mean line per unit length). These parameters are generally very sensitive to the tracing length in the measurements (Thomas, 1982) and are not used in this study. Instead methods from random-process theory are used to describe the spatial variation. According to (Thomas, 1982) the use of random process theory to describe surface roughness originates from theoretical studies of ocean surfaces by Lounget-Higgins in the 1950's. In his work Lounget-Higgins treated the surface height, gradients and curvatures as random time dependent variables. In working with solid surfaces the variables are time independent and the analysis is simplified. In the analysis the profile height 2(z) is treated as a random variable (Thomas, 1982). At a point ©; the profile height z(z1) has a probability density p(2(c1)) Common assumptions in random process theory is that the variables are stationary and ergodic. Stationary means invariance under arbitrary translations of the x-axis, and ergodicity means that p(2(11)) is independent of x, and that p(z(v1), 2(x2)) is only dependent on |r; —:ra| (Thomas, 1982). For stationary ergodic random variables the statistical expectation of a variable at a fixed point is equal to the space average of the variable (Bendat and Piersol, 1986). Thus the ensemble average is equal to the space average: i< if AS ine 201) = Jim + f 2(x)de (2.39) N is the number of realizations or samples and L is the sample length.2. Theory and Literature Review 27 Strictly speaking, the samples obtained from measurements of a real surface are not stationary. It is observed (Thomas, 1982) that for example the R, value (the standard deviation) is dependent on the sample length used in the measurement. However given that the sample length is fixed the measured profile may be analyzed as stationary. This type of non-stationary is described as "stationary within an interval” (Bendat and Piersol, 1986) The autocorrelation function is important in the characterization of surface roughness. Tt an be defined as (Thomas, 1982) (x)2(x + €)) = if 2(a)z(a + 8)de (2.40) where L is the sample length. The value of R(€) at zero shift is the mean square of the sample, that is J = R(0). Note that for a zero-mean variable the standard deviation equals the mean square. ‘The mean line of the measured roughness profiles are removed (0). A normalized autocorrelation function is defined by normalizing with the variance: from the data before processing. Thus in this analysis 0? p(€) = R(E)/R(0) = R(S)/o? (2.41) In Thomas’ (1982) book it is described how the decay rate of the autocorrelation function may be used to characterize spatial structure An alternative to the autocorrelation function is the structure function (Sayles and ‘Thomas, 1977): 5) = Biel) -ae+ol=+ f * (ela) - oe + ))Pae (242) hs The expression describes the mean square difference in height over a distance €, In some references, for example (Feder, 1988), the structure function is called the variance of increments. ‘The advantage with the structure function in comparison to the autororrelation function is that the relation is easily calculated, not limited to the stationary case and independent of the choice of mean line (Thomas, 1982). It is shown in (Sayles and Thomas, 1977) that for stationary (zero-mean) data 5(6) 2fo? — R()] = 207[1 — p(€)] (2.43)28 2. Theory and Literature Review ‘The power spectrum is defined (Bendat and Piersol, 1986): fF Rie oontosnas (2.44) 5 re) = [ae where R(é) is the autocorrelation function and w = 2/A the radial frequency. Usually (uf) ds (2.45) the power spectrum is calculated from the profile data directly by fast Fourier transform (FFT) algorithms. The power spectrum is the representation of the autocorrelation function in Fourier space and therefore includes the same information (Thomas, 1982). From Equation 2.44 it can be shown that the variance is found by integrating the power spectrum (Bendat and Piersol, 1986). o = A) = f° G(ode (2.46) For & measured roughness profile the lower and upper limit in the integral in Equation 1 /L is the lowest frequency measured. The power spectrum for a zero mean sample can be 2.46 is wy and wy respectively where wy is the Nyquist frequency and wo interpreted as the rate of change of the variance with frequency (Bendat and Piersol, 1986). As discussed above, rough surfaces are observed to be non-stationary. The variance of measured profiles are generally dependent on the sample length used. The power spectrum calculated from roughness measurements can be represented by a power law with an exponent close to —2, that is G(w) ~ 2 (Sayles and Thomas, 1978). This form of the spectra indicates that a sample of finite length will not completely represent. the surface (Sayles and Thomas, 1978). The energy in the long wavelengths dominates the shorter wavelengths. Thus the variance of the sample will be dominated by the longest wavelengths. Increasing the sample length the variance will therefore increase. A non-stationary model of surface roughness was presented by Sayles and Thomas (1978). It was shown that by assuming that the variance is linear dependent on the sample length ie. o? ~ L, the power spectrum of the form G(w) = 2rk;/w® may be2. Theory and Literature Review 29 derived. They claimed that a variety of surfaces could be characterized by this power spectrum and that the topothesy ky, a constant with dimension length, would uniquely define their statistical geometry. The aon-stationary model used by Sayles and Thomas (1978) is based on the theory of Brownian motion, Brownian motion, or random walk as it is often called in two dimensions, can be realized by letting a ” particle” move stepwise along a horizontal axis (x) and letting the height difference (Az = 2(;;1) — 2(#,)) be a Gaussian random variable. As explained in (Feder, 1988) the difference in height between two points will be independent of the difference in height of two other points. However, the actual height at one point is not independent of the height in another point. It can be shown (Feder, 1988} that for Brownian motion in two dimension the expected value of the particle position is zero, but its variance is proportional to the sample size |e; — 2 In Brownian motion (Feder, 1988) the height above the mean line at a given « is: 2(z) — 2(a0) ~la—aol"¥ > a9 (2.47) where H = 1/2. The random function 2(z) may be found by choosing a random number ¢ from a Gaussian distribution, multiplying it with the distance |x — ol! and finally adding the reference height. 2(9). Also, Brownian motion is invariant if the horizontal scale is multiplied by a factor @ and the vertical scale z is multiplied by a factor 3”. This scaling property is called self-affinity. As explained in (Feder, 1988) Mandelbrot introduced the concept of fractional Brownian motion as a generalization of the Brownian motion, The random funetion 2(z) in Equation 2.47 is generalized by letting the exponent H (the Hurst exponent) be a real number in the range 0 < H <1. The Brownian motion with H = 1/2 is a special case and has the unique property of independent increments, whereas for H # 1/2 the increments are statistically dependent. It is shown (Feder, 1988) that fractal (or30 ‘Theory and Literature Review fractional) Brownian motion has the following statistical properties: Elz(z) — z(xo)] =0 (2.48) V(x a0) = El(z() ~ 2(¢0))?7] ~ [a — aol?” i.e. the expected value is zero and the variance of increments is proportional to the horizontal distance. With this results it is seen that the variance of both ordinary and fractional Brownian motion is increasing with the sample length, After presenting their scaling law for the power spectrum of rough surfaces, Thomas and Sayles (1978) were met with arguments by Berry and Hannay (1978) that a more general model of the power spectrum G(w) ~ w"" where the exponent n is in the range 1 een thks =60 ase 8 aoe tke =252, euenenense ae Te mgaeocitt Ik =507 ‘Symbols indicate r/ks configurations friction factor, f 8 10" 10° 10° 10! 10 Reynolds number Figure 2.4: Nikuradse’s measurements in rough pipes a) Ra-0.25, Rq=0.41 b) Ra-0.25, Rq=0.32 Figure 2.5: Roughness parameters of test profiles 34Chapter 3 Flow Experiments A major task in this work has been flow experiments to determine the effect of roughness on the friction factor in coated gas pipes. This work was done at Statoil’s K-Lab (Karsto Metering and Technology Laboratory) in two occasions. The first and major survey was done in October-November 1997 and included measurements in 8 test pipes. In September 1998 another survey was done. This chapter describes the experimental set-up, the measurements and the results obtained from the flow measurements. 3.1 Test pipes The flow measurements were performed using 8 specially designed test pipes. They consisted of 6 m long honed carbon steel tubing with flanges, static pressure taps and one 1/2 inch instrument connection. The test pipes were designed for a maximum pressure of 150 bar in the temperature range from —46 °C to 105 °C. All specifications were according to Statoil’s standard for high pressure pipeline carrying hydrocarbons (Drag, 1997). ‘The test pipes were made of STRUCTO 525 CDS tubing with inner diameter 150 mm 3536 3. Flow Experiments and outer diameter 170 mm. The tolerance of the inner diameter was +0.1 mm as delivered from the STRUCTO AB in Sweden. The pipes were fitted with class 1500 bed in the ASME/ANSI lapped flanges and stud ends with ring joint facings as desct B16.5 -1988 standard (ASME, 1988) In each pipe six pressure taps were made consisting of a socket with 1/4inch NPT. threads for instrument connection and a drilled 4 mm hole in the pipe wall. The design of the pressure taps is seen in Figure 3.1. ‘The pressure taps were placed 1 m apart in the same azimuthal position on the pipe wall. The first pressure tap was located 0.5 m from the upstream end of the pipe. In Figure 3.2 a schematic view of a test pipe is shown. In the downstream end of the test pipes an instrument connection with 1/2inch NPT threads and a 12.5 mm drilled hole were made. The design of this connection was similar to the pressure taps except for the dimension, This connection point was rotated 60° in comparison to the pressure taps and placed 0.25 m from the downstream end of the pipe. After welding and drilling pressure taps and instrument connection, the pipes were honed to ensure high quality tolerance of the inner diameter. The test pipes as used in the flow measurements had an inner diameter tolerance of +0.1 mm. The test pipes were pressure tested with water up to 150 bar before flow measurements. Welding, assembling and honing were done by Cylinderservice A/S in Rissa, Norway. ‘The inner diameter of the test sections was measured before and after coating. The measured inner diameters of the test sections are tabulated in Table 3.1. The distance between the pressure taps was measured after the flow tests. The measured distances for all test sections are shown in Table 3.2. 3.2 Coating ‘The inside of the test pipes was treated differently for all 8 pipes. One pipe was left untreated and one was coated with a two-component epoxy coating commonly used in3. Flow Experiments 37 natural gas pipelines (COPON EP2306 H.F.). The remaining 6 pipes were coated with a mixture of the same epoxy coating and glass beads with different diameter distribution. The siz Table 3 and volume concentration of glass beads used in the coating are tabulated in . Before coating the pipes were sandblasted to SA 2.5. ‘The coating was applied by a high speed rotating nozzle which was pulled through the pipe with constant speed by a hydraulic winch. The nozzle was fed with coating at constant rate by a membrane pump. For details about the coating procedure see (Sletfjerding, 1997). During coating all instrument connections were plugged to avoid blasting sand and coating to enter the holes. After coating the plugs were removed and the holes were inspected with video equipment and cleaned if necessary. The quality of the coating was also inspected both visually and with video equipment and was found to be homogeneous. 3.3 Instrumentation ‘The flow measurements were done at Statoil’s K-Lab. K-lab is a ISO-certified laboratory for calibration of flow meters and has a high pressure closed flow loop designed for measurements and calibration of equipment for the natural gas industry. A schematic view of the flow loop at K-Lab is shown in Figure 3.3 The 6 inch test section in the K-Lab flow loop was used in the measurements. The flow loop was fed with dry natural gas from the Karst terminal. The gas was circulated in the loop by a centrifugal compressor. Temperature control was ensured with heat exchangers, and the mass flow was measured with sonic nozzles. ‘The test pipes were assembled in the 6 inch section of the flow loop. Three test pipes were put after one another. Preceding the first test pipe there was an 18 m long straight section with 6 inch piping. A K-Lab designed flow conditioner was placed at the inlet of this straight section. The flow conditioner was designed to ensure fully developed flow38 3. Flow Experiments conditions after 15 pipe diameters (Erdal, 1997). With the 18 m long 6 inch section (© 108 D) after the flow conditioner, the flow was assumed to be fully developed in the test pip: . Because of the small diameter change between the 6 inch piping (D = 138 mm) and the 150 mm test pipes, the first of the three test pipes was not used for measurements. Each test section was instrumented as shown in Figure 3.2. The static pressure drop was measured with the high pressure point 1.5 m from the pipe inlet and the low pressure peint 2,3 or 4 m downstream of the high pressure point. A special designed manifold with valves and steel tubing made it possible to choose the length between the high and low pressure points. The manifold was operated manually. Flexible tubing connected the high and low pressure points with the differential pressure transducers. ‘The absolute pressure was measured in the third pressure tap located 2.5 m from the pipe entrance. The temperature was measured in the 1/2 inch instrument connection 0.25 m from the outlet of the pipe. The temperature probe was located downstream all pressure taps to avoid disturbances from the probe on the pressure readings. ‘Three differential pressure transducers were used in the measurements. A Paroscientific Digiquartz differential pressure transducer was used for low pressure (P ~ 20 bar) and medium high pressure (P ~ 70 bar). This transducer could not be used for high pressure tests (P © 120 bar), because it had a pressure limit of 1200 psi (= 83 bar). The transducer was the model 5303D — 101 with a differential pressure range of 0 — 3 psi (0-207 mbar). The uncertainty of the transducer was -+0.01% of full scale reading, that is +0.0207 mbar. The transducer was connected by a fiber optic cable to a Paroscientific 790 series fiber optic control unit. The output of the control unit was logged on a computer. For all three pressure regimes (low, medium, and high pressure), a Bailey differential transducer PTSDDC was used. The transducer had a differential pressure range of 300 mbar and had an uncertainty of 0.1% of full scale reading. In addition a Rosemount. 3051 CD differential pressure transducer was used in all pressure regimes. This3. Flow Experiments 39 transducer had a pressure range of 0 — 50 mbar and an uncertainty of 0.1% of upper range limit. Both the Bailey and the Rosemount transducers had digital output and were connected to the K-lab Scientific Data Acquisition System (SDAS) as well as a separate computer for logging. ‘The absolute pressure transducer was a Honewell smart pressure transducer with range 0 — 200 bar. The uncertainty of the transducer was 40.1 % of full scale reading (200 bar). The temperature transducer was a PT100 3 mm element mounted in a” pocket” in the 1/2 in instrument connection. The uncertainty of the temperature measurement was £0.2°C. Both the temperature and the pressure transducers had digital output. and were connected to K-lab’s data acquisition system SDAS for logging. ‘The mass flow was measured with reference sonic nozzles. The SDAS logging system was used for sampling of the differential pressure over the nozzles and calculation of the mass flow in the test pipes. The uncertainty in the mass flow measurements in the K-lab loop was £0.3.%. After loading of gas in the loop a sample of the gas was taken and analyzed in a gas chromatograph. Seven different gas compositions were used during the tests. In Table 3.4 the gas compositions are listed. There were only slight variations in the gas compositions. The density and the viscosity in the test section were caleulated from the measured gas composition, temperature and pressure. The density was calculated with the AGA-8 equation of state (Starling, 1992) and the viscosity with a correlation developed by Churg et al. (Chung et al., 1988). During a run the mass flow, temperature and absolute pressure were logged through the SDAS system. The actual logging time was dependent on the mass flow rate because the use of several sonic nozzles would increase the calculation time of the SDAS system. The logging time was the time it took to iterate 100 samples of the mass flow, typically 1-2 minutes. The differential pressure transducers were logged simultaneously to a separate computer.40 3. Flow Experiments 3.4 Measurements 3.4.1 Main test period The measurements were performed in October and November 1997. The test pipes were assembled three in a row in the 6 inch test section of the K-lab flow loop. Measurements were made in the last two test pipes and the first test pipe was used as flow stabilizer. To measure in all 8 test pipes at different line pressures it was necessary to assemble new test pipes in the flow loop several times. All eight pipes were tested at a line pressure of 70 bar with 10 different flow rates. The flow rate was varied by opening and closing different sonic nozzles. The pressure drop was measured over 4 m for every flow rate. In addition the pressure drop was measured over 2 and 3 m for one flow rate for each pipe (two flow rates for test pipe 7). The absolute temperature, pressure, pressure drop and the mass flow rate were measured simultaneously. ‘Two of the test pipes (pipe 1 and pipe 8) were tested with a line pressure of 25 bar. For each pipe 5 different flow rates were used and the absolute temperature, pressure, pressure drop and the mass flow rate were measured ‘Measurements were also made at a line pressure of 120 bar. However after these measurements there were found a film of sulphur in the sonic nozzles. The sulphur film in the nozzles had also earlier been noticed after high pressure measurements. The presence of such a film may lead to biased estimates of the mass rate and the results from these tests were discarded.3. Flow Experiments 41 3.4.2 Second test period Measurements at high line pressures were crucial to obtain Reynolds numbers comparable with operating conditions. Therefore a second survey was made concentrating on high pressure measurements in the test pipe with plain coating (pipe 1). This test pipe was the most interesting for comparison with operational data. ‘The set-up and instrumentation were identical with the main test period except that a turbine meter was used as a mass flow meter in addition to the sonic nozzles. The turbine meter was a 8 in Instrumet type SM-RI-E and a constant calibration factor was used for all measurements. The turbine meter was mounted downstream the test pipes after a 6-8 inch diffuser