Journal of Fluids and Structures 80 (2018) 441–450

Contents lists available at ScienceDirect

Journal of Fluids and Structures

journal homepage: www.elsevier.com/locate/jfs

Numerical study of the fluid-dynamic loading on pipes

conveying fluid with a laminar velocity profile
Gregor Bobovnik *, Jože Kutin
University of Ljubljana, Faculty of Mechanical Engineering, Laboratory of Measurements in Process Engineering, Aškerčeva 6, SI-1000
Ljubljana, Slovenia

article info a b s t r a c t
Article history: The paper deals with the fluid-dynamic forces of laminar fluid flow in long vibrating pipes
Received 27 October 2017 as well as in long curved pipes at rest. It focuses on the contribution of the fluid-dynamic
Received in revised form 14 February 2018 force related to the centrifugal acceleration. The study is based on the results of a CFD
Accepted 9 April 2018
numerical model of an incompressible laminar fluid flow in the pipe, which is deflected
Available online 8 May 2018
in a bending, beam-type mode. The results are presented in terms of the centrifugal
correction factor, which is defined as the ratio between the actual centrifugal-related term
Keywords:
Pipe conveying fluid of the observed fluid-dynamic force and the solution of the one-dimensional model. The
Fluid-dynamic force estimated values of the respective correction factor are presented for a range of vibration
Centrifugal correction factor frequencies, lengths of the pipe and amplitudes of the pipe deflection. The results show
Laminar flow that all the listed parameters affect the centrifugal correction factor. The resulting values
CFD are also compared and assessed with respect to some accessible analytical solutions. The
influence of the predicted values of the centrifugal correction factor on the onset of static
and dynamic instabilities of pipes conveying fluid is also discussed.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction

The fluid-conveying pipe is a fundamental dynamical problem in the field of fluid–structure interactions (Païdoussis,
2014, 2016). This topic has several engineering applications and also serves as a model for understanding more complex
systems and for searching out new dynamic features and phenomena. The modelling of fluid-conveying pipes is typically
focused on an analysis of the dynamic behaviour and stability issues. In the field of Coriolis flow metering, which represents
the application of fluid-conveying pipes for direct measurements of the mass flow rate and the density of the fluids, the aim
of the modelling is to examine the measurement characteristics and different influential parameters (Baker, 2016; Wang
and Baker, 2014).
The use of analytical modelling for fluid-conveying pipes is dependent on the availability of suitable analytical math-
ematical models for the fluid-dynamic loading acting on the pipe. The currently available analytical models have different
ranges of application, because they are based on different assumptions regarding the pipe and the fluid flow through the pipe.
Regarding their treatment of the fluid flow, these models can be divided into one-dimensional models and wave models. See,
e.g., some early contributions to the evolution of the one-dimensional models from Housner (1952), Benjamin (1961) and
Gregory and Païdoussis (1966), etc., and of the wave models from Païdoussis and Denise (1972), Weaver and Unny (1973),
Weaver and Myklatun (1973) and Shayo and Ellen (1974), etc. More exhaustive and recent review of this topic can be found
in Païdoussis (2014, 2016).

* Corresponding author.
E-mail address: gregor.bobovnik@fs.uni-lj.si (G. Bobovnik).

https://doi.org/10.1016/j.jfluidstructs.2018.04.006
0889-9746/© 2018 Elsevier Ltd. All rights reserved.
442 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450

Fig. 1. Sketch of a pipe conveying fluid deflected in a beam-type mode.

Here, the fluid flow can be divided into the mean flow, which refers to the flow in the undeflected pipe, and the perturbed
flow, which considers all the variations relative to the mean flow resulting from the pipe deflections. As will be presented in
the continuation of this section, there are some open questions about the proper application ranges of particular analytical
models for the fluid-dynamic loading, as well as some discrepancies between the results of different authors. In this respect,
the primary aim of the paper is to develop a numerical method that will allow the analytical models to be validated for the
fluid-dynamic loading.
The study presented here is focused on straight fluid-conveying pipes that deflect in a bending, beam-type mode
(see Fig. 1). The pipe deflections w (x, t) and the corresponding perturbed flow field are assumed to be small enough
so that the non-linear effects can be neglected. The fluid-dynamic force per unit length can be written as the sum of
three components that are related to the translational acceleration ∂ 2 w/∂ t 2 , the Coriolis acceleration 2V ∂ 2 w/∂ t ∂ x and the
2
centrifugal acceleration V ∂ 2 w/∂ x2 of the fluid:
∂ 2w ∂ 2w 2∂ w
2
fd (x, t ) = −α0 Mf − α1 2Mf V − α2 Mf V , (1)
∂t 2 ∂t∂x ∂ x2
where Mf is the fluid mass per unit length, V is the average axial velocity of the mean flow and αj , j = 0, 1 and 2 are the
corrections factors that depend on the assumptions of the model. First, we define some dimensionless numbers that will
serve in discussions of the model assumptions:

