Sunteți pe pagina 1din 12

J. Non-Newtonian Fluid Mech.

123 (2004) 269280

A model for restart of a pipeline with compressible gelled waxy crude oil
Malcolm R. Davidsona, , Q. Dzuy Nguyenb , Cheng Changa , Hans Petter Rnningsenc
a

Department of Chemical and Biomolecular Engineering, The University of Melbourne, Parkville, Vic. 3010, Australia b School of Chemical Engineering, The University of Adelaide, Adelaide, SA 5005, Australia c Statoil, N-4035 Stavanger, Norway Received 22 May 2003; received in revised form 19 August 2004; accepted 22 September 2004

Abstract The restart of a pipeline containing gelled waxy crude oil after shutdown is simulated using a two-uid displacement model. The model assumes that the gelled oil is to be displaced by pumping in another uid under constant pressure and that the incoming uid displays Bingham plastic behaviour. The rheological behaviour of the gelled oil is characterised by a time-dependent constitutive equation with a yield stress, based on modelling of a North Sea waxy crude oil. The start-up ow model takes into account the compressibility and longitudinal variations in physical and rheological properties of both uids as they ow along the pipeline. The effects of yield stress and the compressible nature of the gelled oil are effectively handled by introducing the concept of a propagating yield front and initial compression ow, that precede viscous ow of the gelled oil for a successful start-up. The model predicts that start-up ow is subjected to two delay times, one due to the transient yield stress behaviour of the gelled oil and the other due to the compressibility effects. Oil compressibility has a positive effect on ow rate, movement of the uidoil interface and the time taken to clear the gelled oil from the pipe. The effects of other factors such as start-up pressure, initial compression time and liquid hold-up on ow after restart are also examined. The model produces a more realistic prediction of the start-up situation than a previous model developed for incompressible uids. 2004 Elsevier B.V. All rights reserved.
Keywords: Waxy crude oil; Yield stress; Pipeline start-up; Numerical model

1. Introduction Crude oil is a primary and essential energy source throughout most of the world. The declining resources of conventional oil and the increasing demand for petroleum has resulted in the sourcing of more heavy, waxy crude oils. A waxy crude oil contains parafns, which can cause it to form a gel, consisting of wax crystals in a viscous matrix, when the oil is cooled. This circumstance can occur when the pipeline carrying the crude oil is being shut down for operational or emergency reasons in locations with low ambient temperature (e.g. in offshore and Arctic areas). The gelled oil exhibits a complex time-dependent rheology with a yield stress and viscosity, which depend on the time of shearing [13]. Typically, to restart the ow in the pipeline, the gelled

Corresponding author. Tel.: +613 8344 6615; fax: +613 8344 4153. E-mail address: m.davidson@unimelb.edu.au (M.R. Davidson).

oil is displaced by another uid under pressure. The pressure, which must be applied exceeds the usual operating pressure and must be sufciently large to overcome the strength of the gelled oil plug. As ow begins, shearing causes a progressive breakdown in the gel structure leading to further increases in the ow rate. However, the formation of a solid-like gel after shutdown presents the serious risk that restarting the pipeline may not be possible if the static yield stress of the gel is too great for the available applied pressure. A mathematical ow model to predict the time to clear the oil plug from the pipe, as well as determining the restart pressure, requires the solution of the equations of continuity and momentum for the gelled oil and the displacing uid coupled with appropriate time-dependent constitutive equations. Typically, inertia is ignored and the momentum equation consists of a simple force balance relating pressure drop and wall shear stress. Previous models include the use of a structural kinetic approach to describe the time-dependent rheological

0377-0257/$ see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jnnfm.2004.09.007

270

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

Nomenclature a, b c D f G He K L Lc Lf LICF L0 LOGF k LICF M m P Pk Pc Pf PT Q Re t t0 ts v vc vf V z empirical parameters sound speed in gelled oil pipe inner diameter Fanning friction factor mass ow rate Hedstrom number number of subdivisions of the ICF at t = t0 pipe length length of the compression ow region length of the yield front length of the incoming uid initial length of each segment in the subdivision of the gelled oil plug length of the kth segment in the subdivision of the OGF at time t length of each segment in the subdivision of the ICF at time t initial number of subdivisions of the gelled oil plug number of subdivisions of the gelled oil plug in the pipe gauge pressure pressure drop over the kth subdivision pressure drop over the compression ow region pressure drop over the yield front applied pressure drop volumetric ow rate Reynolds number time total duration of compression ow shearing time velocity velocity of the compression ow velocity of the yield front volume distance from the inlet

Subscripts k kth subdivision w wall B Bingham calc calculated value ref scaling value Superscript * dimensionless variable

Greek symbols ratio of yield stress to critical wall stress uid compressibility shear rate plastic viscosity B Bingham plastic viscosity rate constant in Eq. (18) uid density 0 oil density ahead of the yield front f density change across the yield front uid stress c wall shear stress in compression ow s static yield stress B Bingham yield stress y apparent yield stress in Eqs. (17) and (18)

behaviour of the gelled oil [47]. More recently Chang et al. [8] developed a ow model (henceforth referred to as CNR) based on a time-dependent Bingham constitutive equation in which the apparent (time varying) yield stress is related to the three characteristic yield stresses (static, elastic-limit and dynamic) that they use to describe the yielding behaviour of gelled oil. The CNR model assumes that the pipeline, which has been shut down for a period of time, is completely lled with homogeneous gelled oil before the restart. At start-up, another homogeneous uid with simple rheological properties is pumped into the pipe to displace the gelled oil off the pipeline. Both the gelled oil and the displacing uid are assumed to be incompressible in the CNR model. In practice, the oil is not homogeneous in the pipeline after the ow has ceased for a period of time, and the displacing uid and the gelled oil are compressible. Heterogeneity in rheological properties and the oil density before restart result from non-uniform cooling of the oil during shutdown. In addition, shrinkage associated with thermally induced density change can result in the formation of gas plugs during cooling following the release of gases previously dissolved in the oil. Oil compressibility has effects on longitudinal displacement due to pressure-induced density changes. It also causes a delay before the gelled oil plug begins to move, following the application of the restart pressure. The aim of the present work is to extend the CNR two-uid model for incompressible, homogeneous oils to account for compressibility and longitudinal property variation of both the displacing uid and the gelled oil. Consideration of multiple plugs (resulting from a free gas phase present in multiphase pipelines or thermal shrinkage upon cooling) and its consequences is deferred to a later paper.

