Rapid Flow Startup in Filled Horizontal Pipelines

Jose G. Vasconcelos1 and Steven J. Wright2

ABSTRACT

Maintenance or repair of water mains may require the interruption of flow. Re-initiation of flow may

present opportunities for the interaction of air and the accelerating water, resulting in undesirable pressure

transients. This study focuses on air intrusion during flow startup in an initially filled horizontal pipeline

supplied by a constant-head reservoir. Flow was initiated by opening the downstream end, allowing a free

discharge at that point. During the initial phases of flow startup, an air cavity intruded in the upper portion

of the pipe cross section while water exited the pipeline through the lower portion. The initial reservoir

head controlled the rate of water velocity increase and thus the air intrusion advance. Air intrusion occurred

until the water velocity was sufficient to arrest the intrusion. Higher reservoir heads resulted in the sub

sequent expulsion of the intrusion from the pipeline. However, for lower reservoir heads, the intrusion was

not expelled, but re-advanced to a final position that was controlled by shear forces. Criteria are developed

for the flow conditions necessary to arrest the air intrusion in a horizontal pipeline as well as the intrusion

lengths to be expected.

Keywords: Transient flow; Water Pipelines; Open channel flow; Pressurized flow; Intrusion

INTRODUCTION

Water mains are periodically subject to maintenance, requiring draining and subsequent refilling of

sections of the pipeline. A typical concern during such operations is damage to the water mains due to high

pressure transients associated with the interaction between the inflowing water and the air existing in the

pipeline. Even when the filling operation is performed slowly and when air relief valves and vacuum breakers

are distributed along the pipeline, there is still potential for damage if the system has not been designed

properly. One possible cause for pipeline damage is the entrapment of discrete air pockets. Martin (1976)

has demonstrated how the compression of air in discrete pockets can result in high transient pressures within

the filling pipeline. Additionally, waterhammer pressure peaks may result from the sudden expulsion of air

pockets through ventilation points (Albertson and Andrews, 1971).

In spite of the potential for a variety of transient phenomena associated with flow startup, experimental

investigations related to this problem have been limited. Most previous studies have developed numerical

1Professor, Dept. of Civ. and Env. Engrg., Univ. of Brasilia, Campus Darcy Ribeiro, Brasilia, DF, Brazil CEP-70910.E-mail: jvasconcelos@unb.br.

2Professor, Dept. of Civ. and Env. Engrg., Univ. of Michigan, 1351 Beal Av., Ann Arbor, MI 48109. E-mail:sjwright@engin.umich.edu.

1

approaches to simulate the transient flow conditions; comparatively less emphasis has been placed on ex-

perimental investigations that could help to assess the validity of the assumptions used by such numerical

models as well the overall applicability of such models. Liou and Hunt (1996) proposed a rigid-column

model of flow startup in empty pipelines with undulating elevation profiles, assuming a vertical interface

between the air and water phases. They proposed a velocity-based criterion to justify the application of

the rigid-column approach. Experimental investigations were performed with a pipe of internal diameter

D=2.29 cm and length L=6.66 m to verify the model. The pipeline, with downward slopes of 2.66% and

2.25%, was connected to an upstream reservoir. The initial ratio between the pressure head at the reservoir

H and D was greater than 15.4. The proposed model was generally successful in replicating the results from

the experiments. This study did not rigorously verify the proposed criterion for the existence of a vertical

air/water interface.

Another numerical model for the flow startup of pipelines with undulating profiles was presented by

Izquierdo et al. (1999), who also applied the rigid-column approach to each segment of the pipeline, assuming

that water pools were initially present in the low points of the pipeline. The model assumes that a pipe-filling

inflow front forms following flow startup, entrapping an air pocket between the front and the next water

pool. According to the proposed numerical model, the compression of this air pocket results in very high

pressures within the system. An experimental investigation (Fuertes et al., 2000) confirmed the predictions

of the numerical model proposed. Space limitations in the experimental setup required the use of large

pipeline slopes, a condition that may not be representative of typical applications.

Such rigid-column-based models assume that any transition between air and water phases in the pipeline

occurs through a vertical interface. If the pipeline is filled sufficiently slowly, this assumption should not

generally be expected to be valid since the air/water interface would be nearly horizontal in locations where

the pipeline slopes upwards and a free surface flow along the bottom of the pipe would occur along the

downsloping sections. Even during rapid filling of pipelines, the assumption of a nearly vertical air/water

interface may be questionable. Understanding of the interactions between the two fluid phases during typical

filling scenarios is desirable to understand the limitations of such pipeline filling models.

A velocity-based criterion was proposed by Liou and Hunt (1996) to determine whether full-pipe-flow

conditions will exist. This criterion suggests that if the flow velocity U > 0.54√gD, with g the gravitational

acceleration, the flow will be in a pressurized regime with an essentially vertical air/water interface. This

expression represents the velocity of an intruding air cavity cair in a horizontal pipeline initially filled and

with no initial velocity. The implication is that so long as the water velocity exceeds this air intrusion speed,

air cannot move upstream against the water flow and therefore should be expelled in front of the advancing

water front.