and a flow conditioner. A Bailey differential transducer PTSDDC was used for the pressure drop measurements. The transducer had a differential pressure range of 300 mbar and an uncertainty of 0.1% of full scale reading. ‘Three test pipes were assembled in the flow loop downstream the straight section. The two first pipes were used as flow stabilizers and measurements were made in the last pipe (pipe 1) at line pressures of 95 bar and 120 bar. The wall roughness was also measured after the flow test to check if there was any change in the roughness of the coated pipe wall. 3.5 Results ‘The results from the measurements in the main test period are shown in tabulated form in Tables 3.5, 3.6, 3.7, 3.8, 3.9, 3.10, 3.11 and 3.12 for the test pipes 1-8 respectively. The ‘quantities listed are the mass flow rate measured by reference sonic nozzles, the absolute pressure and temperature in the test section, the density and viscosity in the test section calculated from the gas composition data, the nominal distance between the pressure taps and the measured pressure difference with standard deviation. The pressure differences listed in the tables were measured with the Paroscientific transducer.42 3. Flow Experiments ‘The results from the second test period are tabulated in Table 3.13. The quantities are the mass flow measured with the sonic nozzles, the absolute pressure and temperature in the test section, the calculated viscosity and density, the distance between the pressure taps and the pressure difference with standard deviation. The gas used in this ‘measurements had a mol weight of 18.88 kg/kmol. In addition to the sonic nozzles measurements of the mass flow was also made with a turbine meter in the second test period. However the nozzle readings were preferred because no signs of sulphur were found in the nozzles after the measurements. The readings from the turbine meter and the sonic nozzles differed with a maximum of 0.94 % at the highest flow rates. The turbine meter gave higher values of the mass flow rate than the sonic nozzles. Due to the noise to signal ratio in the pressure drop measurements it was difficult to get reliable readings at low Reynolds numbers. At the lowest Reynolds numbers the pressure drops were so low (~ 1 mbar) that the noise to signal ratio was significant. Also, the accuracies of the differential pressure transducers were not good at these low readings. Some of the lowest readings of the pressure drop were therefore discarded from the data set. See the error analysis section for a more detailed discussion. ‘The friction factor was calculated from the momentum equation for pipe flow (Equation 2.1) neglecting the gravity term. AP 1 AP 1 pada Be 190" + A(pU* Ag 0 (3.1) Because the pressure drop measurements were made over a short distance the flow was assumed to be isothermal at the measured temperature T. The compressibility factor Z was assumed to be constant over the distance Az (distance between pressure taps). Using the real gas law p = ¥2° the momentum equation can be rearranged (see for example (Smith et al., 1956)): 2 a2 A Fappar + 5 Mas AP 9 (3.2) Equation 3.2 was used to calculate the friction factor for each measurement series. The3. Flow Experiments 43 Reynolds number was calculated from the definition: and the viscosity was calculated for each point from the Chung et al. (1988) correlation. Normally in gas pipe flow the last term (kinetic energy or acceleration) in Equation 3.1 may be neglected without any significant loss of accuracy. The term is only significant if the Mach-number of the flow is large. In the measurements the Mach number never exceeded 0.08. If the kinetic energy term was neglected an error of maximum 0.9 % (at maximum flow speed) were introduced in the friction factor estimates. ‘Therefor the kinetic energy term was included in the calculation of the friction factor. ‘The results for all measurements are shown in Figure 3.4. In Figure 3.5 the four smoothest pipes are compared and in Figure 3.6 the four roughest pipes are compared. ‘The results from the measurements of the pressure gradient over 2, 3 and 4 meter is shown in Figures 3.7 and 3.8. For test section 7 the pressure gradient test over 2,3 and 4 meter were made for two flow rates. The straight lines are least square fitted to the three points and the slope of the straight line is the linearized pressure gradient in the pipe. 3.6 Error Analysis 3.6.1 Measurement Uncertainty ‘The friction factor was calculated from Equation 3.2. However in the error analysis the kinetic energy term in Equation 3.1 was neglected (see the Results section for the discussion of the size of this term). The friction factor can be expressed in measured quantities: AP/Ar 300? DP (33)44 3. Flow Experiments ‘Thus, the estimate of the friction factor was a function of AP, Ac, p, ra and D. To estimate the total uncertainty in the friction factors the errors in the measured variables were assumed to have normal distribution and to be statistically independent. ‘The ‘otal uncertainty in the friction factor was estimated (Doebelin, 1990): (3.4) where the ;’s are measured variables and the A(z;)’s are the uncertainties in each variable. Combining Equation 3.3 and Equation 3.4 the following relation was obtained: [(Binstan)'s (Aeeaan)'+ (So%ea0) en + (og2Racm) + (gE a0) ] ‘This relationship can be simplified if it is assumed that all uncertainties are given as a fraction of the variable, that is Ap = p x e(p) wl e(p) is the uncertainty in percent. ae = V(C(AP))? + (e(Az))* + (e(e))* + Relrin))* + Ge(D))” (8.6) The variances of the measured quantities must be estimated and the expected variance of the friction factor calculated for each measuring point. To estimate the 95% confidence interval of the friction factor A(x,) is chosen to +2std(z;). In the same manner as above the uncertainty in the Reynolds number was: carrey? = (4)*| (Laon) + (a) + (aw) ] — @n ® |\Dp Du De 7 ARO) — Meciny + (DY + (8) ‘The uncertainty in the pressure drop readings were estimated taking into account the uncertainty in the transducers, the errors due to the measurement set-up and the3, Flow Experiments 45 fiuetuations seen in the recorded signals. For the Parosscientific differential transducer the calibration certificate gave the uncertainty of the transducer as 0.01 % of full scale (207 mbar) For a reading of 1.5 mbar (lower readings were discarded) the worst case uncertainty was therefore 1.4 %. ‘To estimate the error in the experimental set-up the readings of the pressure drop over 2,3 and 4 m were used. The pressure gradient calculated over 4 m was compared to a least square fit of the 2,3 and 4 m readings. As seen in the Table 3.14 the deviation from the least square fitted AP/Azr and the reading over 4 m are generally lower than 2.1%. However, for the test section 5 the deviation was larger (3.8 %). The explanation of this large deviation may be undetected burrs in a pressure tap or a result of corrosion in the pressure taps. For testing the variations in the signals from the pressure transducers the following procedure was used. The time signals recorded were divided in 10 segments and the mean values and standard deviations of each segment were calculated for each segment and compared with the statisties for the whole signal. This test showed that the scatter was below 1 % with the exception of the measurements where the pressure drop was very low (below 1.5 mbar). From these considerations of the sources of error in the pressure drop measurements the total uncertainty was estimated to be within 2.5% of the reading for all pipes with the exception of test pipe 5. Due to the large variations in the AP/Az value at. 2,3 and 4m for this pipe the uncertainty in the pressure drop readings was estimated to 4%. ‘The uncertainty in the mass flow measurements using sonic nozzles was +0.3 % according to K-Lab documentation (K-Lab, 1997). The uncertainty value refers to +2std(rn). For the turbine meter used in the second test period the uncertainty in the mass flow readings was +0.5 % (K-Lab, 1997). ‘The density was calculated from the real gas law p = #2 given the gas composition, the pressure and the temperature. The compressibility factor was calculated using the46 3. Flow Experiments AGA-8 correlation which according to the manual (Starling, 1992) has an uncertainty in the region of interest of +40.1 %. ‘The gas composition was measured with a gas chromatograph and the uncertainty in the methane content was +0.15 %. The pressure was measured in the test section with a transducer with an uncertainty of -£0.1 % of full scale reading (200 bar) which correspond to a worst case uncertainty of +0.8 % for the measurements at low pressure (25 bar). The temperature was measured with a PT100 element and the uncertainty was +0.2 °C that is less than +0.1 %. The resulting uncertainty in the density was estimated to be better than +1 %. The viscosity was calculated from the correlation developed by Chung et al. (Chung et al, 1988) given the gas composition, the pressure and the temperature. The uncertainty in the viscosity estimation by the Chung et al. correlation is given to +1.5 % for single component: gases. For the light natural gases used in the flow experiments the uncertainty in the viscosity estimation will be larger. No relevant uncertainty value for dry natural gas viscosity is given in (Chung et al., 1988). However as noted in (Mecain, 1990) the estimated dry gas viscosity at moderate pressures and temperature from modern correlations is within 2 % of experimental values. Therefor the uncertainty in the calculated viscosity was assumed to be +2 %. ‘The uncertainty in the inner diameter of the test pipes was given to +0.1 mm from the manufacture. After coating, the uncertainty in the measured inner diameter was assumed to be smaller than 40.15 mm, that is £0.1%. The lengths between the pressure taps were measured on each test pipe and the uncertainty in these measurement was +1 mm. or 0.05 % (worst case). In Table 3.15 the estimated uncertainties in each variable are listed together with the worst case uncertainty in the friction factor and in the Reynolds number. The uncertainty estimates for the friction factor and the Reynolds numbers are calculated from Equations 3.5 and 3.8. The total uncertainty in the friction factor was estimated to 2.8 % for all pipes with the exception of test pipe 5 which had an uncertainty of 4.2 %. The total uncertainty in the calculated Reynolds number was estimated to 2.0 %.3. Flow Experiments 47 3.6.2 Wall tap errors Using wall taps the measured static pressure is expected to be different from the true static pressure due to the presence of the tap (Chue, 1975). A hole of finite size in the pipe wall will disturb the flow in its immediate vicinity. In addition the presence of burrs on the edge of the hole will disturb the flow. In the measurements the errors due to the hole size were expected to influence the absolute pressure measurements only. The measured pressure drop should be independent of the hole size as long as the taps were uniform. However any burrs on the edge of the pressure taps would influence both the absolute and pressure drop measurements. According to Shaw (Shaw, 1959) the errors in the static pressure due to a finite hole size may be estimated as: (3.9) where ¢(P) is the error, 7 is the wall shear stress, d is the hole diameter, v is the kinematic viscosity, p is the density and d* is the non-dimensional hole diameter (scaled by inner variables). Shaw showed that for pressure holes with I/d ratio of 1.5 to 6 (in the test pipes the hole depth ! = 10 mm and d= 4 mm, t/d = 2.5) the pressure error reach an asymptotic value of 2% = 2.75 for d* > 750. However, the maximum d* in Shaw's work was 800. In (Benedict, 1984) it described by the relation {2 = 0,269(d*)3 ‘as shown that the static tap error could be In our measurements the d* was in the range 2100 — 27000 and the maximum error due to the hole size was estimated as“£) ~ 10. The error was never larger than 0.01 % of the measured absolute pressure and was therefore neglected in comparison to the uncertainty of the transducer. The error analysis was based on flow in smooth pipes. The errors in rough pipes are expected to be of the same order of magnitude. Benedict (Benedict, 1984) reported that the static tap errors in rough channels were one half to two thirds of the errors found in smooth channels.48 3. Flow Experiments ‘The error due to burrs or imperfections of the pressure holes may cause large errors in the static pressure readings. Shaw (Shaw, 1959) showed that for an ¢/d ratio of 1/31.7 (€ is she burr height and d is the hole diameter) the error measured was e(P) = 87. for a d* ~ 300. For a 4 mm pressure tap a ¢/d ratio of 1/31.7 correspond to € = 126 jm. His measurements indicate that the error increases for increasing d*. For the absolute pressure measurements the errors due to burrs were neglected for two reesons. First, even if the errors were one order of magnitude larger than the errors due to the finite holes size they would be small compared to the uncertainty of the transducers. Secondly, in preparing the test sections for the measurements all pressure taps were inspected with video equipment and cleaned if necessary. ‘The presence of burrs on the edge of the taps would have an effect on the pressure drop readings. Taking into account that the pressure drop was measured mainly over 4 m (© 27D) the error due to a potential burr on the second pressure tap may be estimated. ‘The wall shear stress in fully developed pipe flow is related to the pressure gradient as: tw =e (3.10) ure drop was AP = AP’ + e(P;) where AP’ was the true pressure drop and e(,) was the error due to a burr at pressure tap 2. From Assuming that the measured pres Equation 3.10: AP = ~(AP' + e(P,)) = en 107 Tw If the error due to burr is assumed to be e(P;) = 87 (referring again to Shaw's work (Shaw, 1959) and ¢/d = 1/31.7) then AP/AP’ = 107/115 = 0.93. A burr on the first tap will cause a similar error (7 %). To avoid and control burr error the pressure taps were carefully examined and cleaned after coating. In addition the measurements of the pressure gradient at 2, 3 and 4 m gave a check of the linearity and constancy of the pressure gradient. As seen in Table 3.14 the3. Flow Experiments 49 difference between the pressure gradient measured at 4 m and the value calculated from ‘measurements on 2,3 and 4 m is generally about 2 % or less with the exception of the test pipe 5 which shows a deviation of nearly 4%.50 3. Flow Experiments Table 3.1: Inner diameter of test pipes Pipe | D (uncoated) [mm] | D (coated) [mm] 1 150.08 150.00 2 150.10 150.10. 3 150.13 150.11 4 150.11 150.04 5 150.06 149.92 6 150.07 149.90 7 150.08 149.83 8 150.05 149.95 ‘Table 3.2: Measured distance between pressure taps Pipe | Tap 2-4 [mm] | Tap 2-5 [mm] | Tap 2-6 [mm] 1 | 2000.5 3000.0 4002.5 2 2002.0 3000.5 4002.5 3 | 2002.0 3001.0 4002.0 4 | 2000.5 3001.5 4002.0 5 | 1999.5, 2998.0 3999.5 6 | 2002.0 3002.0 4002.0 7 | 2001.0 3002.0 4003.0 8 | 2000.0 3001.5 4001.53. Flow Experiments Table 3.3: Coating of test-sections Pipe | Coating | Glass-bead size | Volum coating/beads 1 | yes - - 2 | no : - 3 | yes 0 — 50m Qe1 4 |yes 40 — 70 um 2:1 5 yes 70 — 110 pm. 2:1 6 | yes 90 = 150 pm 7 [yes 100 — 200 mm 8 | yes 50 — 105 pm. Qe Table 3.4: Gas compositions in mol percent Klab1 | Klab2 | Klab3 | Klab4 | Klab5 | Klab6 | Klab7 No 1.05 1.13 1.13 1.18 117 1.21 1.21 co, |o94 [095 [096 [09s 093 }109 | 1.01 Q 80.84 | 8147 | 81.86 | 81.84 | 8295 | 84.17 | 84.27 Cr 15.74 [15.15 [1482 [14.79 | 13.99 | 12.73 | 12.72 C3 13 1.18 1.13 113 0.96 0.72 0.71 iC, 0.05 0.05 0.04 0.04 0.04 0.03 0.03 nC, |0.08 | 0.07 | 0.06 [007 | 0.06 | 0.05 | 0.05 Cs. | < 0.01 | < 0.01 | <0.01 | <0.01 | <0.01 | <0.01 | <0.01 Weight | 19.06 | 18.95 | 18.88 | 18.99 | 18.71 | 18.51 | 18.48 513. Flow Experiments Table 3.5: Pipe 1 measurements m |p [Tt |p M Ax | AP | sta(AP) {ks /s) | {bar} | C)_| tke /m*] | [ke /sm?] | {m} | [mbar] | [mbar] 3.043 | 72.12 | 36.82] 62.22 | 1.29E-05 | 4.0 |0.80 | 0.20 8.960 | 71.76 | 37.64] 61.59 | 1.29E-05 | 4.0 | 5.19 | 0.12 14.83 | 71.82 | 37.72] 61.62 | 1.29E-05 | 4.0 | 13.39 | 0.25 17.77 | 72.02 | 37.59} 61.86 | 1.29E-05 | 4.0 | 18.88 | 0.32 20.50 | 71.40 | 37.44 | 61.30 | 1.298-05 | 4.0 | 24.98 | 0.34 23.30 | 71.31 | 37.36 | 61.24 1.29E-05 | 4.0 | 32.39 | 0.51 25.86 | 70.62 | 37.09 | 60.65 | 1.28E-05 | 4.0 | 40.17 | 0.56 28.49 | 70.38 | 37.10 | 60.40 | 1.28E-05 | 4.0 | 48.42 | 0.62 31.15 | 70.21 | 36.81 | 60.33 | 1.28F-05 | 4.0 | 57.51 | 0.66 32.90 | 69.62 | 36.62 | 59.81 | 1.28E-05 | 4.0 | 64.24 | 0.70 32.89 | 69.58 | 36.58 | 59.79 | 1.28E-05 | 3.0 | 46.08 | 0.51 32.87 | 69.53 | 36.56 | 59.74 | 1.28E-05 | 2.0 | 32.03 | 0.29 2.974 | 25.41 | 36.30) 19.51 | 1.14E-05 | 4.0 | 2.00 | 0.06 3.939 | 25.41 | 36.62 | 19.49 | 1.15E-05 | 4.0 | 3.39 | 0.07 5.820 | 25.18 | 36.93} 19.29 | 1.15E-05 | 4.0] 7.13 | 0.11 7.678 | 25.08 | 36.93] 19.20 | 1.15E-05 | 4.0 | 12.10 | 0.18 9.358 | 24.66 | 36.92] 18.86 | 1.15E-05 | 4.0 | 17.92 | 0.273. Flow Experiments 53 Table 3.6: Pipe 2 measurements th P |T |p a Ax| AP | std(AP fee /s} | [bar] | (°C]_| [ke /m®] | [kg /sm? | [m] | [mbar] | [mbar] 3.046 | 71.77 | 36.30] 61.72 | 1.29B-05 | 4.0 ]0.93 | 0.18 5.999 | 72.33 | 37.26 | 61.95 | 1.29E-05 | 4.0 | 3.12 | 0.15 8.966 | 72.09 | 37.44] 61.66 | 1.29B-05 | 4.0 6.96 | 0.16 14.61 | 71.01 | 37.43 | 60.60 | 1.29F-05 | 4.0 | 19.09 | 0.28 17.65 | 71.79 | 37.45 | 61.35 | 1.29B-05 | 4.0 | 27.56 | 0.39 20.45 | 71.42 | 37.24] 61.06 | 1.29E-05 | 4.0 | 37.34 | 0.44 23.04 | 70.78 | 37.15 | 60.46 | 1.29E-05 | 4.0 | 47.93 | 0.48 25.86 | 70.82 | 36.95 | 60.57 | 1.29B-05 | 4.0 | 60.58 | 0.56 28.46 | 70.54 | 36.87 | 60.32 | 1.28B-05 | 4.0 | 73.59 | 0.70 32.86. | 69.79 | 36.47 | 59.72 | 1.28E-05 | 4.0 | 98.45 | 0.91 32.86 | 69.79 | 36.47 | 59.72 | 1.28E-05 | 3.0 | 72.58 | 0.60 32.86 | 69.79 | 36.47 | 59.72 | 1.28B-05 | 2.0 | 48.21 | 0.273. Flow Experiments Table 3.7: Pipe 3 measurements m |P [Tt |p H Ax| ap | sta(aP) (ke /s] | (bar) | PC] | tke /m*] | (ke /sm?] | {ml} | [mbar] | [mbar] 33.79 | 71.06 | 36.22| 61.36 | 1.29B-05 | 4.0 | 105.96 | 0.75 33.79 | 71.06 | 36.22} 61.36 | 1298-05 | 2.0 | 52.13 | 0.21 33.79 | 71.06 | 36.22] 61.36 | 1298-05 | 3.0 | 81.21 | 0.57 31.98 | 71.64 | 36.44 | 61.88 | 1.29E-05 | 4.0 | 94.25 | 0.76 29.39 | 72.13 | 36.59 | 62.30 | 1.29B-05 | 4.0 | 78.98 | 0.54 26.80 | 72.66 | 36.95 | 62.71 | 1.295-05 | 4.0 | 64.88 | 0.57 24.00 | 73.05 | 37.28 | 62.98 | 1.3E-05 | 4.0 | 51.72 | 0.48 20.80 | 72.15 | 37.45 | 62.03 | 1.29B-05 | 4.0 | 39.39 | 0.38 18.26 | 73.6 | 37.55 | 63.43 | 1.30B-05 | 4.0 | 29.55 | 0.33 15.21, | 73.29 | 37.73 | 63.06 | 1.30B-05 | 4.0 | 20.51 | 0.24 9.213 | 73.54 | 37.73 | 63.31 | 1.30B-05 | 4.0] 7.45 | 0.16 3.117 | 73.67 | 37.28 | 63.60 | 1.30B-05 | 4.0 |0.90 | 0.173. Flow Experiments 55 Table 3.8: Pipe 4 measurements m |P iT |p H Ax| AP | sta(AP) tke /s] | [bar] | PC] | (ke /m®] | [kg /sm4] | [mn] | [mbar] | [mbar] 33.77 | 71.37 | 36.23 | 61.35 | 1.29-05 | 2.0 | 64.36 | 0.27 33.77 | 71.37 | 36.23 | 61.35 | 1.298-05 | 3.0 | 96.23 | 0.65, 33.77 | 71.37 | 36.23] 61.35 | 1.298-05 | 4.0 | 128.25 | 0.88 29.33 | 72.36 | 36.76 | 62.15 | 1.298-05 | 4.0 | 94.98 | 0.74 26.73 | 72.88 / 36.88) 62.62 | 136-05 | 40 | 7815 | 0.74 23.76 | 72.71 | 37.04 | 62.40 | 1.298-05 | 4.0 | 62.10 | 0.46 21.03 | 73.20 | 37.11 | 62.86 | 1.30E-05 | 4.0 | 48.04 | 0.47 18.01 | 73.02 | 37.26 | 62.64 | 1.30E-05 | 4.0 | 35.39 | 0.36 15.19 | 73.55 | 37.43 | 63.10 | 1.30E-05 | 4.0 | 24.78 | 0.35 9.194 | 73.81 | 37.53 | 63.32 | 1.30E-05 | 4.0 |8.93 | 0.16 6.114 | 73.72| 37.33| 63.30 | 