2. Model for pipeline restart Consider a horizontal pipe of inner diameter D, lled over length L with gelled oil at the time of start-up (t = 0). The oil has rheological properties and density, which may vary along the pipe at t = 0, and which are a function of the local thermal history and resting time between shutdown and restart. However, for simplicity, the gelled oil is assumed to have

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

271

uniform properties initially (i.e. assume that non-uniformity only arises after restart), although the methodology presented here is easily applied when there is an initial variation of properties. Flow in the pipeline is restarted by pumping in an incoming uid (ICF) to displace the gelled oil, which becomes the outgoing uid (OGF). The ICF may be water or an oil with the same properties as those of the gelled oil before shutdown. For simplicity, we assume the ICF to be a time-independent Bingham uid or a Newtonian uid in the special case of zero yield stress. The interface between the ICF and OGF is taken to be at since any radial variation in the interfacial position is small compared with the pipe length. When a constant total pressure drop is applied, so that the pipe wall shear stress is higher than the initial static yield stress along the pipe and the gel structure begins to break down, the ow will eventually start and the OGF is displaced by the ICF. The ow of each uid consists of an unsheared central plug surrounded by shearing ow in the annular space between the plug ow and the pipe wall. CNR give an account of two-uid displacement when both uids are homogeneous and Bingham plastic at any time. 2.1. Yield front concept and compression ow
Fig. 1. Schematic of compression ow.

Prediction of the start-up ow, and the applied pressure drop PT required to achieve it, is usually based on the assumption that the gelled oil yields everywhere throughout the pipe simultaneously, so that PT > 4 s L/D where s denotes the static yield stress of the oil, assumed to be uniform. However, in practice, yielding is expected to occur near the pipe wall and be accompanied by a compression ow as illustrated in Fig. 1. On application of pressure at the inlet of the pipe at t = 0, the gel within a narrow region (length Lf ) deforms under the stress (Fig. 1a). If the wall shear stress exceeds the static yield stress of the gel, the gel breaks down over this region, which then moves along the pipe as a yield front (Fig. 1b). The high inlet pressure compresses the yielded part of the gelled oil, which allows the ICF, which is also compressed, to enter the pipe. When the yield front reaches the end of the gelled oil plug (at t = t0 , say), the entire plug and ICF move together with the same mass ow rate (Fig. 1c). Fig. 1d illustrates the sudden pressure drop across the yield front, and the more gradual pressure variation expected through the compression ow behind it. Ahead of the front, the oil velocity is zero and the pressure is constant and equal to the exit pressure, assumed to be atmospheric. The above concept of a yield front and compression ow has been discussed in an unpublished TNO report in 1998 on yielding of waxy crude oil. Referring to Fig. 1d, the overall pressure drop PT consists of the combined pressure drops over the compression ow region ( Pc ) and the yield front ( Pf ); that is, PT = Pc + Pf . (1)

If we ignore property variations within the yielded segment, then Pc = 4c Lc D (2)

where c denotes the wall shear stress related to the compression ow. Based on a standard linear approximation for small disturbances (but applied to nite changes), the pressure drop across the yield front can be divided into an inviscid component (vf vc ) and a component (4 s Lf /D) required to overcome the static yield stress over the length of the yield front, so that Pf = 0 vf vc + 4s Lf . D (3)

This represents the force balance across the yield front (on a control volume of length Lf containing the front, moving with the velocity of the front), and can be obtained by extending the inviscid analysis for a pressure wave (e.g. [9]; p. 137) to include wall shear stress in the control volume. Here 0 denotes oil density at atmospheric pressure ahead of the front, and vf and vc denote the velocity of the yield front and the compression ow, respectively. The value of vc used in Eq. (3) is that which occurs just behind the front; it depends on the compressibility of the oil and is approximated as vc f = vf 0 (4)

where f denotes the density change across the yield front. Eq. (4) derives from a standard analysis of mass balance

272

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

through a pressure wave front for small changes ([9]; p. 137), but applied here to the yield front. The pressure drop Pc required to maintain the compression ow depends on vc , the plug length, and on the rheology of the oil. The compression ow is not calculated here, but it is assumed that the pressure drop is sufcient to drive the yield front along the full length of the plug. The minimum overall pressure drop for restart is discussed in Section 4.2. The denition of compressibility () based on nite changes of state is = V/V , P (5)

where V/V is the relative volume change when the pressure change is P. If we consider changes from the external (atmospheric) conditions, then it is easy to deduce the relation between density () and local gauge pressure as = 0 . 1 P (6)
Fig. 2. Schematic of shearing time for each segment at t = t0 .