2

The results presented in the investigation by Guizani et al. (2005) point to a different mechanism to

determine the existence of full pipe flow conditions. Experiments were performed in a horizontal, 94-mm

diameter, 14.6-m long acrylic pipeline in which flow was suddenly initiated in an empty pipe set on various

small, downward slopes. These experiments were performed such that U > cair, but showed that even

in these cases a broad filling front was observed. Under certain conditions, the entire pipe was not in a

pressurized state even as the flow exited the downstream end of the pipe. The authors also observed that

the length of the inflow front, measured from the forward part of the advancing water to the point where

the water reached the pipe crown, increased as the filling front moved downstream. The authors concluded

that a more satisfactory criterion for full-pipe-flow regime should be based on the pipe conveyance. In other

words, the inflow discharge must be sufficiently large to exceed the conveyance capacity, Qc, of the conduit

in a free-surface-flow regime. The shape of the inflow front resembled a dam-break front propagating over a

dry bed with friction, as described by Whitham (1955).

There have also been limited investigations on a related type of pipeline problem: the rapid startup of

filled pipelines. Rapid startup may be defined as occurring when the time required for the valve opening is

much smaller than the time required for the flow to be established in pipeline. Wylie and Streeter (1993)

propose a startup time Tw = LV/g∆H, in which Tw is the time required for a flow to be accelerated from

zero to a velocity V within a frictionless pipeline with length L and driven by a constant pressure head of

∆H. Friction tends to increase Tw, which can be obtained by integrating the pipeline momentum equation

over time using the rigid-column assumption. Standard numerical modeling approaches for rapid startup of

filled pipelines also invoke the rigid-column assumption and therefore completely disregard any interaction of

the air with the flow. Probably the simplest example of this flow condition is the case of a pipeline connected

upstream to a reservoir while at the downstream end a valve blocks the flow that would otherwise discharge

freely to the atmosphere. Townson (1991) pointed out that as the water flow is suddenly initiated by quickly

opening the valve, air advances upstream occupying the upper portion of the conduit, while water flows

beneath it. The air intrusion is in the form of a gravity current, and moves upstream with a velocity of

cair ≈ 0.5√gD. As the water accelerates in the pipeline and when U > cair, the air intrusion will be halted

and subsequently pushed downstream. The issue is particularly important in non-horizontal pipelines due

to the possibilities for the air to advance to a point where it becomes entrapped as discrete pockets. No

experimental investigation has been performed to corroborate this flow phenomena. An important question is

whether the previous analyses for air intrusion dynamics can be applied to typical conditions associated with

flow initiation of liquid flow in pipelines either for conditions of an initially empty pipeline or in an initially

full pipeline that would have exposure to air at atmospheric pressure at the downstream end. Another

relevant question is whether rigid-column models can always be applied for rapid startup of filled pipelines.

3

It seems intuitive that if the startup time of the pipeline is too long and the air intrusion velocity is of the

same order of magnitude as the final steady velocity of the pipeline, by the time that the flow is established,

the length of the pipeline intruded by air is too significant to be disregarded in the analysis. In such cases,

a different modeling framework should be applied to the problem.

Air intrusion velocity

In order to understand the basis for an intrusion velocity-based criterion, the theory of Benjamin (1968)

is considered. Benjamin formulated the conditions for the propagation of a large air volume in a draining

horizontal pipeline by applying a momentum analysis in a frame of reference moving with the intrusion.

His analysis was primarily for intrusions into rectangular ducts but he included some partial analyses for

circular cross-sections. In particular, Benjamin showed that there was a unique solution for which the cavity

advanced steadily (at constant velocity) without energy dissipation. For a circular pipe, this occurs with the

relative water depth behind the front, y/D, of 0.563 for which Ucav/√gD = 0.542. Here Ucav is the intrusion

velocity relative to the approaching water flow. In the case of the flow startup of filled pipelines, as there

is no initial approaching inflow, Ucav is equivalent to cair. Benjamin also discussed a family of solutions for

which the intrusion front would be sharply curved, similar to the zero-dissipation case, but followed by a

hydraulic jump-like transition to the depth y. Townson (1991) presents a plot of the solution for this case

in a circular pipe but does not give the equations. The solution can be presented in terms of the half angle

α = cos−1(2y/D− 1) in which 2α is the angle subtended from the center of the pipe to the free surface. The

solution is given by:

c2airgD

=[π − α− sinα cosα

π

]2[1− 13π (3 sinα+ 3(π − α) cosα− sin3 α)

π−α+sinα cosαπ

[2− π−α+sinα cosα

π

] ](1)

and is plotted in Figure 1. It has been suggested by Cardle and Song (1988) that the energy-dissipating

solution can only occur if y/D > 0.681; otherwise the Froude Number associated with the conditions behind

the jump at the intrusion front will be subcritical and therefore incapable of maintaining a steady flow.