2V R ω R2 R ωR
Re = , Reω = , εL = , εC = , (2)
ν ν L c
where ν is the fluid kinematic viscosity, c is the fluid speed of sound, ω is the angular frequency of the pipe, R is the
pipe internal radius and L is the pipe length. The Reynolds number Re represents the ratio of the inertial forces to the
viscous forces within the mean flow field; the velocity profile approaches the uniform distribution for turbulent flow at
high Reynolds numbers (Re ≫ 104 ), but the parabolic velocity profile is achieved for a fully developed flow in a straight
circular pipe under laminar flow conditions (Re < 2000). The vibrational Reynolds number Reω represents the ratio of the
inertial forces to the viscous forces within the perturbed flow field; the viscous effects become less significant at higher
vibrational Reynolds numbers. The dimensionless numbers εL and εC represent the magnitudes of the pipe length effects
and the fluid compressibility effects, respectively, on the perturbed flow field.
Kutin and Bajsić (2014) estimated the correction factors αj of the uniform and laminar mean-flow velocity profiles for
different circumferential mode shapes of the straight pipe. This study is based on a mathematical model that assumes inviscid
perturbed flow, so it is valid for sufficiently high vibrational Reynolds numbers Reω . The results for the beam-type pipe, for
the case of the uniform mean-flow velocity profile, can be written as:
( )2 ( )2
∂2 1 1 ωR ∂2 1 1 ωR
α0 = 1 + R + 2
, α1 = 1 + R + 2
,
4 ∂ x2 4 c 4 ∂ x2 2 c
2
∂2 3 ωR
( )
1
α2 = 1 + R2 2 + , (3)
4 ∂x 2 c
and, similarly, for the case of the laminar mean-flow velocity profile:
( )2 ( )2
∂2 11 ωR 5 2 ∂2 7 ωR
α0 = 1 + R2 + , α1 = 1 + R + ,
4 ∂ x2 4 c 12 ∂ x2 12 c
( )2
2 5 ∂2 43 ωR
α2 = + R2 2 + , (4)
3 8 ∂x 24 c
 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450 443

where the effects of the pipe length and the fluid compressibility up to the order εL2 and εC2 , respectively, are taken into
account. The correction factors α0 = α1 = α2 = 1, which are very often used in the mathematical models of fluid-conveying
pipes, are valid for relatively long pipes conveying an incompressible fluid (εL2 ≪ 1, εC2 ≪ 1) with a uniform mean-flow
velocity profile (Re ≫ 104 ). The laminar mean-flow velocity profile (Re < 2000) also leads to α0 = α1 = 1 for εL2 ≪ 1 and
εC2 ≪ 1, but the centrifugal-acceleration term is decreased by the factor α2 = 2/3.
This result is different from the correction factor of α2 = 4/3 proposed by Guo et al. (2010) and Hellum et al. (2010)
following a definition of the momentum correction factor. Their approach would be correct if the non-uniform mean flow
entirely followed the pipe vibration (see Kutin and Bajsić, 2014 for more details). As also confirmed by numerical simulations
in the current paper, such an omission of the effects of the secondary flow does not generally agree with the actual flow
conditions within fluid-conveying pipes.
A boundary case in the dynamics of pipes conveying fluid is the static buckling instability, which occurs at zero
frequency (Païdoussis, 2014). At this point, the problem reduces to the fluid-dynamic loading acting on deflected, curved
pipes at rest. There are some available analytical solutions for such a case, e.g., Gammack and Hydon (2001) and Roberts
(2004). The fluid-dynamic force per unit length of a slightly curved pipe at rest can be written as:

2 ∂ 2w
fd (x) = −α2 Mf V , (5)
∂ x2
where the centrifugal correction factor for the incompressible fluid and the pipe length up to the order εL reads as:
∂
( )
5 2 11
α2 = − + Re R . (6)
3 Re 1080 ∂x
The term related to Re−1 becomes significant at very small Reynolds numbers (Re ≪ 1), which corresponds to the so-called
Stokes or creeping flow effects. At moderate Reynolds numbers and for long pipes (εL ≪ 1) the centrifugal correction factor
has a value of α2 = 5/3. This means that the curved pipe at rest is subjected to a significant increase in the centrifugal
correction factor for laminar flow conditions, which is in contrast to the results for high-frequency vibrating pipes where
α2 = 2/3.
The purpose of this paper is to analyse the centrifugal correction factor in laminar mean-flow conditions for steady-state
curved pipes and vibrating pipes by using CFD (computational fluid dynamics) modelling. The method employed for this
analysis is based on a similar approach to that used by Bobovnik et al. (2004, 2005) to estimate the velocity profile effects
on the Coriolis-acceleration term of fluid-dynamic loading.
The paper is organized as follows. Section 2 describes the applied CFD numerical model for the vibrating pipe and the
steady-state curved pipe. The description encompasses the discretization of the model, the boundary and initial conditions
and the post-processing algorithm applied for the estimation of the centrifugal correction factor. Section 3 presents the
numerical results of the centrifugal correction factor for different vibration frequencies and pipe lengths. The results also
demonstrate how the value of the centrifugal correction factor is influenced by accounting for the viscous shear stress acting
on the wall of the pipe. Section 4 discusses the implications of our findings on to the onset of static and dynamic instabilities
of pipes conveying fluid.

2. Numerical model

2.1. Governing equations

This section presents the fundamental equations used to describe the fluid that are solved with the CFD program Star-CD
v4.18 using the finite-volume method (Ferziger and Perić, 1999). Based on our assumptions of isothermal and incompressible
laminar fluid flow the continuity and momentum equations are as follows:
∂
∫ ∫
ρ dV + ρ (v − vs ) · ds = 0, (7)
∂t V S

∂
∫ ∫ ∫ ∫
ρ vdV + ρ v (v − vs ) · ds = µ grad v + (grad v)T − pI · ds + fb dV ,
( ( ) )
(8)
∂t V S S V

where ρ is the fluid density, v is the fluid velocity vector, V is the finite volume, S is the surface enclosing the finite volume,
vs is the surface velocity vector, p is the pressure, I is the unit tensor, µ is the fluid dynamic viscosity and fb represents the
vector of the forces acting on the finite volume. In the case of steady-state simulations the first term on the left-hand side of
Eqs. (7) and (8) is equal to zero.
The unsteady terms are discretized using a second-order, three-time-levels implicit scheme, while the convective and
diffusive terms are approximated using the central differencing scheme. The pressure–velocity coupling is handled by the
SIMPLE algorithm.
444 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450

Fig. 2. Moving grid strategy (i represents the index of the axial section).

2.2. Computational domain

The computational domain consists of the curved/vibrating pipe section and the additional inlet and outlet sections (see
Fig. 1). The radius of the pipe R equals 0.5 mm and the length L of the steady-state curved pipe is varied from 12.5 mm to
400 mm (L/R = εL−1 = 25 to 800), whereas only two specific lengths of 50 mm and 200 mm (εL−1 = 100 and 400) are
considered in the transient vibrating pipe model. The lengths of the inlet and outlet sections are 10 mm. The cross-section
of the pipe is divided into 1816 cells with M = 80 cells forming the circumference. The deflected part of the domain is split
into N = 500 uniform sections in the axial direction (908,000 cells), whereas the entire computational domain, including
the inlet and outlet sections, is represented by 1,089,600 cells.