In general, the compressibility factor may change with temperature, pressure and oil composition. However, we take to be constant and equal to the value corresponding to the oil in the initial uncompressed state, which is the simplest case. We assume that Pf in Eq. (3) is dominated by the inviscid pressure drop at the yield front, with the gel-yielding component making a much smaller contribution. This will be valid for sufciently small values of Lf (i.e. when the front approximates a step change) and for sufciently high pressure behind the front for compression ow there to be signicant. In that case the velocity of the yield front moving through the gelled oil may be approximated by the speed of sound (c) as vf c = dP d =
=0

1 0

(7)

using Eq. (6). The time required for the yield front to traverse the length of the gelled oil plug is then simply calculated from the estimated (constant) value of vf as t0 = L . vf (8)

With sufciently short elements, the oil can be treated as incompressible with uniform properties within each segment. However, at any given time the axial segments can have different uniform properties. If the gelled oil plug is initially homogeneous, as is assumed here, each of the M segments will have the same properties at t = 0, but these will change with time, as will the length and velocity of each segment due to the compressibility effect (Fig. 2a). When t = t0 , each longitudinal segment of gelled oil has experienced a different shearing time, depending on its position and the velocity (vf ) of the yield front traversing it. Typically, the rst segment has been sheared for time t0 , and the last segment (which is just starting to move) has not been sheared. The shearing time for the kth segment at t = t0 can be obtained by subtracting the time required for the yield front to propagate to the kth segment, and can be approximated as tsk (t0 ) = t0
zk

dz , z1 vf

(9)

The compression ow is not calculated here. The approximate representation above simply provides a plausible basis for estimating vf and hence t0 ; however, any reasonable choice of t0 can be provided as input to the start-up model described below. 2.2. Simplied compression process Assume that the applied pressure is sufciently high for the gelled oil plug to move as a whole at time t0 . The pressure required will be determined as part of the solution to the ow model. To account for the longitudinal variation in the oil properties when t t0 , the gelled oil plug at t = 0 is divided into M axial segments of equal lengths L0 = L/M.

where z1 and zk denote the axial locations at t = 0 of the midpoints of the 1st and kth segments, respectively. Based on the assumption of constant vf , the shearing time for the kth segment at t = t0 can be approximately calculated according to the linear relation (Fig. 2b) as follows tsk (t0 ) = Mk t0 . M1 (10)

In the following, dimensionless forms of the equations are based on length scale D, velocity scale vref = s /0 , and time scale tref = D/vref . The pressure and stresses are scaled with s , and the density scale is taken to be 0 . Dimensionless quantities are denoted by an asterisk. For exam ple Eqs. (6)(8) become * = 1/(1 * P* ), v = 1/ and f t0 = (L/D) , respectively, where * = s .

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

273

2.3. Displacement after compression After the yield front has passed through the gelled oil (t t0 ), both the ICF and OGF move at the same mass ow rate. Each OGF segment is tracked using a time stepping procedure, which calculates position, length, density and pressure drop, taking into account the dependence of the rheology on the shearing time for each segment. Consistent with Eq. (10), the shearing time for the kth OGF segment at time t is tsk (t) = t k1 t0 . M1 (11)

where L either denotes the scaled length of an ICF segk ment, or equals LOGF . The volumetric ow rate Q for k k the kth segment can be computed from the wall shear stress wk at a given time. Mass conservation requires that the ow rate should satisfy
k Q = G , k

(16)

With increasing time, OGF segments will be progressively cleared off the pipe, reducing their number. The total length of OGF remaining determines the (increasing) length of ICF in the pipe. The ICF is also divided into segments but these do not need to be tracked in time because the ICF rheology is assumed to be time-independent. Instead, the ICF length is segmented into equal parts at each time step, with the number of segments increased by one at each step to maintain accuracy as the ICF length increases. Thus, the length of an ICF segment at any time t t0 is given by LICF =
OGF L m LICF k=1 Lk = , K+i K+i

where the mass ow rate G* is constant. The volumetric ow rate and mass ow rate have been scaled with D2 vref and 0 D2 vref , respectively. The numerical solution procedure in volves a double iteration: an inner iteration to calculate wk (and hence the pressure drop) over each segment for a given mass ow rate, and an outer iteration to determine the value of G* , which leads to the correct overall pressure drop. The details of this procedure are given later. 2.4. Rheology of OGF and ICF The rheological properties of both OGF and ICF are assumed to be independent of pressure. The rheology of the gelled oil (OGF) is modelled based on a North Sea waxy crude oil [7] which can be reasonably described [8] by a Bingham equation with a time-dependent apparent yield stress ( y ) and a constant plastic viscosity () in terms of the shearing time ts . In dimensionless form, the relationship is
= y (ts ) + He1/2 , y (ts ) =
y (0)y () 1+ ts

(12)

where LOGF is the length of the kth gelled oil segment for k t t0 , m the number of OGF segments remaining in the pipe, K is the number of ICF segments chosen at t = t0 , and the index i denotes the time step. The mass ow rate at any time t t0 must be determined so that the calculated pressure drop over the whole pipe equals the actual pressure drop applied. The overall pressure drop is equal to the sum of the pressure drops ( Pk = Pk1 Pk ) over all uid segments where the pressures Pk are dened at the ends of the segments. In dimensionless form, the length of the kth gelled oil segment for t t0 is given by LOGF = k L (1 Pk ), 0 (13)

> y (ts ),

(17) (18)

+ y (),

and the average density by


k =

1 , 1 Pk

where y (0) and y () denote the apparent yield stress (scaled with s ) at ts = 0 and ts , respectively, and 2 /2 is a Hedstrom number for the oil. Eq. (18) He = D 0 s indicates that the yield stress decays non-linearly with time according to a second-ordered structure breakdown kinetics [8] at a rate characterised the rate constant * = tref , which is a function of the waxy gelled oil structure at restart. The yield stress at ts = 0 equals the static yield stress of the gelled oil (0) = 1), and () < (0) because of the break(hence y y y down in gelled oil structure as the shearing time increases. The ICF is taken to be a time-independent Bingham uid with the constitutive equation = B + HeB 1/2