¿From Figure 1, provided y/D <≈ 0.8, the intrusion celerity will not be substantially different from that of

the zero energy dissipation case. However, as the intrusion thickness, δair, decreases, cair begins to decrease

dramatically, suggesting that U lower than 0.542√gD would be required to prevent thin air intrusions.

Observations in the experiments discussed below indicate that the jump at the intrusion front is typically

undular in nature, resulting in the formation of free surface waves behind the intrusion front. It was often

observed that the crests of these waves would touch the crown of the pipe, resulting in the formation of a

small entrapped air pocket and the breakdown of the nearly steady intrusion front.

Since U in the pipeline is initially zero, any non-zero cair will result in the intrusion of air upstream into

4

the pipe. Initially, the intrusion appears to be of the energy conserving type with a nearly equal exchange

flow of water downstream and air upstream. However, as U begins to increase, the occurrence of critical flow

at the free overfall at the pipe outlet fixes the depth-discharge relation there and forces the intrusion to adapt

to one of the possible energy-dissipating solutions. As the intrusion proceeds further upstream, the presence

of shear forces along the water interface with air and the pipe walls create a flow profile similar to a drawdown

profile in gradually varied flow and the intrusion becomes progressively thinner with a corresponding decrease

in intrusion celerity according to Figure 1.

Depending on the rate of increase in U , one of two events follow:

1. U increases until it reaches cair, after which the air intrusion is swept out of the pipe;

2. the intrusion thickness reaches a sufficiently small value that the free surface wave crests behind the

intrusion front touch the pipe crown resulting in a breakdown of the intrusion. In the experiments

described below, this second outcome only occurred for relatively long air intrusions.

Since cair ∼√gD, there are implications relating to D and the relative potential for air intrusion. For

example, in the experiments of Liou and Hunt (1996) with a 2.29 cm-diameter pipe, the intrusion celerity for

a zero-energy-dissipation solution is only 0.25 m/s and for any significant driving head, the intrusion would

likely only propagate a small distance upstream before being reversed by the accelerating flow. However,

if the pipe diameter is increased to 2 m such as may be the case in a large water distribution main, the

intrusion celerity would be increased by about an order of magnitude and the effects of air intrusion can be

expected to be much more pronounced.

OBJECTIVES

An experimental investigation was performed to examine the problem of rapid flow startup in initially

filled pipelines subject to air intrusion from the downstream end upon flow initiation. One goal of this

investigation was to characterize the advance of the air intrusion as the flow initiates and to correlate

this intrusion characteristics with the supply reservoir head. The second goal of this investigation was to

determine whether the criterion for full-pipe-flows should be based on velocity or on conveyance. Although

the issues are likely to be equally important for flow startup in initially empty pipelines, it was much easier

to draw conclusions from experiments performed in an initially full pipeline. Also, even though the more

interesting applications of this investigation would be to the case of undulating pipelines, it was decided to

conduct experiments with a horizontal pipeline as a first step at examining some of the key issues of the

problem.

5

METHODOLOGY

The experimental setup consisted of a L = 14.1m D = 9.4cm acrylic pipeline laid on a horizontal

slope. The pipeline was connected to an upstream reservoir with a cross-sectional area of 2.96 m2. The

initial reservoir water level was used as the primary variable to generate the range of flow responses. The

pipeline was free of air prior to flow initiation; the blockage of the flow was provided by a rubber plug at

the downstream end. This plug could be removed in a fraction of a second and an upstream air intrusion

formed immediately with a well-defined intrusion front, indicating negligible effect of any irregularities in the

plug removal from the system response, and therefore, the experimental results were quite repeatable. The

pipeline was supported in such way that following the removal of the plug, free discharge conditions occurred

at the downstream outlet to the pipe. A sketch of the experimental apparatus used in the investigation is

presented in Figure 2.

A list of experimental measurements devices used in the experiments is presented below, along with x̄,

which is the distance from the upstream reservoir x normalized by L:

1. Three piezo-resistive pressure transducers manufactured by ENDEVCO, model 8510-B, located at

• Pipe invert x̄ = 0.184

• Pipe invert x̄ = 0.872

• Pipe crown x̄ = 0.879

2. Two MicroADV (Acoustic Doppler Velocimeter) probes manufactured by SONTEK located at x̄ =

0.262, referred to as the upstream ADV, and x̄ = 0.887, referred to as the downstream ADV.

3. Digital camcorder(30 frames per second) recording the advance of the intrusion between the 13.1 m

and the 14.1 m stations. In certain experiments, the video recorder was moved to follow the front in

cases where the intrusion propagated several meters upstream.

The MicroADV probe introduced a local resistance to the flow due to form drag on the probe head but

the projected area of the probe head is much less than the cross-sectional area of the pipe. The sensor head

was placed as close to the pipe wall as possible so that the 5 cm focal length would place the sample volume

close to the pipe centerline. Since there is very little boundary layer growth during the short duration of a

typical experiment, it is presumed that the probe measures very close to the cross-sectionally averaged flow

velocity which would be the output from one-dimensional rigid column flow models.