1.30E-05 | 40] 3.94 | 014 3.096 | 73.63 | 36.87| 63.38 | 1308-05 | 4.0 | 1.04 | 0.133. Flow Experiments Table 3.9: Pipe 5 measurements m |p [tr [op # Ax | AP | sta(AP) {kg /s} | [bar] | [°C)_| ke /m®] | (kg /sm?] | [m} | [mbar] | [mbar] 33.54 | 71.00 | 36.36 | 60.94 | 1298-05 | 2.0 | 65.79 | 0.30 33.54 | 71.00 | 36.36 | 60.94 | 1298-05 | 3.0 | 99.68 | 0.64 33.54 | 71.00 | 36.36 | 60.94 | 1.29E-05 | 4.0 | 137.09 | 1.05 29.23 | 72.18| 36.8 | 61.96 | 1298-05 | 4.0 | 102.89 | 0.77 26.61 | 72.63 | 37.09 | 62.31 | 1.29B-05 | 4.0 | 84.35 | 0.55 23.87 | 73.06 | 37.08 | 62.74 | 1.30B-05 | 4.0 | 67.32 | 0.60 21.01 | 73.17 | 37.32 | 62.76 | 1.30B-05 | 4.0 | 52.11 | 0.50 15.10 | 73.23 | 37.52 | 62.75 | 1.308-05 | 4.0 | 26.79 | 0.28 9.071 | 72.95 | 37.51] 62.47 | 1.30B-05 | 4.0 9.64 | 0.19 6.084 | 73.42 | 37.44 | 62.97 | 1.30E-05 | 4.0 | 4.30 | 0.13 3.102 | 73.80 | 36.84] 63.55 | 1.30E-05 | 4.0 |113 | 0.11 0.7661 | 73.17 | 34.32 | 63.83 | 1.29E-05 | 4.0 [0.18 | 0.133. Flow Experiments 57 Table 3.10: Pipe 6 measurements fia Pp |T |p M Ax| AP | std(AP) tha /se} | thar) | FC) | th /m’j | the/sm) [fm | fmbar) | mbar) 3.060 72.76 | 36.67 | 62.58 1.29E-05 | 4.0 | 1.31 0.13 5.994 72.30 | 37.24 | 61.93 1.29E-05 | 4.0 | 4.81 0.12 8.996 72.35 | 37.46 | 61.90 1.29E-05 | 4.0 | 10.76 | 0.17 14.89 72.31 | 37.58 | 61.83 1.29E-05 | 4.0 | 29.65 | 0.36 17.63 | 71.72 | 37.38 | 61.32 1.29E-05 | 4.0 | 41.92 | 0.41 20.51 | 71.65 | 37.29 | 61.28 1.29E-05 | 4.0 | 57.06 | 0.78 23.29 71.53 | 37.14 | 61.21 1.29E-05 | 4.0 | 73.84 | 0.55 26.01 71.25 | 37.01 | 60.98 1.29E-05 | 4.0 | 92.60 | 0.52 28.45 70.58 | 36.77 | 60.40 1.28E-05 | 4.0 | 112.04 | 0.74 32.80 69.77 | 36.37 | 59.73 1.28E-05 | 4.0 | 151.15 | 0.95 32.80 69.77 | 36.37 | 59.73 1.28E-05 | 3.0 | 113.63 | 0.65 32.80 69.77 | 36.37 | 59.73 1.28E-05 | 2.0 | 75.87 | 0.213. Flow Experiments Table 3.11: Pipe 7 measurements m |P |T |p M Ax| AP | sta(AP) tke /s) | {bar} | [°C)_| (ke /m°] | [ke /sm? | [m] | [mbar] | [mbar] 14.21 | 68.71 | 36.65 | 59.42 | 1.28E-05 | 2.0 | 15.05 | 0.09 14.21 | 68.71 | 36.65 | 59.42 | 1.28E-05 | 3.0 | 22.72 | 0.15 14.21 | 68.71 | 36.65 | 59.42 | 1.288-05 | 4.0 | 30.35 | 0.26 8.627 | 69.11 | 36.67 | 59.82 | 1.28E-05 | 4.0 | 11.14 | 0.15 8.627 | 69.11 | 36.67 | 59.82 | 1.28E-05 | 3.0] 8.31 | 0.16 8.627 | 69.11 | 36.67 | 59.82 | 1.28E-05 | 2.0 | 5.51 | 0.06 2.909 | 69.20 | 36.13 | 60.08 | 1.28E-05 | 4.0 | 1.30 | 0.18 0.726 | 69.12 | 33.52 | 60.90 | 1.27E-05 | 4.0 | 0.21 | 0.18 31.83. | 67.18 | 35.98 | 58.12 | 1.27E-05 | 4.0 | 154.81 | 0.60 30.10 | 67.70 | 36.35 | 58.52 | 1.27E-05 | 4.0 | 137.56 | 0.65 27.63 | 68.10 | 36.57 | 58.85 | 1.27E-05 | 4.0 | 115.38 | 0.61 25.11 | 68.43 | 36.7 | 59.13 | 1.27E-05 | 4.0 | 94.67 | 0.53 22.48 | 68.73 | 36.93 | 59.36 | 1.28E-05 | 4.0 | 75.76 | 0.44 19.7 | 68.80 | 37.08 | 59.38 | 1.28E-05 | 4.0 | 58.56 | 0.493. Flow Experiments Table 3.12: Pipe 8 measurements th Pp |t |p w Ax| AP | std(AP) the /s} | foar) | °C} _[ the /m°h | fog /sm*) | tr) | mbar) | bar] 2.974 | 70.80 | 36.37 | 60.74 1.28E-05 | 4.0 | 1.06 0.11 5.976 | 72.48 | 37.24 | 61.29 1.29E-05 | 4.0 | 4.04 0.13 8.959 | 72.41 | 37.35 | 61.18 1.29E-05 | 4.0 | 9.25 0.16 14.41 70.46 | 37.41 | 59.28 1.28E-05 | 4.0 | 24.83 | 0.31 17.31 | 70.83 | 37.39 | 59.64 1.29E-05 | 4.0 | 35.81 | 0.41 20.12 | 70.69 | 37.25 | 59.55 1.28E-05 | 4.0 | 48.59 | 0.52 22.84 70.54 | 37.12 | 59.44 1.28E-05 | 4.0 | 62.98 | 0.62 25.37 | 69.85 | 36.95 | 58.84 1.28E-05 | 4.0 | 78.65 | 0.61 28.04 | 69.83 | 36.76 | 58.88 1.28E-05 | 4.0 | 96.33 | 0.67 32.18 | 68.69 | 36.33 | 57.92 1.27E-05 | 4.0 | 129.65 | 0.85 32.18 | 68.69 | 36.33 | 57.92 1.27E-05 | 3.0 | 96.10 | 0.76 32.18 | 68.69 | 36.33 | 57.92 1.27E-05 | 2.0 | 65.73 | 0.21 3.061 | 26.07 | 35.90 | 20.08 1L14E-05 | 4.0 | 3.16 0.06 4.028 | 25.91 | 36.30 | 19.92 LI15E-05 | 4.0 | 5.58 0.07 5.911 | 25.51 | 36.64 | 19.58 1.15E-05 | 4.0 | 12.39 | 0.12 7.847 | 25.59 | 36.79 | 19.62 1.15E-05 | 4.0 | 22.05 | 0.18 9.607 | 25.27 | 36.77 | 19.36 1.15E-05 | 4.0 | 33.48 | 0.25 5960 3. Flow Experiments Table 3.13: Pipe 1 measurements, second test period mm =|P T |p L Ax} AP | std(AP) tke/s} | thar] | O)_| tha/m) | ths 7m? | fm | mbar | mbar] 31.62 | 119.29 | 37.22 | 110.85 | 1.58E-5 4.0 | 31.66 | 0.10 37.04 | 119.85 | 37.16 | 111.56 | 1.59B-5 4.0 | 42.53 | 0.18 47.45 | 120.03 | 36.86 | 111.97 | 1.59B-5 4.0 | 66.93 | 0.32 42.41 | 120.35 | 36.98 | 112.23 | 1.59E-5 4.0 | 54.37 | 0.18 51.89 | 118.66 | 36.66 | 110.65 | 1.58E-5 4.0 | 78.48 | 0.37 55.81 | 116.57 | 36.10 | 108.83 | 1.58E-5 4.0 | 88.58 | 0.55 23.25 | 91.82 | 38.24 | 81.29 L.40E-5 4.0 | 23.56 | 0.07 30.52 | 90.94 | 37.93 | 80.54 1.39E-5 4.0 | 40.14 | 0.15 37.90 | 90.83 | 37.62 | 80.58 1.39E-5 4.0 | 59.33 | 0.27 41.53 | 90.76 | 37.43 | 80.60 1.39E-5 4.0 | 69.98 | 0.35 34.24 | 90.88 | 37.74 | 80.57 1.39E-5 4.0 | 49.69 | 0.21 26.76 | 90.83 | 38.09 | 80.35 1.39E-5 4.0 | 31.28 | 0.11 19.25 | 90.94 | 38.27 | 80.38 1.39E-5, 4.0 | 16.69 | 0.053. Flow Experiments Table 3.14: Error estimate of AP measurements Section | AP (fitted) [mbar] | AP (4 m) [mbar] | Dev. [%] 1 =16.09 = 16.05 =0.5 2 25,12 =24,60 =2.07 3 26.91 26.48 1.60 4 —31.92 —32.05 0.41 5 35.65 34.28 3.84 6 -37.64 37,77 0.35 Ta ~7.64 7.58 -0.78 7 -2.81 -2,78 1.08 8 31.94 —32.40 1.44 Table 3.15: Uncertainty in variables, main test period Variable | Uncertainty [%] AP 2.5 (4) Ar 0.05 D oO p 1 th 0.3 H 2 T oO P 1 f ~ 2.8 (4.2) Re ~262 3. Flow Experiments Wa" NPT Pipe wall Figure 3.1: Static pressure tap Test pipe 6 m long, ID=150 mm T 0sm im im Im im 1m 0.25 m 0.25m Figure 3.2: Schematic view of a test pipe3. Flow Experiments 63 x atop sous one wine Compre / 5 | Has riter |S | 1 Dry 9a supply Figure 3.3: Schematic view of the K-Lab loop64 Flow Experiments 0.025; 0.02} 015} friction factor, f 01 0.008! 5 10 15 20 25 30 Reynolds number [x 16] Figure 3.4: Friction factor; o and #: coated pipe, x: steel pipe, +: pipe 3, <: pipe 4, © pipe 8, ©: pipe 5, V: pipe 6, A: pipe 73. Flow Experiments friction factor, f Pipe 1 Pipe 2 oor oor oor oor fal jon EERE orf oot o.oo} = °° 0.008) 3 10 1 2 2 00 a Pipe 3 Pipe 4 oo oo] gq acdsee < OOM Lbs a oor oor oot oot 008 0008 5 10S aa 31015 20a Reynolds number [x 126] Figure 3. Friction factor pipes 1,2,3/4 6566 friction factor, 3. Flow Experiments 0.022 0.02) core o.or6 0.022} 0.02 ora} o.016| Pipe 8 Pipe 5 0.022) 0.02) 018} noo 0 oo 6 00000 pn ae 0.016} 5 10 18 20 25 S10 18 20 2 Pipe 6 Pipe 7 0.022| 4 6 aAaaaa 0.02] ve vovvey 9 oe} 0.016) 5 10 18 20 2% 5 10 18 20 2 Reynolds number [x 166] Figure 3.6: Friction factor pipes 8,5,6,73, Flow Experiments 67 Coated pipe Stee! pipe 29) “ol ~40) ~«o| 60) a -20) 60) = ~100] -20) : 5 2 3 4 a 3 4 aS Pipe 3 Pipe 4 a 40 40) 60] -60) 20) -20) 100 400 -120) 140 ~120 2 3 + 2 3 + x [m] Figure 3.7: Test of AP/Ax in pipes 1-468 3, Flow Experiments AP [mbar] Pipe 8 Pipe 5 50 ~100 150 2 3 3 + Pipe 6 -20} 2 3 “ 3 4 ax (m] Figure 3.8: Test of AP/Ax in pipes 5-8Chapter 4 Pipeline data Pipeline data obtained from long-distance pipelines in the Norwegian gas export infrastructure is presented in this chapter. The data are obtained from Statoil which operate the major part of the pipeline infrastructure in Norway on behalf of several licensees. 4.1 Capacity tests ‘The pipeline data presented are from capacity tests on operating pipelines (Sjgen, 1998). ‘The capacity tests include measurements of inlet and outlet pressure, temperature, flow rate and gas composition. The capacity tests were carefully planned in order to achieve steady state flow conditions during the tests. In this respect the data are more accurate than data from normal (non-steady state) operating conditions. The operator uses the capacity test data together with a pipeline flow simulator to tune” an equivalent roughness value for the pipeline. The flow simulator solves the one-dimensional momentum and energy equations for pipe flow. The tuned roughness value will be affected by sources of error in measurements as well as the limitations of the 6970 4, Pipeline data flow simulator. However, the measurements are of high quality and particularly the subsea pipelines included in the data set have few disturbing elements (few significant bends, no compressor stations). ‘Therefore the comparison between the pipeline data and our experimental work is relevant. ‘Table 4.1 shows the type of pipelines which were included in the analysis. Also shown in the table is the friction factor correlation used by the operator to tune the equivalent roughne: 1965) and "CW” refers to the Colebrook-White correlation (Colebrook, 1939). Table 4.2 shows the reported data from the tests. The data given are the length of the pipelines, s in each pipeline. "AGA" refers to the AGA friction factor correlations (Uhl, inlet and outlet pressure measured, volumetrie flow rate at standard conditions, gas mole weight, pipe diameter. Only the results from the main test periods are given (some of the tests included more than one measurement point). Table 4.3 shows the reported gas compositions. ‘Table 4.4 shows the data calculated for this study. The data listed are the mean pressure in the pipeline, the mean ambient temperature, the density at mean pressure and temperature, the mass flow rate, the gas gravity, the gas viscosity and the compressibility factor (Z) at mean pressure and temperature. ‘The Z-factor was calculated with the AGA-8 correlation (Starling, 1992) and the viscosity from the Chung et al. correlation (Chung et al., 1988). Table 4.5 shows calculated Reynolds numbers from the capacity tests The values are based on the viscosity calculated at mean pressure and temperature (see Table 4.4). 4.2 Modelling ‘To model steady state flow in gas pipelines the momentum and energy equations for one-dimensional flow (Equations 2.1 and 2.2) should be solved. Here, to calculate a mean friction factor in each pipeline, a simplified analysis is made where the flow is assumed to be isothermal at the mean ambient, temperature and only the momentum equation is4. Pipeline data a considered. The momentum equation for pipe flow (Equation 2.1) reads: aP 1 nefu , d(o0?) “ie tPgsina + pls 4 OE Assuming that the pipeline is horizontal and that the temperature and compressibility factor are constant at average values (J and Z), the momentum equation can be combined with the real gas law p = ¥2% and integrated over a length L (Tian and Adewumi, 1994): 1 D (P2 - Pz ;D" + Faint Tp Diez Pe BP)” (4.1) Equation 4.1 gives an analytical relation between the mass flow, inlet and outlet pressures, mean temperature, pipe geometry, gas properties and the friction factor. The equation can be solved by Newton-Raphson iteration. However, the kinetic energy term is usually very small in comparison with the wall friction term in a gas pipeline. Neglecting the kinetic energy term Equation 4.1 can be written (Equation 1.1): In this study the difference in the results obtained from the two equations (Equation 4.1 and 1.1) was negligible and Equation 1.1 was preferred for its simplicity. The variations in pipeline profile was not included in this analysis. However, the height difference between the inlet and the outlet of the pipelines was included. The pressure drop due to gravity was added to the inlet pressure with the term gAh where the p is the mean density of the gas in the pipeline and Ah is the difference in height between the inlet and the outlet. Equation 1.1 is a simplified model of steady state flow in gas pipelines. Both the assumption of isothermal flow and the assumption that the variations in height along the pipeline can be neglected are serious limitations of the model. To check the accuracy of Equation 1.1, the pressure drop in the gas pipelines were calculated and compared with the measured values,72 4. Pipeline data Three different friction factor correlations were used in the analysis. The Weymouth friction factor correlation (Equation 2.12) is an early model of friction factors in gas pipelines (Katz and Lee, 1990) (the pipe diameter has dimension meter): 0.00941 Dis ‘The Weymouth correlation does not include wall roughness and is therefor expected not to be accurate when the Reynolds number in the pipeline is large. In the gas industry the Colebrook-White correlation (Equation 2.27) and the AGA (Uhl, 1965) correlations (Equations 2.33 and 2.34) are commonly used. These correlations are discussed in the literature review (page 15 and 21 respectively) and are shown below. [Ee ogy 251 he au G Plos(ae VF t 37D) AGA: Vi = Fr2log(Re VF) — 0.9 = Fy2log (eZ / FE = Qlog(£) + 1.74 = 2log(7.412) The Reynolds number was calculated from the definition Re = 2? = 4.3 Results ‘The pressure drop in the pipelines was calculated from Equation 1.1 using the data given in the capacity tests. The pipe geometry including the tuned equivalent roughness values (Sioen, 1998), the gas mole weight, the mass flow, the mean ambient temperature, the mean Z-factor and the inlet pressure was given as input and the outlet pressure was calculated. ‘Table 4.6 shows the measured and calculated output pressures. The measured output pressures are given in column 2 and the calculated output pressures using the three different friction factor correlations (Weymouth, Colebrook-White and AGA) are given in4, Pipeline data 73 the column 3, 4 and 5. In the table the results corresponding to the friction factor correlation used in the tuning of the equivalent wall roughness (see Table 4.1) are underlined. The calculated output pressures using the appropriate friction factor correlation (underlined results) are generally within +0.6 bar of the measured pressure. ‘The Weymouth equation gave a very poor prediction of the pressure drop. The calculations show that the pressure drop in the pipelines can be estimated with reasonable accuracy using Equation 1.1, even though the calculation model does not. include heat transfer effects and pressure drop due to variation of heights along the pipeline. The same model was therefor used to evaluate the mean friction factor in the pipelines. ‘The mean friction factor in each pipe was calculated from Equation 1.1. The viscosity value at mean pressure and temperature was used to calculate the Reynolds number. In Figure 4.1 the results are shown. It is seen that the calculated friction factor in subsea B, C, B and H) can be modelled by the Colebrook-White correlation with a sand-grain roughness of 2.0 um, The onshore coated pipelines (pipe D coated pipelines (pipe and F) appear to have a higher value of sand grain roughness. The study also includes data from one bare steel offshore pipe (pipe G) which has the highest friction factor of all the pipelines in the study.74 Table 4.1: Pipeline data in the study Pipe | Type of pipe _| f A Offshore coated AGA B Offshore coated cw C | Offshore coated | AGA D Onshore coated AGA E Offshore coated AGA F | Onshore coated | AGA G Offshore not coated | CW H Offshore coated cw ‘Table 4.2: Reported data, main test periods 4. Pipeline data Pipe|Length|P, |P. | Q, My D thin} [ tar) | toar) | tossim?/a) | the /Amot | tn) A 812.4 108.42 | 85.59 | 22.33 18.84 0.9664 B 303.5 166.26 | 145.59 | 45.15 17.31 0.9664 Cc 619.0 107.97 | 94.16 | 20.15 18.87 0.9664 D 48.5 65.22 | 63.64 | 20.15 18.87 1.034 E 619.0 129.85 | 86.80 06. 18.07 0.9664 F 48.5 72.03 | 67.45 | 38.06 18.07 1.034 G 227 136.3 | 112.1 | 18.5 18.60 0.6698 H 812.8 146.7 | 95.5 38.5 18.52 0.96644. Pipeline data 75 Table 4.3: Reported gas compositions Pipe |A |B |c |p |B F G lH CG 83.50 | 93.13 | 82.85 | 82.85 | 88.282 | 88.282 | 83.37 | 87.42 Co 11.62 | 3.746 | 13.12 | 13.12 | 7.895 | 7.895 | 14.0 | 6.97 Cs 2.43 | 0.615 | 144 [144 | 1.264 | 1.264 | 0.83 | 213 iC, [0.17 | 0.338 | 0.097 | 0.097 | 0.215 | 0.215 | 0.03 | 0.30 nC, | 0.23 | 0.065 | 0.19 [0.19 [0.127 | 0.127 | 0.05 | 0.25 iCs _| 0.027 | 0.065 | 0.023 | 0.023 | 0.037 | 0.037 | 0.01 | 0.06 nC; | 0.022] 0.015 | 0.02 | 0.02 | 0.019 | 0.019 | 0.01 | 0.03 Co | 0.008 | 0.108 | 0.023 | 0.023 | 0.071 | 0.071 |- | 0.10 No 1.43 | 1.574 | 1.28 | 1.28 | 1.544 | 1.544 | 0.92 | 1.37 co, | 0.56 | 0.331 | 0.95 | 0.95 | 0.541 | 0.541 | 0.75 | 1.38 Weight | 18.84 | 17.31 | 18.87 | 18.87 | 18.07 | 18.07 | 18.60 | 18.52 Table 4.4: Data calculated for the study Pipe | P T p th 7 i Zz bar] | IK) | tke/m*] | (ke/s] | | bePas] | A | 97.0 | 280.15 | 109.84 | 205.5 | 0.652| 147 | 0.72 B | 155.93 | 280.15 | 158.66 | 383.5 |0.598| 19.2 | 0.73 © | 101.07 | 281.15 | 114.32 | 185.4 } 0.651] 15.1 | 0.71 D | 64.43 | 283.15 | 63.62 | 185.4 | 0.651| 12.1 | 0.81 E | 108.33 | 282.15 | 112.66 | 334.1 | 0.624| 15.2 | 0.74 F_ | 69.74 | 284.15 | 64.78 | 334.1 | 0.624] 12.4 | 0.82 G | 124.2 | 280.65 | 145.4 | 167.6 | 0.642| 17.6 | 0.68 H_ [121.1 | 278.65] 1389 | 348.4 |0.639|16.9 | 0.7076 4, Pipeline data ‘Table 4.5: Calculated Reynolds numbers Table 4.6: Calculated outlet pressure i rm Pipe | Re =24, 1.84107 2.6310" 1.6110" 1.89% 10" 2.9010" 3.32107 1.81x107 mlol=lelolalal> 2.72107 Pipe | P2(given) | P2(Weymouth) | P2(CW) | P2(AGA) [bar] {bar] {bar] | [bar] A [85.59 | 79.05 85.02 | 84.95 B [145.59 | 139.64 146.03 | 144.69 co fosie | 9117 93.76 | 94.15 D 63.64 63.55 63.76 | 63.77 E | 86.80 | 68.86 84.97 | 86.61 F [6745 | 66.66 67.64 | 67.77 G {112.10 | 106.94 1.01 | 112.77 H [95.51 | 76.97 95.99 | 97.714. Pipeline data Ne Friction factor, 75| Offshore pipeline, coated Onshore pipeline, coated Ofishore pipeline, stee! C-W equation, ks= 2.0 m| Jorx 10 20 30 40 Reynolds Number [x 1e6] Figure 4.1: Friction factor in gas pipelines 178Chapter 5 Roughness Measurements A major object of the present work was to study the effect of roughness on the pressure drop in gas pipelines. Therefor the wall roughness of the test pipes used in the experimental work and line pipes used in gas pipelines was measured. The aim has been to ccrrelate the wall roughness with the measured pressure drop in the pipes. This chapter describes the apparatus used to measure the wall roughness and the results obtained. 5.1 Instrument ‘To measure the roughness of the pipe wall a stylus instrument of the type Perthometer S3P was used. The function of a stylus instrument is to record height variations of a surface. The stylus instrument had a drive unit PGK and pickup MFW-250. The stylus tip had a radius of curvature of 5 pm and an opening angle of 90°. A schematic view of the measurement equipment is shown in Figure 5.1. ‘The stylus instrument gave as output both roughness parameters (Ry, Rg, Rz) and the measured profile, The output from the stylus instrument was transferred through the 7980 5, Roughness Measurements extemal interface to a computer for processing. Depending on the roughness of the surface measured, the vertical measuring range of the stylus instrument was chosen to either +25 ym or +250 ym. The instrument digitized the vertical measuring range in 32768 intervals. Therefore the vertical resolution was given by 25 jm /16384 = 1.5 nm or 250 jrm /16384 = 15 nm depending on the vertical measuring range chosen. The measuring range +25 jum was used whenever possible. ‘The cut-off wavelength is the largest roughness wavelength recorded by the instrument. The cut-off wavelength, A., of the instrument could be chosen to 0.08, 0.25, 0.8 or 2.5 mm. The total tracing length in each run was 7A., where ¢ is the cut-off wavelength (seven samples was recorded in each run). The traced length was digitized in 8064 intervals thus the resolution in the horizontal direction was given by [,/8064, where |, was the tracing length. ‘The appropriate cut-off wavelength was chosen according to DIN 4768 (Sander, 1991) depending on the roughness of the surfaces. The recommendations for the choice of cut-off wavelengths are shown in Table 5.1. However, because of the limitations of the instrument the maximum cut-off wavelength used was 2.5 mm. A Gaussian phaso-corrected high-pass filter (DIN 4777, 1990) was employed to ensure that wavelengths larger than the cut-off wavelength was filtered out. As a result 5 filtered samples were produced in each run, ‘The smallest resolved wavelength was dependent on the resolution in the horizontal direction and the stylus geometry. The horizontal resolution gives the smallest resolved wavelength as Amin = 2h/8064 from sampling theory. The Amin is equivalent to 1/f, where Jn is the Nyquist frequency as defined in the sampling theory for time series (Svardstrém, 1987). In Table 5.2 the smallest resolved wavelength is listed for each cut-off wavelength. Accerding to DIN 4768 (Sander, 1991) a stylus tip with a radius of curvature of 5 jim was recommended for measuring surfaces with R, > 2ym and Ry > 0.4um. In the measurements the lowest value of Ry was approximately 1 jum, thus the choice of stylus5, Roughness Measurements 81 should be appropriate. However the effect of the stylus on the recorded profile should be noticed. The spherical shape of the stylus tip will produce a smoother measured profile than the actual surface, especially at small length scales. Because of this smoothing of the stylus tip acts as a low-pass filter (Thomas, 1982), and will effectively reduce the aliasing effect of the smaller wave-length irregularities on the measured profile. To sum up this short discussion of the equipment, it should be noted that the stylus method for measuring surface roughness is basically a comparison method (Sander, 1991). The result from measurements should only be compared with measurements done under comparable conditions (e.g. cut-off wavelength) and similar equipment. 