(14)

> B ,

(19)

where Pk = (Pk1 + Pk )/2 is the scaled average pressure in the segment. Eq. (14) also applies to the ICF with an ICFspecic value for compressibility * , and the numerator replaced by the ICF/OGF density ratio at atmospheric pressure. In keeping with the CNR and other models, the momentum equation for ow at any instant in the pipeline is reduced to a simple quasi-steady force balance between wall shear stress and pressure drop (i.e. inertia is ignored). Thus, the scaled wall shear stress in a segment is given by wk = Pk , 4 L k

where B is the scaled Bingham yield stress, HeB = 2 /2 is a Bingham Hedstrom number, and is the D 0 s B B constant Bingham plastic viscosity.

2.5. Flow of ICF and OGF For the kth segment at time t* , the wall shear stress in Eq. (15) can be expressed in terms of the bulk average velocity v as k
wk = Pk fk k v2 k , = 4 Lk 2

(15)

(20)

274

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

where fk is a Fanning friction factor for the segment. When the ow is laminar, the friction factor for the ICF can be calculated using the BuckinghamReiner equation for pipe ow of a time-independent Bingham uid (e.g. [10]) given by 16 fk = ReBk He4 HeBk Bk 1+ 3 6ReBk 3fk Re7 Bk
1/2

and OGF segments it is fk = 10a Re0.193 , k where a = 1.378(1 + 0.146 exp(2.9 105 Hek )). (24) (23)

(21)

where ReBk = HeB k v is the Bingham Reynolds number k is the Bingham Hedstrom number for and HeBk = HeB k B an ICF segment. Thus, for a given ow rate (and hence v ), k Eq. (21) can be solved iteratively for fk , which then yields wk and Pk from Eq. (20). For laminar ow of the OGF, CNR use nite differences over the radius of the pipe to calculate the radial variation in velocity (and hence the average velocity) from a given wall shear stress. An iterative procedure then yields the inverse relationship of wall shear stress from a given average velocity. However, this nite difference calculation is unnecessary here (and in the CNR paper) because the ow is taken to be quasi steady and the OGF constitutive Eqs. (17) and (18) are of Bingham form with the apparent yield stress and plastic viscosity being independent of shear rate, and hence the radial position. Instead, as for the ICF, we can use Eq. (20) together with the BuckinghamReiner equation for OGF segments at each time t* in which case

Here we use a unied expression for friction factor, developed by Darby and Melson [11], which is valid for both laminar and turbulent ow. For both ICF and OGF segments it is
b b fk = (fkL + fkT ) 1/b

(25)

where fkL and fkT denote laminar (Eqs. (21) or (22)) and turbulent friction factors (Eq. (23)), respectively, and the exponent b = 1.7 + 4 104 . Rek (26)

Since Eq. (23) does not reduce to the (Newtonian) Blasius equation for friction factor in the limit of zero yield stress, it is taken to apply when Hek 1000 [11]. For Hek < 1000 the uid may be considered to be Newtonian with a transition from laminar to turbulent ow occurring when Rek = 2300, in which case fk = 0.079Re0.25 k
16 Rek

Rek < 2300 Rek 2300, (Blasius equation) (27)

fk =

16 Rek

1+

He4 Hek k 3 6Rek 3fk Re7 k

(22)

where Rek = He1/2 k v and Hek = Hek yk (tsk ). The more k general CNR nite difference approach can be used with other constitutive relations (e.g. Casson), or when the rheological parameters vary with radial distance due to shear rate dependence. Radial dependence of rheological properties could also result before start-up from non-uniform cooling at different radial positions in the pipe during shutdown. After start-up, a simple estimate based on thermal diffusivity shows that radial heat transfer between the oil and the surroundings is not signicant within the time scale of startup (based on a 10-h start-up and a 300-mm pipe diameter); consequently, rheological parameters can be taken as independent of radial position throughout start-up if they are assumed to be radially uniform at the beginning (as is the case here). As the gelled oil yields and ows along the pipe, its ow rate gradually increases and its apparent viscosity progressively decreases due to structure breakdown. It is thus possible that at some stage the ow regime in a segment will change from laminar to turbulent, if the applied pump pressure is sufciently high. Such a transition from laminar to turbulent ow would be gradual rather than sudden, as it is for Newtonian uids [11]. For turbulent ow, Darby and Melson [11] developed an empirical expression for friction factor based on an analysis of Hanks and Dadia [12]. For both ICF

Although there will be no sudden transition from laminar to turbulent ow for a yield stress uid, there is a critical Reynolds number Rekc above which the ow ceases to be laminar [13], given by Rekc = Hek 8k 1 4 4k + k 3 3 (28)

where k is the ratio of the yield stress to a critical wall stress satisfying k (1 k )
3

Hek . 16, 800

(29)

Although it is not used in the ow calculation, Eq. (28) will be used to identify the progression to turbulence in the pipeline. In the present model, we consider only smooth pipes for simplicity; this is the common approach for pipe ow simulation. However, in reality, wax deposition from crude oil can result in an effective pipeline roughness. The effect of wall roughness can be easily incorporated into the model by modifying the friction factor Eq. (23).