Prior to the removal of the plug valve all the measurement devices were started. Following the plug

valve removal, water flow began and air intruded into the pressurized water column as a gravity current with

the shape of the dissipation free solution described by Benjamin (1968). The movement of the air/water

6

interface was tracked by using the digital video recorder to track the location where the water detached

from the pipe ceiling. Distances were determined from a tape measure placed along the top of the pipeline.

Pressure measurements with the pressure transducers were calibrated each experimental run from water

levels in stilling wells located at the 2.9 m and 12.4 m stations. In most experiments, the duration was

generally short (less than 20 seconds), and no inflow was introduced into the upstream reservoir after flow

initiation as the upstream water level was not altered significantly during the experimental run. In other

experiments with longer duration, a metered inflow was admitted to the supply reservoir simultaneously

with the flow initiation such that the upstream head on the pipeline was maintained at a constant elevation;

this was possible due to a preliminary experiment that determined the steady state discharge through the

pipeline at that reservoir head.

Different initial H values were used in the experiments, which are presented in this paper normalized by

D. Most experimental runs were repeated at least three times to ensure consistency in the measurements.

EXPERIMENTAL RESULTS AND ANALYSIS

Initial experiments were performed to study the initial advance of the air intrusion as a function of H/D.

Figure 3 illustrates the advance of the air recorded with the video camera normalized by L. In the graph,

the air intrusion trajectories for H/D =2.13, 4.26 and 9.04 are included. Figure 3 also presents the absolute

velocity recorded by the upstream ADV probe for the same experimental conditions.

Immediately after the plug was removed the air layer filled approximately the upper half of the pipe

cross section while water exited the pipe in the lower half as described by Townson (1991). This would be

consistent with the dissipation-free solution proposed by Benjamin. Eventually the air intrusion halted its

advance, and the air/water interface reversed direction. As Figure 3 shows, as H/D increased, the time

required for the air intrusion to halt decreased. The time and location of the interface reversal for H/D

= 2.13 could not be determined as it took place at a location further upstream in the pipeline, outside

the recording field of the digital camera. Figure 3 indicates that the air intrusion halt is associated with

the instant when the flow velocity in the pipeline upstream from the air intrusion became larger than the

predicted celerity of the cavity, given by Benjamin (1968) as 0.542√gD. This is also in agreement with

Townson (1991) who predicted that the advance of the air/water cavity in such conditions could eventually

be halted by the acceleration in the water velocity upstream from the cavity.

Observations of the shape of the intrusion front indicated that for longer intrusions (Lair > 1 m), the

nature of the front changed from a loss-free to a dissipative front shape. Interfacial waves started to form

on the interface behind the head after which a transition to an intrusion with a smaller thickness δair was

observed. No abrupt transition between the two states was observed. Assuming that the smaller δair with

7

dissipation follows equation 1, cair will only approximately be given by 0.54√gD. The intrusion thickness,

δair, then tended to decrease with further propagation distance.

Following the halt of the air intrusion, the shape of the air/water interface became more complex, no

longer characterized by a distinct front but instead by an air wedge. This result is well known from studies

on salinity intrusions in that the gravity-current shape is only maintained so long as the intrusion propagates

against an oncoming flow; otherwise a saline wedge forms (Keulegan, 1966). For the runs with higher H/D,

this wedge was pushed downstream and the air intrusion was expelled from the pipeline. For H/D = 2.13,

even though the velocity U recorded by the upstream ADV probe was larger than cair, the water inflow

did not completely expel the air intrusion. Although the initial arrest of the intrusion is consistent with

the Benjamin intrusion theory, the subsequent flow is without a distinct head and the theory is apparently

not applicable, challenging the generality of the intrusion velocity criterion to describe the air/water front

displacement. The investigation by Guizani et al. (2005) reached a similar conclusion for the case of the

rapid filling of a horizontal, empty pipeline. Pressurized flows cannot be maintained if the pipeline is able to

convey a given flow in an open-channel flow regime. In the present experiments, in order for open-channel

flow regime to exist, a significant energy slope would be required, which can be provided by the free discharge

condition at the pipe outlet.

Figure 4 illustrates the pressure measurements performed for H/D = 2.13. Immediately after the removal

of the plug, high-frequency transient pressures were recorded, which disappeared after T ∗ ≈ 0.14 (T ≈ 2 s).

After that time, the head in the pipe invert at the 2.6 m station remained above the pipe crown throughout

the run, with a gradual decrease in H as the upstream reservoir water level decreases with the outflow. This

head was also greater than the head at the invert of the 12.3 m station due to the energy losses between the

two locations. At the 12.3 m station, the pressure at the invert decreased from the initial level of 0.20 m

before the flow was initiated to a value around 0.10 m after the flow becomes established. Then, the pressure

gradually decreased until the moment when the air intrusion arrived at that station, when the pressure

head dropped rapidly due to the rapid decrease in the flow depth caused by the intrusion. For T ∗ > 0.82

(T > 12 s) the halt of the air intrusion caused an increase of the water depth and pressure head at the

12.3 m station. However, visual observations indicated that the flow did not return to a pressurized regime

throughout the pipeline, which is confirmed by the pressure head remaining below the pipe diameter at the

12.3 m station. The transducer at the pipe crown, on the other hand, had a pressure head slightly above 0.10

m prior to the flow initiation. After the flow initiation, the recorded pressure gradually dropped to values

slightly below atmospheric pressure. This is due to the pressure variation at the front of the air intrusion

where non-hydrostatic pressure variations prevail (Benjamin, 1968). Following the air intrusion arrival, the

pressure head became atmospheric, with small fluctuations in the order of ±2 mm. (approximately the

8

resolution of the pressure transducer).