5.2 Measurements The wall roughness of the test pipes was measured on three occasions. The wall roughness was measured directly after the coating in the paint shop, after the flow tests at the K-Leb facility and finally at the Department of Petroleum Engineering and Applied Geophysics in Trondheim four weeks after the flow tests. The two first measurements were done at "field” conditions while the last were done inside a laboratory. The last measurements are considered the most reliable due to the more “controlled” environment. However, the differences between these measurements are small. The wall roughness of the pipe used in the second test period were measured after the flow test in October 1998. To compare the results of the flow measurements done in the main and second test period a significant change in the wall roughness was not desirable. The wall roughness of coated full-scale pipes was measured at the Bredero-Price site in Farsund, Norway, in April 198. Two measurements were done; one on a recently coated pipe (coated the same day) and one on a pipe which had been stored for 5 months. The roughness profile was measured at the bottom of the pipe in the axial direction in all measurements. In the test pipes the wall roughness was measured in the streamwise82 5. Roughness Measurements direction, however no significant difference was noted on test runs in the opposite direction. In the full-scale pipes the streamwise direction was not defined. Before the measurements the surfaces were cleaned for dust with a wet cloth or compressed air. ‘The wall roughness in all pipes was measured with 20 separate runs. Thus for each pipe 100 samples of the roughness profile was produced. ‘The drive unit of the stylus instrument was moved after each run. ‘The measurements were done in a 1 m long section close to the end of the pipes. The small dimensions of the test pipes made it difficult to reach more than 1 m inside the pipes. 5.3 Roughness of test pipes In Figures 5.2, 5.3 and 5.4 the Ry, Ry and R, values measured in the test pipes at three different occasions are plotted. The steel pipe is denoted pipe 2 and the coated pipes 1,3,4,8,5,6,7 according to increasing bead size in the coating (see Table 3.3). Note that the first measurement of the steel pipe (week 38) was done before the pipe was exposed to water in pressure testing. After the pipe was exposed to water it was stored outside (coast climate) for two weeks before the flow tests. The difference in roughness values of the steel pipe from the first measurement. compared to the latter two are due to corrosion. ‘The measured values of the Ry, Ry and R, of the test pipes are tabulated in Table 5.3. ‘The data are from the measurement done inside the laboratory building in week 51 1998. ‘The roughness parameters are mean values of all 100 samples. Given also is the standard deviation of the measurements and the mean bead size of the glass beads in the coating. ‘The cut-off wavelengths in the measurements were A, = 0.8 mm for the pipe coated with plain coating and A, = 2.5 mm for the other pipes. In the table the coated pipes are listed according to increasing bead size in the coating. The roughness parameters generally increase with increasing bead size, but the R, value of pipe 8 seems to be low. ‘The roughness of the coated pipe measured after the second flow test is shown in Table5. Roughness Measurements 83 5.4. The wall roughness parameters differ slightly from the values measured after the first flow test (Table 5.3). The trend indicates a smoother surface after the second flow test. However the variation is within one standard deviation in all roughness parameters. Figures 5.5, 5.6, 5.7 and 5.8 show examples of the measured roughness profiles. In Figure 5.5 the coated pipe and the steel pipe are compared. The scaling of the axis is equal and the steel pipe appears to have a rougher surface than the coated pipe. In the following three figures (Figures 5.6, 5.7 and 5.8) test pipes 3,4,8,5,6,7 are compared with equal scaling on the axis. As seen in the figures the roughness of the profiles is increasing according to the bead size in the coating. The axis scaling in these three figures are different from the Figure 5.5, In Figures 5.9 and 5.10 the structure functions (Equation 2.42) calculated from the measured roughness profiles are compared. The plotted structure functions are averages over 100 samples. The axis scaling on all 8 plots are equal and the test pipes are shown with respect to increasing roughness. According to Equation 2.43 the structure function will reach a limit of 20? (note that & = R,) for large eparations (£ > 00). This limiting value at large separations is increasing according to the increasing Ry of the test sections (see Table 5.3 for the R, values). In Figures 5.11 and 5.12 the power spectral density (Equation 2.44) calculated from the measured roughness profiles are plotted. The power spectral densities are averages over 100 samples and are calculated using the Welch algorithm in Matlab (MATLAB, 1992) with a Hanning window. The axis scaling is equal in all 8 plots. The value of the power spectral density at low frequency (large length scales) is increasing according to the increasing roughness of the surfaces. Comparing the power spectral density (PSD) of the plain coated pipe and the bare steel pipe (Figure 5.11) it is seen the coated pipe shows a clear white noise behavior at large fraquencies (small length scales). This is not seen in the PSD of the steel pipe. This white noise behavior is due to the smoothing effect of the coating at the smallest length scales.84 5. Roughness Measurements In Table 5.6 the Hurst exponent resulting from fitting a straight line to the log-log plots 49) are shown. The two different methods of estimating the Hurst exponent give non-consistent estimates. of the structure function and the power spectrum (see Equation ‘The Hurst exponent estimated from the power spectral density are generally larger than 1 and therefore non-defined. ‘The Hurst exponents obtained from the structure function are close to one. As noted by Schmittbubl et al. (1995) the variable bandwidth method (similar to the structure function method) is limited to Hurst exponents between 0 and 1, and the method will not give exponents larger than 1. If Hurst exponents close to 1 are found, Schmittbubl et al. (1995) recommend to use the power spectrum method. The differing result between the power spectrum method and the structure function method makes the fractal analysis of the surface roughness difficult. The results indicate that the surfaces measured in this work do not have a fractal scaling. However, the power spectrum describes how the variance of a profile change with frequency (Bendat and Piersol, 1986). Therefor the slope of the power spectrum is an important characteristic of a surface, In the correlations chapter (Chapter 6) the parameter H' defined from the power specter similarly to the Hurst exponent (G(w) ~ w~#'+)) is used as a measure of the slope of the power spectrum. The numerical values of the H' parameter is equal to the Hurst exponents tabulated in Table 5.6. 5.4 Roughness of full-scale pipes ‘Table 5.5 shows the measured values of Ra, R. and Ry from measurements on full scale pipes. The cut-off wavelength the measurements was 0.8 mm. Figure 5.13 shows examples of the measured profiles from the full-scale pipes (line pipes). Line pipe 1 was coated the same day as the measurements were performed and line pipe 2 had been stored outside for 5 months when the measurements were done. The axis scaling in Figure 5.13 is similar to Figure 5.5 (coated test pipe) for comparison.5. Roughness Measurements 85 In Figure 5.14 the structure function and the power spectral density function from the measurements on the line pipes are compared. The scaling in the plot are similar to Figures 5.9 and 5.11 and the line pipe results may be compared with those from the test When the measurements on line pipe 1 and 2 are compared no signs of aging after 5 months storage is found. ‘The measured roughness parameters are practically identical, but the power spectrum calculated from the measurements on line pipe 1 shows more noise for high frequencies than line pipe 2. The measurements on line pipe 1 were done close to the production line at the Bredero Price site and the measurements on line pipe 2 were done outside in the storage area. Vibrations may be the cause of the noise at high frequencies in the power spectrum for line pipe 1 Comparison of the measurements on the line pipes and on the coated test pipe (pipe 2) show that the coated test pipe is smoother than the full-scale pipes. However the statistical variations in the measurements on the full-scale pipes are larger and the test pipe roughness is within one standard deviation of the full-scale pipe roughness values. ‘The test pipes were honed before coating which may explain why the coated test pipes is smoother than the full-scale pipes. The larger statistical variations in the full-scale pipe measurements may also be due to environmental influence (background noise). 5.5 Comparison with Nikuradse’s rough pipes ‘To compare the current work with the measurements of Nikuradse (1933) and his sand-grain roughness the, Ry, Ry and R; values of a sand-grain surfaces must be estimated. Assuming that the sand-grains in Nikuradse’s experiments were perfect sphi to R, =0.28d and R, = 0.375d where d is the diameter of the spheres. The roughness parameters are calculated as surface averages. The R, value of sand-grain roughness and that the grains were packed side-by-side the roughness parameters was found should be equal to the grain diameter.86 Roughness Measurements In Figure 5.15 the measured roughness parameters Ry, Ry and R, are shown as a function of the mean bead size of the glass beads mixed in the coating. For comparison with Nikuradse’s measurements straight lines are least square fitted to the data to find the constant of proportionality between dieaq and the measured roughness parameter. ‘The straight lines in the Figure 5.15 are Ry =0.16dyeoa, Rg = 0.20dbead, Re = 0.77dbead- ‘The glass beads used are not of a single size and rather large deviations from the fitted straight lines are seen in the figure. The packing of the glass beads on the test section walls was not as dense as side-by-side and the constants of proportionality should be smaller than the values calculated above for Nikuradse’s experiments. ‘The sand grain sizes in Nikuradse’s experiments in rough pipes was in the range 0.1 — 1.6 mm (see Table 2.1). The glass beads added in the coating of the test pipes in this work was much smaller. The mean bead size was in the range 25 — 150 ym. Nikuradse used small diameter pipes and varied his relative roughness '/k from 15-507. In this work the relative roughness r/k (using R; as the roughness measure) in the glass bead roughened pipes was in the range 614-2562, that is, considerable higher values than investigated by Nikuradse.5. Roughness Measurements Table 5.1: Recommendations of cut-off wavelengths from DIN 4768 Roughness of surface Cut-off wavelength Re (uml Ry (um Ae [mm] 0.1 < R. < 0.5 | 0.02 < R, < 0.1 | 0.25 0.5 1.83, and smooth flow was found when log(kz) < 0.55. The roughness function for the rough test pipes is shown in Figure 6.5. The equivalent sand-grain roughness in the pipes and R,/H" are used as roughness measures. When the top plot in this figure is compared with the roughness function of Nikuradse (Figure 6.3) it is seen that the steel pipe (marked x) does not enter the transition zone at the value of log(k}) = 1.83 as found in Nikuradse’s sand-grain roughened pipes. For the coated pipes (roughened with glass beads) only the test pipe number 8 (marked with a square) show signs of having entered the transition zone. This pipe shows a transition behavior more compatible with Nikuradse’s sand-roughened pipes. The roughness function f(k*) plotted with Rj/H' as the roughness measure (Figure 6.5, bottom plot) shows a good collapse of the data from the seven rough tests pipes. Such a collapse of the data using a directly measured roughness parameter is remarkable. The figure indicates that the steel pipe (marked x) and the smoothest of the glass-bead roughened pipes (marked +) behave differently from the other pipes. The reason may be that the surface roughness of the rougher pipes are dominated by the glass beads and a, sand-grain type of transition can be expected. In the smoothest glass bead roughened pipe the glass beads are imbedded in the coating film because of the small size of the beads and the structure of the surface is not dominated by the shape of the beads. ‘Therefore a more “natural” transition comparable with the steel pipe can be expected for6. Correlations 107 this pipe. Steel pipes have been shown (Smith et al., 1956) to have a more abrupt transition (a lower peak in the roughness function, refer Figure 6.3) between smooth and rough flow than sand-grain roughened pipes. 6.2 Smooth test pipe The friction factor in the test pipe coated with coating only was lower than the other 7 test pipes. No well defined rough flow characteristics were seen in the measured friction factor (Figure 6.1). Also shown in Figure 6.1 is the Zagarola friction factor correlation for smooth pipe (Equation 2.31). When this correlation is used as a reference the measured friction factor in the coated pipe indicates that the coated surface is not hydraulically smooth. The friction factor is therefore a function of both roughness and Reynolds number which is typical for the transition region between smooth and rough flow (Nikuradse, 1933). The results from measurement series 1 and 2 show somewhat different behavior. The low pressure result from the first test (marked o in Figure 6.1) departs from the smooth flow behavior and these results can be found to fit the Colebrook-White equation with an ecuivalent sand-grain roughness ky 1 jm (Sletfjerding et al., 1998) (the reference is reproduced in the appendix). ‘The friction factor results from the second flow test (at higher pressures) show a more smooth behavior than the results from the first test. The roughness measurement on the coated surface showed that the surface roughness of the coated pipe in the second flow test was smoother than in the first flow test (Tables 5.3 and 5.4), a result that was rather surprising. Aging of the epoxy-coating may be an explanation for this behavior. In Figure 6.6 the measured friction factor in the “smooth” pipe is plotted with error bars according to the estimated error (see error analysis section page 43). The figure shows that the deviations between the two sets of data are within the estimated uncertainty.108 6. Correlations However, the downward trend in the friction factor from the second test series is not seen in the result from the first test series. This behavior is probably due to errors in the mass flow measurements. It was found during the tests that the results from the turbine meter deviated from the readings of the reference sonic nozzles. The difference between the two readings increased with the flow rate and was maximum 0.94 % at the highest flow rates. ‘This deviation is larger than the specified accuracy of the mass flow measurements. 6.3 New correlations ‘To model the deviations in the friction factor from smooth to rough behavior an equation combining smooth flow and rough flow is needed. The Colebrook-White equation has been the standard correlation for this purpose, but Zagarola (Zagarola, 1996) showed that the smooth friction law of Prandtl (on which the Colebrook-White equation was based) was not accurate at high Reynolds numbers. However, the measurements of the flow in the rough test pipes show that the friction factor correlation for rough flow presented by Nikuradse is valid at high Reynolds numbers. Combining the smooth friction law presented by Zagarola (Equation 2.31) and the rough friction correlation of Nikuradse (Equation 2.25) a semi-empirical friction factor correlation valid in both smooth, transitional and rough regime was obtained: 1 1.55 k, —1.89 log(/—— $1.06) 6.5) (Fa Histon es + gE) (5) ‘The combination of the two equation was done as in the Colebrook-White equation, where the transition is governed by a linear interaction between the viscous length scale and the wall roughness. In Equation 6.5 the equivalent sand-grain roughness was used as the roughness measure as in Nikuradse’s (1933) work. In Figure 6.6 it is shown that the friction factor in the coated test pipe can be modelled by Equation 6.5 with an equivalent sand-grain roughness of ky = 0.7 pm and ky = 0.25 im for the first and second test series respectively. The deviations between Equation 6.56. Correlations 109 and the experimental points are within the experimental uncertainty in the frietion factor (42.8 %) except for the measurement point at the highest Reynolds number (Re = 30.6 x 10°) where the deviation is 4.2 %. In the seven test pipes that reached fully rough flow condition correlations between the directly measured wall roughness and the friction factor can be found (Equations 6.2 and 63). Combining the smooth flow correlation of Zagarola (Equation 2.31) with the rough flow correlation based on R,/H" (Equation 6.2) a new Colebrook-White type equation VG 1.89 log( 1.55 + (Fl 3) (6.6) was obtained. 