3. Numerical procedure The numerical scheme for calculating the ow in the pipeline involves of two levels of iteration at each time step. First, the bulk mass ow rate G* is guessed (an appropri-

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

275

ate value is that calculated at the previous time step). The friction factor fk is then calculated by iteration for each longitudinal ICF and OGF segment at this time step using Eqs. (21)(27), with a mean velocity evaluated from the continuity Eq. (16) and shearing time from Eq. (11). The pressure drop over each segment is then evaluated from Eq. (20) and the length and density of each segment are updated from the pressure distribution using Eqs. (12)(14). The location of each segment, and the ICFOGF interface, is determined relative to the downstream end of the OGF plug. The overall pressure drop is then obtained by summation of the pressure drops of all ICF and OGF segments. The difference between the calculated and the applied total pressure drop is calculated, and the mass ow rate is adjusted according to that difference using a bisection method, and the procedure is repeated. The process is continued until the difference between the pressure drops is negligible. If the applied pressure is not sufcient to start the ow at a given time t* , then the mass ow rate G* iterates to zero and the calculated pressure drop becomes the minimum value required for start-up at that time. This minimum P* progressively de creases towards a limiting value as y approaches y () at long shearing time (Eq. (18)), so that a delayed start-up may occur at some time t > t0 . If the applied pressure drop is less * limit, which corresponds to the equilibthan the critical P rium yield stress y (), then start-up will not occur at any time. In the simulation, the total number of OGF segments eventually decreases with time as OGF segments are cleared off the pipe one by one. The shearing history and the position of each OGF segment is traced in order to determine the timevariation of their properties. At each time step the location of the downstream end of the OGF is updated (according to the calculated value of G* ) if it has not reached the end of the pipe, otherwise the mass of OGF that is cleared off the pipe is calculated. Particular treatment must be taken for the OGF segment at the pipe outlet, as only part of this segment may have been cleared off the pipe, and the segment length must be adjusted when calculating the total pressure drop.

Like CNR, we consider a pipeline with length/inside diameter ratio L/D = 32787. The pipeline is assumed to be lled with gelled oil initially (100% hold-up). The dimensionless compressibility coefcient is taken to be * = 3.606 106 based on compressibility measurements (1.127 107 Pa1 ) of Cawkwell and Charles [14] for an ambient temperature of about 18 C, scaled according to inverse of the OGF static yield stress of CNR (32.0 Pa). In that case the scaled initial compression time t0 = 62.3 (based on Eq. (8)). The number of axial segments (M) into which the OGF is divided initially is taken to be 100 and the number of ICF segments (K) at t = t0 (when the yield front has passed through the pipeline) is taken to be 20. Increasing M to 1000 and K to 200 resulted in negligible differences in predicted mass ow rate and clearance time. 4.1. Comparison with previous incompressible oil phase model (CNR) Fig. 3 shows the simulated start-up predictions with the current model and the previous (CNR) model for the same ap-

4. Results and discussion To illustrate the use of the numerical model, we choose dimensionless parameters based on the same properties for the gelled oil and the displacing uid, and the same pipeline, as was considered by Chang et al. [8]. In that case, the OGF is a waxy crude oil from North Sea whose rheological behaviour can be described by Eqs. (17) and (18). The correspond ing dimensionless parameters are y (0) = 1; y () = 0.172; * = 0.00406; He = 5.16 105 . The ICF is assumed to be a time-independent Bingham uid (Eq. (19)) with B = 0.172 5 (i.e. / = 1). The ICF/OGF density raand He = 5.16 10 B tio at atmospheric pressure is taken to be 1.

Fig. 3. Comparison of (a) mass ow rate and (b) ICFOGF interface position from the inlet, predicted by the current and the previous (CNR) model for an applied pressure drop PT / s = 1.375 105 .

276

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

plied pressure drop ( PT / s = 1.375 105 ). There are clear differences between the results from these two models. Fig. 3a compares the computed mass ow rate varying with time from start-up to complete displacement of the gelled oil. With the CNR model, which is based on an incompressibility assumption for both ICF and OGF, ow begins immediately after the pressure difference is applied. The CNR model also assumes that transition from laminar to turbulent ow occurs abruptly at the critical Reynolds number given by Eq. (28); this results in a sudden variation in ow rate at transition. In contrast, the present model exhibits a smooth transition as it uses a unied expression (Eq. (5)), which is valid for both laminar and turbulent ow [11]. Both the CNR model and the present model without compressibility (* = 0) are in close agreement except in the transition region. With the current model in which the compressibility of both uids is taken into account, a delayed start-up is observed due to the compression process after the pressure difference is applied (see Fig. 3a). The delay time t0 is dependent on the gelled oil properties such as bulk density and compressibility. Here Eq. (8) is assumed to apply, in which case t0 = (L/D) . Since the oil density and volumetric ow rate in the current model are computed for each longitudinal segment individually, this model is able to describe a progressive transition between laminar to turbulent ow from segment to segment, which is more realistic than the CNR model for which transition to turbulence occurs everywhere simultaneously in each of the ICF and OGF. The nal ow rate at breakthrough predicted by the current model is about 17.6% higher than that obtained from the CNR model for the same applied pressure difference PT / s = 1.375 105 . Fig. 3b shows the location of the ICFOGF interface moving with time during start-up computed using both models. As discussed above, while the OGF plug starts to move immediately after the pressure is applied when the oil is incompressible, it begins to move only after a dimensionless time of 62.3 for the current model with compressibility included (* = 3.606 106 ). This is the time taken for the yield front to reach the end of the pipe. During this time, the interface has moved forward about 8140 pipe diameters as the result of compression of the OGF plug. The time required to clear the OGF off the pipe is 1652 with compressibility included and 1915 for the previous incompressible model. Thus, without taking compressibility into account, the predicted clearance time is about 16% higher. In Fig. 4, the pressure distribution along the pipeline is displayed at three different times after start-up simulated with the current model when PT / s = 1.0625 105 . The variable Z indicates the location of the ICFOGF interface. Generally, the slope of the pressure distribution increases gradually with time in the ICF region behind the interface due to the increase of ow rate, but decreases gradually with time in the OGF region due to the time-dependent yield stress decay. The curve on each side of the interface is not linear, as oil properties are dependent on axial position due to compressibility (ICF and OGF) and shear history (OGF) effects. For the CNR model,