2.3. Boundary and initial conditions

In the simulations of the vibrating pipe it is assumed that its centreline displacements are equal to the first mode shape
of an Euler-beam with fixed ends (Blevins, 2016):
w(x, t) = C W (x) sin(2π ft)
( x) ( x ) cosh λ − cos λ ( ( x) ( x ))
1 1 (9)
W (x) = cosh λ1 − cos λ1 − sinh λ1 − sin λ1
L L sinh λ1 − sin λ1 L L
where λ1 = 4.730 is the eigenvalue for the first mode shape and C = A/W (L/2) is the factor related to the vibrational
amplitude at the middle of the pipe A. The nominal value of A equals L/5000, and this is increased up to L/10 when the effects
of the pipe deflection are studied for the steady-state curved pipe. The vibration frequency f is changed between 10 Hz and
300 Hz (Reω = 15.7 to 471). A single oscillation period of the pipe is modelled with 100 time steps, and each simulation is
run for 500 time steps (five oscillation periods). In the simulation of the steady-state curved pipe its centreline displacement
is not time dependent and equals w (x) = C W (x).
A moving-grid simulation of the fluid flow is performed to capture the fluid dynamics in the vibrating pipe, i.e., the
position of the pipe centreline in accordance with Eq. (9) is prescribed for each time step of the computational process. The
grid motion is prescribed in such a manner that the moved vertices of a particular cross-section of the pipe remain normal
to the centreline (see Fig. 2). The same strategy is also used in the construction of the computational grid in the steady-state
curved pipe model.
The simulations are performed for an incompressible water-like fluid with the fluid density ρ = 1000 kg/m3 and the
dynamic viscosity µ = 1 · 10−3 Pa·s. The parabolic laminar velocity profile with an average velocity V = 1 m/s (Re = 1000)
is prescribed at the inlet of the computational domain, while the constant pressure boundary is imposed at the outlet cross-
section. In the vibrating pipe simulations the initial field is set to the steady-state field of the fluid flow in the undisturbed
straight pipe.

2.4. Estimation of the centrifugal correction factor (α2 )

As previously explained, the constituted numerical grid divides the vibrating pipe into N sections along its length and
each section is surrounded by M boundary elements on its circumference. So the total force acting on the section i in the y-
direction for an arbitrary time step can be written as:
M
∑
Fi = pi,k sy,i,k , i = 1 . . . N , (10)
k=1

where pi,k is the pressure on boundary element k of section i and sy,i,k is the y-component of the surface vector of the boundary
element k of section i.
The calculated fluid-dynamic force in the vibrating pipe comprises the effects of the static pressure and the dynamic
components related to the translational, Coriolis and centrifugal accelerations. To obtain the centrifugal correction factor α2 ,
 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450 445

Fig. 3. Schematic representation of the five-step algorithm used for the estimation of the centrifugal correction factor (α2 ) for the vibrating pipe.

all the listed contributions, except the one arising from the centrifugal acceleration, have to be eliminated. In this context
another numerical model of the vibrating pipe is also prepared, with the only difference being that the zero inlet velocity
(V = 0 m/s) is imposed at the inlet of the computational domain. Using the results of the simulations, the elimination of
the undesired acceleration components is accomplished with the following five-step algorithm (Fig. 3).
Step 1: The static pressure contribution is eliminated by subtracting the mean pressure pi around the circumference of
section i from the total fluid-dynamic force (Eq. (10)). The resulting force per unit length acting on the section i of width
δ (= L/N) is given by:
∑M
Fi − pi k=1 sy,i,k
fi =
˜ , i = 1 . . . N. (11)
δ
Step 2: To remove the components due to translational accelerations the difference between the forces per unit length
flow
fi ) and from the model with the zero
resulting from the model with the prescribed inlet velocity of 1 m/s at the inlet (˜
zero−flow
inlet velocity (˜
fi ) is calculated for each time step:
flow zero−flow
∆˜
fi = ˜
fi fi
−˜ . (12)

Step 3: The contribution due to the Coriolis force is eliminated by observing ∆˜

fi at the instants when the Coriolis acceleration
equals zero, which is when the displacement of the pipe reaches its amplitude (the top or bottom position). This implies
that for each oscillation period two values of ∆˜ fi are obtained. Unfortunately, it is not possible to pinpoint exactly the
corresponding times, because of the time integration scheme and the finite time step size used in the numerical model.
Therefore, the value of ∆˜fi is taken at times tj , j = 1 . . . 2Z (Z is the number of oscillation periods) for which the solutions
correspond best to the top/bottom position of the vibrating pipe. For the force difference ∆˜ fi extracted at time tj the following
j
notation is used ∆fi = ∆fi tj .
( )
˜ ˜
Step 4: The centrifugal correction factor for each axial section i with its centre located at xi (see Fig. 2) is calculated as:
j
∆˜ fi
α2j ,i = 2 ∂ 2 w (x,tj ) ⏐
⏐ , i = 1 . . . N , j = 1 . . . 2Z . (13)
−Mf V ∂ x2
⏐
x=xi
j
Step 5: In the last step the centrifugal correction factor α2 is determined at x = L/2. Because the middle of the pipe at x = L/2
coincides with the border between the two central sections of the pipe (i = N /2 ± 1), its value is determined as:

α2j ,N /2−1 + α2j ,N /2+1

α2j = . (14)
2
j
The final value of the centrifugal correction factor (α2 ) is obtained by averaging the α2 in the last three oscillation periods of
j
the pipe (the last six calculated values of α2 ).
446 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450

Fig. 4. Variation of the centrifugal correction factor for different vibration frequencies and two different lengths of the pipe (εL−1 = 100 and 400;
A/L = 2 · 10−4 ).