Further experiments were performed to study the advance and retreat of the air phase in more detail

for lower H/D ratios. These experiments recorded the advance of the air intrusion for H/D values of 1.38,

1.60, and 1.81 for longer time periods, and the results are shown in Figure 5. These were the experiments

in which a metered inflow was supplied to the upstream reservoir so that the reservoir level was maintained

at a constant level. The results confirmed the observation in the previous experiments with H/D = 2.13

that for small H/D, the intrusion was not totally removed by the inflow. For H/D = 1.81, the accelerating

water flow halted the intrusion at T ∗ ≈ 0.69 (T ≈ 10 s), and initially, the cavity is pushed backwards rapidly

as the air intrusion front collapsed, similarly to what was recorded with H/D = 9.04 and H/D = 4.26.

However, after T ∗ ≈ 1.37 (T ≈ 20 s), the retreat of the cavity slowed down, and the location where the water

detached from the pipe ceiling stabilized at approximately 0.60 m from the pipe outlet. The same trend was

observed for the smaller H/D cases, with the final cavity location further upstream in the pipeline. After

the stabilization of the cavity location, the flow became steady in the case of constant supply reservoir head.

The time required for the stabilization of the air cavity was quite long in some cases.

If discharge conditions influence the final location of the air/water interface, then this implies that a

conveyance-based criteria should be applied after the halt of the air intrusion. This was confirmed by

experiments performed with H/D=3.32 conducted over a much longer duration of T ∗ = 6.85 (T = 100 s) to

provide a final steady-state U close to that suggested as the limiting velocity for an intrusion by Liou and

Hunt (1996). The results in Figure 6 for an experiment in which H slowly declined indicate the same general

behavior of the previous experiments in the sense that an air cavity advanced in the pipeline but was totally

expelled before T ∗ = 0.82 (T = 12 s). However, as the time progressed and H decreased, the location of

the air/water interface again begins to moves upstream, particularly after T ∗ = 5.14 (T = 75 s). The initial

arresting of the intrusion occurs at a velocity that is consistent with the Benjamin relation. Visually, the

following sequence of events was observed: First, a relatively large-diameter intrusion proceeded upstream,

becoming progressively thinner until a wave on the free surface touched the crown of the pipe. The intrusion

then rapidly disintegrated and the air front rapidly retreated towards the downstream end . Finally, after

approximately T ∗ = 5.8 (T = 55 s), an intrusion with a very small thickness but lacking the free surface

wave behind it such as was observed in the initial intrusion began to slowly migrate upstream, eventually to

a position further upstream. This upstream advance started when the water flow velocity measured at both

ADV probes was greater than cair. It appears that this latter phase was controlled by the shear forces in

the system and represents the adjustment of the air cavity to these forces in essentially the same sense as

salinity intrusions form in estuaries.

In summary, these experiments indicate that a velocity-based criterion to determine whether pressurized

9

flow conditions will exist throughout the pipeline is valid until the pressure gradient between the upstream

reservoir and the discharge point has accelerated the water such that U > cair. After that time, even though

U > cair, the air intrusion was not entirely removed from the pipeline. Moreover, the shape of the intrusion

evolved from the original curved shape that resembled a density current into more of a wedge-like interface.

Clearly, since the term ”air intrusion” in the sense of the solution by Benjamin used to define the behavior of

the air phase within the pipeline is no longer adequate, the air wedge is referred to simply as the air/water

interface. The initial intrusion as described by Benjamin can be considered to be a balance between buoyancy

and inertia forces while the final state is dominated by shear forces.

A significant implication of these findings is related to the anticipated scale effects of the described air

intrusions in actual pipelines. Since air intrusion celerities are scaled by√gD, such air interactions are

likely to be much more significant in large diameter pipelines, particularly, in cases when the H/D ratio

and the friction within the system cannot accelerate the flow to above the intrusion velocity. Even under

the circumstance of a rapid initiation of flow, the possibility for air intrusions to develop during pipe filling

conditions exist in a variety of different geometric configurations. Under these conditions, the ability of

a numerical model that relies solely on a rigid-column formulation may be greatly compromised. A more

suitable modeling approach would allow for the simulation of the flow regime transition between free surface

and pressurized flow such as proposed by Vasconcelos et al. (2006). This would be even more important in

typical circumstances where pipelines are filled slowly in order to avoid strong transients. However even these

approaches have limitations to the particular problem of air intrusions where the strongly curved interface

at the intrusion front limits the applicability of models that rely on the shallow water equations. Additional

research on extending shallow water equation models to the air intrusion problem is required to understand

the inherent limitations and possible methods to adapt such models to these problems.