0.683) ReV7 Using Equation 6.3 for the rough flow regime (that is Ry as the roughness measure) a similar correlation was derived. Vi =1.89 106s + (BtS)™) 627) ‘The exponent in the relative roughness term in Equations 6.5, 6.6 and 6.7 is a result of the difference in the leading constants of the smooth flow and the rough flow correlations. However Equation 6.2 can also be expressed as : Vj- sam Equation 6.8 is obtained by forcing the leading constant to be 1.89 and least-square 1.8910g(—) + 0.45 = 1.89 log( 2882 (6.8) fitting the second constant to the experimental data. The maximum deviation between the experimental points and Equation 6.8 was 1.5 %. Using this equation for the rough flow regime a new Colebrook-White type correlation was derived. =1.89 oy re + Poll (69) T Vi * 0.865D Equation 6.9 combines the smooth flow urnen for high Reynolds number flow with the fully rough correlation for the test pipes in this study. It constitutes a simple way of predicting the friction factor in smooth and rough pipes at high Reynolds numbers. ‘The correlations presented above for the friction factor in smooth and rough pipes are of the Colebrook-White type. The transition between smooth and rough flow is governed by110 Correlations 4 linear interaction between the roughness measure and the viscous length scale. However, Uhl et al. (1965) stated that the transition between smooth and rough flow as modeled by the Colebrook-White equation was too gradual for steel pipelines. This indicates that the linear model of Colebrook-White for the transition region is not appropriate. Assuming that the interaction between the wall roughness and the viscous length scale is governed by a higher order expression (y, = [C,k" + C2(2)"]/*, with reference to Equation 2.26), a more abrupt transition between smooth and rough flow is obtained, Using this model for the transition region Equations 6.10 and 6.11 were obtained (using k, and R,/H’ as roughness measures). 1 7 (6.10) T__ 1.89, 4 Balt pa ot py + A a ‘The roughness function including all pipes is shown in Figure 6.7. Equation 6.10 is plotted with n = 1 and n =2in the figure. It is shown that n =2 give a more abrupt transition between smooth and rough flow. Haaland (1983) used a similar approach in his explicit friction factor correlation (Equation 2.28) to model the transition region. However, his proposed friction factor was based on the Prandtl law of flow in smooth pipes, which has been shown to be inaccurate for high Reynolds numbers (Zagarola, 1996). Following his arguments an explicit approximation to Equation 6.10 can be derived. Zagarola’s correlation for smooth flow (Equation 2.31) can be approximated (the deviations are within +£0.5 % in the range Re =1 x 10° — 1 x 108) 1 log Fe (6.12) vi Combining Equation 6.12 with the rough flow correlation of Nikuradse (Equation 2.25) an explicit approximation to Equation 6.10 was obtained, (6.13)6. Correlations 111 Similarly an explicit correlation based on R,/H’ as the roughness measure was derived. oul Seen + (aL ym) 0.865D (14) Equations 6.13 and 6.14 enables simple explicit estimations of the friction factor either from the equivalent sand-grain roughness or from the directly measured wall roughness. 6.4 Pipeline data When comparing the pipeline data with the experimental data the differences in the systems should be addressed. The experimental work is aimed at measuring the friction {actor in the pipes without the influence of additional pressure drop which are commonly found in operating pipelines. Uhl et al. (1965) identified and discussed the effect of additional pressure drop in a pipeline. They listed several such ”minor” losses in their assumed order of importance; weld beads, bends, sags and crests, deposits of solids, fittings, roughness changes, compressor station effects and suspensoids (water, heavy components). All these factors will influence the flow in the pipelines to deviate from *idealistic” conditions In an offshore coated pipeline most of these minor loss contributions are of little significance. The pipelines are continuously laid on the sea bottom with no bends. The bending of the pipeline due to sea bottom topography will generally cause large radius of curvature bends because of the stiffness of the pipe, and the additional resistance due to such bends is small. The gas transported is very clean, and the effect of solids or suspensoids can be neglected However the weld beads between the pipe sections will influence the pressure drop in the pipeline. Uhl et al. (1965) suggested two techniques for estimation of the loss due to welds. Both these seriously overestimate the pressure drop due to bends (their Figures C-15 and C-16). Idelchik (1986) (based on experimental work in gas pipelines) proposed12 6. Correlations that the increase in the friction factor due to welds in a pipe may be estimated: Zo A 3/2 (6.15) Seta where l, is the weld spacing and 6 is the weld height. The constant C’ is a function of the weld spacing and is equal to 0.52 for a weld spacing of 10 m (Idelchik, 1986). From Equation 6.15 the additional pressure drop due to welds in the coated pipelines in the study (6 ~ 3mm, ly = 10 m) was 1 — 2 % of the mean friction factor. Figure 6.8 shows the pipeline data when Equation 6.15 is used to correct: for the pressure drop due to welds. The friction factor in coated offshore pipelines shows a smooth flow behavior. Also shown in the figure is the Colebrook-White equation with a equivalent sand-grain roughness of 1.5 zm. The deviation between the measured friction factor and the correlations are within +2 % for both the smooth (Zagarola) and Colebrook-White correlation. ‘The pipeline data also includes measurements from gas pipelines on land. However, noticeable difference is observed between the friction factor in pipelines on land and off is the additional losses due to bends shore (Figure 6.8). One reason for the differenc and welds. The laying of a gas pipeline on land makes it necessary to include bends in the pipeline. Because pipeline profiles were not available, no correction for the effect of bends was done. The quality of the welds is better in offshore pipelines compared to pipelines on land because of the laying technique. This will cause a larger pressure drop due to welds in onshore pipelines. ‘The results from the onshore pipelines indicate that a constant shift in the friction factor rather than a change in surface roughness has occurred. Therefor, a drag factor as used. in the AGA correlations will probably be an appropriate way of modeling the deviation from the smooth pipe friction factor. Using a drag factor of 0.94 in the AGA correlations reasonable agreement with the field data was obtained. However, with only two separate tests it is difficult to analyze this results. The data set also included one test in a pipeline without coating. The friction factor inCorrelations 113 this pipeline is shown in Figure 6.8. If it assumed that the flow in the pipe is in the fully rough regime the equivalent sand-grain roughness for the steel pipe is ky = 12 jam (correction for welds included). Without the correction for welds the equivalent sand-grain roughness is ky = 13.3 ym, Only one measurement point is given for the steel pipeline, which limits the possibilities for conclusions about the flow in bare steel pipelines.14 6. Correlations ‘Table 6.1: Equivalent sand-grain roughness compared with mean bead size and measured roughness Pipe | ky (um | deeaa [um] | Ra [ym] | Ry [um] | Re (uml | Ro/H’ [um] Coated | — 108 [iat [615 [0.93 Steel | 21 244 [366 | 2128 | 2.79 3 QT 25 4.73 5.98 29.27 3.42 4 62 55 10.65 [13.28 | 62.84 | 9.03 8 76 78 190 [1457 | 62.66 | 10.33 5 87 90 16.08 | 1882 | 72.10 | 13.07 6 12 | 120 20.95 | 25.35 | 98.32 | 19.20 7 181 | 150 23.68 | 31.02 | 192.10 | 25.02 Table 6.2: Constants in the curve fit of (1/f)"/? versus log(r/k) k A |B Ra 1.90 | 0.51 Ry 2.06 | 0.10 R 2.53 | 0.04 RH! | 1.94) 0.26 ky (fitted) | 2.04 | 1.67 ky (given) | 2.00 | 1.74 Table 6.3: Constants in the curve fit of k, versus the measured roughness k A Ra | 6.75 R, | 545 R 1.34 Rq/H!' | 7.20.025; aoa a We ke= 181um voy 9 9 9 tore kee 142m oo 8 8 eee ke BHM 3 em Bo 0 eS kee 76m oP & +++ Soars ee 3 5 bob he 27m = ee oT) oot . O29 oP P.Q0, 9.008 5 10 15 20 5 30 Reynolds number [x 1¢6] Figure 6.1: Equivalent sand-grain roughness of test pipes; o and +: coated pipe, x: steel pipe, +: pipe 3, <: pipe 4, O: pipe 8, o: pipe 5, V: pipe 6, A: pipe 7, line: Zagarola frietion factor (smooth pipe) 15Ra Rq 95 9 % q 8s| as 2 8 a , : 75 75 7 7 é 68 65) 35 4 45 3 35 4 45 g € Rz Rg = 9 9 as 85| a e a 73] 78 7 7 65 “ 5 35 3 35 «ORS 35 4 45 log(rik) Figure 6.2: Correlations between (1//)'/? and log(r/k) for the different roughness measures 116112 14 16 18 2 22 24 26 28 3 log(rks) 25 g 2 z Bis ‘S14 08 O05 1 15 2 25 3 35 logtks") Figure 6.3: Correlations from Nikuradse’s work; Top figure: Correlation between (1/f)'/? and log(r/k,), Bottom figure: the roughness function 1?ks [micron] Ra Rq 200 200 5 180) 160) 100] 7 oo) i 60) * +0] * % 10 20 30 % 020 Rz Rg 200 200 rT) E 160] +100) K 100) 50) * 50] ot 4 ° 0 100 «150 ° 10 2 30 Measured roughness [micron] Figure 6.4: Correlation of k, versus measured roughness 11825) z 4 29 eB <8 + ALERM RS damper nates a 15) g © 1 O85 05 1 15, 2 25 3 35 logtks*) 1 = ° Bs + SERN tant g a S05 = to 0s 1 15 2 25 3 35 log(k"), k=Ra/H’ Figure 6.5: Roughness function of the test pipes; Top figure k = k,, bottom figure k = 2, /H', x: steel pipe, +: pipe 3, <: pipe 4, O: pipe 8, o: pipe 5, V: pipe 6, A: pipe 7 119© First measurement series * Second series P=120 bar 2 Second series P=95 bar — Smooth (Zagarola) ~' Equation 6.5, ks= 0.7 ym ~~ Equation 6.5, ks= 0.2 um friction factor, 0.006 5 10 16 20 25 30 Reynolds number [x 1¢6] Figure 6.6: Friction factor smooth coated pipe, error bars indicate the uncertainty in the friction factor 12025) (1f)" — 2 logtks/r) Figure 6.7: Roughness function and transition region; o and *: coated pipe, x: steel pipe, +: pipe 3, <: pipe 4, O: pipe 8, o: pipe 3, V: pipe 6, A: pipe 7, line: Equation 6.10 dashed line: Equation 6.10 n=2. 121an Friction factor, 75| 10 20 Reynolds Number [x 166] jose Offshore pipeline, coated Onshore pipeline, coated Offshore pipeline, steel C-W equation, ke= 1.5 um| ‘Smooth (Zagarola) Figure 6.8: Pipeline data corrected for the effect of welds 122Chapter 7 Discussion In the previous chapters the flow in offshore gas pipelines is shown to be characterized by high Reynolds numbers (Re > 1 x 10°). As discussed in the literature review, there is presently a debate among researchers on how to analyze turbulent flow in pipes at high Reynolds numbers. Unfortunately, we were not able to measure the velocity profile in the test pipes and it is therefore difficult to contribute in this debate. The correlation presented by Zagarola (1996) (Equation 2.31) has been used as the reference for smooth flow at high Reynolds numbers. ‘The Zagarola correlation gives friction factors 4 % higher than the Prandtl law of flow in smooth pipes for Re = 2 x 10”. ‘To verify the behavior of the friction factor in the transition region with the results from the present experimental work proved to be difficult. The smoothest pipe in the study (the pipe with plain coating) showed only smalll deviations from the smooth law indicating that only the very early part of the transition was covered (Figure 6.7). The other 7 pipes showed typically rough flow behavior. No measurements covered all three regions (smooth, transition and rough). Therefore, it was difficult to establish the type of model most suitable for the transition. The coated pipe seemed to follow the linear transition model (Equation 6.10, n = 1). ‘The results from the steel pipe indicated a more ion (Equation 6.10, n = 2) abrupt trai 123124 7. Discussion ‘The lack of data in the transition region was rather surprising. The experiments were designed in such a way that data in the smooth, transitional and rough flow regimes should be obtained because of the spread in the surface roughness of the 8 test pipes. Nevertheless, we failed to get sufficient data in the transition region. Interestingly, the “failure” shows how sensitive the flow at high Reynolds numbers is to small changes in surface roughness. The measured Ry in the coated pipe and the smoothest of the glass-bead roughened pipe (pipe 3) was 1.4 jum and 6.0 pm respectively. The equivalent sard-grain roughness (a hydraulic measure of the roughness) of the two pipes was 0.7 jum and 27 jam. ‘The uncertainty in the calculated equivalent sand-grain roughness was larger for the pipe costed with smooth coating compared to the 7 *rough” pipes. First of all, in the coated pipe a model for the transition was needed in order to calculate an equivalent sand-grain value. The fact that only parts of the transition was covered in the measurement led to uncertainties in the choice of model and the calculated roughness value. Secondly, in the transition region the flow is governed by an interaction of the wall roughness and the viscous length scale. Probably, due to the dual length scales, very small changes in wall roughness will give large effects on the flow. Accordingly one should expect larger spread in she measured data in this region, which can explain the differences between the results in she smooth coated pipe for test series 1 and 2. To more accurately define the fully rovgh flow behavior of the smooth coated pipe, measurements at even higher Reynolds numbers is needed. To find the equivalent wall roughness of coated pipelines has been a major object of this study. ‘The values found in the literature vary significantly. Blevins (1992) states that the equivalent sand-grain roughness of welded steel tubes with "new, smooth centrifugally applied enamel” is in the range 9 — 60 jum. Asante (1996) reported that a representative value for coated and uncoated gas pipelines was 6.4 jan and 19.1 jum respectively. These values has been used since 1993 in the NOVA Gas Transmission pipeline system. Earlier the values used was 10.2 jm and 16.5 jm (Asante, 1996). In this study it was found that the friction factor in offshore coated gas pipelines was well7. Discussion 125 represented by the smooth correlation presented by Zagarola (1996) (or Equation 6.5 with ky = ) if the additional pressure drop due to welds were taken into account. Without the correction for welds Equation 6.5 with k, = 0.7 jam was an appropriate model. When using the Colebrook-White equation, the equivalent sand-grain roughness of coated offshore gas pipelines was found to be 1.5 ym and 2.0 ym with and without the corrections for welds. ‘The analysis of the offshore gas pipelines showed that internal coating successfully reduced the friction factor to its smooth flow limit. Therefore, the use of smoother coatings will probably have a limited effect on the pressure drop in large diameter pipelines. However, in the test pipe coated with smooth coating, the friction factor deviated from smooth flow behavior and a smoother coating could probably reduce the pressure drop even further. ‘The additional pressure drop in pipelines due to welds was found to be 1-2 % of the wall friction pressure drop. Accordingly a reduction of the pressure drop of 1-2 % could be achieved if the weld beads were leveled and coated. However, Figure 6.8 indicates that the correction used (Equation 6.15) may overestimate the pressure drop due to welds in offshore coated pipelines because the friction factor tends to be lower than what is found from the smooth correlation. Further study of the pressure drop due to welds in large diameter pipelines should be made. ‘The results from this study show that internal coating effectively reduce the pressure drop and increase the transport capacity of a pipeline. Using the equivalent sand-grain roughness for a steel pipe (k, = 19.1 pm) as reported by Asante (1996) and k, = 2.0 jam for the coated pipes, it is found that the friction factor at typical operating conditions (Re = 2 x 107) is reduced by 18 % in a 40 inch (D = 0.9664) pipeline due to internal coating. The corresponding increase in transport capacity is 11%. In a pipe with D=0.5 m the reduction in friction factor is 25 % leading to an increase in transport capacity of 16 %. Because the friction factor is dependent on the ratio k/D, the drag reducing effect of126 Discussion coatings will increase as the diameter decreases. The large difference in the measured friction factor between the steel pipe and the coated pipe in the experimental work shows the potential for the use of internal coatings also in gas production. The inner diameter of the test pipes are directly comparable to typical tubing sizes in gas wells. In the paper in the appendix (Sletfjerding, 1998), the potential for the use of coatings for drag reduction in gas well tubing is discv ed. In high rate gas wells the tubing pressure drop can be reduced significantly if internal coating is used, ‘The experimental work on the flow in the pipes roughened with glass-beads showed the well-known constancy of the friction factor at high Reynolds numbers. The rough surfaces were made by adding glass beads with a quite wide distribution of sizes in order to create a rough surface with a more random nature than the sand-grain surfaces of Nikuradse (1933). Nikuradse used a single height parameter (sand-grain diameter) to describe the wall roughness of his rough pipes. However, because he used single diameter sand grains the texture of the wall roughness was implicitly given. For random surfaces, which are definitely more complex in texture than sand-grain roughness, the use of a. single height parameter to describe the surface roughness is more problematic. Some measure of the texture should be included in the description of the wall roughness to model the fluid- surface interaction accurately. Some authors, for example (Townsin et al., 1985), have used a combination of height parameters and horizontal parameters to form an “equivalent” height parameter. The procedure recommended by Townsin et al. (1985) was followed in this work without success. They used a correlation length defined from the autocorrelation of the measured roughne: s profile as a horizontal roughness measure and A, as a height measure. In this work a combination of a height parameter (Rg) and a texture parameter (H’) was used to describe the surface roughness of the rough pipes. The use of the parameter R,/H’ is not reported in the literature, but was found to correlate far better with the measured friction factors than single height parameters (Ry, Ry or Re). The equivalent sand-grain roughness in the test pipe coated with smooth coating was found to be 0.7 jum in the first test series. Calculating the equivalent sand grain7. Discussion 127 roughness in this pipe from the expression k, = 7.2Rq/H' a value of 6.7 ym is obtained. ‘The difference between the two estimates is large and the use of the Rg/H' parameter as a roughness measure for the "smooth” pipe gives overestimated friction factors. The reason for the difference may be the uncertainty in the estimated equivalent sand-grain roughness in the smooth pipe. Measurements at even higher Reynolds numbers is needed in order to obtain a well-defined value of the equivalent sand-grain roughness of the coated pipe. In engineering the use of the equivalent sand-grain roughness to describe rough walls has been the standard approach since Nikuradse presented his rough pipe flow results (Nikuradse, 1933). In several wall-bounded flows of engineering interest such as gas pipelines, vessel hulls, propellers, turbine engines and aircraft, the wall roughness is an important parameter in determining the performance. The results from this work represent a more direct alternative than the equivalent sand-grain roughness approach to escimate the influence of wall roughness on rough wall flows. Therefore, the presented correlations have potential applications not only to rough pipe flow but also to other rough wall flows.128Chapter 8 Conclusions Flow measurements in glass bead roughened test pipes