For the estimation of the centrifugal correction factor α2 in the steady-state curved pipe model steps 2 and 3 are omitted
in the above described algorithm as the fluid force only contains the contributions related to the static pressure and to the
centrifugal acceleration. Also, the indexing in j is irrelevant as the pipe position remains fixed in time, which results in a
single estimate of the centrifugal correction factor.
The described algorithm is realized by using the post-processing abilities of STAR-CD (custom Fortran routine for step 1)
in combination with LabVIEW (steps 2, 3, 4 & 5).

2.5. Uncertainty of the estimated centrifugal correction factor

The uncertainty of the estimated α2 depends in general on the temporal and spatial discretization of the numerical
model, on the convergence criterion applied in the simulation and on the repeatability (scatter) of its calculated values.
The influences of the applied spatial and temporal discretization are reflected directly in the calculated pressure field in the
deflected pipe as well as in the estimation of α2 by using the five-step algorithm. As already discussed, the uncertainty of
α2 related to the applied algorithm arises from the inability to obtain its value exactly at the instant for which the resulting
pressure field reflects the top/bottom position of the pipe (step 3) and due to the fact that its final value is obtained by
averaging the values from the two central sections of the pipe (step 5).
To estimate the uncertainty of α2 three additional models of the vibrating pipe with εL−1 = 100, f = 100 Hz and
A/L = 2 · 10−4 were prepared in addition to the basic model: the convergence criterion is restricted for one order of
magnitude in the first model, the axial length of the sections dividing the pipe in the axial direction is reduced by a factor of 2
in the second model and the time step size is reduced by a factor of 2 in the third model. The value of α2 estimated using the
basic model equals 0.939. The first and second models do not produce any significant change of its value (less than 0.001),
whereas when applying the model with the reduced time step its value changes by about 0.003.
The repeatability of α2 in the last three modelled oscillation periods is estimated using the experimental standard
deviation. It increases approximately linearly from 0 to 0.005 for frequencies from 0 Hz to 300 Hz, respectively (the
repeatability for the pipe length εL−1 = 400 is for a factor of 2.5 larger). It was also confirmed that none of the above so-called
refined models had any notable influence on the estimated repeatability.

3. Numerical results

Fig. 4 shows the variation of the estimated centrifugal correction factor α2 with the vibration frequency for two different
lengths of the pipe (εL−1 = 100 and 400) and the two limiting values at 5/3 and 2/3 defined by Eqs. (6) and (4) for the laminar
flow in the long steady-state curved pipe and in the long vibrating pipe, respectively. The values at 0 Hz, which represent
the steady-state curved pipe, show a significant difference between the values of α2 for the two modelled lengths of the
pipe. For the longer pipe its value corresponds to the prediction of Eq. (6), whereas it is considerably smaller (α2 = 0.94)
for the shorter pipe. As shown, α2 quickly reduces with the increasing frequency for both modelled lengths and there are no
significant differences between the two lengths of the pipes at frequencies larger than 50 Hz. The latter is expected, because
also the difference between the predicted correction factors for both pipes using Eq. (4) is very small (less than 0.002). At
300 Hz α2 is about 0.8, which is still larger than the limiting value of 2/3. However, it could be expected that it will further
decrease for higher frequencies. Unfortunately, because of the very high scattering of the predicted α2 , no reliable results
can be shown for frequencies above 300 Hz.
 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450 447

Fig. 5. Variation of the centrifugal correction factor, estimated by including/excluding viscous shear forces for different vibration frequencies (εL−1 = 100,
A/L = 2 · 10−4 ).

Fig. 6. Variation of the centrifugal correction factor for different lengths of the steady-state curved pipe (A/L = 2 · 10−4 ).