PREDICTING THE LOCATION OF THE AIR/WATER INTERFACE

The problem of the startup of a filled pipeline with free-discharge conditions at the downstream end is

sufficiently simple that an approximate analytical solution can be obtained for cases when H/D is relatively

large and Lair/L relatively small, i.e., when the balance between buoyancy and inertial forces is dominant.

The flow can be represented by the advance of a gravity current against a rigid column that accelerates due

to the pressure gradient between the upstream reservoir and the free discharge at the downstream end. In

this case, the celerity of the cavity can be estimated with an expression of the form cair = Ci

√gD. Although

the coefficient Ci depends on the cavity thickness (equation 1) it is assumed for the sake of simplicity to have

a constant value of Ci = 0.54, which corresponds to an air cavity occupying greater than about 20 percent

of the pipe diameter as previously discussed. Disregarding the local acceleration of the flow approaching the

10

pipe inlet, the velocity in the pipeline can be calculated in a rigid-column fashion as:

dU

dt=

g

Lpw

(H − (f

Lpw

D+K)

U2

2g

)(2)

In the above equation Lpw is the length of the pressurized water column, f is the Darcy Weisbach friction

factor and K represents the sum of all the local loss coefficients. Constant reservoir pressure head is assumed

and the variation of Lpw due to the advance of the air intrusion is neglected making the results only valid

for Lair/L ≪ 1. So long as the normalized pressure head H/D is sufficiently large (about 3 or 4), the time

required for the cavity advance, halt and reversal will be small. For such conditions, Lair/L ≪ 1. After

introducing a “total resistance” coefficient given by R = fL/D + K, the analytical solution of equation 2

becomes:

U(T ) =

√2gH

Rtanh

( T

2Lpw

√2gHR

)(3)

The results from equation 3 are compared with the experimental results in Figure 7. These results were

obtained for a value of f = 0.027, consistent with previous measurements of steady state flow within the

pipeline, and an estimated value for entrance loss of K = 0.5. The observed and predicted velocities agree

well, particularly for the larger values of H/D. With this pipe velocity, and assuming a constant celerity of

the air cavity given by Vcavity = 0.54√gD, the extent of the cavity can be obtained by:

Lair =

∫ T

0

(cair − U(t)

)dt (4)

The advance of the air intrusions observed in the experiments with H/D values of 9.04, 4.26 and 2.13 is

compared with the results from from equation 4 in Figure 8. The interval of integration was chosen so that

the value of Lintrusion was always positive. For the largest H/D the observed and predicted trajectory of the

air intrusion agreed reasonably well, but as H/D decreased to 4.26 this agreement was not so satisfactory

This is explained by the assumptions introduced in the calculation, especially the assumption of constant

cair. Nevertheless, the discrepancy in the maximum Lair was less than 15%.

It is very important to reiterate that this simplified approach to calculate the extent of the air intrusion

upon the startup of the filled pipeline is valid only for large values of H/D, probably above 3 or 4. For

smaller values, the value of Lair/L may no longer be negligible. In addition, the assumption of a constant

value for the celerity coefficient Ci may not also hold.

For situations where a free surface is maintained in the pipeline at steady state, it should be possible to

estimate the location of the air/water interface for single-sloped pipelines discharging freely. This estimation

11

is based on the calculation of the flow profile using a gradually varied flow framework. The location where

the water level touches the pipe crown signals the point where pressurized flow begins.

In order to compare the length of Lair measured in the experiments with this theoretical prediction it was

decided to apply a computer model to simulate the gradually varying flow. To predict the location where

there is a transition in the flow regimes, the model was adapted to include a conceptual model that accounts

for flow regime transitions. The Preissmann slot (Cunge and Wegner, 1964) or the TPA model (Vasconcelos

et al., 2006) are possible alternatives to simulate such transitions. The calculation starts at the downstream

end of the pipeline, where critical flow conditions are assumed. Then, according to Henderson (1966), the

depth at upstream locations can be calculated with the expression:

dy

dx=

So − Sf

1− Fr2(5)

in which So is the pipeline bed slope, Sf is the energy slope (implemented here with the Manning equation

with constant Manning coefficient) and Fr is the Froude number in the cross section. As the flow depth

reaches the pipe crown and pressurized flows are expected, the Preissmann slot concept (Cunge and Wegner,