at high Reynolds numbers showed that the friction factor was constant for values of ky/t* > 63,where hy is the equivalent the sand-grain roughness and [+ is the viscous length scale. The fully rough friction factor correlation of Von Karman and Nikuradse was shown to be an appropriate model for the friction factor in rough pipes also for high Reynolds number flow. The fully rough friction factor in the test pipes was well correlated with the directly measured wall roughness, and friction factor correlations based on measured wall roughness were presented. The parameter Rg/H', which is a combination of a height parameter and a texture parameter, was proposed as the roughness measure that gave the best correlation between the measured friction factor and the wall roughness of the test pipes. Ry is the root-mean-square roughness and H' is related to the slope of the power spectrum of the measured roughness profile. Combining the friction factor laws for smooth and rough flow, correlations for smooth, transitional and rough flow in pipes were presented. The correlations were presented both in terms of the traditional equivalent sand-grain roughness and the directly measured wall roughness. The presented correlations are believed to model high Reynolds number flow in smooth and rough pipes more accurate than existing correlations. Explicit 129130 Conclusions approximations to the implicit friction factor correlations were also presented to simplify the calculations. ‘The friction factor in coated gas pipelines was found to be well modeled by the hydraulically smooth friction factor correlation when the additional pressure drop due to welds was accounted for. Thus the wall roughness in the pipelines had negligible effect on the pressure drop. Without correcting for the pressure drop due to welds the friction factor in offshore coated pipelines could be modeled by Equation 6.5 with ky = 0.7 yum. If the Colebrook-White correlation is preferred the equivalent sand-grain roughness of coated offshore pipelines was found to be 1.5 um and 2.0 um respectively with and without correction for the additional pressure drop due to welds. These values of the equivalent sand-grain roughness are significantly lower than what is cited in the literature. ‘The flow experiments and the pipeline data documented the drag reducing effect of in:ernal coating in gas pipes. In the flow experiments the friction factor in the coated pipe was 31 % lower than in the steel pipe at a Reynolds number of 1 x 10’. The reduction in friction factor corresponded to an increase in transport capacity of 21 % due to internal coating. The pipeline data showed that the internal coating reduced the friction factor in the pipelines to smooth flow behavior for Reynolds numbers up to 3 x 107.References Adewumi, M. A. (1997). Natural gas transportation issues. Journal of Petroleum Technology, 49(2):139. Asante, B. (1996). Hydraulic and cost of service analysis for internal pipe coatings. In The International Pipeline Monitoring and Rehabilitation Seminar, pages 145-152, Houston Texas. Hart's Pipeline Digest and Systems Integrety. ASME (1988). Pipe: flanges und flanged fittings. The American Society of Mechanical Engineers, ASME/ANSI B16.5 edition. Barabasi, A. and Stanley, H. (1995). Fractal Concepts in Surface Growth. Cambridge University Press. Barenblatt, G. (1993). Scaling laws for fully developed turbulent shear flows. Part 1 Basic hypothesis and analysis. Journal of Fluid Mechanics, 248:513-520. Barenblatt, G., Chorin, A., and Prostokishin, V. (1997). Scaling laws for fully developed turbulent flow in pipes. Applied Mechanics Review, 50(7):413-429. Barenblatt, G. and Chorin, A. J. (1998). Sealing of the intermediate region in wall-bounded turbulence: The power law. Physics of Fluids, 10(4):1043-1044. Barenblat, G. and Prostokishin, V. (1993). Scaling laws for fully developed turbulent. shear flows. Part 2. Processing of experimental data. Journal of Fluid Mechanics, 248:521-529. 131132 References Bendat, J. and Piersol, A. (1986). Random Data, Analysis and Measurement Procedures. John Wiley and Sons, New York, second edition. Benedict, R. P. (1984). Fundamentals of Temperature, Pressure and Flow Measurements. John Wiley and Sons Inc., New York. Berry, M. and Hannay, J. (1978). Topography of random surfaces. Nature, 273:573. Blevins, R. D. (1992). Applied Fluid Dynamics Handbook. Krieger Publishing Company, Florida USA. Brown, S. R. and Scholz, C. H. (1985). Broad bandwidth study of the topography of natural rock surfaces. Journal of Geophysical Research, 90(B14):12575~12582. Chue, S. (1975). Pressure probes for fluid measurements. Progress in Aerospace Science, 16(2):147-223. Chung, T., Ajlan, M., Lee, L., and Starling, K. (1988). Generalized multi-parameter correlation for nonpolar and polar fluid transport properties. Industrial and Chemical Engineering Research, 27:671-679. Colebrook, C. (1939). Turbulent flow in pipes, with particular reference to the transition regime between smooth and rough pipe laws. Institution of Civ. Eng. Journal, 11:133-156. paper no. 5204. DIN 477 (1990). DIN 4777, Profilfilter zur anwendung in elektrischen tastschnittgeraten. Deutsche Norm. Doebelin, E. O. (1990). Measurement systems. McGraw Hill, New York. Drag, K. A. (1997). Design av 6m pipe spool NINU/K-lab. Technical report, Proserv AYS, Norway. Erdal, A. (1997). Computational analysis of the flow field downstream of flow conditioners. PhD thesis, Norwegian University of Science and Technology, Department of Applied Mechanics, Thermodynamics and Fluid Dynamics.References 133 Feder, J. (1988). Fractals. Plenum Publishing Company, New York. Ganti, S. and Bhushan, B. (1995). Generalized fractal analysis and its application to engineering surfaces. Wear, 180:17-34 Gregory, G., Aziz, K., and Moore, R. (1979). Computer design of densi Journal of Petroleum Technology, (January):40-50. hase pipelines. Grigson, C. (1984). Nikuradse’s experiment. AIAA Journal, 22(7):999-1001. Grigson, C. (1992). Drag losses of new ships caused by hull finnish. Journal of Ship Research, 36(2):182-196. Haaland, S. (1975). Calculation of pressure and temperature along a gas pipeline taking into account heat transfer and the Joule-Thomson effect. Technical report, Institutt, for Aero- og Gassdynamikk, Norges Tekniske Hogskole. Haaland, S. (1983). Simple and explicit formulas for the friction factor in turbulent pipe flow. Journal of Fluids Engineering, 105:89-90. Hinze, J. (1962). Turbulent pipe flow. Mechanique de la turbulence, (108):129-165. Hinze, J. (1975). Turbulence. McGraw-Hill Inc., New York. Idelchik, 1. (1986). Handbook of Hydraulic Resistance. Hemisphere Publishing Corporation, New York. K-Lab (1997). Description of the primary calibration system and the traceability chain at K-Lab. KarstoMetering and Technology Laboratorium, Statoil, Haugesund, Norway. Katz, D. L. and Lee, R. L. (1990). Natural Gas Engineering. McGraw-Hill, Inc. New York. Koch, C. and Smith, L. (1976). Loss sources and magnitudes in axial-flow compressors. Journal of Engineering for Power, 98:411-423. Kut, S. (1975). Internal and external coating of pipelines. In First International Conference on the Internal and External Protection of Pipes. Paper Ad.134 References Massey, B. (1990). Mechanics of fluids. Chapman and Hall Ltd. London, 6th edition. MATLAB (1992). MATLAB, High-Performance Numeric Computation and Visualization Software. wersion 4.2. The Math Works Inc., Mass. USA. Matthew, G. (1994). Turbulent flow in pipes: a historic speculation. Proc. Instn. Civ. Eng., 106:311-316. Mecain, W. D. (1990). The Properties of Petroleum Fluids, Pennvwell Publishing Company, Tulsa Oklahoma. Millikan, C. A. (1938). A critical discussion of turbulent flows in channels and circular tubes. In Proceedings of the 5th International Congress of Applied Mechanics, pages 386-392. Moody, L. (1944). Friction factors for pipe flow. Transaction of the ASME, 66:671-634. Nikuradse, J. (1932). Gesetzmessigkeiten der turbulenten stromung in glatten rohren. In Forschungsheft 356, volume B. VDI Verlag Berlin. Translated in NASA TT F-10, 359, 1966. Nikuradse, J. (1933). Stromungsgesetze in rauhen rohren. In Forschungsheft 361, volume B. VDI Verlag Berlin. Translated in NACA Technical Memorandum nr. 1202, 1950. OED (1998). Environment "98 - The Norwegian Petroleum sector. Ministry of Petroleum and Energy (OED), Oslo Norway. Patel, V. (1998). Perspective: Flow at high Reynolds number and over rough surfaces - Achilles heel of CFD. Journal of Fluids Engineering, 120:434-444. Powe, R. and Townes, H. (1973). Turbulent structure in rough pipes. Journal of Fluids Engineering, pages 255-262. Provder, T. and Kunz, B. (1996). Application of profilometry and fractal analysis to the characterization of coatings surface roughness. Progress in Organic Coatings, 27-219-226,References 135 Purushothaman, K. (1993). Reynolds number effects on the momentum flur in turbulent boundary layers. PhD thesis, Yale University. Raupach, M., Antonia, R., and Rajagopalan, S. (1991). Rough wall turbulent boundary layers. Appl. Mech. Rev., 44(1). Reynolds, O. (1883). An experimental investigation of the circumstances which determine whether the motion of water shall be direct or sinuous, and of the law of resistance in parallel channels. Phil. Trans. Roy. Soc., 174:935-982. Robertson, J., Martin, J., and Burkhart, T. (1968). Turbulent flow in rough pipes. I and EC Fundamentals, 7(2):253-265. Sander, M. (1991). A Practical Guide to the Assessment of Surface Texture. Feinpruf GmbH, Gottingen Germany. Sayles, R. and Thomas, T. (1977). The spatial representation of surface roughness by means of the structure function: A practical alternative to correlation. Wear, 42:263-276. Sayles, R. and Thomas, ‘T. (1978). Surface topography as a nonstationary random process. Nature, 271:431-434. Schlichting, H. (1979). Boundary layer theory (Seventh Edition). McGraw-Hill Inc. New York. Translated by Dr. J. Kestin. Schmittbuhl, J., Vilotte, J.-P., and Roux, S. (1995). Reliability of self-affine measurements. Physical Review E, 51(1):131-147. Shaw, R. (1959). The influence of hole dimensions on static pressure measurements. Journal of Fluid Mechanics, 7(4):550-564. Simonsen, I., Hansen, A., and Nes, O. M. (1998). Using wavelet transforma for Hurst exponent determination. Physical Review E, 58(3):2779-2787. Singh, G. and Samdal, O. (1988). Internal coating justified by operating costs. Oil and Gas Journal, pages 50-55.136 References Sjoen, K. (1998). Capacity tests of gas pipelines in the Norwegian gas infrastructure. Private communications. Sletfjerding, B. (1997). Innvendig maling av test-seksjoner med epoxy-belegg. Technical report, Norges Teknisk Naturvitenskapelige Universitet, Institutt for Petroleumsteknologi og Anvendt Geofysikk. Sletfjerding, B. (1998). Friction factor in coated gas pipelines and well tubing. In SPE European Petroleum Conference, Student paper contest, Doctoral division, Den Hague, Netherlands. SPE Paper 52059 Sletfjerding, E., Gudmundsson, J. S., and Sjven, K. (1998). Flow experiments with high pressure natural gas in coated and plain pipes; comparison of transport capacity. In Proceedings of the 30th Pipeline Simulation Interst Group (PSIG) Annual Meeting, Denver, Colorado. Smith, R., Miller, J., and Ferguson, J. (1956). Flow of natural gas through experimental pipelines and transmission lines. Bureau of Mines, Monograph 9. American Gas Association. Starling, K. E. (1992). Compressibility factors of natural gas and other related hydrocarbon gases. Transmission Measurement Committee Report no. 8. American Gas Association. Svardstrém, A. (1987). Tillémpad Signalanalys. Studentlitteratur, Lund Sweden ‘Tennekes, H. (1968). Outline of a second order theory for turbulent pipe flow. AAIA Journal, 6(9):1735-1740, ‘Tennekes, H. and Lumley, J. (1987). A first course in turbulence. The MIT Press, Cambridge Mass. ‘Thomas, T. (1982). Rough Surfaces. Longman group Limited, London. Tian, S. and Adewumi, M. (1994). Development of analytical design equation for gas pipelines. SPEPF, pages 100-106.References 137 ‘Townsin, R., Spencer, D., Mosaad, M., and Patience, G. (1985). Rough propeller penalties. Society of Naval Architects and Marine Engineers (SNAME) Transaction, 93:165-187. Uhl, A. (1965). Steady flow in gas pipelines. Institute of Gas Technology, Technical Report no. 10. American Gas Association. Warburton, C. (1974). Surface roughness of graphite and its effect on friction factor. Proc. Instn. Mech. Engrs, 188(457-460). Zagarola, M., Perry, A., and Smits, A. (1997). Log laws or power laws: The scaling in the overlap region. Physics of Fluids, 9(7):2094-2100. Zagarola, M. V. (1996). Mean-flow scaling of turbulent pipe flow. PhD thesis, Princeton University, USA.138Appendix A Published papers ‘Two conference papers resulting from my doctoral work have been published. ”Flow experiments with high pressure natural gas in coated and plain pipes; Comparison of transport capacity” was presented at the 30th Annual meeting of the Pipeline Simulation Interest Group (PSIG) in Denver, Colorado, October 1998 Friction factor in coated gas pipelines and well tubing” was presented in the student paper contest, doctoral division, at the SPE European Petroleum Conference (EUROPEC) in The Hague, Netherlands, October, 1998. Both papers are reproduced in the following pages. 139140PSIG 30th Annual Meeting Denver, Colorado, October 28-30, 1998 FLOW EXPERIMENTS WITH HIGH PRESSURE NATURAL GAS IN COATED AND PLAIN PIPES Comparison of transport capacity Elling Sletfjerding, Jon S. Gudmundsson, Karl Sjoen) Department of Petroleum Engineering and Applied Geophysics Norwegian University of Science and Technology 7084 Trondheim, Norway (statoil, Transport and Systems Technologies, 4035 Stavanger, Norway Abstract An experimental study of pressure drop in pipes at high Reynolds numbers has been carried out to determine the effect of roughness on the transport capacity of natural gas pipelines. In a high pressure flow loop the pressure drop was measured in 6 m {19.7 ft] long test pipes with 150 mm [5.9 in] inner diameter. The measurement of pressure drop of a coated pipe and a bare steel pipe showed the drag reducing effect of pipeline coating. ‘The measurements were performed at Reynolds numbers which are typical for subsea pipelines in the North Sea (Re ~ 10"). At a Reynolds number of 1 x 10” the friction factor for the coated pipe was 31 % lower than for the steel pipe, which corresponded to an increase in the transport capacity of 21 %. Roughness values obtained by direct measurements on the pipe wall were compared to the roughness values obtained from flow tests (equivalent sand-grain roughness k,). For the steel pipe the measured roughness parameter R: (mean peak to valley height) corresponded well with ky. In the ‘ky was not easily defined because the flow never reached fully jons at the maximum Reynolds number.Introduction Internal coatings have been used with success in gas pipelines since the 1950's [1]. Economic studies [2] show that the typical pay-back time for the investment in internal coating is 3 ~ 5 years due to improvements in pipeline hydraulics. It is well known that internal coatings reduce the friction in the pipeline and therefor reduce the operating cost of compressors. In Norway 35 % of offshore generated power [3] is used for gas export compressors. The use of coatings to reduce the operating cost of compressors is therefor important, In addition the coatings protect the pipe wall against corrosion and reduce the need for maintenance of the pipeline [1]. Also, pigging, operations are improved. ‘The aim of this study was to study the friction factor of coated pipes. In Norway internal coatings are used in the export pipelines which transport. gas from the Norwegian shelf to continental Europe. The object has been to establish more accurate friction factor correlations for coated gas pipelines and thus being able to predict the capacity of large diameter trunk lines accurately. ‘The flow in offshore gas pipelines is characterized by large Reynolds numbers (Re ~ 10") due to the low viscosity and the relative high density at typical ‘operating pressures (100 ~ 180 bar). From the classical Colebrook-White friction factor correlation [4] itis seen that even very minute irregularities on the pipe wall will have a significant effect on the friction in the pipeline at high Reynolds numbers. However the measurements from which the Colebrook-White correlation was developed reached a Reynolds number of 1x 10° as maximum, one decade lower than what is typically encountered in offshore gas pipelines. Our aim has been to investigate the influence of roughness on the friction factor for high Reynolds numbers. ‘To characterize hydraulic roughness of pipelines the equivalent sand-grain roughness has traditionally been used. The concept refers to the rough pipe experiments of Nikuradse [5], and as common practice the hydraulic properties of a pipeline are compared to Nikuradse’s work to "back-track” the equivalent roughness. Equivalent sand-grain roughness of coated pipes are also given in handbooks for example (6), but the values cited vary significantly. In addition to finding the equivalent sand-grain roughness we have measured the wall roughness of the coated pipes. ‘The paper is organized in five parts: After this introduction a brief literature review is presented to sum up earlier relevant work and to present the theory. ‘Then our flow experiments and the roughness measurements are presented. In the results chapter the results from the experimental work are presented and.discussed. The major findings of this work are summed in the conclusions. Background In horizontal pipeline flow the main cause of energy loss is the friction between the pipe wall and the moving gas. The steady state frictional pressure drop may be calculated from the well known Darcy-Weisbach Equation: 2 @ Equation 1 is the remaining parts of the governing equation for pipeline flow when the kinetic energy term and the elevation term are neglected. ‘The influence of the friction factor on the pipeline capacity may be seen from Equation 3, which can be derived by combining Equation 1 with the real gas law (Equation 2). Note that Equation 3 is similar to the Weymouth flow equation with the exception that no friction factor correlation is inserted. M,P. @) ZRT ; ) a= 7, From Equation 3 it is seen that the volumetric rate of flow is proportional with the friction factor to the minus 1/2 power ( Qo ~ f-"/2) and from Equation 1 the pressure drop in the pipeline is proportional to the friction factor (dP ~ f). Thus a reduction in the friction factor of say 10% will increase the throughput of the pipeline by 5% if the same pressure drop is applied, or reduce the pressure drop by 10 % if the flow rate is constant. ‘The main references for friction factors in pipes have been the work of Nilsuradse on flow in smooth and sand roughened pipes [5] (7). From these experimental works friction factor relations for smooth pipes (Equation 4) and rough pipes (Equation 5) were presented. The smooth pipe law (Equation 4) was originally presented by Prandtl and the rough pipe relation (Equation 5) by von Karman [8]. However the constants in both relations are adjusted to fit,the data of Nikuradse. T f Olog(Re VF) — 0.8 = @) VE Blog z-) + 14 = Dlogt \Nikuradse showed that for low Reynolds numbers (Re) the friction factor was a function of Re only, then at sufficiently high Reynolds numbers the friction factor became a function of the wall roughness only. The first case was called