The results in Fig. 4 were based on the definition of fluid forces according to Eq. (10), which accounts only for the pressure
induced forces. Actually, in the case of a viscous fluid the resulting fluid forces also include the viscous shear forces acting on
the wall of the pipe. In Fig. 5 the variation of α2 , which accounts for the shear force y- component, is shown in comparison with
the original case in which the shear force is not considered. The comparison is made for the vibrating pipe with εL−1 = 100
and A/L = 2 · 10−4 for different vibration frequencies. If the shear force is accounted for, the values of α2 are lower for all
the observed frequencies. This is caused by the secondary flow that is generated when the fluid passes through the deflected
pipe. The shear force caused by the secondary flow opposes the y-component of the fluid force calculated from the pressure
field, so the resulting fluid force magnitude and, consequently, also the estimated values of the centrifugal correction factor
α2 are smaller. However, at higher frequencies, being closer to the assumption of the inviscid-perturbed flow, the observed
difference gets smaller. At vibration frequencies higher than 50 Hz the estimated values for α2 differ by less than 0.04.
To study in more detail the observed difference for the two different lengths of the pipe shown in Fig. 4, additional steady-
state simulations covering the wider range of εL−1 were performed (Fig. 6). It is clear that the value of εL−1 has an important
influence on the centrifugal correction factor. For long pipes (εL−1 ≥ 400) α2 approaches the value of 5/3 (α2 = 1.65 at
εL−1 = 400 and α2 = 1.66 at εL−1 = 800) given by Eq. (6), whereas it is significantly smaller in the case of shorter pipes,
i.e., up to 0.94 for the pipe with εL−1 = 25. Based on these results we can assume that the validity of Eq. (6) is limited to
relatively long pipes.
Furthermore, the influence of the amplitude of the pipe deflection on the centrifugal correction factor in the steady-state
curved pipe is investigated by varying A/L from 2 · 10−4 to 0.1. According to the results shown in Fig. 7 the range lower than
about A/L = 5 · 10−3 can be treated as linear in terms of fluid flow perturbations, because the value of α2 remains practically
unchanged. On other hand, the non-linear effects on the perturbed flow become important because α2 changes considerably
448 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450

Fig. 7. Variation of the centrifugal correction factor for different amplitudes of the pipe deflection for the steady-state curved pipe (εL−1 = 200).

at higher values of A/L. The variation of α2 is also caused partially by the alteration of the fluid-dynamic force distribution
along the length of the pipe.

4. Instabilities of pipes conveying fluid

We proceed with the analysis of the influence of the centrifugal correction factor magnitude on the stability characteristics
of pipes conveying fluid. Let us assume a relatively long pipe with flexural rigidity EI and mass per unit length Mp , which
vibrates in the bending mode and is modelled in terms of the Euler–Bernoulli beam theory. Including the fluid flow effects
the equation of motion can be written as:
∂ 4w 2∂ w
2
∂ 2w ( ) ∂ 2w
EI + α2 Mf V + 2Mf V + Mp + Mf = 0. (15)
∂x 4 ∂x 2 ∂ x∂ t ∂t2
Two cases of boundary conditions are studied, i.e., a pipe clamped at both ends:
∂w
w= = 0 at x = 0, x = L, (16)
∂x
and a cantilever pipe clamped at one end and free at the other:
∂w ∂ 2w ∂ 3w
w= = 0 at x = 0; = = 0 at x = L. (17)
∂x ∂x 2 ∂ x3
A general complex solution of Eq. (15) is considered of the form:
4

Aj eiλj x eiωc t ,
∑
w(x, t) = (18)
j=1

where ωc is the complex angular frequency, λj are the complex eigenvalues and Aj are the complex constants. Substituting
Eq. (18) into (15) leads to equation for four λj being dependent on unknown ωc :
2
EI λ4j − α2 Mf V λ2j − 2Mf V ωc λj − Mp + Mf ωc2 = 0.
( )
(19)

Considering Eq. (18) and four boundary conditions defined by Eq. (16) or (17) leads to a homogeneous set of four linear
equations in Aj . For non-trivial solutions, the determinant of the coefficients has to vanish, thus yielding a transcendental
equation in the complex frequency ωc . Its positive real part represents the natural frequency of the pipe (Re(ωc ) = ω) and its
imaginary part represents the vibrational damping (Im(ωc ) > 0) or the amplification (Im(ωc ) < 0). The critical flow velocity
V cr is defined as the smallest flow velocity at which the system becomes unstable, i.e., when Im(ωc ) becomes negative.
With respect to the generalization of the presented results, we define the mass ratio β and the dimensionless flow velocity
υ:
√
Mf Mf
β= , υ = VL . (20)
Mp + Mf EI
 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450 449

Fig. 8. Dimensionless critical flow velocity υcr versus the mass ratio β for three different values of the centrifugal correction factor.