Fig. 4. Pressure distribution along a pipeline during start-up predicted with the current model for an applied pressure drop PT / s = 1.0625 105 .

however, the curves on each side of the interface would be straight lines as that model assumes that the properties of the ICF and OGF are independent of the axial position. 4.2. Effect of restart pressure Model start-up predictions of mass ow rate and position of ICFOGF interface under different applied pressure differences are displayed in Fig. 5. As is expected, lower PT values result in lower mass ow rates (Fig. 5a) and less rapid displacement of the gelled oil plug (Fig. 5b). The initial compression time is the same (t0 = 62.3) for all start-up pressures. Fig. 5a shows that for PT / s 90,625 the ow does not start immediately after the compression; instead there is additional delay associated with a progressive breakdown of the gelled oil structure, which causes a decrease in the yield stress of gelled oil in the last segment to a sufcient level for it to be overcome by the applied pressure. The delay time is larger for smaller values of PT and it is found that when PT / s 22,781 the delay time is innite (i.e. start-up ow never begins). The concept of start-up with delay has previously been put forward in the CNR paper and corresponds to the conditions where the applied pressure difference lies between two critical limits corresponding to the static yield stress y (0) and the equilibrium yield stress y () of the gelled oil [8]. The minimum applied pressure drop above which restart will occur can be calculated directly from the equilibrium yield stress as 4L y ()/D. In the present example, the corresponding minimum value of PT / s equals 22,531, which agrees (to within numerical error) with the limiting value of 22,781 determined from the simulation. For a successful start-up, the ow rate is non-zero (G > 0) and the calculated pressure difference ( Pcalc ) along the pipeline balances the wall shear stress (Eq. (15)) and equals the applied pressure difference PT . However, when G = 0 0 we set Pcalc = Pcalc , which is the value calculated assuming that the wall shear stress equals the apparent yield 0 stress (i.e. Pcalc is the minimum pressure difference re-

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

277

Fig. 5. Predicted (a) mass ow rate and (b) ICFOGF interface position from the inlet for various applied pressure drops. Symbols ( ) and ( ) mark the beginning of turbulence in the ICF and the time after which the entire ICF is turbulent, respectively. Symbol ( ) marks the beginning of turbulence in the OGF, and ( ) marks the time after which the entire OGF is turbulent.
0 quired for ow to begin). While PT Pcalc there is no 0 ow initially. However, the value of Pcalc decreases, together with the apparent yield stress (Eq. (18)), towards a 0 limiting value Pcalc () as the time ts elapsed after the passage of the yield front increases (referred to herein as the shearing time, regardless of whether ow is occurring). Even0 0 tually PT = Pcalc , provided PT > Pcalc (), where0 thereafter. As upon ow begins and PT = Pcalc > Pcalc 0 indicated above, Pcalc ()/s = 22, 781 for the present test case. Fig. 5b shows that for PT / s 75,000, not only is there a delay after the passage of the yield front before ow begins, but also the ICFOGF interface position appears to recede during the delay period. This apparent backward movement of the interface without ow occurs because the calculated 0 pressure difference prior to start-up equals Pcalc , which decreases with time as discussed above, thereby resulting in a predicted re-expansion of the oil along the pipe after its initial compression. However, in reality, since the pressures in the pipeline prior to start-up will be lower than the calculated values, which are based on the yield stress, only values of the predicted interface position from the onset of ow are meaningful in general.

The condition Rek Rekc (see Eq. (28)) is used to identify the occurrence of turbulence in the pipeline (note that this condition is not used in the ow calculation, and is used for identication purposes only). The corresponding times at which turbulence occurs are marked in Fig. 5a with symbols showing the beginning of turbulence in the ICF ( ) and OGF ( ), and the time after which the entire ICF ( ) and OGF ( ) are turbulent. For PT / s > 59375 the symbols ( ) and ( ) are effectively coincident so the spread of turbulence in the ICF can be regarded as almost instantaneous for those cases. Turbulence begins rst in the ICF since it has a lower yield stress (B = 0.172) than that of the OGF (0.172 y 1). For PT / s = 137500, the ICF is turbulent immediately ow begins. As the applied pressure drop is reduced, turbulence appears at later times until, for PT / s = 43750, the ICF and OGF remain laminar throughout, except for the onset of turbulence in the ICF just as the OGF is cleared from the pipe (not shown in Fig. 5a). At this lowest applied pressure when the pipeline is laminar at all but the last moments before clearance, the spread of ICF turbulence is much slower and cannot be regarded as instantaneous. In that case the elapsed t/tref from the onset of turbulence to pipe clearance is 286 but during that time turbulence had spread to only 8% of the pipe length. In Fig. 6, the distribution of Reynolds number, and critical Reynolds number (Eq. (28)), in the pipeline is shown for the times corresponding to symbol locations in Fig. 5a when PT / s = 1.0625 105 . The Reynolds number is proportional to the mass ow rate and inversely proportional to the plastic viscosity. It is uniform along the pipe because = B has been chosen, and it increases with time as the mass ow rate increases. The step change in critical Reynolds number Rec marks the location of the ICFOGF interface. The value of Rec is lower in the ICF because the yield stress is lower there than in the OGF as discussed above. The decrease in Rec with distance along the ICF or OGF is associated with the decrease in density along the pipe resulting from the pressure drop. The Rec values decrease with time in the OGF because the yield stress is decreasing in the gelled

Fig. 6. Reynolds number and critical Reynolds number distribution along the pipe for an applied pressure drop PT / s = 1.0625 105 at times corresponding to symbol locations in Fig. 5a.