hydraulically smooth flow and the latter rough flow. Colebrook and White [4] presented additional experimental results and developed a correlation for the friction factor (Equation 6) valid also in the transitional regime in between the smooth and rough flow. (6) ‘The Colebrook- White equation has been extensively used since it was presented. However, the implicit character of the equations has ereated a need for a simpler explicit correlation. Many such correlations have been presented and one such is the Haaland equation [9] (Equation 7). i Haaland claimed that this equation is especially suited for gas pipelines if n= 3, With n= 1, the equation is a very good approximation to the Colebrook-White correlation. Most of the implicit friction factor relations are based on the Colebrook-White correlation and the accuracy are therefore at, best similar. | @ In the AGA report ” Steady Flow in Gas Pipelines” Uhl et al. [10] presented friction factor correlations for smooth and rough flow in gas pipelines (Equations 8 and 9 respectively). vi Fy2log(Re 7) = 09 = Flog 8) Vi Plog GF) + 1.74 = 2log(741) 9) ‘The correlations are derived based on data from operating pipelines and from the Bureau of Mines Monograph ” Flow of Natural Gas through ExperimentalPipe Lines and Transmission Lines” [11]. The correlations were similar to Nikuradse’s in form but the constants were adjusted to fit new experimental data. In addition the drag factor Fy and the effective roughness ke were introduced to account for the additional drag from bends, welds etc. In their work Uhl et al. stated that the transition between smooth and rough flow in the Colebrook-White equation was too "slow” and that a more abrupt transition was more suitable for gas pipelines. Recently a study of the flow in smooth pipes at high Reynolds numbers at Princeton University showed that the Prandtl law of flow in smooth pipes (Equation 4) was not accurate for high Reynolds numbers (12). A friction factor correlation of the same form as the Prandtl law was presented (Equation 10). £889 og(Re Vf) ~ 0.3577 (10) This new correlation differs by as much as 5% at, Re = 30 x 10° compared to Prandtl law of flow in smooth pipes (Equation 4). ‘The authors have studied the influence of roughness on the friction factor at Reynolds numbers above 1 x 10°. Flow experiments were performed in 8 pipes of varying wall roughness. Here we present the result from measurements of friction factors in two of the eight pipes, a coated pipe and a lightly corroded Pipe. Experimental work ‘The flow experiments were performed at the Karst Metering and Technology Laboratory (K-Lab). This is a Statoil operated research and calibration laboratory. The measurements were performed in a high pressure flow loop which was fed with natural gas from export pipelines. ‘The gas was kept at constant temperature with heat exchangers, and the rate of flow was measured with sonic nozzles. The flow rate was kept constant by a centrifugal ‘compressor in the loop and the gas composition were measured with a gas chromatograph. ‘Test sections 6 m (19.7 ft] long with an internal diameter of 150 mm [5.9 in} were made of honed steel tubing to ensure high quality tolerance of the inner Giameter. The test sections were assembled in the flow loop downstream a 18 1m (59.1 ft] long, 6 in straight section. Upstream the straight section a flow conditioner was placed to ensure fully developed conditions in the test sections.The pressure drop in the test sections was measured with 4 mm [0.16 in] wall taps. For the measurement of differential pressures a high precision Paroscientific digital differential pressure transducer with an accuracy of 0.01 % of full scale reading was used. The absolute pressure and temperature were ‘measured in each test section. Figure 1 shows schematically the design and the instrumentation of the test section. ‘One test section was coated with two-component epoxy coating (COPON EP2306 HLF). The coating was applied with a high speed rotating nozzle which was pulled through the pipe at constant speed by a hydraulic winch. ‘The nozzle was fed with coating at a constant rate by a membrane pump. During the sand-blasting and coating operation all instrument connections were plugged to avoid sand and coating to enter the holes. After coating the instrument, connections were cleaned and inspected for burrs. The quality of the coating was inspected with video equipment. Another identical test section was left untreated after honing and was exposed to water in pressure testing. The pipe was left in open air (coast climate) to corrode for two weeks before flow tests. The corrosion on the pipe wall of this pipe was not severe and the authors believe the surface is generally smoother than most corroded pipes. The pipe was honed and therefor very smooth before being exposed to water and air. ‘The roughness of the inside pipe wall of both pipes was measured with a profilometer (Perthomer SP3) with a stylus radius of 5 jm [2 x 10~* in} immediately after the flow tests. Several runs were made in the axial direction of flow for both surfaces. The results from the profilometer were given as standard roughness measures (Ra, Rg, Rz). The definition of the roughness parameters and the cut-off wavelengths used in the measurements was according to industry standards (DIN 4768). Results ‘The coated test section was tested at line pressures of 25 and 70 bar with several flow rates. The bare steel test pipe was tested with a line pressure of 70 bar. The gas compositions for both tests are shown in Table 1. The friction factors were calculated from Equation 1. The density of the gas was calculated using the AGA-8 correlation [13], and the viscosity was calculated from the ee et al. correlation [14]. The viscosity was used only to calculate the Reynolds number. ‘The friction factor versus Reynolds numbers for both pipes are shown in Figure 2. The total uncertainty in the calculated friction factor‘was estimated to be better than +2.5 %. The main causes of error were the viscosity and density estimates and the differential pressure readings. ‘The equivalent sand grain roughness for the steel pipe was calculated using Equation 5, Rearranging this equation the equivalent sand-grain roughness was calculated as: wo V7 rate (qa) where the value of the frietion factor is the value at rough conditions. For the steel pipe the friction factor was constant for high Reynolds numbers and the value of the friction factor at rough condition was well defined. However, for the coated pipe the value of the friction factor at rough conditions was not ‘well defined because the friction factor was not constant at the highest Reynolds numbers. In this case the Colebrook-White equation was used to calculate the equivalent sand-grain roughness. The result of the fitting between experimental results and the Colebrook-White equation is shown in Figure 2. The roughness measurements of the inner pipe walls of both pipes and the calculated equivalent sand-grain roughness are shown in Table 2. As seen in Figure 2 the transition between smooth flow and fully rough flow in the bare steel pipe is poorly modelled by the Colebrook-White equations. The transition seem to be more abrupt. Accordingly the AGA friction factor correlations (Equations 8 and 9 with Fy = 1) were plotted using the same value of the equivalent sand-grain roughness. These equations give an abrupt transition. However, for the coated pipe the Colebrook-White equation gave a ‘good fit to the experimental data. The abrupt transition between smooth and rough flow in steel pipes was also reported in the Institute of Gas Technology report on steady state flow in gas pipelines (10]. ‘The transport capacity for the two pipes may be compared using Equation 3 ‘and the equivalent sand grain roughness values found. Neglecting the diameter difference caused by the applied coating (the difference in the diameter due to the coating is comparable with the uncertainty in the diameter measurements, see Table 2), and assuming that the same pressure drop is applied in both pipes the difference in the transported mass may be calculated: teste — [Fae Tse \ Feed ©) ‘The relative difference as a function of the Reynolds number is plotted in Figure 3. As seen in the figure the difference in transport capacity is moderate (5%) for Re = 1 x 10" because the effect of roughness on the friction factor is small. However with increasing Reynolds numbers the effect of the wall roughness in the steel pipe is significant and the difference in transport capacity at Re = 1 x 10" is 21 %.Assuming that the same amount of gas is transported in each pipe the difference in the pressure drop may be found directly from the Equation 1. [dP/dreoates _ dPooated _ fooatea [Parle APaect — Foteat ‘The difference in pressure drop as a function of the Reynolds number is plotted in Figure 4. The pressure drop in the two pipes is equal for low Reynolds numbers, but as the Reynolds number is increasing the pressure drop in the coated pipe will be lower than in the steel pipe due to the difference in friction factor. At Re = 1 x 10? the pressure drop in the,coated pipe is 31 % lower than in the steel pipe. (13) Conclusions ‘* The experimental results showed that an internal coating gave a significant reduction in the frictional pressure drop in gas pipes. In the coated pipe the pressure drop was 31 % lower than in the steel pipe at a Reynolds number of 1 x 10". ‘© The transport capacity of the coated pipe was greater than that of the steel pipe due to the lower frictional drag, At a Reynolds number of 1 x 10" the mass rate in the coated pipe was 21 % greater than in the steel pipe given the same pressure drop. © The equivalent sand-grain roughness of a coated pipe was estimated about 1m (4 x 10-* in]. The equivalent sand-grain roughness of the steel pipe was 21 jm (8.3 x 10~* in}. Direct measurements of the surface roughness of the steel pipe showed that the R. value corresponded very well with the calculated equivalent sand-grain roughness. '* The Colebrook-White equation was found to be appropriate for the friction factor of the coated pipe. In the steel pipe the AGA correlations ‘were more appropriate because of a more abrupt transition between smooth and rough flow. Acknowledgments This research is sponsored by "Den Norske Stats Oljeselskap” (STATOIL) Gas ‘Technology. The support and the cooperation of Statoil is greatly appreciated. The authors would like to thank the staff at K-lab for their support. and. encouragement in all stages of the experimental work.References {1] S.Kut, Internal and external coating of pipelines. In First International Conference on the Internal and Eternal Protection of Pipes, 1975. Paper AA. [2] Gurdial Singh and Ove Samdal. Internal coating justified by operating costs. Oil and Gas Journal, pages 50-55, Apr. 1988. [3] Environment °98 - The Norwegian Petroleum sector. Ministry of Petroleum and Energy, Oslo Norway. (In Norwegian), [4] C.F. Colebrook. Turbulent flow in pipes, with particular reference to the transition regime between smooth and rough pipe laws. Institution of Civ. Bng. Journal, 11:133-156, 1939. paper no. 5204. [5] J. Nikuradse. Stromungsgesetze in rauhen rohren. In Forschungsheft 361, volume B. VDI Verlag Berlin, Jul. /Aug. 1933. Translated in NACA. ‘Technical Memorandum nr. 1292, 1950. [6] Robert D. Blevins. Applied Fluid Dynamics Handbook. Krieger Publishing Company, Florida USA, 1992. (7] J. Nikuradse. Gesetzmessigkeiten der turbulenten stromung in glatten rohren, In Forschungsheft 356, volume B. VDI Verlag Berlin, Sept./Oct. 1932. Translated in NASA TT F-10, 359, 1966. (8] Hermann Schlichting. Boundary layer theory (Seventh Edition). McGraw-Hill Inc. New York, 1979. ‘Translated by Dr. J. Kestin, [9] S.B. Haaland. Simple and explicit formulas for the friction factor in turbulent pipe flow. Journal of Fluids Engineering, 105:89-90, March 1983. [10] A.E. Uh. Steady flow in gas pipelines. Institute of Gas Technology, ‘Technical Report no. 10, 1965. American Gas Association, [11] RV. Smith, J.S. Miller, and J.W. Ferguson. Flow of natural gas through experimental pipelines and transmission lines. Bureau of Mines, Monograph 9, 1956, American Gas Association. [12] Mark V. Zagarola. Mean-flow scaling of turbulent pipe flow. PhD thesis, Princeton University, 1996, [13] Kenneth E. Starling. Compressibility factors of natural gas and other related hydrocarbon gases. Transmission Measurement Committee Report no. 8, 1992, American Gas Association. 14] TH. Chung, M. Ajlan, L.L. Leo, and K. Starling. Generalized iB ‘multi-parameter correlation for nonpolar and polar fluid transport properties. Industrial and Chemical Engineering Research, 27:671-679, 1988,Nomenclature Ebr E PPP P 7 RORUSS [SO ED NE ONES, Pipe cross-sectional area Pipe diameter Drag factor Friction factor (Moody) Roughness Pipe line length Gas mol weight: Mass flow of gas Pressure Differential pressure Volumetric flow rate Universal gas constant Pipe radius Reynolds number Re = #22 = 2 Arithmetical mean roughness Root mean square roughness lean peak to valley height femperature ‘Mean velocity Axial coordinate Compressibilty factor Gas density Dynamic viscosity subscript, standard conditions subscript, upstream end subscript, downstream end superscript, mean value 10‘Table 1: Gas Compositions Coated pipe [mol %) | Steel pipe [mol %] ™ 113 118 CO, 0.95, 0.95 a 81.47 81.84 Cr 15.15, 14.79 Cs 118 113 iC 0.05, 0.04 ny 0.07 0.07 Con < 0.01 < 0.01 Weight {mol] | 18.95 18.89 ‘Table 2: Roughness parameters and pipe diameter Coated pipe Steel pipe Ra (yma) | 1.02 (4.0 x 10-8 in) | 2.36 (9.3 x 10-* in) R, (um] | 1.32 (6.2 x 10-% in) | 3.65 (1.4 x 107* in) Rz (ym) | 5.79 (2.3.x 10-4 in) | 21.66 (8.5 x 10-* in) ky um] [1.1 (43x10-F in) [21 (83x 10-4 in) D (mm) | 150.0 (5.9 in) 150.1 (6.91 in) uTest pipe 6 m long T Flow Ay 2 Siatie pressure holes ap. P Figure 1: Instrumentation of test section 0.014; 0.013) 0.012 0.011! 0.01 0.009} 0.008} k.=210-06 0.007! Figure 2: Friction factor plot; *: coated pipe, x: steel pipe, line: Colebrook- White equation fitted, dashed: AGA equations, dash-dot: smooth pipe law 2Mata! et 25 ha . 15 a 1.05, 10 10" Reynolds number Figure 3: Transport capacity in coated pipe compared to steel pipes puted from friction factor correlations, *: measurements 09 a y ios o7 7 — ; os 10! 10 Reynolds number Figure 4: Pressure drop in coated pipe compared to steel pipe; line: computed from friction factor correlations, *: measurementsAuthors Elling Sletfjerding is a doctoral (Dr. Ing.) student in the Department of Petroleum Engineering and Applied Geophysies at the Norwegian University of Science and Technology. Sletfjerding received his engineering (Siviling.) degree in applied mechanics from the Royal Institute of Technology in Stockholm (Sweden) in 1994. He is doing research on pressure drop in large diameter gas pipelines. Jon 8. Gudmundsson is professor of petroleum engineering, involved in research on fluid flow and gas hydrates, B.Sc. from Heriot-Watt: University in Edinburgh and Ph.D. from Birmingham University in Birmingham (England), both in chemical engineering. Worked previously on geothermal engineering for the National Energy Authority in Reykjavik and the Geothermal Program at Stanford University. Karl Sjoen is chief engineer within transport and systems technology in Statoil (Den Norske Stats Oljeselskap). He is also part time professor in thermo- and fluiddynamies at the Norwegian University of Science and Technology (NTNU). Karl Sjen has a M.Sc. in mechanical engineering (1978) and a, Ph.D. in aero- and gas dynamies (1983) from NTNU. He has been involved in research, development and operation of gas transport infrastructure for 15 years. u°E 52059 iction Factor in Coated Gas Pipelines and Well Tubing ‘1g Sletfjerding, SPE, Department of Petroleum Engineering and Applied Geophysics -wegian University of Science and Technology gh 198, Solely of Paoloum Engineers he. paper was propre for presentation atthe 1998 SPE European Pettbum Sronce eld in The Haguo, Tha Netorans, 20-22 Octobr 1998, ‘apo was sel for presetation by an SPE Program Commitee Sing rovow ofan contained in an abstract suited by the 3), Contr cre paper, as presented, have ot been reviewed by he 3 ot Petrlnum Engioars and are sje! oorecton by the ars). ‘lal as presente. coos not necasarly reflect ary poston ol he vol Pevoleum Engineers, Is fiers, or members. Papers press st mosis are sujet opubleaton review by Easovil Commits of he {vol Parolaum Engioars.Electonc repeduciondistouion, o storage 4 pat of hs paper for commercal purposes witout the wien conse of ‘eet of Peteteum Engineers pronted. Permission rpreauoe in pet ‘tod to an abst a not more than S00 woes uso may ret be 4. The abstract must contain conspicuous acknowledgment of where and by ‘tho papar was sesoroa. Wo Loan, SPE, P.O, Box 83080, srs, TX. 75088-3836, USA. fx 01-972 052435 aract experimental study of pressure drop in pipes at high nolds numbers was performed to determine the fric- factor of cated gas pipes. The results enable more arate predictions of long distance gas pipeline capacity, demonstraze the drag reduction potential of internal ting in gas wells he friction ‘actor of the coated pipe was 31% lower «1 in the bare stec! pipe at a Reynolds number of 1107 tha reducticn in the friction factor means 21 % higher is rate at the same pressure drop if internal coating is in a horizontal pipeline. ulations cf flow in vertical gas wells in the offshore U field showed that internal coating can reduce the ssure drop significantly in high rate gas wells. A re- tion in the tubing pressure drop from 18.9 bar to 16.6 at a production rate of 2.2 x 10° Sm® /day was poss- when internal coating was used. At higher rates the xetion in the pressure drop was even larger. As a result need for compression for gas export may be reduced. ‘duction smal coatings for drag reduction have been used with sess in gas ripelines* since the 1950°s. Economic stud- show that the typical pay-back time for the investment ternal coating is 3 ~ 5 years due to improvements in aline hydravlies. It is well known that internal coat- ‘reduce the friction in the pipeline and therefor reduce operating cost of compressors. In Norway? 