The clamped–clamped tube loses stability by divergence, the static-type instability (ω = 0), at sufficiently high flow velocity.
The corresponding critical flow velocity can be written as:
2π
υcr = √ . (21)
α2
Therefore, υcr under laminar flow conditions at zero vibrational frequency with α2 = 5/3 is about 23% lower than υcr
estimated for the uniform flow assumption with α2 = 1. The clamped-free tube loses stability by flutter, the dynamic-type
instability (ω ̸ = 0), at a sufficiently high flow velocity. Fig. 8 shows the dependence of the corresponding critical flow velocity
on the mass ratio β for different values of α2 . If the vibrational Reynolds number is relatively small, it was shown that the
centrifugal correction factor can reach up to α2 = 5/3 under laminar flow conditions and the system loses stability at a lower
flow velocity than estimated for the uniform flow assumption with α2 = 1 (e.g., about 35% lower for β = 0.5). Otherwise,
if the vibrational Reynolds number is relatively high, the centrifugal correction factor can reach down to α2 = 2/3 under
laminar flow conditions and the system loses stability at higher flow velocity than estimated for the uniform flow (e.g., about
73% higher for β = 0.5).

5. Conclusions

This paper presents a numerical study of the centrifugal correction factor α2 of the fluid-dynamic loading related to the
centrifugal acceleration in curved pipes at rest and in vibrating pipes conveying laminar fluid flow. A CFD numerical model
in combination with the developed post-processing algorithm was used to estimate the centrifugal correction factor. Such
model presents an improvement compared to the analytical models, because it is not restricted to some limiting cases;
e.g., inviscid fluid flow, sufficiently long pipes, small vibrational amplitudes, high vibrational Reynolds numbers or steady-
state case. Here, the model was successfully applied for the calculation of the centrifugal correction factor of a viscous fluid
flow for a wide range of tube lengths, vibration frequencies and vibrational amplitudes as well as by accounting for the
viscous shear forces acting on the wall of the pipe. It was proven that all these factors may significantly affect the magnitude
of the centrifugal correction factor.
To be more specific, the value of the centrifugal correction factor was analysed for an incompressible fluid at Re = 1000
for different vibration frequencies (Reω = 15.7 to 471) and lengths of the pipe (εL−1 = 25 to 800). The deflection of the pipe
was assumed to be equal to the bending beam-type mode shape. The nominal amplitude of the pipe centreline deflection
was equal to A/L = 2 · 10−4 . An additional study made for the steady-state curved pipe (εL−1 = 200) showed that the range
of deflection amplitudes up to about A/L = 5 · 10−3 can be treated as linear in terms of fluid flow perturbations because the
estimated value of the centrifugal correction factor remains practically unchanged in that range.
The results show that value of α2 is strongly dependent on the vibration frequency of the pipe. Its value decreases with
frequency. At frequencies higher than 50 Hz (Reω = 78.5) its value is almost independent of the modelled length of the
pipe (εL−1 = 100 and 400) and equals nearly 0.8 at 300 Hz. A further decrease at higher frequencies is expected. In addition,
it was established that accounting for the viscous shear stress acting on the wall of the pipe, the actual value of α2 is, in
general, smaller than the one obtained just by considering the pressure forces. However, very small differences are observed
at higher frequencies, being closer to the assumption of the inviscid-perturbed flow. The numerically estimated values of
the correction factor α2 at high frequencies are relatively close to the value of 2/3 predicted by Kutin and Bajsić (2014) by
assuming inviscid perturbed flow in long vibrating pipes.
450 G. Bobovnik, J. Kutin / Journal of Fluids and Structures 80 (2018) 441–450