1964) is applied to account for the flow regime transition. This results of the piezometric head profiles

obtained with this simple approach were compared for the cases when H/D values were 1.38, 1.60 and 1.81

and presented in Figure 9. The flow state was simulated immediately after steady flow conditions were

achieved in the pipe, based on velocities recorded by the ADV probes. The velocities obtained for the

different H/D values were UH/D=1.38=0.45 m/s, UH/D=1.60=0.50 m/s and UH/D=1.81=0.55 m/s. Generally,

the predicted location where the flow becomes pressurized is further upstream of the location in which the

flow-regime transition occurred in the experiments, as presented in Table 1. The non-unique depth-discharge

relationship as the flow depth approaches the pipe crown may help explain this discrepancy. In fact, the

location in which y/D=0.92, corresponding to the depth where a circular pipe yields the maximum discharge

in free surface flow fashion according to the Manning equation, was much closer to the location where the

air-water interface is observed. This is also qualitatively consistent with the observations that were made

during the experiments. As the air-water interface migrated slowly upstream to its final steady state position,

the interface did not form a smooth transition tangential to the pipe crown. Rather a curved interface with a

thin air layer that was not laterally uniform developed at the location of the flow regime transition. However,

this interface did not resemble the intrusion interface in the earlier phase of the experiments in that the free

surface waves behind the front were not observed. It is interesting to speculate whether the same behavior

would be observed if the duct were rectangular in geometry or whether the particular observations were

a consequence of the circular geometry. It is also possible that the behavior was partially due to surface

12

tension effects.

CONCLUSIONS

The experiments conducted in this study were chosen to illuminate various issues associated with the

relative motion of the air-water interface during starting flow in a horizontal pipeline. It was observed that

the loss-free intrusion analyzed by Benjamin (1968) and applied to the flow startup problem by Townson

(1991) does adequately represent the flow conditions during the initial phases of air intrusion. However,

as the intrusion proceeds upstream, the loss-free condition cannot be maintained and the air intrusion

transitions to an intrusion with energy dissipation and a smaller intrusion thickness. If the water velocity

increases sufficiently rapidly, the intrusion celerity, even with energy dissipation, does not significantly vary

from the loss-free solution and the upstream extent of the intrusion can be fairly easily predicted. If the

upstream intrusion distance is more significant, then the decrease in the intrusion thickness can result in

the disintegration of the intrusion due to the free surface waves behind the front touching the crown of the

pipe. The loss of the head-like structure results in the rapid downstream migration of the air water interface.

During this phase of the motion, the intrusion dynamics analyzed by Benjamin are no longer valid. This

confirms the observations by Guizani et al. (2005) that the intrusion velocity criterion proposed by Liou and

Hunt (1996) for maintaining a sharp interface between air and water can only be applied when the air is

actually intruding into the water phase. Therefore, the application of the Benjamin intrusion criterion to the

problem of rapid filling of an initially empty pipe does not appear to be justified, a conclusion also reached

by Guizani et al. (2005).

Some experiments were conducted with low reservoir heads and the accelerating flow was unable to

completely expel the intrusion from the pipeline. In these cases, the downstream migration of the air/water

interface following the breakdown of the intrusion front was halted and the interface began to once again

migrate in the upstream direction. For experiments in which the reservoir head was held constant, the

eventual upstream location of the intrusion was controlled by shear forces in the pipeline. A gradually varied

flow analysis could be used to predict the upstream extent of the intrusion. The non-unique depth-discharge

behavior associated with the pipe geometry and frictional effects hampered the ability to make accurate

comparisons of the observed and predicted intrusion lengths. However, if the location where the predicted

y/D on the gradually varied flow profile was 0.92 (depth at maximum discharge according to the Manning

equation) was used as the location of the interface front, then quite acceptable predictions were achieved.

This is consistent with the experimental observations that the interface front did not approach tangent to

the pipe crown but rather that a small sharply curved interface was observed at the front.

13

APPENDIX I. NOTATION

The following symbols are used in this paper.

cair = Celerity of the air cavity intrusion

D = Pipe diameter

f = Friction coefficient

Fr = Froude number

g = Gravity acceleration

H = Pressure head at the reservoir

K = Sum of local losses coefficients

L = Length of the pipeline

Lair = Length of the air intrusion front

Qc = Conveyance capacity of a pipe in free-surface-flow regime

R = Total loss coefficient = fL/D +K

So = Pipeline bed slope

Sf = Energy slope

T = Time

Tw = Frictionless pipeline startup time

U = Flow velocity

Ucav = Velocity of a cavity intrusion relative to the approaching inflow

x = Distance along the pipeline from the upstream reservoir

x̄ = Distance from the upstream reservoir normalized by pipeline length

y = Water depth

α = half angle subtended from the center of the pipe to the free-surface

δair = Thickness of air intrusion front

14

REFERENCES

Albertson, M.L. and Andrews, J.S.(1971) Transients caused by air release. in Control of Flow in Closed

Conduits J.P. Tullis ed. Colorado State University.

Benjamin, T.B.(1968). Gravity currents and related phenomena. J. Fluid Mech., 31(2):209–248, 1968.

Cardle, J.A. and Song, C.S.S. (1988) Mathematical modeling of unsteady flow in Storm Sewers. Int. J.

Eng. Fluid Mech. 1 (4) 495-518.

Cunge, J.A. and Wegner, M.(1964). Integration numerique des equations d’ecoulement de Barre de St.