278

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

Fig. 7. Time to clear the gelled oil from the pipeline as a function of applied pressure difference.

oil, and they are independent of time in the ICF where the yield stress is constant. Fig. 6 shows turbulence beginning rst at the ICFOGF interface and then rapidly spreading upstream with the ICF becoming completely turbulent when t/tref = 171. This occurs over a time interval less than 2.5. Turbulence in the OGF occurs rst at the downstream end when t/tref = 248 and then develops upstream until the OGF is completely turbulent when t/tref = 309. The time required to completely displace the gelled oil from the pipeline is an important parameter in pipeline restart. As shown in Fig. 7 the predicted clearance time, determined as the time when the IGFOGF interface reaches the end of the pipe, decreases with increasing applied pressure differences PT . The dependence of the clearance time on PT is most pronounced in the region of low PT , corresponding to the start-up with delay mode. However, as discussed above, the clearance time will become innitely large if PT / s is reduced toward the critical value of 22,781, below which start-up would be impossible. 4.3. Effect of oil compressibility Fig. 8 shows simulated results for four constant values of the compressibility coefcient (* ) of the displaced gelled oil at a xed total pressure drop of PT / s = 1.0625 105 . For the case of incompressible oil (* = 0), since the oil density is not affected by pressure, there is no initial compression ow and hence no axial displacement due to compressibility. The initial compression time t0 = t0 /tref is zero, of course, * = 0. When * is increased by one order of magnitude for with values 3.606 107 , 1803 106 and 3.606 106 , t0 increases and takes values 19.7, 44.0 and 62.3, respectively. However, the displacements shown in Fig. 8 do not start immediately after the initial compression due to transient yield stress behaviour, except for the highest compressibility value considered (* = 3.606 106 ). Flow of the gelled oil starts at dimensionless time 78.9 for * = 3.606 107 and at 66.2 for * = 1.803 106 , but immediately after the compression at 62.3 for * = 3.606 106 . The corresponding dimension-

Fig. 8. Predicted (a) mass ow rate and (b) ICFOGF interface position from the inlet for various values of compressibility * when the applied pressure drop PT / s = 1.0625 105 .

less start-up times after the initial compression are 59.2, 15.9 and 0. Flow in the incompressible case (* = 0) also does not begin immediately after start-up but proceeds only after a 73.8 time delay. The immediate start-up, after compression ow, for the largest compressibility value is consistent with the notion that increasing the compressibility effect can allow ow to restart with a lower wall shear stress [8]. The results in Fig. 8 also show that the nal mass ow rate increases by 11% and the clearance time falls by 12% as the dimensionless compressibility is increased by an order of magnitude from 3.606 107 to 3.606 106 . A reduction in clearance time with increasing compressibility agrees with predictions of Cawkwell and Charles [6]. Note that, in practice, may be not constant and can be affected by factors such as pressure, temperature and oil composition. However, provided that is known as a function of pressure and temperature, this dependence can be incorporated easily into the current model. 4.4. Effect of liquid hold-up Fig. 9 shows the effect of different liquid hold-ups on the ow rate and displacement of the gelled oil. Liquid hold-up is dened as fraction of the pipe length initially occupied by the gelled oil. It is assumed that before the restart, the OGF plug is located immediately after the inlet. When the liquid

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

279

Fig. 9. Predicted (a) mass ow rate and (b) ICFOGF interface position from the inlet for various values of initial oil hold-up when the applied pressure drop PT / s = 1.0625 105 .

Fig. 10. Predicted (a) mass ow rate and (b) ICFOGF interface position from the inlet for various values of initial compression time when the applied pressure drop PT / s = 1.0625 105 .

hold-up is less than 1 (i.e. the initial length of the oil plug is shorter than the pipe length), the OGF plug must rst move forwards to the outlet before the OGF starts to be cleared off the pipe. The ow rate at any time is determined by the relative lengths of the OGF and ICF, together with the OGF yield stress at that time, which is greater than (or equal to at large times in this case) the ICF yield stress. For 100% liquid hold-up, the ow rate (Fig. 9a) increases monotonically with time throughout the start-up process as more of the higher yield stress OGF is displaced from the pipe and the yield stress of the remaining OGF decreases with continuous shear. For 75% hold-up, the ow rate increases initially as before (until t/tref = 230), then decreases slightly due to the increased length of (and increased wall friction over) the ICF as it enters the pipe, pushing the OGF towards the outlet. After the end of the gelled oil plug has reached the outlet (at t/tref = 525), the ow rate nally increases gradually as the OGF is displaced from the pipe by the ICF, which has a lower yield stress. For liquid hold-ups of 50 and 25%, the initial length of the OGF is sufciently small for the increased length of the ICF to cause the ow rate to decrease with time initially. After the end of the OGF plug arrives at the outlet, the ow rate starts to increase gradually, in the same way as for other values of liquid hold-up.