At the lower end of the range of the vibration frequencies the correction factor is noticeably influenced by the pipe length.
For the steady-state curved pipe (the limiting case when the vibration frequency equals zero) its value increases from 0.94
to 1.66 if εL−1 is varied from 25 to 800, respectively. The estimated correction factor for the pipe lengths εL−1 ≥ 400 agrees
well with the value of 5/3, which is assumed by Gammack and Hydon (2001) or by Roberts (2004) for long pipes.
The values of the centrifugal correction factor are important in the study of the static (buckling) and dynamic (flutter)
instabilities of fluid-conveying pipes. An increase or a decrease of its value compared with unity for a uniform flow indicates
the onset of the instabilities at lower or higher fluid velocities, respectively. In clamped–clamped pipes, the critical velocity
related to the static instability of the pipe is inversely proportional to the square root of the centrifugal correction factor.
The critical velocity under laminar conditions is therefore for about 23% lower than the one estimated for the uniform flow.
The onset of the dynamic instabilities of a clamped-free pipe depends on the vibrational Reynolds number (frequency of
vibration). For small vibrational Reynolds numbers (α2 → 5/3), the instability occurs at lower fluid velocities, whereas at
higher vibrational Reynolds numbers (α2 → 2/3) the instability occurs at higher fluid velocities relative to the uniform flow
assumption.

References

Baker, R.C., 2016. Flow Measurement Handbook: Industrial Designs, Operating Principles, Performance, and Applications, second ed.. Cambridge University
Press, New York.
Benjamin, T.B., 1961. Dynamics of a system of articulated pipes conveying fluid, I, Theory. Proc. R. Soc. A 261, 457–486.
Blevins, R.D., 2016. Formulas for Dynamics, Acoustics and Vibration. John Wiley &Sons Ltd., Chichester.
Bobovnik, G., Kutin, J., Bajsić, I., 2004. The effect of flow conditions on the sensitivity of the Coriolis flowmeter. Flow Meas. Instrum. 15, 69–76.
Bobovnik, G., Kutin, J., Bajsić, I., 2005. Estimation of velocity profile effects in the shell-type Coriolis flowmeter using CFD simulations. Flow Meas. Instrum.
16, 365–373.
Ferziger, J.H., Perić, M., 1999. Computational Methods for Fluid Dynamics. Springer-Verlag, Berlin.
Gammack, D., Hydon, P.E., 2001. Flow in pipes with non-uniform curvature and torsion. J. Fluid Mech. 433, 357–382.
Gregory, R.W., Païdoussis, M.P., 1966. Unstable oscillation of tubular cantilevers conveying fluid, I, Theory. Proc. R. Soc. A 293, 512–527.
Guo, C.Q., Zhang, C.H., Païdoussis, M.P., 2010. Modification of equation of motion of fluid-conveying pipe for laminar and turbulent flow profiles. J. Fluids
Struct. 26, 793–803.
Hellum, A.M., Mukherjee, R., Hull, A.J., 2010. Dynamics of pipes conveying fluid with non-uniform turbulent and laminar velocity profiles. J. Fluids Struct.
26, 804–813.
Housner, G.W., 1952. Bending vibration of a pipeline containing flowing fluid. J. Appl. Mech. 19, 205–208.
Kutin, J., Bajsić, I., 2014. Fluid-dynamic loading of pipes conveying fluid with a laminar mean-flow velocity profile. J. Fluids Struct. 50, 171–183.
Païdoussis, M.P., 2014. fluid-Structure Interactions: Slender Structures and Axial Flow, vol. 1, second ed.. Elsevier Academic Press, London.
Païdoussis, M.P., 2016. fluid-Structure Interactions: Slender Structures and Axial Flow, vol. 2, second ed.. Elsevier Academic Press, London.
Païdoussis, M.P., Denise, J.-P., 1972. Flutter of thin cylindrical shells conveying fluid. J. Sound Vib. 20, 9–26.
Roberts, A.J., 2004. Shear dispersion along circular pipes is affected by bends, but the torsion of the pipe is negligible. SIAM J. Appl. Dyn. Syst. 3, 433–462.
Shayo, L.K., Ellen, C.H., 1974. The stability of finite length circular cross-section pipes conveying inviscid fluid. J. Sound Vib. 37, 535–545.
Wang, T., Baker, R., 2014. Coriolis flowmeters: A review of developments over the past 20 years, and an assessment of the state of the art and likely future
directions. Flow Meas. Instrum. 40, 99–123.
Weaver, D.S., Myklatun, B., 1973. On the stability of thin pipes with an internal flow. J. Sound Vib. 31, 399–410.
Weaver, D.S., Unny, T.E., 1973. On the dynamic stability of fluid-conveying pipes. J. Appl. Mech. 40, 48–52.