Venant par un schema implicite de differences finies. Application au cas d’une galerie tantot en charge

tantot a surface libre.” La Houille Blanche, 1: 33–39, 1964.

Fuertes, V. S., Arregui, F., Cabrera, E. and Iglesias, P. L. (2000) Experimental setup of entrapped air pockets

model validation. Proceedings of the 8th International Conference on Pressure Surges - Safe Design and

Operation of Industrial Pipe Systems, BHRA, The Hague, Netherlands, pp. 133-145.

Guizani,M., Vasconcelos,J. G., Wright, S. J. and Maalel,K.(2005). Investigation of rapid filling of empty

pipes. Intelligent Modeling of Urban Water Systems, Ch. 14: 463–482, 2006.

Henderson,F. M.(1966). Open Channel Flow. Prentice Hall, Upper Saddle River, NJ, 1966.

Izquierdo,J., Fuertes,V.S., Cabrera,E., Iglesias,P. L. and Garcia-Serra,J.(1999). Pipeline start-up with en-

trapped air. J. Hydr. Res., 37(5): 579–590, 1999.

Keulegan,G. H.(1966). The mechanism of an arrested saline wedge. Estuary and Coastline Hydrodynamics,

A. Ippen, ed., McGraw-Hill, chapter 11, 1966.

Liou,C. P. and Hunt,W. A.(1996). Filling of pipelines with undulating elevation profiles. J. Hydr. Engrg.,

122(10): 534–539, 1996.

Martin,C. S. (1976). Entrapped Air in Pipelines. Proc. 2nd Int. Conf. Pressure Surges - BHRA England:

F2-15-F2-25

Townson,J. M.(1991). Free-Surface Hydraulics. London, UK, 1991.

Vasconcelos,J. G., Wright,S. J. and Roe,P. L.(2006). Improved simulation of flow regime transition in sewers:

The two-component pressure approach J. Hydr. Engrg. , 132(6): 553-562, 2006

Whitham,G. B.(1955). The effects of hydraulic resistance in the dam-break problem Proc. Royal Soc. London

- Series A, 227(1170): 399–407, 1955.

15

Wylie,E. B. and Streeter, V. L. (1993). Fluid Transients in Systems Prentice-Hall, Upper Saddle River, NJ,

pp: 95–99, 1993.

Zukoski,E. E.(1966). Influence of viscosity, surface tension, and inclination on motion of long bubbles in

closed pipes. J. Fluid Mech, 25(4): 821–837, 1966.

16

List of Tables1 Observed and predicted flow regime transition stations for steady flow and different H/D . . 18

17

TABLE 1. Observed and predicted flow regime transition stations for steady flow and different H/D

(1) (2) (3) (4)

H/D Flow regime transition station Location whereObserved (m) Predicted (m) y/D=0.92 (m)

1.38 9.7 7.5 9.91.60 11.3 9.8 11.51.81 13.4 11.2 12.5

18

List of Figures1 Velocity of the front advance in a circular pipe as a function of the half-angle α . . . . . . . . 202 Representation of the experimental apparatus used in the experiments on the startup of filled

pipelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Absolute flow velocity U measured at x̄=0.262 compared with cair = 0.54

√gD and compared

with the advance of the air cavity for different H/D ratios. Note that for larger H/D themaximum advance is linked to the instant when U > cair. Time is normalized by L/

√gD . . 22

4 Pressure head measurements for the experiments with H/D=2.13 at three locations in thepipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5 Advance of the air cavity after rapid flow initiation for H/D=1.38, 1.60 and 1.81 . . . . . . . 246 Measurement of U with two ADV probes and the advance of the intrusion flow velocity for

the case when H/D=3.32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Comparison between U measured with the ADV at the x̄ =0.262 and the predictions from

equation 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268 Comparison between the length of the air intrusion over time with the predicted length from

equation 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 Predicted free-surface profiles using gradually varied flow calculation for H/D=1.38, 1.60 and

1.81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

19

FIG. 1. Velocity of the front advance in a circular pipe as a function of the half-angle α

20

FIG. 2. Representation of the experimental apparatus used in the experiments on the startup of filledpipelines

21

FIG. 3. Absolute flow velocity U measured at x̄=0.262 compared with cair = 0.54√gD and compared

with the advance of the air cavity for different H/D ratios. Note that for larger H/D the maximumadvance is linked to the instant when U > cair. Time is normalized by L/

√gD

22

FIG. 4. Pressure head measurements for the experiments with H/D=2.13 at three locations in thepipeline

23

FIG. 5. Advance of the air cavity after rapid flow initiation for H/D=1.38, 1.60 and 1.81

24

FIG. 6. Measurement of U with two ADV probes and the advance of the intrusion flow velocity for thecase when H/D=3.32

25

FIG. 7. Comparison between U measured with the ADV at the x̄ =0.262 and the predictions fromequation 3.

26

FIG. 8. Comparison between the length of the air intrusion over time with the predicted length fromequation 4.

27

FIG. 9. Predicted free-surface profiles using gradually varied flow calculation for H/D=1.38, 1.60 and1.81

28