35% of offshore generated electrical power is used for gas export compressors. ‘The use of coatings to reduce the operating cost is therefor important. In addition the coatings' pro- tect the pipe wall against corrosion and reduce the need for maintenance of the pipeline. In this study the potential of internal coating for drag reduction in gas wells is discussed. Internal coating is not commonly used in gas well tubing. However, if coatings is. used in a gas well the pressure available at the wellhead will be higher. Thus the energy in the gas stream is higher and the need for compression will be reduced. ‘The gas from the Troll field is transported in pipelines without compression to the onshore processing plant at Kollsnes. The driving mechanism is the available pressure at the platform deck. Compressors will be installed when the well head pressure is too low to move the gas onshore. Reduction of the pressure drop in the wells may therefor delay the investments in and installation of compressors. ‘The flow in producing gas wells and offshore pipelines is characterized by large Reynolds numbers (Re ~ 10") due to the low gas viscosity and the relative high density at typical operating pressures (+100 bar). From the classi cal Colebrook-White friction factor correlation® it is seen that even very minute irregularities on the pipe wall will have a significant effect on the friction in the pipe at high Reynolds numbers. Therefor the potential of reducing the wall friction with internal coatings is large in gas wells and pipelines, To characterize hydraulic roughness in pipes the equiva- lent sand-grain roughness has traditionally been used. The concept refers to the rough pipe experiments of Nikuradse® in 1933, and as common practice the hydraulic proper- ties of a pipe are compared to Nikuradse’s work to "back- track” the equivalent roughness. Equivalent sand-grain roughness of coated pipes are given in handbooks® but the values cited vary significantly. In this work both the equiv- alent: sand-grain roughness and the directly measured wall roughness of a coated pipe and a bare steel pipe are given. ‘The paper is organized in five parts: After this intro- duction a brief literature review is presented to sum up ‘earlier relevant work and to present the theory. Then the ‘experimental work is presented. In the fourth part appli- cations of the experimental results in gas pipelines and gas wells are discussed. ‘The major findings of this work areELLING SLETFJERDING SPE 52059 sumed in the conclusions :kground + steady state pressure drop in a pipe (such as well ing) may be calculated from the governing momentum, ation” and the real gas law. Strictly, the energy equa- 1 is also required when there is heat. transfer to the condings. Here the flow is treated as isothermal flow + given temperature (the mean temperature) and the rey equation is neglected tation 1 show how the pressure gradient 42 in single se flow is dependent on three different mechanisms: tion against the pipe wall, gravity and change in the 2tic energy cf the flow due to acceleration. In a vertical well all three mechanisms must be taken into account. vever in a ges pipeline the friction pressure loss is dom- ing and the last two terms in the Equation 1 are often lected. { solving Equation 1 for flow in a pipe a model for the sion factor is needed. The main references for fri ors in pipes have been the work of Nikuradse®® on +in smooth and sand-roughened pipes. From these ex- mental works friction factor relations for smooth pipes uation 3) and rough pipes (Equation 4) were presented. smooth pipe law (Equation 3) was originally presented >randt! and the rough pipe relation (Equation 4) by von man (see Schlichting’s book®). However the constants coth relations are adjusted to fit the data of Nikuradse. Rev 2014) (3) 2.0 log(Re V/f) — 0. 2log( ) + 1.74 = log (7.417) @ uradse showed that for low Reynolds numbers (Re) the tion factor was a function of Re only, then at suficiently 1 Reynolds numbers the friction factor became a func- \ of the wall roughness only. The first case was called raulically sraooth flow and the latter rough flow. olebrook and White" presented additional experimen- results and developed a correlation for the friction fac- (Bquation 5) valid also in the transitional regime in seen the smooth and rough flow regimes. 251 ~Poe Rey © ‘The Colebrook- White equation has been extensively used since it was presented, In the AGA report ” Steady Flow in Gas Pipelines” Uhl et al presented friction factor correlations for smooth and rough flow in gas pipelines (Equations 6 and 7 respec- tively). Ve Fy2log(Re /f) - 0.9 (6) VE Dios) +174= Blog( tL) a The correlations are derived based on data from operat- ing pipelines and from the Bureau of Mines Monograph? "Flow of Natural Gas through Experimental Pipe Lines and Transmission Lines”. The correlations were similar to Nikuradse’s in form but the constants were adjusted to fit new experimental data. In addition the drag factor Fy and the effective roughness ke were introduced to account for the additional drag from bends, welds etc. In their work Uhl et al. stated that the transition between smooth and rough flow in the Colebrook-White equation was too grad- ual and that a more abrupt transition was more suitable for gas pipelines. ‘The values of the equivalent sand-grain roughness of dif- ferent pipe material may be found in handbooks (for ex- ample Blevins 1992). However the values given have large variations. At high Reynolds numbers the roughness value ‘used in the friction factor calculation has large effect. To accurately calculate the pressure drop or capacity in a gas pipe the value of the equivalent roughness should be chosen with care. For gas pipelines Uhl et al. estimated the values of the surface roughness of steel pipes and plastic lined pipes as seen in Table 1. In an extensive work involving field data from 144 gas wells Peffer'? et al. found that an absolute roughness value of 45.7 jm (0.0018 in) was a representative value for the tubing roughness. Experiments As part of this study flow experiments were performed at the Karsto Metering and Technology Laboratory (K- Lab) in Norway. This is a Statoil operated research and calibration laboratory. The measurements were performed ina high pressure flow loop which was fed with natural gas from export pipelines. ‘Test sections 6 m long with an internal diameter of 150 ‘mm (5.9 in) were made of honed steel tubing to ensure high quality tolerance of the inner diameter. The test sections were assembled horizontally in the flow loop downstream a 18 m long, 6 in straight section. The pressure drop in the test sections was measured with 4 mm wall taps. The absolute pressure and temperature were also measured in52059 FRICTION FACTOR IN COATED GAS PIPELINES AND WELL TUBING 3 test section. ‘The experimental work is described in Alin Sletfjerding’® et al ae test section was coated with two-component epoxy PON EP2306 H.F.). The coating was applied with a speed! rotating nozzle which was pulled through the at constant speed by a hydraulic winch. The nozzle fed with coating at a constant rate by a membrane . othe: identical test section was left untreated after ing ard was exposed to fresh water. The pipe was left pen air (coast climate) to corrode for two weeks before tests. ‘The corrosion on the pipe wall of this pipe not severe and the surface was smoother than most; oded pipes. ae rorghness of the inside pipe wall of both pipes was sured with a profilometer immediately after the flow 3. Several runs were made in the axial direction of for both surfaces. ‘The results from the profilometer > given as standard roughness measures (Ra, Ry, Rz). definition of the roughness parameters and the cut-off lengths used in the measurements was according to istry standards (DIN 4768). easurements of the pressure drop was done at line pres- 25 bar and 70 bar. The Reynolds number was varied 12x 10-2 107. The friction factor versus Reynolds bers for both pipes are shown in Figure 1. In the fig- both the Colebrook-White and the AGA friction factor lations are shown. The Colebrook-White correlation 1s to have a too gradual transition region to model the jon factor of the steel pipe (as noted in Uhl"et al.). sever for the coated pipe it gave a good fit. he equivalent. sand grain roughness for the steel pipe calculated using Equation 4. Rearranging this equa- ‘the equivalent sand-grain roughness was calculated ky =107 VF rar ® sre the value of the friction factor is the value at rough ditions. For the steel pipe the friction factor was con- it for high Reynolds numbers and the value of the fric- factor at rough condition was well defined. However, the coated pipe the value of the friction factor at rough ditions was not well defined because the friction factor not constant at the highest Reynolds numbers. In this » the Colebrook-White equation was used to calculate equivalent sand-grain roughness. hhe roughness measurements of the inner pipe walls of h pipes and the calculated equivalent sand-grain rough sare shown in Table 2. The value of the equivalent d-grain roughness of the coated pipe found in this study lower than the value recommended by AGA (see Ta~ 1). Hewever the sand-grain roughness of the steel pipe our experimental study is comparable with the AGA Applications In this part the results from the experimental work are applied to gas pipeline flow and flow in gas wells. The experimental results could direcily be applied to horizon- tal pipeline flow assuming that the ow in the pipeline is characterized by Reynolds numbers similar to those in the experimental work. The measurements in the steel pipe and the coated pipe are used to estimate the effect of coating on the transport capacity and the pressure drop in a pipeline. ‘To study the effect of coatings in gas wells the equiva- lent sand-grain roughness of the coated pipe found in the experimental work is used. Flow simulations in a typical gas well at the Troll field is perfcrmed to compare the well head pressure in a coated well and a well with steel tubing. Gas pipelines. In horizontal pipeline flow the main cause of energy loss is the friction between the pipe wall and ‘the moving gas. The steady state frictional pressure drop can be calculated from the well known Darcy-Weisbach Equation: a 7 (} Equation 9 is the remaining parts of the governing equa- tion for pipeline flow (Equation 1) when the kinetic energy and the elevation terms are neglected. ‘The influence of the friction factor on the pipeline ca- pacity can be seen from Equation 10, which is derived by combining Equation 9 with the real gas law (Equation 2) (10) Equation 10 shows that the volumetric flow rate is pro- portional with the friction factor to the minus 1/2 power (Qo~ f-¥). “Assuming the same pressure drop the difference in eapac- ity in an internally coated pipeline and a bare steel pipeline can be estimated. The ratio of the mass flow in the coated pipeline to the mass flow in the bare steel pipeline can be expressed as (ref. Equation 10): ieoated _ ay From the experimental results (Figure 1) it was found that the friction factor in the coated pipe is 31 % lower than in the steel pipe at a Reynolds number of 1 x 107. From Equation 11 the transport capacity of the coated pipe at ‘this Reynolds number is 21 % larger than in the steel pipe.ELLING SLETFJERDING SPE 52059 higher Reynolds numbers the difference in transport acity will be even larger. milarly the pressure drop in a coated pipe and a bare 4 pipeline can be compared assuming that the same 's flow is pvt through the two pipelines. From Equa- 1-10 the ration between the pressure drop in the two atines can be expressed: APeoated steel ‘BPaeet ~ Fret Bessel) a + ratio (Patee/Deooted) in Equation 12 must be less than if the same inlet pressure is applied in both pipes. + experimental results showed that the friction factor he coated pipe was 31 % lower than in the steel pipe Xe = 1x 10. According to Equation 12 the difference he pressure drop will be even larger than this value. he economic benefits of drag reducing internal coatings pipelines is well known"? in the literature. How- data on the hydraulic behavior of internal coatings at ‘conditions typically found in offshore gas pipelines is «se. Our experimental data enable more accurate pre- ions of the capacity (and the pressure drop) of coated pipelines feoated 3 wells. To investigate the drag reducing effect of coat- in gas wells, simulations of the pressure drop in a pro- ing well at che Troll gas field were performed. A two se simulation model was used because the production showed that there was some liquid in the gas stream. + simulations were based on measurements of flow rate, tom hole pressure, well head pressure, temperature and composition, ‘The well was slightly deviated with a true ical depth of 1365 m and a length along tubing of 1422 The inner diameter of the tubing was 175 mm (6.9 in), ‘gas had molweight of 17.8 kg /kmol. he simulations were done in the program PIPESIM' the geometry of the well was modeled in accordance 1 the final completion schematics. ‘The roughness of tubing was chosen to 46 jm following the recommen- ons of Peffer!? et al. The AGA correlations (Equations ad 7) were used for the friction factor. The Beggs and I revised hok/-up correlation was used in the two-phase Jel. The hold-up factor was chosen so that the simula- 's matched the data from the production test. he equivalent sand-grain roughness of the coated pipe yan) was found from the experimental work. This was in the simulation model and the wellhead pressure onstant bottomhole pressure (148 bar) was simulated. «flow rate was varied from 1.2x 10°—3.2x 10° Sm?/day. are 2 show che wellhead pressure as a function of the “rate, At a rate of 2.2 x 10° Sm°/day the wellhead ssure of the coated well was 2.3 bar higher than the with wall roughness of 46 jum. At the highest rate of x 10° Sm? /day the difference in the wellhead pressure 44 bar. ‘The tubing roughness of 46 jum given by Peffer!? et al. was found after comparing field data from several gas wells. ‘The value of the wall roughness in the coated pipe (1.1 um) obtained from our experimental work was more idealized and may not be directly comparable. However, an effec- tive wall roughness lower than 5 ym should be possible to obtain in a gas well. Simulations for a wall roughness value of 5 jm are also shown in Figure 2. ‘A second set of simulations were done to investigate the drag reducing effect of internal coating after pressure de- cline in the reservoir. The wellhead pressure was simulated varying the bottomhole pressure from 148 — 100 bar. The flow rate was held constant at a rate of 2.2 x 10° Sm°/day. ‘The results from the simulation is shown in Figure 3. At bottomhole pressure of 148 bar the wellhead pressure in the coated pipe was 2.3 bar higher than in the well with ordinary tubing. ‘The difference in wellhead pressure in- creased to 3.6 bar at a bottomhole pressure of 100 bar. The amount of liquid present in the well stream will influence the frictional drag at the pipe wall. In the sim- ulations shown in Figure 2 the flow was in single phase in a major part of the well. In the simulations shown in Figure 3 the amount of liquid in the gas stream increased ‘as the bottomhole pressure decreased. ‘The liquid hold- up fraction at a bottombole pressure of 100 bar reached a maximum of 0.035, ‘The results from the simulations show that the pressure drop in vertical gas wells may be reduced if internal coat- ings are used. In high rate gas wells where the wall friction is comparable in size with the pressure drop due to grav~ ity the effect is significant. The simulations show that the drag reducing effect of coatings is larger for low reser pressures Conclusions ‘* Experimental results showed that the friction factor in a horizontal coated pipe was 31 % lower than in a steel pipe at a Reynolds number of 1 x 10% © Ata Reynolds number of 1 x 10"the mass rate in the coated pipe was 21 % higher than in the steel pipe given the same pressure drop. © Simulations of the flow in a high rate gas well in the ‘Troll field showed that the pressure drop in the tub- ing may be reduced significantly if internal coating is, used. © The drag reducing effect of coatings in gas well tub- ing is increasing with increasing rate and decreasing bottom hole pressure.£52059 FRICTION FACTOR IN COATED GAS PIPELINES AND WELL TUBING 5 nenolature A = Pipe cross-sectional area, m? Pipe diameter, m Drag factor Friction factor (Moody) Gravitational constant, m*/s Sand-grain roughness, jm Effective roughness, mm Pipe line length, m Gas mol weight, kg/kmol Mass flow of gas, kg/s Pressure, bar [Pa] Differential pressure, Pa Pressure drop, Pa Volumetric flow rate, m® /d Universal gas constant, J/kmol K Pipe radius, mm Reynolds number Re = 22 = 2 Arithimetical mean roughness, gm Root mean square roughness, jam Mean peak to valley height, jam ‘Temperature, K Mean velocity, m/s Axial coordinate, m Compressibility factor Inclination angle, degrees Gas density, kg/m® Dynamic viscosity, Pa-s subscript, standard conditions subscript, upstream end subscript, downstream end superscript, mean value nowledgments 3 doctoral research is sponsored by “Den Norske Stats selskap” (STATOIL). The author would like to thank fessor Jon Steinar Gudmundsson, NTNU and chief en- ‘er Karl Sjoen in STATOIL who have encouraged and ported this research, ‘The support and interest from staff at K-Lab and ‘Trond Saksvik in STATOIL's Troll organization is greatly appreciated. erences SKut: “Internal and external coating of pipelines,” In First Internetional Conference on the Internal and Exter- nal Protection of Pipes, (1975). Paper Ad ‘Singh, G. and Samdal, O.: “Internal coating justified by ‘operating costs,” Oil and Gas Journal (Apr. 1988) 50-55. “Environment '98 - The Norwegian Petroleum sector,” Ministry of Petroleum and Energy, Oslo Norway. (In Nor- wegian). Colebrook, C: Turbulent flow in pipes, with particu lar reference to the transition regime between smooth and rough pipe laws,” Institution of Civ. Eng. Journal (1939) 11, 133-156. paper no. 5204. 5, Nikuradse, J: “Stromungsgesetze in rauhen rohren,” In Forschungsheft 961, volume B. VDI Verlag Berlin, (Jul/Aug. 1983). ‘Translated in NACA "Technical Mem- orandum nr. 1282, 1950. 6. Blevins, R. D.: Applied Fluid Dynamics Handbook. Krieger Publishing Company, Florida USA (1992) 7. Massey, Bi: Mechanics of fluids. Chapman and Hall Ltd. London, 6th edition (1990), 8, Nikuradse, Ji: “Gesetamessigheiten der turbulenten stro- ‘mung in glatten robren,” In Forschungsheft 856, volume B. VDI Verlag Berlin, (Sept/Oct. 1952). Translated in NASA, ‘TT F-10, 359, 1966. 9. Schlichting, H.: Boundary layer theory (Seventh Baition) McGraw-Hill Inc. New York (1979). Translated by Dr. J Kestin 10. Uhl, Ac “Steady flow in gas pipelines,” Institute of Gas ‘Technology, Technical Report no. 10 (1965). American Gas Association 11, Smith, R., Mille, J. and Ferguson, J: “Plow of natu- ral gas through experimental pipelines and transmission lines," Bureau of Mines, Monogeaph 9 (1956). American Gas Assocation. 12, J.W.Pefler, Miller, M. and Hil, A: “An improved method for calculating bottombole pressures in flowing gas wells with liquid present,” SPE Production Engineering (Nov. 1988) 645-65. 48, Sletfjerding, E., Gudmundsson, J. $. and Sjoen, K.s “Flow experiments with high pressure natural gas in coated and plain pipes; comparison of transport capacity,” In Proceed- ings of the 30th Pipeline Simulation Interst Group (PSIG) Annual Meeting, Denver, Colorado, (October 1998) 14, Baker Jardine Petroleum Engincering Software. PIPESIM Jor Windows, (1997). SI Metric Conversion Factors bar x1.0E +05 Pa ft x3.048E-01 = m in x2.54E +01 mm Author Elling Sletfjerding is a doctoral student (Dr. Ing.) in the Department of Petroleum Engineering and Applied Geo- physics at the Norwegian University of Science and ‘Tech- nology. Sletfjerding received his engineering degree (Sivil- ing.) in applied mechanics from the Royal Institute of ‘Technology in Stockholm (Sweden) in 1994. He is doing research on pressure drop in large diameter gas pipelines.ELLING SLETFIERDING ‘SPE 52059 le 1. Roughness values of gas pipelines given in Uhl et Pipe type and condition | k. [um Bare steel pipe (new) 127 = 19 after air exposure 6 months | 25.4 32 after air exposure 12 months | 38 after air exposure 24 months | 445 Plastic lined pipe 51-76 Table 2. Roughness parameters of test pipes Coated pipe | Steel pipe + is 2 a = as ‘Re [pan] | 1.02 2.36 al Ry ym) | 1.32 3.65 Fig, 2 Well head pressure vs. flow rate Re (yan) | 5.79 21.66 , [uma] [11 21 Will head preseure oar s 2 “ooo to 20 toa Bottom hole posure bar 3 Well head pressure at constant rate oor ‘Smooth 1 0 Reynolds number 9. 1.- Friction factor plot; *: coated pipe, x: stee! pipe, 1e: Colebrook-White equation fitted, dashed: AGA equa- on, dash-dot: smooth pipe law