The results in Fig. 9 also show that the compression time, after which ow can begin, is predicted to decrease as the hold-up is decreased. This occurs because the distance travelled by the yield front is shorter when the OGF hold-up is lower. Reducing the liquid hold-up is predicted to lower the total time required to clear all of the OGF from the pipe (Fig. 9b). This is expected since a smaller oil plug offers less resistance to ow. 4.5. Effect of initial compression time For all results presented so far, the initial compression time (t0 ) is calculated from the sound speed, based on the OGF properties at its normal state. In real restart operations, the compression process is inuenced by many factors such as temperature and contents, and the compression time may be different from this computed value. Fig. 10 shows the effect of initial compression time on the predicted mass ow rate and displacement. Results are presented for four different compression times at a xed pressure difference PT / s = 1.0625 105 . Although the effect on the initial stage of the start-up is signicant, the results in Fig. 10a indicate that the mass ow rate becomes independent of compression time after a dimensionless time of 1000. Fig. 10b shows the effect of compression time on the movement of

280

M.R. Davidson et al. / J. Non-Newtonian Fluid Mech. 123 (2004) 269280

the ICFOGF interface. If the compression takes no time (not possible in reality), the interface starts to move only after a delay time of about 12. For compression times of 191, 382, and 573, the OGF plug moves immediately after the compression process. Overall, the displacement time reduces from 2045 to 1916 (6.3%); but the total clearance time increases from 2045 to 2488 (21.6%) as the dimensionless compression time increases from 0 to 573.

incompressible and homogeneous, with a friction loss based on a power-law rheology with no yield stress. We believe this approach is rather conservative and may be suitable for pipe sizing but is less realistic for start-up prediction purposes. In the present model, with the focus of pipeline restart, the oil compressibility and the time-varying yield stress play signicant roles during the initial transient ow of the gelled oil.

5. Conclusion The start-up model presented incorporates the effects of both the yield stress and thixotropic behaviour, and the compressibility of the gelled oil after a period of pipeline shutdown. Under an applied pressure, ow can commence after a delay period due to compression ow associated with a yield front traversing the length of the oil plug and time-dependent breakdown of the gel structure. The lag time is a function of applied pressure, gelled oil properties and length of the gelled oil plug. As the gelled oil ows in the pipe, its properties vary continuously along the pipe length due to non-uniform shear history. Longitudinal variations of the uid properties are considered by the model, which is able to describe a progressive transition between laminar to turbulent ow for each uid element. This results in the predicted spread of turbulence from the downstream to the upstream end of the gelled oil when transition in the ow regime occurs. Compared to the previous model based on incompressible two-uid displacement, the present model gives a more realistic and improved prediction of the start-up ow of gelled oil. The simulated results show that oil compressibility effectively result in an increase in the nal oil ow rate and a decrease in the clearance time, which is often observed in real start-up situations. Although the simulation assumes the gelled oil follows the Bingham plastic behaviour and a structural kinetic model, the model is sufciently robust and can easily accommodate other rheological models for both the gelled oil and the displacing uid. The model will be further extended to include more complex factors, such as multiple plugs separated by gas lled sections and multiphase behaviour with variable compressibility due to thermal shrinkage and non-uniform cooling during shutdown. Ajienka and Ikoku [15] developed a design method for waxy crude oil pipelines based on the highest value of the pressure drops required for (i) normal steady ow of nongelled oil at the operating throughput, (ii) restarting ow of the gelled oil at 50% operating ow rate and, (iii) breaking the gelled oil. Their model assumes the oil to be single-phase,

Acknowledgement The rst author wishes to thank Professor D. Boger for introducing him to this problem. The authors are also grateful to Statoil for permission to publish this paper.

References
[1] L.T. Wardhaugh, D.V. Boger, Flow characteristics of waxy crude oils: application to pipeline design, AIChE J. 37 (6) (1991) 871885. [2] C. Chang, D.V. Boger, Q.D. Nguyen, The yielding of waxy crude oils, Ind. Eng. Chem. Res. 37 (1998) 15511559. [3] C. Chang, Yielding of Waxy Crude Oil and Numerical Simulation of Gelled Oil Pipeline Start-up. Ph.D. Thesis, The University of Melbourne, Australia, 1998. [4] D.C.H. Cheng, W. Whittaker, in: Hydrotransport 2, The Second International Conference on the Hydraulic Transport of Solids in Pipes, University of Warwick, BHRA Fluid Engineering, 2022 September, 1972, pp. 441. [5] J. Sestak, M.E. Charles, M.G. Cawkwell, M. Houska, Start-up of gelled crude oil pipelines, J. Pipelines 6 (1987) 1524. [6] M.G. Cawkwell, M.E. Charles, An improved model for start-up of pipelines containing gelled crude oil, J. Pipelines 7 (1987) 4152. [7] H.P. Rnningsen, Rheological behaviour of gelled, waxy North Sea crude oils, J. Petr. Sci. Eng. 7 (1992) 177213. [8] C. Chang, Q.D. Nguyen, H.P. Rnningsen, Isothermal start-up of pipeline transporting waxy crude oil, J. Non-Newtonian Fluid Mech. 87 (1999) 127154. [9] J.F. Douglas, J.M. Gasiorek, J.A. Swafeld, Fluid Mechanics, third ed., Longman, Singapore, 1995. [10] G.W. Govier, K. Aziz, The Flow of Complex Mixtures in Pipes, Krieger Publisher Co., Florida, 1977. [11] R. Darby, J. Melson, How to predict the friction factor for ow of Bingham plastics, Chem. Eng. Dec. 28 (1981) 5961. [12] R.W. Hanks, B.H. Dadia, Theoretical analysis of the turbulent ow of non-Newtonian slurries in pipes, AIChE J. 17 (3) (1971) 554557. [13] R.W. Hanks, The laminar-turbulent transition for uids with a yield stress, AIChE J. 9 (3) (1963) 306309. [14] M.G. Cawkwell, M.E. Charles, Start-up of pipelines containing gelled crude oils: comparison of improved model and pilot pipeline data, J. Pipelines 7 (1989) 265280. [15] J.A. Ajienka, C.U. Ikoku, Criteria for the design of waxy crude oil pipelines: maximum pump (horsepower) pressure requirement, J. Petr. Sci. Eng. 13 (1995) 8794.

S-ar putea să vă placă și